VU User’s Manual Users

User Manual: Pdf

Open the PDF directly: View PDF .
Page Count: 368

Download
Open PDF In Browser	View PDF

VU User's Manual

Copyright © 2002 Sony Computer Entertainment Inc.
All Rights Reserved.
SCE Confidential

SCE CONFIDENTIAL

VU User's Manual Version 6.0

© 2002 Sony Computer Entertainment Inc.
Publication date: April 2002
Sony Computer Entertainment Inc.
1-1, Akasaka 7-chome, Minato-ku
Tokyo 107-0052 Japan
Sony Computer Entertainment America
919 East Hillsdale Blvd.
Foster City, CA 94404, U.S.A.
Sony Computer Entertainment Europe
30 Golden Square
London W1F 9LD, U.K.
The VU User’s Manual is supplied pursuant to and subject to the terms of the Sony Computer Entertainment
PlayStation® license agreements.
The VU User’s Manual is intended for distribution to and use by only Sony Computer Entertainment licensed
Developers and Publishers in accordance with the PlayStation® license agreements.
Unauthorized reproduction, distribution, lending, rental or disclosure to any third party, in whole or in part, of
this book is expressly prohibited by law and by the terms of the Sony Computer Entertainment PlayStation®
license agreements.
Ownership of the physical property of the book is retained by and reserved by Sony Computer Entertainment.
Alteration to or deletion, in whole or in part, of the book, its presentation, or its contents is prohibited.
The information in the VU User’s Manual is subject to change without notice. The content of this book is
Confidential Information of Sony Computer Entertainment.
and PlayStation® are registered trademarks, and GRAPHICS SYNTHESIZERTM and
TM
EMOTION ENGINE are trademarks of Sony Computer Entertainment Inc. All other trademarks are property
of their respective owners and/or their licensors.
®

© SCEI
-2-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

About This Manual
The "VU User's Manual" describes the operational functions of the VPU (Vector Operation Unit) built into the
Emotion Engine. For information on data and microprogram transfers to the VPU, refer to the "EE User's
Manual".
- Chapter 1 "VU Overview" describes the configurations of the VPU and the VU (the core of the VPU), the
differences between the two VPUs (VPU0/VPU1) that are embedded, and the operation modes.
- Chapter 2 "Data/Calculation Basic Specifications" describes the numerical data formats used by the VU,
rounding-off operations in calculations, and exception specifications. Note that the description does not
fully conform to the requirements of the IEEE 754 standard.
- Chapter 3 "Micro Mode" describes the architecture and operation of micro mode, in which the VU operates
as a stand-alone processor.
- Chapter 4 "Micro Mode Instruction Reference" describes the individual micro mode instructions.
- Chapter 5 "Macro Mode" describes the architecture and operation of macro mode, in which VPU0 operates
as a coprocessor of the EE Core. This chapter also explains how to control VPU1 from the EE Core.
- Chapter 6 "Macro Mode Instruction Reference" describes the individual macro mode instructions.
- Chapter 7 "Appendix" gives other information, including sample micro programs.
Changes Since Release of 5th Edition
Since release of the 5th Edition of the VU User’s Manual, the following changes have been made.
Note that each of these changes is indicated by a revision bar in the margin of the affected page.
Ch. 1: VU Overview
• Information about the derivation of the M series polynomial was added to “RANDU” in section 1.1.2.
Lower Execution Unit, on page 17.
Ch. 3: Micro Mode
• A correction was made to the description following the figure in Section 3.4.9. XGKICK Pipeline, on
page 52.
Ch. 4: Micro Mode Instruction Reference
• A correction was made to the “Operation Code” figure in the EATANxz reference on page 134.
• A correction was made to “Example” in the IBGTZ reference on page 164.
• A correction was made to “Example” in the IBLEZ reference on page 165.
• A correction was made to “Example” in the IBLTZ reference on page 166.
• A correction was made to “Mnemonic” in the ISW reference on page 173.
• Information about the XGKIXK pipeline and XGKICK synchronization was added to “Remarks” in
the XGKICK reference on page 196.
• Information was added to “Operation” in the XITOP reference on page 197.
Ch. 6: Macro Mode Instruction Reference
• A correction was made to “Operation” in the QMTC2 reference on page 238.

© SCEI
-3-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

(This page is left blank intentionally)

© SCEI
-4-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Glossary
Term
EE
EE Core
COP0
COP1
COP2
GS
GIF
IOP
SBUS
VPU (VPU0/VPU1)
VU (VU0/VU1)
VIF (VIF0/VIF1)
VIFcode
SPR
IPU
word
qword
Slice
Packet
Transfer list
Tag
DMAtag
GS primitive
Context
GIFtag
Display list

Definition
Emotion Engine. CPU of the PlayStation 2.
Generalized computation and control unit of EE. Core of the CPU.
EE Core system control coprocessor.
EE Core floating-point operation coprocessor. Also referred to as FPU.
Vector operation unit coupled as a coprocessor of EE Core. VPU0.
Graphics Synthesizer.
Graphics processor connected to EE.
EE Interface unit to GS.
Processor connected to EE for controlling input/output devices.
Bus connecting EE to IOP.
Vector operation unit.
EE contains 2 VPUs: VPU0 and VPU1.
VPU core operation unit.
VPU data decompression unit.
Instruction code for VIF.
Quick-access data memory built into EE Core (Scratchpad memory).
EE Image processor unit.
Unit of data length: 32 bits
Unit of data length: 128 bits
Physical unit of DMA transfer: 8 qwords or less
Data to be handled as a logical unit for transfer processing.
A group of packets transferred in serial DMA transfer processing.
Additional data indicating data size and other attributes of packets.
Tag positioned first in DMA packet to indicate address/size of data and address
of the following packet.
Data to indicate image elements such as point and triangle.
A set of drawing information (e.g. texture, distant fog color, and dither matrix)
applied to two or more primitives uniformly. Also referred to as the drawing
environment.
Additional data to indicate attributes of GS primitives.
A group of GS primitives to indicate batches of images.

© SCEI
-5-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

(This page is left blank intentionally)

© SCEI
-6-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Contents
1. VU Overview ......................................................................................................................................................................... 15
1.1. VPU Structure ................................................................................................................................................................ 16
1.1.1. Upper Execution Unit ........................................................................................................................................... 16
1.1.2. Lower Execution Unit ........................................................................................................................................... 17
1.1.3. Floating-Point Registers ........................................................................................................................................ 18
1.1.4. Integer Register....................................................................................................................................................... 18
1.1.5. VU Mem.................................................................................................................................................................. 18
1.1.6. Micro Mem.............................................................................................................................................................. 19
1.2. VU Execution Mode ..................................................................................................................................................... 20
1.3. VU Operation Status ..................................................................................................................................................... 21
1.3.1. Ready State .............................................................................................................................................................. 21
1.3.2. Run State ................................................................................................................................................................. 21
1.3.3. Stop State................................................................................................................................................................. 21
1.4. VU Usage ........................................................................................................................................................................ 23
1.4.1. VU1 Usage Outline................................................................................................................................................ 23
1.4.2. VU0 Usage Outline................................................................................................................................................ 23
2. Data/Calculation Basic Specifications................................................................................................................................ 25
2.1. Data Format ................................................................................................................................................................... 26
2.1.1. Floating-Point Values ............................................................................................................................................ 26
2.1.2. Fixed-Point Values ................................................................................................................................................. 26
2.2. Rounding Off Floating-Point Values .......................................................................................................................... 27
2.3. Exception Processing .................................................................................................................................................... 28
2.4. Differences from IEEE 754......................................................................................................................................... 29
3. Micro Mode............................................................................................................................................................................ 31
3.1. Micro Mode Register Set............................................................................................................................................... 32
3.1.1. Floating-Point Registers ........................................................................................................................................ 32
3.1.2. Integer Registers ..................................................................................................................................................... 32
3.1.3. ACC Registers......................................................................................................................................................... 33
3.1.4. I Register.................................................................................................................................................................. 33
3.1.5. Q Register................................................................................................................................................................ 34
3.1.6. R Register ................................................................................................................................................................ 34
3.1.7. P Register................................................................................................................................................................. 34
3.2. Micro Instruction Set Overview .................................................................................................................................. 35
3.2.1. Upper Instructions ................................................................................................................................................. 35
3.2.2. Lower Instructions ................................................................................................................................................. 37
3.3. Flags................................................................................................................................................................................. 39
3.3.1. MAC Flags............................................................................................................................................................... 39
3.3.2. Status Flags (SF) ..................................................................................................................................................... 39
3.3.3. Clipping Flags (CF) ................................................................................................................................................ 40
3.3.4. Flag Set Instructions .............................................................................................................................................. 40
© SCEI
-7-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.3.5. Flag Changes for Each Instruction.......................................................................................................................41
3.3.6. Flag Changes for Exception Occurrences...........................................................................................................41
3.4. Pipeline Operation .........................................................................................................................................................44
3.4.1. Hazards.....................................................................................................................................................................44
3.4.2. Upper Instruction and Lower Instruction...........................................................................................................44
3.4.3. Priority for Writing to a Register ..........................................................................................................................44
3.4.4. FMAC Pipeline........................................................................................................................................................45
3.4.5. FDIV Pipeline .........................................................................................................................................................46
3.4.6. EFU Pipeline ...........................................................................................................................................................48
3.4.7. IALU Pipeline .........................................................................................................................................................49
3.4.8. Conditional Branching and Pipeline.....................................................................................................................50
3.4.9. XGKICK Pipeline ..................................................................................................................................................52
3.5. Micro Subroutine Execution.........................................................................................................................................53
3.5.1. How to Execute a Micro Subroutine ...................................................................................................................53
3.5.2. How to Terminate a Micro Subroutine ...............................................................................................................53
3.5.3. Operation of Execution and Termination...........................................................................................................53
3.6. Other Functions .............................................................................................................................................................55
3.6.1. Data Transfer with VU Mem/Micro Mem .........................................................................................................55
3.6.2. Debug Support Function.......................................................................................................................................55
4. Micro Mode Instruction Reference .....................................................................................................................................57
4.1. Micro Mode Instruction Set..........................................................................................................................................58
4.1.1. Types of Upper Instruction...................................................................................................................................58
4.1.2. Types of Lower Instructions .................................................................................................................................59
4.1.3. Operation Fields for Micro Instructions .............................................................................................................61
4.2. Upper Instruction Reference ........................................................................................................................................65
ABS : Absolute Value...................................................................................................................................................66
ADD : Add ....................................................................................................................................................................67
ADDi : Add to I Register ............................................................................................................................................68
ADDq : Add to Q Register .........................................................................................................................................69
ADDbc : Broadcast Add .............................................................................................................................................70
ADDA : Add; to Accumulator ...................................................................................................................................71
ADDAi : Add I Register; to Accumulator.................................................................................................................72
ADDAq : Add Q Register; to Accumulator .............................................................................................................73
ADDAbc : Broadcast Add; to Accumulator.............................................................................................................74
CLIP : Clipping Judgment ...........................................................................................................................................75
FTOI0 : Convert to Fixed Point.................................................................................................................................77
FTOI4 : Convert to Fixed Point.................................................................................................................................78
FTOI12 : Convert to Fixed Point ..............................................................................................................................79
FTOI15 : Convert to Fixed Point ..............................................................................................................................80
ITOF0 : Convert to Floating-Point Number............................................................................................................81
ITOF4 : Convert to Floating-Point Number............................................................................................................82
ITOF12 : Convert to Floating-Point Number..........................................................................................................83
ITOF15 : Convert to Floating-Point Number..........................................................................................................84
MADD : Product Sum.................................................................................................................................................85
MADDi : Product Sum; with I Register ....................................................................................................................86
MADDq : Product Sum; by Q Register.....................................................................................................................87
© SCEI
-8-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDbc : Broadcast Product Sum.......................................................................................................................... 88
MADDA : Product Sum; to Accumulator................................................................................................................ 89
MADDAi : Product Sum; by I register, to Accumulator ........................................................................................ 90
MADDAq : Product Sum; by Q Register, to Accumulator.................................................................................... 91
MADDAbc : Broadcast Product Sum; to Accumulator ......................................................................................... 92
MAX : Maximum Value .............................................................................................................................................. 93
MAXi : Maximum Value ............................................................................................................................................. 94
MAXbc : Maximum Value .......................................................................................................................................... 95
MINI : Minimum Value .............................................................................................................................................. 96
MINIi : Minimum Value ............................................................................................................................................. 97
MINIbc : Minimum Value .......................................................................................................................................... 98
MSUB : Multiply and Subtract.................................................................................................................................... 99
MSUBi : Multiply and Subtract; with I Register.....................................................................................................100
MSUBq : Multiply and Subtract; by Q Register .....................................................................................................101
MSUBbc : Broadcast Multiply and Subtract...........................................................................................................102
MSUBA : Multiply and Subtract; to Accumulator .................................................................................................103
MSUBAi : Multiply and Subtract; with I Register, to Accumulator ....................................................................104
MSUBAq : Multiply and Subtract; by Q Register, to Accumulator.....................................................................105
MSUBAbc : Broadcast Multiply and Subtract; to Accumulator ..........................................................................106
MUL : Multiply ...........................................................................................................................................................107
MULi : Multiply by I Register...................................................................................................................................108
MULq : Multiply by Q Register................................................................................................................................109
MULbc : Multiply by Broadcast ...............................................................................................................................110
MULA : Multiply; to Accumulator ..........................................................................................................................111
MULAi : Multiply by I Register, to Accumulator ..................................................................................................112
MULAq : Multiply by Q Register, to Accumulator ...............................................................................................113
MULAbc : Broadcast Multiply by broadcast, to Accumulator.............................................................................114
NOP : No Operation.................................................................................................................................................115
OPMULA : Vector Outer Product..........................................................................................................................116
OPMSUB : Vector Outer Product...........................................................................................................................117
SUB : Subtract ............................................................................................................................................................118
SUBi : Subtract I Register..........................................................................................................................................119
SUBq : Subtract Q Register ......................................................................................................................................120
SUBbc : Broadcast Subtract......................................................................................................................................121
SUBA : Substract; to Accumulator ..........................................................................................................................122
SUBAi : Subtract I Register; to Accumulator.........................................................................................................123
SUBAq : Subtract Q Register; to Accumulator......................................................................................................124
SUBAbc : Broadcast Subtract; to Accumulator .....................................................................................................125
4.3. Lower Instruction Reference......................................................................................................................................126
B : Unconditional Branch..........................................................................................................................................127
BAL : Unconditional Branch with Saving Address ...............................................................................................128
DIV : Divide ...............................................................................................................................................................129
EATAN : Arctangent ................................................................................................................................................130
EATANxy : Arctangent ............................................................................................................................................132
EATANxz : Arctangent ............................................................................................................................................134
EEXP : Exponent ......................................................................................................................................................136
© SCEI
-9-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ELENG : Length....................................................................................................................................................... 137
ERCPR : Reciprocal Number .................................................................................................................................. 138
ERLENG : Reciprocal Number of Length............................................................................................................ 139
ERSADD : Reciprocal Number .............................................................................................................................. 140
ERSQRT : Reciprocal Number of Square Root.................................................................................................... 141
ESADD : Sum of Square Numbers ........................................................................................................................ 142
ESIN : Sine ................................................................................................................................................................. 143
ESQRT : Square Root............................................................................................................................................... 144
ESUM : Sum of Each Field...................................................................................................................................... 145
FCAND : Test Clipping Flag ................................................................................................................................... 146
FCEQ : Test Clipping Flag....................................................................................................................................... 147
FCGET : Get Clipping Flag..................................................................................................................................... 148
FCOR : Test Clipping Flag....................................................................................................................................... 149
FCSET : Setting Clipping Flag................................................................................................................................. 150
FMAND : Test MAC Flag Check ........................................................................................................................... 151
FMEQ : Test MAC Flag Check............................................................................................................................... 152
FMOR : Test MAC Flag Check............................................................................................................................... 153
FSAND : Test Status Flag Check ............................................................................................................................ 154
FSEQ : Test Status Flag Check................................................................................................................................ 155
FSOR : Test Status Flag............................................................................................................................................ 156
FSSET : Set Sticky Flags........................................................................................................................................... 157
IADD : ADD Integer ............................................................................................................................................... 158
IADDI : Add Immediate Value Integer ................................................................................................................. 159
IADDIU : Add Immediate Integer ......................................................................................................................... 160
IAND : Logical Product ........................................................................................................................................... 161
IBEQ : Conditional Branch...................................................................................................................................... 162
IBGEZ : Conditional Branch................................................................................................................................... 163
IBGTZ : Conditional Branch................................................................................................................................... 164
IBLEZ : Conditional Branch ................................................................................................................................... 165
IBLTZ : Conditional Branch.................................................................................................................................... 166
IBNE : Conditional Branch...................................................................................................................................... 167
ILW : Integer Load with Offset Specification ....................................................................................................... 168
ILWR : Integer Load ................................................................................................................................................. 169
IOR : Logical Sum ..................................................................................................................................................... 170
ISUB : Integer Subtract............................................................................................................................................. 171
ISUBIU : Immediate Value Integer Subtract ......................................................................................................... 172
ISW : Integer Store with Offset............................................................................................................................... 173
ISWR : Integer Store ................................................................................................................................................. 174
JALR : Unconditional Jump with Address Saving ................................................................................................ 175
JR : Unconditional Jump........................................................................................................................................... 176
LQ : Load Qword...................................................................................................................................................... 177
LQD : Load Qword with Pre-Decrement.............................................................................................................. 178
LQI : Load with Post-Increment............................................................................................................................. 179
MFIR : Move from Integer Register to Floating-Point Register......................................................................... 180
MFP : Move from P Register to Floating-Point Register..................................................................................... 181
MOVE : Transfer between Floating-Point Registers............................................................................................ 182
© SCEI
-10-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MR32 : Move with Rotate.........................................................................................................................................183
MTIR : Move from Floating-Point Register to Integer Register .........................................................................184
RGET : Get Random Number ................................................................................................................................185
RINIT : Random Number Intialize.........................................................................................................................186
RNEXT : Next Random Number ...........................................................................................................................187
RSQRT : Square Root Division ...............................................................................................................................188
RXOR : Random Number Set .................................................................................................................................189
SQ : Store Qword with Offset .................................................................................................................................190
SQD : Store Qword with Pre-Decrement ..............................................................................................................191
SQI : Store with Post-Increment..............................................................................................................................192
SQRT : Square Root ..................................................................................................................................................193
WAITP : P Register Syncronize ...............................................................................................................................194
WAITQ : Q Register Syncronize .............................................................................................................................195
XGKICK : GIF Control ...........................................................................................................................................196
XITOP : VIF Control................................................................................................................................................196
XTOP : VIF Control .................................................................................................................................................198
5. Macro Mode .........................................................................................................................................................................199
5.1. Macro Mode Register Set............................................................................................................................................200
5.1.1. Floating-Point Registers ......................................................................................................................................200
5.1.2. Integer Registers ...................................................................................................................................................200
5.1.3. Control Registers..................................................................................................................................................200
5.1.4. Special Registers ...................................................................................................................................................205
5.2. Macro Instruction Set Overview................................................................................................................................206
5.2.1. MIPS COP2 Instructions ....................................................................................................................................206
5.2.2. Coprocessor Transfer Instructions ....................................................................................................................206
5.2.3. Coprocessor Branch Instructions ......................................................................................................................206
5.2.4. Coprocessor Calculation Instructions ...............................................................................................................206
5.2.5. Micro Subroutine Execution Instructions ........................................................................................................208
5.3. Flags...............................................................................................................................................................................209
5.4. Macro Mode Pipeline ..................................................................................................................................................210
5.4.1. Pipeline Structure of Macroinstructions............................................................................................................210
5.4.2. Hazards in Macro Mode......................................................................................................................................210
5.4.3. Macroinstruction Operation ...............................................................................................................................211
5.4.4. Operation when Transferring Data with EE Core..........................................................................................211
5.4.5. Operation when Executing a Micro Subroutine ..............................................................................................215
5.4.6. Micro Subroutine and Data Transfer Operations............................................................................................216
5.4.7. Q Register Synchronization ................................................................................................................................218
5.4.8. Notes on Other Pipeline Operations.................................................................................................................218
5.5. VU1 Control.................................................................................................................................................................220
5.5.1. MIPS COP2 Condition Signal............................................................................................................................220
5.5.2. MIPS COP2 Control Register ............................................................................................................................220
5.5.3. Floating-Point Registers ......................................................................................................................................221
5.5.4. Integer Registers ...................................................................................................................................................222
5.5.5. Control Registers..................................................................................................................................................222
6. Macro Mode Instruction Reference..................................................................................................................................225
6.1. Macro Instruction Operation Code...........................................................................................................................226
© SCEI
-11-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

6.1.1. Macro Instruction Operation Type ................................................................................................................... 226
6.1.2. Macro Instruction Operation Field ................................................................................................................... 227
6.2. Macro Instruction Set ................................................................................................................................................. 229
BC2F : Branch on COP2 Conditional Signal......................................................................................................... 230
BC2FL : Branch on COP2 Conditional Signal ...................................................................................................... 231
BC2T : Branch on COP2 Conditional Signal ........................................................................................................ 232
BC2TL : Branch on COP2 Conditional signal ...................................................................................................... 233
CFC2 : Transfer Integer Data from VU to EE Core ........................................................................................... 234
CTC2 : Transfer Integer Data from EE Core to VU ........................................................................................... 235
LQC2 : Floating-Point Data Transfer from EE Core to VU .............................................................................. 236
QMFC2 : Floating-Point Data Transfer from VU to EE Core........................................................................... 237
QMTC2 : Floating-Point Data Transfer from EE Core to VU .......................................................................... 238
SQC2 : Floating-Point Data Transfer from VU to EE Core............................................................................... 239
VABS : Absolute Value............................................................................................................................................. 240
VADD : Add.............................................................................................................................................................. 241
VADDi : Add to I Register ...................................................................................................................................... 242
VADDq : Add to Q Register ................................................................................................................................... 243
VADDbc : Broadcast Add ....................................................................................................................................... 244
VADDA : Add to Accumulator .............................................................................................................................. 245
VADDAi : Add I Register to Accumulator ........................................................................................................... 246
VADDAq : Add Q Register to Accumulator ........................................................................................................ 247
VADDAbc : Broadcast Add to Accumulator........................................................................................................ 248
VCALLMS : Start Micro Sub-Routine.................................................................................................................... 249
VCALLMSR : Start Micro Sub-Routine by Register............................................................................................. 250
VCLIP : Clipping Judgment ..................................................................................................................................... 251
VDIV : Divide............................................................................................................................................................ 252
VFTOI0 : Conversion to Fixed Point..................................................................................................................... 253
VFTOI4 : Conversion to Fixed Point..................................................................................................................... 254
VFTOI12 : Conversion to Fixed Point .................................................................................................................. 255
VFTOI15 : Conversion to Fixed Point .................................................................................................................. 256
VIADD : Add Integer............................................................................................................................................... 257
VIADDI : Add Immediate Value Integer .............................................................................................................. 258
VIAND : Logical Product ........................................................................................................................................ 259
VILWR : Integer Load .............................................................................................................................................. 260
VIOR : Logical Sum.................................................................................................................................................. 261
VISUB : Integer Subtract.......................................................................................................................................... 262
VISWR : Integer Store .............................................................................................................................................. 263
VITOF0 : Conversion to Floating-Point Number................................................................................................ 264
VITOF4 : Conversion to Floating-Point Number................................................................................................ 265
VITOF12 : Conversion to Floating-Point Number.............................................................................................. 266
VITOF15 : Conversion to Floating-Point Number.............................................................................................. 267
VLQD : Load with Pre-Decrement ........................................................................................................................ 268
VLQI : Load with Post-Increment.......................................................................................................................... 269
VMADD : Product Sum........................................................................................................................................... 270
VMADDi : Product Sum; with I Register .............................................................................................................. 271
VMADDq : Product Sum; with Q Register ........................................................................................................... 272
© SCEI
-12-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDbc : Broadcast Product Sum.....................................................................................................................273
VMADDA : Product Sum; to Accumulator...........................................................................................................274
VMADDAi : Product Sum; with I Register, to Accumulator ..............................................................................275
VMADDAq : Product Sum; with Q Register, to Accumulator ...........................................................................276
VMADDAbc : Broadcast Product Sum; to Accumulator ....................................................................................277
VMAX : Maximum Value .........................................................................................................................................278
VMAXi : Maximum Value ........................................................................................................................................279
VMAXbc : Maximum Value.....................................................................................................................................280
VMFIR : Transfer from Integer Register to Floating-Point Register..................................................................281
VMINI : Minimum Value .........................................................................................................................................282
VMINIi : Minimum Value ........................................................................................................................................283
VMINIbc : Minimum Value .....................................................................................................................................284
VMOVE : Transfer between Floating-Point Registers .........................................................................................285
VMR32 : Vector Rotate.............................................................................................................................................286
VMSUB : Multiply and Subtract ..............................................................................................................................287
VMSUBi : Multiply and Subtract with I Register...................................................................................................288
VMSUBq : Multiply and Subtract; Q Register........................................................................................................289
VMSUBbc : Broadcast Multiply and Subtract........................................................................................................290
VMSUBA : Multiply and Subtract; to Accumulator..............................................................................................291
VMSUBAi : Multiply and Subtract; with I Register, to Accumulator .................................................................292
VMSUBAq : Multiply and Subtract; with Q Register, to Accumulator ..............................................................293
VMSUBAbc : Broadcast Multiply and Subtract; to Accumulator .......................................................................294
VMTIR : Transfer from Floating-Point Register to Integer Register .................................................................295
VMUL : Multiply........................................................................................................................................................296
VMULi : Multiply; by I Register...............................................................................................................................297
VMULq : Multiply; by Q Register............................................................................................................................298
VMULbc : Broadcast Multiply .................................................................................................................................299
VMULA : Multiply; to Accumulator .......................................................................................................................300
VMULAi : Multiply by I Register; to Accumulator ...............................................................................................301
VMULAq : Multiply by Q Register; to Accumulator ............................................................................................302
VMULAbc : Broadcast Multiply; to Accumulator.................................................................................................303
VNOP : No Operation..............................................................................................................................................304
VOPMULA : Vector Outer Product.......................................................................................................................305
VOPMSUB : Vector Outer Product........................................................................................................................306
VRGET : Get Random Numbers............................................................................................................................307
VRINIT : Random Number Initial Set ...................................................................................................................308
VRNEXT : New Random Numbers.......................................................................................................................309
VRSQRT : Square Root Division ............................................................................................................................310
VRXOR : Random Number Set ..............................................................................................................................311
VSQD : Store with Pre-Decrement.........................................................................................................................312
VSQI : Store with Post-Increment ..........................................................................................................................313
VSQRT : Square Root ...............................................................................................................................................314
VSUB : Subtract .........................................................................................................................................................315
VSUBi : Subtract I Register ......................................................................................................................................316
VSUBq : Subtract Q Register ...................................................................................................................................317
VSUBbc : Broadcast Subtract...................................................................................................................................318
© SCEI
-13-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBA : Subtract; to Accumulator ........................................................................................................................ 319
VSUBAi : Subtract I Register; to Accumulator ..................................................................................................... 320
VSUBAq : Subtract Q Register; to Accumulator .................................................................................................. 321
VSUBAbc : Broadcast Subtract; to Accumulator.................................................................................................. 322
VWAITQ : Q Register Synchronize ....................................................................................................................... 323
7. Appendix.............................................................................................................................................................................. 325
7.1. Sample Micro Programs ............................................................................................................................................. 326
7.2. EFU Processing........................................................................................................................................................... 353
7.3. Micro Subroutine Debugging .................................................................................................................................... 362
7.3.1. Debug Flow .......................................................................................................................................................... 362
7.3.2. Notes on Re-execution........................................................................................................................................ 363
7.4. Throughput / Latency List ........................................................................................................................................ 365

© SCEI
-14-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

1. VU Overview
The VU is a vector ALU that efficiently performs four-element floating-point vector calculations. It is part of
the VPU, along with the VU Mem (VU Data Memory) and the VIF (compressed data expansion engine).
Two VPUs are mounted on the EE as shown in Figure 1-1.
VPU0

VPU1
VU1

VU0
4KByte

Micro Mem1
16KByte

VUcore

VU Mem1

4KByte

16KByte

VIF0

VIF1

GIF

VU Mem0

EFU

VUcore

VU regs.

VU regs.

EE
core

Micro Mem0

Main Memory

Figure 1-1 VPU System Outline
VU0 is joined to the EE Core as COP2 via a coprocessor connection. It assists the EE Core in non-stationary
geometry processing. It has a 4-Kbyte instruction memory (MicroMem0) and a 4-Kbyte data memory (VU
Mem0).
VU1 operates independently, and is chiefly in charge of background stationary geometry processing. VU1 has
an Elementary Function Unit (EFU), as well as a 16-Kbyte instruction memory (MicroMem1) and a 16-Kbyte
data memory (VU Mem1). VU1 is also connected to the GIF (the interface unit to the Graphics Synthesizer),
and the GIF control instruction (XGKICK instruction) is mounted. VU1's floating-point and integer registers
are mapped to VPU0's VU Mem0.

© SCEI
-15-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

1.1. VPU Structure
Figure 1-2 is a block diagram of the VPU including the VU.
Micro instruction fetch unit

Micro Mem

Vector Unit :VU

special
registers

4KByte or 16KByte

64

63

VI16` VI31

0

Upper Instruction Lower Instruction

bold line : 128 bit

32

32

16

COP2
control
registers

16
32
BRU

EFU

IALU

RANDU/etc

FDIV

(COP2 data registers)

LSU

Lower Execution Unit

FMACx

VF00-VF31

Upper Execution Unit
FMACy

floating
registers

0

FMACz

QMTC2
/ LQC2

127

FMACw

QMFC2
/ SQC2

integer
registers
VI00` VI15

CFC2

32
CTC2

16

Vector Processing Unit:VPU

VU Mem
4KByte or 16KByte

VIF
External Units

Figure 1-2 VU Outline Block Diagram
The VPU consists of the VU, VU Mem (VU Data Memory), and the VIF (compressed data expansion engine).
The VU loads data in 128-bit units (single-precision floating-point number x 4) from the VU Mem, performs
calculations according to micro programs in the VU's internal MicroMem, and stores the results in the VU Mem.
The VU Mem may be used as a temporary area depending on the micro program.
Micro programs employ 64-bit length LIW (Long Instruction Word) instruction sets. They can concurrently
execute a floating-point product-sum calculation in the upper 32 bits (the Upper instruction field) and a floatingpoint division or integer calculation in the lower 32 bits (the Lower instruction field).
There are 32 128-bit floating-point registers (single-precision floating-point x 4). There are 16 16-bit integer
registers.

1.1.1. Upper Execution Unit
FMAC
This unit adds, subtracts, multiplies, and does product-sum operations on floating-point numbers. Four
units are mounted in order to efficiently execute four-element vector calculations: FMACx, FMACy,
FMACz, and FMACw. To increase the efficiency of pipeline processing, the latency of the instructions that
use the FMAC has been unified at four cycles.

© SCEI
-16-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

1.1.2. Lower Execution Unit
FDIV
This unit performs self-synchronous high-speed floating-point division/square root calculations. It uses a
single-precision floating-point value as input, then stores the calculation result in the dedicated Q register.
The next FDIV instruction cannot be executed while the FDIV is executing. The FDIV stalls if this is
attempted.
LSU
This unit controls Load/Store to and from VU Mem.
Load/Store must be performed in units of 128 bits, but the x, y, z, and w field units can be masked.
There are two ways of specifying addresses. The first specifies a base (Integer) register and offsets in the
operation code field. The second specifies the base register only, not the offsets, and performs postincrementing or pre-decrementing. Incrementing/Decrementing is +1/-1 in 128-bit word address units.
IALU
This unit performs 16-bit integer calculations.
Loop counter calculation and Load/Store address calculation are performed using the Integer registers.
BRU
This unit controls jumping and conditional branching.
Most instructions specify PC-relative addresses for the jump target address. The offset is specified by an 11bit immediate value, so it is possible to jump within a range of 8 Kbytes before and after the PC. The JR and
JALR instructions are register indirect jump instructions, which use the data in a register as an absolute
address.
Conditional branching is performed by a comparison with one or two Integer registers. When doing
conditional branching based on the results of floating-point calculations, the results of the AND operation
with the MAC flag, status flag, or clipping flag and the appropriate mask value are temporarily stored in an
Integer register. Branching is performed by comparing with this register.
RANDU
This unit generates floating-point random numbers in the range +1.0
fixed point count)
Note: D flag is not set for 0/0.

MAC / status flag

Sticky flag

I flag = 1
I flag = 1
D flag = 1
Ox/Oy/Oz/Ow/O flag = 1
Ux/Uy/Uz/Uw/U flag = 1
Zx/Zy/Zz/Zw/Z flag = 1
None

IS flag = 1
IS flag = 1
DS flag = 1
OS flag = 1
US flag = 1
ZS flag = 1
None

© SCEI
-28-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

2.4. Differences from IEEE 754
The following are differences between VU calculations and the IEEE 754 standard:
• Precision
The VU supports only single-precision floating-point values.
• Rounding off
Of the four IEEE 754 rounding-off modes, the VU performs rounding similar to a 0 direction truncation.
Since the least significant bit may vary, this method is not exactly the same as IEEE 754.
• Non-numeral/infinity/non-normalized number
The VU does not support non-numerals, infinity, and non-normalized numbers. Therefore, exceptions
related to these items are not generated.
• Overflow/underflow
Overflow and underflow are detected only by the overflow and underflow of the exponent. Underflow in
particular is not considered to lose the precision prescribed by IEEE 754.
• Trap for exception
IEEE 754 recommends that a trap be settable for a calculation result exception. In the VU, processing
continues by just setting a flag. If a trap is necessary, use a flag check instruction after the calculation
instruction.

© SCEI
-29-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

(This page is left blank intentionally)

© SCEI
-30-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3. Micro Mode
In Micro mode, the VU executes microinstruction programs (micro subroutines) in MicroMem as a stand-alone
processor. Using the VU in micro mode maximizes the degree of parallelism, and the highest performance can
be achieved.
Microinstructions are 64-bit long LIW instructions. It is possible to specify both an Upper instruction and a
Lower instruction at the same time. The Upper instruction controls four floating-point product-sum ALUs
(FMAC), and the Lower instruction controls one floating-point division/square root ALU (FDIV), one loadstore unit (LSU), and one integer operation unit (IALU), etc. A maximum of six units can execute concurrently.

© SCEI
-31-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.1. Micro Mode Register Set
In micro mode, 32 floating-point registers, 16 integer registers, and other special registers can be used. These
registers are explained below.

3.1.1. Floating-Point Registers
There are 32 floating-point registers, VF00 - VF31. These are 128-bit long vector registers, which consist of
four single-precision floating-point fields.
The four 32-bit fields have a little-endian arrangement. From the least to the most significant bit, they are
referred to as field x, field y, field z, and field w.
32 bits
127
96

VF00
VF01
VF02
VF03

VF00w
VF01w
VF02w
VF03w

VF31

VF31w

32 bits
95

32 bits
64

VF00z
VF01z
VF02z
VF03z
:
:
VF31z

63

32 bits
32

31

0

VF00y
VF01y
VF02y
VF03y

VF00x
VF01x
VF02x
VF03x

VF31y

VF31x

VF00 is the constant register. Its fields are set to the following values:
• VF00x: 0.0

(Single-precision floating number)

• VF00y: 0.0

(Single-precision floating number)

• VF00z: 0.0

(Single-precision floating number)

• VF00w: 1.0

(Single-precision floating number)

3.1.2. Integer Registers
There are 16 16-bit long integer registers, VI00 - VI15. VI00 is the constant register and is set to 0.

© SCEI
-32-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

15

VI00
VI01
VI02
VI03
VI04
VI05
VI06
VI07
VI08
VI09
VI10
VI11
VI12
VI13
VI14
VI15

0

0 Register

Stack pointer (recommended)
Link register (recommended)

3.1.3. ACC Registers
The ACC registers are accumulators for floating-point product-sum calculations. Four registers exist for four
product-sum ALUs: ACCx, ACCy, ACCz, and ACCw. The ACC registers not only work as destinations of
instructions such as ADDA and MULA but also store the intermediate results of the vector outer product
calculated by OPMULA and OPMULB.
Stalls due to data dependency, described later in this document, do not occur to the ACC register.

3.1.4. I Register
The I register is a 32-bit single-precision floating-point register in which immediate values are stored. When the
I bit (bit 63) of the Upper OP field is set, the content of the Lower OP field is written to the I register at the T
stage of the instruction as a single-precision floating-point number. It is used by the next instruction to be
executed, such as ADDi/MULi. This operation can be described with the LOI pseudo instruction.
No stalls due to data dependency are generated for the I register. Figure 3-1 illustrates a pipeline operation
example in which the I register is used.
clock cycle
MUL.xyzw VF05, VF10, VF20
LOI 100.0 (floating point count imm value)

1

2

3

4

5

6

M

T

X

Y

Z

S

M

T

X

Y

Z

S

M

T

X

Y

Z

S

M

T

X

Y

Z

S

MULi.xyzw VF10, VF11, I
NOP (MOVE VF00, VF00)

7

8

Figure 3-1 Usage Example of I Register
For more information on pipeline operation, see "3.4. Pipeline Operation".

© SCEI
-33-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.1.5. Q Register
The Q register is a floating-point register in which results of division, square root operations, and square root
division are stored. Because these calculations differ in latency from other floating-point calculations, a special
register (the Q register) is needed.
No stalls due to data dependency are generated for the Q register, and it is necessary to use the WAITQ
instruction for synchronization.

3.1.6. R Register
The R register is a 23-bit register in which random number values are stored.

3.1.7. P Register
The P register is a 32-bit register in which the EFU instruction results are stored. Because elementary function
calculations differ in latency from other floating-point calculations, a special register (the P register) is needed.
No stalls due to data dependency are generated for the P register, and it is necessary to use the WAITP
instruction for synchronization.

© SCEI
-34-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.2. Micro Instruction Set Overview
A microinstruction is a 64-bit LIW (Long Instruction Word). It can specify instructions in the Upper instruction
field (the upper 32 bits) and the Lower instruction field (the lower 32 bits) independently.
Instructions that use floating-point product-sum ALUs (FMAC) are usually specified in the Upper instruction
field, and other instructions are specified in the Lower instruction field.
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

Upper Instruction Field
Upper Instruction
32 bits
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower Instruction Field
Lower Instruction
32 bits

3.2.1. Upper Instructions
The Upper instructions are mainly related to floating-point calculations. There are 59 instructions.
Category
Instruction
Function
Floating- point ABS
absolute
calculation
ADD
addition
ADDi
ADD broadcast I register
ADDq
ADD broadcast Q register
ADDbc
ADD broadcast bc field
ADDA
ADD output to ACC
ADDAi
ADD output to ACC broadcast I register
ADDAq
ADD output to ACC broadcast Q register
ADDAbc
ADD output to ACC broadcast bc field
SUB
subtraction
SUBi
SUB broadcast I register
SUBq
SUB broadcast Q register
SUBbc
SUB broadcast bc field
SUBA
SUB output to ACC
SUBAi
SUB output to ACC broadcast I register
SUBAq
SUB output to ACC broadcast Q register
SUBAbc
SUB output to ACC broadcast bc field
MUL
multiply
MULi
MUL broadcast I register
MULq
MUL broadcast Q register
MULbc
MUL broadcast bc field
MULA
MUL output to ACC
MULAi
MUL output to ACC broadcast I register
MULAq
MUL output to ACC broadcast Q register
MULAbc
MUL output to ACC broadcast bc field
MADD
MUL and ADD
MADDi
MUL and ADD broadcast I register
MADDq
MUL and ADD broadcast Q register
MADDbc
MUL and ADD broadcast bc field
© SCEI
-35-

SCE CONFIDENTIAL

Category

Floatingpoint/fixedpoint
conversion

Clipping
judgment

Instruction
MADDA
MADDAi
MADDAq
MADDAbc
MSUB
MSUBi
MSUBq
MSUBbc
MSUBA
MSUBAi
MSUBAq
MSUBAbc
MAX
MAXi
MAXbc
MINI
MINIi
MINIbc
OPMULA
OPMSUB
NOP
FTOI0
FTOI4
FTOI12
FTOI15
ITOF0
ITOF4
ITOF12
ITOF15
CLIP

VU User's Manual Version 6.0

Function
MUL and ADD output to ACC
MUL and ADD output to ACC broadcast I register
MUL and ADD output to ACC broadcast Q register
MUL and ADD output to ACC broadcast bc field
MUL and SUB
MUL and SUB broadcast I register
MUL and SUB broadcast Q register
MUL and SUB broadcast bc field
MUL and SUB output to ACC
MUL and SUB output to ACC broadcast I register
MUL and SUB output to ACC broadcast Q register
MUL and SUB output to ACC broadcast bc field
maximum
MAX broadcast I register
MAX broadcast bc field
minimum
MINI broadcast I register
MINI broadcast bc field
outer product MULA
outer product MSUB
no operation
float to integer, fixed point 0 bit
float to integer, fixed point 4 bits
float to integer, fixed point 12 bits
float to integer, fixed point 15 bits
integer to float, fixed point 0 bit
integer to float, fixed point 4 bits
integer to float, fixed point 12 bits
integer to float, fixed point 15 bits
clipping

© SCEI
-36-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.2.2. Lower Instructions
The Lower instructions are floating-point division, integer calculation, transfer between registers, flag operation,
branching, and elementary function calculation, and other control instructions. There are 69 instructions listed
below.
The NOP instruction is not included in the Lower instructions. If necessary, use meaningless instructions such
as MOVE VF00, VF00 in place of NOP.
Category
Floating-point
division
Integer
calculation

Registerregister transfer

Load/Store

Random
numbers

Synchronization
Flag operation

Branching

Instruction
DIV
SQRT
RSQRT
IADD
IADDI
IADDIU
IAND
IOR
ISUB
ISUBIU
MOVE
MFIR
MTIR
MR32
LQ
LQD
LQI
SQ
SQD
SQI
ILW
ISW
ILWR
ISWR
LOI
RINIT
RGET
RNEXT
RXOR
WAITQ

Function
floating divide
floating square-root
floating reciprocal square-root
integer ADD
integer ADD immediate
integer ADD immediate unsigned
integer AND
integer OR
integer SUB
integer SUB immediate unsigned
move floating register
move from integer register
move to integer register
move rotate 32 bits
Load Quadword
Load Quadword with pre-decrement
Load Quadword with post-increment
Store Quadword
Store Quadword with pre-decrement
Store Quadword with post-increment
integer load word
integer store word
integer load word register
integer store word register
Load immediate value to I register (pseudo instruction)
random-unit init R register
random-unit get R register
random-unit next M sequence
random-unit XOR R register
wait Q register

FSAND
FSEQ
FSOR
FSSET
FMAND
FMEQ
FMOR
FCAND
FCEQ
FCOR
FCSET
FCGET
IBEQ

flag-operation status flag AND
flag-operation status flag EQ
flag-operation status flag OR
flag-operation set status flag
flag-operation MAC flag AND
flag-operation MAC flag EQ
flag-operation MAC flag OR
flag-operation clipping flag AND
flag-operation clipping flag EQ
flag-operation clipping flag OR
flag-operation clipping flag set
flag-operation clipping flag get
integer branch on equal
© SCEI
-37-

SCE CONFIDENTIAL

Category

EFU transfer
EFU
synchronization
Vector
elementary
function

Scalar
elementary
function

External unit
control

VU User's Manual Version 6.0

Instruction
IBGEZ
IBGTZ
IBLEZ
IBLTZ
IBNE
B
BAL
JR
JALR
MFP
WAITP

Function
integer branch on greater than or equal to zero
integer branch on greater than zero
integer branch on less than or equal to zero
integer branch on less than zero
integer branch on not equal
branch (PC relative address)
branch and link (PC relative address)
jump register (absolute address)
jump and link register (absolute address)
move from P register
wait P register

ESADD
ERSADD
ELENG
ERLENG
EATANxy
EATANxz
ESUM
ERCPR
ESQRT
ERSQRT
ESIN
EATAN
EEXP
XGKICK
XTOP
XITOP

Elementary-function Square and ADD
Elementary-function Reciprocal Square and ADD
Elementary-function Length
Elementary-function Reciprocal Length
Elementary-function ArcTAN y/x
Elementary-function ArcTAN z/x
Elementary-function Sum
Elementary-function Reciprocal
Elementary-function Square-root
Elementary-function Reciprocal Square-root
Elementary-function SIN
Elementary-function ArcTAN
Elementary-function Exponential
Kick external unit (GIF)
Read VIF data (TOP register)
Read VIF data (ITOP register)

© SCEI
-38-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.3. Flags
VU flags are roughly divided into 3 types: MAC flags which show floating-point calculation results, status flags
which show total calculation results, and clipping flags which show clipping judgment results.

3.3.1. MAC Flags
The MAC flags show the FMAC calculation results of the Upper instruction (or the macroinstruction
corresponding to it). There are four flag types, z, s, u and o, and each of them has 1 bit corresponding to the
four floating-point calculation units, FMACx, FMACy, FMACz, and FMACw (16 bits in total).
15

14

13

12

11

10

O field
Ox

Oy

09

08

07

06

U field

Oz

Ow

Ux

Uy

4 bits

Uz

05

04

03

S field
Uw

Sx

Sy

4 bits

Sz

02

01

00

Z field
Sw

Zx

4 bits

Zy

Zz

Zw

4 bits

Z field: Zero flag
This flag is set to 1 when the calculation results are 0, and set to 0 when the calculation results are non-zero.
S field: Sign flag
This flag is set to 1 when the calculation results are negative, and set to 0 when the calculation results are
positive or 0.
U field: Underflow flag
This flag is set to 1 when the calculation results cause an underflow.
O field: Overflow flag
This flag is set to 1 when the calculation results cause an overflow.
The flags of ALUs that do not operate are cleared to 0. In the ADD.xyz instruction, for example, the Zw, Sw,
Uw, and Ow flags become 0 since FMACw does not operate.
Also, the MAC flags do not change when NOPs are executed.
When an overflow or underflow occurs, the Z and S flags are set according to the clamped results (+MAX/MAX/+0/-0).

3.3.2. Status Flags (SF)
Status flags consist of the following 12 bits. There are flags to show MAC flag status, flags to show invalid
calculation, and flags to show their accumulation (sticky flag).
11

10

09

08

07

06

05

04

03

02

01

00

DS

IS

OS

US

SS

ZS

D

I

O

U

S

Z

Z: Zero flag
This flag is set to 1 when any of the Zx, Zy, Zz, and Zw bits of the MAC flags is set to 1.
S: Sign flag
This flag is set to 1 when any of the Sx, Sy, Sz, and Sw bits of the MAC flags is set to 1.
© SCEI
-39-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

U: Underflow flag
This flag is set to 1 when any of the Ux, Uy, Uz, and Uw bits of the MAC flags is set to 1.
O: Overflow flag
This flag is set to 1 when any of the Ox, Oy, Oz, and Ow bits of the MAC flags is set to 1.
I: Invalid flag
This flag is set to 1 when a 0/0 calculation is executed by the DIV instruction or a root calculation of
negative number is executed by the SQRT/RSQRT instruction.
D: Zero division flag
This flag is set to 1 when 0 division (except 0/0) is performed by the DIV/RSQRT instruction, and set to 0
when the SQRT instruction is executed regardless of the results.
ZS/SS/US/OS/IS/DS: Sticky flag
Six flags, ZS, SS, US, OS, IS, and DS, are called Sticky Flags, and indicate the accumulation values of the Z,
S, U, O, I, and D flags, respectively. For example, the ZS flag value becomes the logical OR of the Z flag
value and the ZS flag value from the results of the most-recently performed instruction.
In product-sum calculations, a single instruction, such as MADD/MSUB, performs addition and subtraction
following multiplication. Addition and subtraction results are reflected in the Z/ZS/S/SS/U/US/O/OS
flags, and multiplication results are reflected only in the ZS/SS/US/OS flags.

3.3.3. Clipping Flags (CF)
The clipping flags are set according to the results of the clipping judgment (CLIP) instruction.
-x flag
Set to 1 when x < -|w|.
+x flag
Set to 1 when x > +|w|.
-y flag
Set to 1 when y < -|w|.
+y flag
Set to 1 when y > +|w|.
-z flag
Set to 1 when z < -|w|.
+z flag
Set to 1 when z > +|w|.
There are four sets of clipping flags, as shown below. The clipping flag is shifted 6 bits to the left each time the
CLIP instruction is executed, so clipping information for the four most recent vertices is always maintained.
23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

3rd previous
judgment

2nd previous
judgment

Previous
judgment

Current
judgment

-z +z -y +y -x +x -z +z -y +y -x +x -z +z -y +y -x +x -z +z -y +y -x +x
6 bits

6 bits

6 bits

6 bits

3.3.4. Flag Set Instructions
The following instructions (Lower instructions) are used to set the flags to a specific value.
FSSET instruction: Sets the status flag.
FCSET instruction: Sets the clipping flag.
The effect on the flag of the simultaneous Upper instruction is ignored and; the flag reflects only the results of
the flag setting instructions.

© SCEI
-40-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.3.5. Flag Changes for Each Instruction
The flag changes for each micorinstruction are shown in the table below.
"-" means "no flag change" and "X" means "flag change according to calculation results".
For the MAC flags, x, y, z, and w are grouped in the same column. Instructions of the same kind are grouped
with "*", for example ADD*.
Upper Instruction
Instruction
ABS
ADD*
ADDA*
CLIP
FTOI*
ITOF*
MADD*
MADDA*
MAX*
MINI*
MSUB*
MSUBA*
MUL*
MULA*
NOP
OPMULA
OPMSUB
SUB*
SUBA*
Lower Instruction
Instruction
DIV
RSQRT
SQRT
FCSET
FSSET
Others

MAC Flag
OUSZ
---XXXX
XXXX
---------XXXX
XXXX
------XXXX
XXXX
XXXX
XXXX
---XXXX
XXXX
XXXX
XXXX
MAC Flag
OUSZ
-------------------

Status Flag
DIOUSZ
DIOUSZ
SSSSSS
------------XXXX
--XXXX
--XXXX
--XXXX
--------------------------------XXXX
--XXXX
--XXXX
--XXXX
----------------------XXXX
--XXXX
--XXXX
--XXXX
--XXXX
--XXXX
--XXXX
--XXXX
------------XXXX
--XXXX
--XXXX
--XXXX
--XXXX
--XXXX
--XXXX
--XXXX

Clipping Flag

Status Flag
DIOUSZ
DIOUSZ
SSSSSS
XX---XX---XX---XX----X---0X-------------XXXXXX
----------------

Clipping Flag

X
-

X
-

3.3.6. Flag Changes for Exception Occurrences
Flag changes in the basic instructions (except MADD/MSUB)
The following table shows the results and flag changes for instructions other than MADD or MSUB in
which calculation exceptions are generated.

© SCEI
-41-

SCE CONFIDENTIAL

Exception
No exception
Overflow
Underflow

VU User's Manual Version 6.0

Calculation
Results
X
MAX
0

MAC Flag
Status Flag
Z* S* U* O* Z S U
X X 0
0 X X 0
0 X 0
1
0 X 0
1 X 1
0
1 X 1

O
0
1
0

Sticky Flag
ZS SS US OS
X X - X 1
1 X 1
-

X: 1 or 0 according to the calculation results -: Same as past values
MAC Flag is set to 0 when the corresponding FMAC unit is not used.

Flag change in MADD instruction
In the MADD instruction, addition/subtraction with the accumulator is performed following the
multiplication. There is a double exception generation factor in one instruction, and flags are set in
compliance with both the multiplication and addition/subtraction. Therefore, flag settings in this instruction
are complicated.
Exception Factor
Multiplication
ACC
0/Normalized
No exception
value
0/Normalized
value
0/Normalized
value

Calculation
Results

X

OVF

+/-MAX

UDF

-

+/-MAX
No exception
+MAX
OVF(+MAX)
+MAX
OVF(-MAX)
-MAX
OVF(+MAX)
-MAX
OVF(-MAX)
+/-MAX
UDF
Addition/Subtraction
OFV/UDF

+/-MAX
+MAX
-MAX
+MAX
-MAX
+/-MAX
+/-MAX/0

X: 1 or 0 according to the calculation results of addition

MAC Flag
Status Flag
Z* S* U* O* Z S U
X X 0 0 X X 0
0

Sticky Flag
O ZS SS US OS
0 X X -

X

0

1

0

X

0

1

-

X

-

1

X X

0

0

X

X

0

0

X

X

1

-

0 X 0 1
0 0 0 1
0 1 0 1
0 0 0 1
0 1 0 1
0 X 0 1
X X X X

0
0
0
0
0
0
X

X
0
1
0
1
X
X

0
0
0
0
0
0
X

1
1
1
1
1
1
X

b
1
c

c
b
1
b
1
c
c

1
a

1
1
1
1
1
1
a

-: Same as past values

a: Logical OR of the flag value that shows calculation results of addition and the past flag values
b: Logical OR of the flag value that shows calculation results of multiplication and the past flag values
c: Logical OR of the flag values that show calculation results of addition and multiplication and the past flag values

MAC Flag is set to 0 when the corresponding FMAC unit is not used.

© SCEI
-42-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Flag changes in MSUB instruction
The flag changes in the MSUB instruction are almost the same as those in the MADD instruction, but
positive and negative signs are changed when an overflow occurs to multiplication.
Exception Factor
Multiplication
ACC
0/Normalized
No exception
value
0/Normalized
value
0/Normalized
value

Calculation
Results

X

MAC Flag
Status Flag
Z* S* U* O* Z S U
X X 0 0 X X 0

O
0

Sticky Flag
ZS SS US OS
X X - -

OVF

+/-MAX

0

X

0

1

0

X

0

1

-

X

- 1

UDF

-

X

X

0

0

X

X

0

0

X

X

1 -

+/-MAX
-MAX
+MAX
-MAX
+MAX
+/-MAX
+/-MAX/0

0
0
0
0
0
0
X

X
1
0
1
0
X
X

0
0
0
0
0
0
X

1
1
1
1
1
1
X

0
0
0
0
0
0
X

X
1
0
1
0
X
X

0
0
0
0
0
0
X

1
1
1
1
1
1
X

b
1
c

c
1
b
1
b
c
c

1
a

+/-MAX
No exception
+MAX
OVF(+MAX)
+MAX
OVF(-MAX)
-MAX
OVF(+MAX)
-MAX
OVF(-MAX)
+/-MAX
UDF
Addition/Subtraction
OFV/UDF

X: 1 or 0 according to the calculation results of addition

1
1
1
1
1
1
a

-: Same as past values

a: Logical add of the flag value that shows calculation results of addition and the past flag values
b: Logical add of the flag value that shows calculation results of multiplication and the past flag values
c: Logical add of the flag values that show calculation results of addition and multiplication and the past flag values
MAC Flag is set to 0 when the corresponding FMAC unit is not used.

© SCEI
-43-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.4. Pipeline Operation
3.4.1. Hazards
VU calculations are performed concurrently via the pipeline, but some operations may stall occasionally under
the conditions specified below. For concrete operations, see sections "3.4.4. FMAC Pipeline" and following.
DIV Resource Hazards
An instruction (DIV/SQRT/RSQRT) that uses the floating-point divider unit when another instruction of
this type is being executed.
EFU Resource Hazards
An instruction (such as ESIN) that uses the EFU when another instruction of this type is being executed.
Floating-Point Register Data Hazards
An instruction that uses a floating-point register when another instruction that uses the register as the
destination is being executed (until the value is fixed).
Data hazard checks are performed independently in each field of x/y/z/w. VF00 is a constant register, and
is not subject to hazard checks.
Integer Register Data Hazards
An instruction that uses the values of an integer register when a load/store instruction to the integer register
is being executed.
Since integer calculation latency is 1 clock, data hazards due to calculations are not generated. VI00 is not
subject to hazard checks.
Data hazards are not generated for the special registers such as ACC, I, Q, P, and R. It is possible to make the
Q and P registers synchronize with each other by using the WAITQ/WAITP instruction.
When transferring data to an external unit (GS) by the XGKICK instruction, stalls continue until the data can be
transferred.

3.4.2. Upper Instruction and Lower Instruction
In micro mode, there is an Upper instruction pipeline and a Lower instruction pipeline. The Upper and Lower
instructions are issued concurrently, so both pipelines stall if hazards occur to either of them.

3.4.3. Priority for Writing to a Register
When the Upper and Lower instructions write data to the same register at the same time, priority is given to the
Upper instruction and the result of the Lower instruction is discarded. Moreover, when a coprocessor transfer
instruction (COP2) writes data at the same time, it is given priority.
COP2 Transfer Instruction> Upper Instruction> Lower Instruction
The above is performed in register units, so note that the result of the Lower instruction is discarded even when
the Upper and Lower instructions write data to different fields as shown in the following example.
© SCEI
-44-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADD.xy VF01, VF01, VF23

MOVE.w VF01, VF09

3.4.4. FMAC Pipeline
Figure 3-2 illustrates the FMAC pipeline. All instructions except some of the Lower instructions (DIV, SQRT,
RSQRT, integer calculation, and conditional branching) are executed according to this pipeline. Load/Store of
integer registers also follows this pipeline.
The MAC flag, status flag and clipping flag, which show the results of the calculation, are set at the S stage.

access MicroMem
write back

read VPU register

M

T

X

Y

Z

S
execute3

execute2
execute1
Figure 3-2 FMAC Pipeline
An example of the FMAC pipeline is illustrated below.
clock cycle

7

8

9

10

1

2

3

4

5

6

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

11

Instruction

1

MUL.xyzw VF05, VF10, VF20
NOP (MOVE VF00, VF00)

2

MULA.xyzw ACC, VF11, VF21
NOP (MOVE VF00, VF00)

3

MADDA.xyzw ACC, VF12, VF22
NOP (MOVE VF00, VF00)

4

MADD.xyzw VF06, VF13, VF23
NOP (MOVE VF00, VF00)

5

MUL.xyzw VF07, VF05, VF24
NOP (MOVE VF00, VF00)11

6

MUL.xyzw VF08, VF15, VF25
NOP (MOVE VF00, VF00)

S
S

Figure 3-3 FMAC Pipeline Operation Example
Figure 3-3 illustrates a normal operation example of the FMAC pipeline. Calculation results of Instruction 1 are
used for Instruction 5, but stalls do not occur because other instructions are being executed between them.

© SCEI
-45-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

clock cycle

Instruction

1

MUL.xyzw VF10, VF10, VF10
NOP (MOVE VF00, VF00)

2

ADDy.x VF10x, VF10x, VF10y
NOP (MOVE VF00, VF00)

3
4
5
6

stall

ADDw.z VF10z, VF10z, VF10w
NOP (MOVE VF00, VF00)

1

2

3

4

5

6

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

Ts
Ts

Ts
Ts

Ts
Ts

Ms
Ms

Ms
Ms

Ms
Ms

non-stall

NOP
MOVE.z VF20z, VF10z

7

8

9

10

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

Ts
Ts

Ts
Ts

Ts
Ts

T
T

X
X

Ms
Ms

Ms
Ms

Ms
Ms

M
M

T
T

stall

NOP
NOP (MOVE VF00, VF00)

11

M
M

NOP
NOP (MOVE VF00, VF00)

Ts: T stage stall

Ms: M stage stall

Figure 3-4 Stalls due to FMAC Pipeline Data Hazards
Figure 3-4 shows an example of FMAC pipeline stall. VF10x of Instruction 1 output is used for Instruction 2.
Therefore, execution of Instruction 2 is delayed until Instruction 1 reaches the S stage.
Between Instruction 3 and Instruction 4, data hazards occur to VF10z.
Stalls do not occur between Instruction 2 and Instruction 3. The output VF10x of Instruction 2 and the input
VF10z and VF10w of Instruction 3 are in the same register, but data hazards are not generated since they are in
different fields. (Data hazards are generated in the macroinstruction.)
Stalls due to data dependency do not occur between the Upper instruction and the Lower instruction in the
same instruction, though this is not shown in this example. For the relationship with COP2 instructions, see
"5.4.4. Operation when Transferring Data with EE Core".

3.4.5. FDIV Pipeline
The following figures illustrate the pipelines related to the floating-point division unit.
write back

access MicroM em
read VU register
M

T

D1

D2

D3

D4

D5

D6

F

self time DIV/SQRT execute

Figure 3-5 FDIV Pipeline (DIV / SQRT)
access M icroM em

write back

read VPU register
M

T

D1

D2

D3

D4

D5

D6

D7

D8

D9

DA

DB

DC

F

self tim e RSQRT execute

Figure 3-6 FDIV Pipeline (RSQRT)
The next DIV/SQRT/RSQRT instruction stalls with the generation of resource hazards during execution of
D1- D6 stage of DIV/SQRT instruction and D1- DC stage of RSQRT instruction.

© SCEI
-46-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

clock cycle

Instruction

1

NOP
DIV Q, VF10, VF20

2

NOP
NOP (MOVE VF00, VF00)

3

NOP
NOP (MOVE VF00, VF00)

4

NOP
NOP (MOVE VF00, VF00)

5

NOP
NOP (MOVE VF00, VF00)

6

NOP
NOP (MOVE VF00, VF00)

7

NOP
NOP (MOVE VF00, VF00)

8

MULq.xyzw VF13, VF23, Q
NOP (MOVE VF00, VF00)

1

2

3

4

5

6

7

8

9

10

11

M
M

T
T

X
D1

Y
D2

Z
D3

S
D4

D5

D6

F

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

Figure 3-7 FDIV Pipeline Operation Example
Figure 3-7 illustrates a normal operation example of the FDIV pipeline. The DIV calculation results can be
received and calculation can be performed by inserting six or more other instructions between the DIV
instruction (Instruction 1) and the MULq instruction (Instruction 8) which uses the Q register, the results of the
DIV instruction, after the DIV instruction (Instruction 1).
clock cycle
Instruction

1
2

NOP
DIV Q, VF10, VF20
MULq.xyzw VF12, VF22, Q
NOP (MOVE VF00, VF00)

3

MULq.xyzw VF13, VF23, Q
DIV Q, VF11, VF21

4

NOP
NOP (MOVE VF00, VF00)

5

NOP
NOP (MOVE VF00, VF00)

6

NOP
NOP (MOVE VF00, VF00)

1

2

3

4

5

6

7

8

9

M
M

T
T

X
D1

Y
D2

Z
D3

S
D4

D5

D6

F

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

Ts
Ts

Ts
Ts

Ts
Ts

Ts
Ts

Ts
Ts

Ms Ms
Ms Ms

Ms Ms Ms
Ms Ms Ms

10

11

T
T

X
D1

Y
D2

Z
D3

M
M

T
T

X
X

Y
Y

M
M

T
T

X
X

M
M

T
T

Figure 3-8 FDIV Pipeline Continuous Execution Example
Figure 3-8 shows an example of FDIV pipeline continuous execution. The next DIV instruction (Instruction 3)
is started during the DIV instruction (Instruction 1) execution, and Instruction 3 stalls until the D6 stage of
Instruction 1 ends.
Although Instruction 2 is a MULq instruction, data dependency is not checked regarding the Q register, so the
Q register value previously obtained is used here, not the calculation results of Instruction 1.
The MULq of Instruction 3 uses the calculation results of Instruction 1, which was written to the Q register at
the F stage, due to stalls.

© SCEI
-47-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

clock cycle

Instruction

1
2

NOP
DIV Q, VF10, VF20

1

2

3

4

5

6

7

8

9

M
M

T
T

X
D1

Y
D2

Z
D3

S
D4

D5

D6

F

M
M

Ts
Ts

Ts
Ts

Ts
Ts

Ts
Ts

Ts
Ts

Ts
Ts

Ms
Ms

Ms
Ms

Ms
Ms

Ms
Ms

Ms
Ms

Ms
Ms

MULq.xyzw VF13, VF23, Q
WAITQ

3

NOP
NOP (MOVE VF00, VF00)

4

NOP
NOP (MOVE VF00, VF00)

5

NOP
NOP (MOVE VF00, VF00)

6

NOP
NOP (MOVE VF00, VF00)

10

11

T
T

X
X

Y
Y

Z
Z

M
M

T
T

X
X

Y
Y

M
M

T
T

X
X

M
M

T
T
M
M

Figure 3-9 WAITQ Instruction Operation Example
Figure 3-9 illustrates an example of synchronization using the WAITQ instruction. Due to the WAITQ
instruction of Instruction 2, subsequent instructions stall until the output of the DIV instruction (Instruction 1)
is fixed. The results of Instruction 1 can be used from the Upper instruction of Instruction 2.

3.4.6. EFU Pipeline
Figure 3-10 illustrates the EFU pipeline. During execution of N1 - Nn-1 stages, the next elementary function
calculation instruction stalls with the generation of resource hazards. Unlike the FDIV pipeline, the resource
hazards are not generated during execution of the Nn stage, the final stage to be executed.
access MicroMem
write back
read VPU register

M

T

N1

..................

N2

Nn-1

Nn

P

EFU execute

Figure 3-10 EFU Pipeline

1
2

1

2

3

4

5

......

13

14

15

16

17

M
M

T
T

X
N

Y
N

Z
N

......

N

P

M
M

Ts
Ts

Ts
Ts

Ts
Ts

......

T
T

X
N

Y
N

Z
N

S
N

Ms
Ms

Ms
Ms

Ms
Ms

......

M
M

T
T

X
X

Y
Y

Z
Z

M
M

T
T

X
X

Y
Y

M
M

T
T

X
X

M
M

T
T

clock cycle

Instruction
NOP
ESADD P, VF10
NOP
ESADD P, VF10

3

NOP
MFP

4

NOP
NOP (MOVE VF00, VF00)

5

NOP
NOP (MOVE VF00, VF00)

6

NOP
NOP (MOVE VF00, VF00)

Figure 3-11 EFU Pipeline Continuous Execution Example
© SCEI
-48-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Figure 3-11 illustrates a pipeline operation example in which the EFU instruction is continuously executed.
Since the ESADD instruction (Latency 11, Throughput 10) is executed by Instruction 1 and a new ESADD
instruction (Instruction 2) is executed before the end of the current ESADD instruction, stalls are generated at
the T stage of Instruction 2. Unlike the FDIV pipeline, the stall is cleared at the Nn stage (one stage before the
P stage).
clock cycle
Instruction

1
2

NOP
ESADD P, VF10
NOP
MFP

3

NOP
WAITP

4

NOP
MFP

5

NOP
NOP (MOVE VF00, VF00)

6

NOP
NOP (MOVE VF00, VF00)

1

2

3

4

5

6

......

13

14

15

M
M

T
T

X
N

Y
N

Z
N

S
N

......

N

P

M
M

T
T

X
X

Y
Y

Z
Z

M
M

Ts
Ts

Ts
Ts

Ts
Ts

......

T
T

X
X

Y
Y

Ms
Ms

Ms
Ms

Ms
Ms

......

M
M

T
T

X
X

M
M

T
T
M
M

Figure 3-12 WAITP Instruction Operation Example
Figure 3-12 illustrates a synchronization example using the WAITP instruction. Due to the WAITP instruction
of Instruction 3, a stall occurs and continues until the end of the ESADD instruction of Instruction 1 and is
cleared at the Nn stage, then the succeeding instruction is executed.
Similar to the Q register, the data dependency is not checked in the P register. In the MFP instruction of
Instruction 2, the P register values gained prior to Instruction 1 are used. In the MFP instruction of Instruction
4, the P register values gained from Instruction 1 are used as a result of the stall.

3.4.7. IALU Pipeline
Figure 3-13 illustrates the IALU pipeline, which performs Integer calculation.
access MicroMem
write back

read VPU register

M

T

X

y

z

S
dummy2

dummy1
execute

Figure 3-13 Actual IALU Pipeline
IALU execution ends in one cycle, but the y stage and z stage exist as dummy stages in order to adjust the timing
with the FMAC pipeline. Although the results are actually stored in the integer register at the S stage, there is no
latency in effect for the dummy stages since the results are bypassed from the X/y/z/S stage to the T stage.
The latency for the dummy stages appears only when the microinstruction calculation results are transferred
after the Integer register has been written at the S stage by the CFC2 instruction (a coprocessor transfer
macroinstruction).

© SCEI
-49-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Instruction

clock cycle

1

LOOP:
MULA.xyzw ACC, VF20, VF10
IADD VI04, VI03, VI02

2

MADDA.xyzw ACC, VF21, VF11
IADDI VI05, VI05, -1

3

MADDA.xyzw ACC, VF22, VF12
IADD VI06, VI05, VI04

4

MADD.xyzw VF14, VF23, VF13
IADD VI07, VI06, VI04

5

MULA.xyzw ACC, VF20, VF10
LQI.xyzw VF10, (VI04++)

1

2

3

4

5

6

M
M

T
T

X
X

Y
y

Z
z

S
S

M
M

T
T

X
X

Y
y

Z
z

S
S

T
T

X
X

Y
y

Z
z

S
S

M
M

T
T

X
X

Y
y

Z
z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

M
Calculation results bypass
M

Calculation results bypass

7

8

9

10

11

S
S

Figure 3-14 IALU Pipeline Bypass Example
Figure 3-14 illustrates an operation example of the IALU pipeline. The IADD instruction of Instruction 3 uses
the results of Instruction 1 and Instruction 2, so bypasses are generated between the y stage of Instruction 1 and
the T stage of Instruction 3, and between the X stage of Instruction 2 and the T stage of Instruction 3.
Similarly, bypasses are generated between the z stage of Instruction 1 and the T stage of Instruction 4, and
between the X stage of Instruction 3 and the T stage of Instruction 4.

3.4.8. Conditional Branching and Pipeline
Figure 3-15 illustrates an example of pipeline operation, which accompanies conditional branching.
Instruction

1

Loop:
MULA.xyzw ACC, VF23, VF13
IADDI VI03, VI03, -1

2

MADDA.xyzw ACC, VF20, VF10
LQI.xyzw VF10, (VI04++)

3

MADDA.xyzw ACC, VF21, VF11
IBNE VI03, VI00, Loop:

4

Branch delay slot
MADD.xyzw VF14, VF22, VF12
SQI.xyzw VF14, (VI05++)

5

Loop:
MULA.xyzw ACC, VF23, VF13
IADDI VI03, VI03, -1

6

MADDA.xyzw ACC, VF20, VF10
LQI.xyzw VF10, (VI04++)

7

MADDA.xyzw ACC, VF21, VF11
IBNE VI03, VI00, Loop:

8

Branch delay slot
MADD.xyzw VF14, VF22, VF12
SQI.xyzw VF14, (VI05++)

clock cycle

1

2

3

4

5

6

M
M

T
T

X
X

Y
y

Z
z

S
S

7

8

9

10

11

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
y

Z
z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

Figure 3-15 Operation Example of Integer Calculation Branching Instruction
In Figure 3-15, a loop has been created between Instruction 1 and Instruction 4. A conditional branch is
performed in Instruction 3 according to the results of the Lower instruction of Instruction 1, and it branches to
the beginning of the loop (Instruction 5) after the one-instruction branch delay slot (Instruction 4).
As mentioned above, a one-instruction slot is necessary between a branch condition setting instruction and a
condition branching instruction. Flag check instructions (FCAND, FCEQ, FCGET, FCOR, FMAND, FMEQ,
FMOR, FSAND, FSEQ, and FSOR) are an exception, and a conditional branch instruction can be placed
immediately after them.

© SCEI
-50-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

clock cycle
Instruction

1
2
3
4

MUL.xyzw VF05, VF10, VF20
NOP (MOVE VF00, VF00)

1

2

3

4

5

6

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
y

Z
z

S
S

M
M

T
T

X
X

Y
y

Z
z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

ADD.xyzw VF06, VF11, VF21
NOP (MOVE VF00, VF00)
ADD.xyzw VF07, VF12, VF22
NOP (MOVE VF00, VF00)
ADD.xyzw VF08, VF13, VF23
NOP (MOVE VF00, VF00)

5

ADD.xyzw VF09, VF14, VF24
FSAND VI01, SF, 0xfff:

6

ADD.xyzw VF10, VF15, VF25
IBNE VI01, VI00, Label:

7

Branch delay slot
NOP
NOP (MOVE VF00, VF00)

8

MUL.xyzw VF05, VF20, VF10
LQI.xyzw VF10, (VI04++)

7

8

9

10

11

Figure 3-16 Example of Floating-Point Calculation Branch Instruction (1)
Figure 3-16 shows an example of branching according to floating-point calculation results. A conditional branch
is performed according to the results of Instruction 1, and the calculation results are written to the status flag at
the S stage. The logical AND of the status flag and the immediate value is written to VI01 in Instruction 5, after
inserting Instructions 2 to 4. Then, the conditional branch is performed according to the value of VI01 in
Instruction 6.
Instruction

clock cycle

4

5

6

Y
Y

Z
Z

S
S

7

8

9

10

11

12

13

14

1

MUL.xyzw VF05, VF10, VF20
NOP (MOVE VF00, VF00)

2

ADD.xyzw VF06, VF11, VF05
NOP (MOVE VF00, VF00)

Ts
Ts

Ts
Ts

T
T

X
X

Y
Y

Z
Z

S
S

3

ADD.xyzw VF07, VF12, VF06
NOP (MOVE VF00, VF00)

Ms
Ms

Ms
Ms

M
M

Ts
Ts

Ts
Ts

Ts
Ts

T
T

X
X

Y
Y

Z
Z

S
S

4

ADD.xyzw VF08, VF13, VF23
NOP (MOVE VF00, VF00)

Ms
Ms

Ms
Ms

Ms
Ms

M
M

T
T

X
X

Y
Y

Z
Z

5

ADD.xyzw VF09, VF14, VF24
FSAND VI01, SF, 0xfff

M
M

T
T

X
X

Y
y

6

ADD.xyzw VF10, VF15, VF25
IBNE VI01, VI00, Label:

M
M

T
T

X
X

M
M

T
T

7

Branch delay slot
NOP
NOP (MOVE VF00, VF00)

8

Label:
MUL.xyzw VF05, VF20, VF10
LQI.xyzw VF10, (VI04++)

M
M

Figure 3-17 Example of Floating-Point Calculation Branch Instruction (2)
Figure 3-17 shows an example in which stalls are generated due to data hazards. As a result, the status flag
referred to by Instruction 5 shows the calculation results of Instruction 2.
As mentioned above, when reading the status flag, it is necessary to pay attention to the timing.

© SCEI
-51-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.4.9. XGKICK Pipeline
The stage structure of the XGKICK pipeline that activates GIF transfer via PATH1 is basically the same as that
of the FMAC pipeline, but the XGKICK instruction performs a special pipeline operation that makes the
subsequent instructions stall on the T stage when continuously executed.
Instruction

1
2
3

clock cycle

MUL.xyzw VF03, VF10, VF20
XGKICK
MUL.xyzw VF04, VF11, VF21
XGKICK
MUL.xyzw VF05, VF012, VF22
NOP (MOVE VF00, VF00)

4

MUL.xyzw VF06, VF13, VF23
NOP (MOVE VF00, VF00)

5

MUL.xyzw VF07, VF14, VF24
NOP (MOVE VF00, VF00)

6

MUL.xyzw VF08, VF15, VF25
NOP (MOVE VF00, VF00)

1

2

3

4

5

6

7

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

Ts
Ts

Ts
Ts

Ts
Ts

Ts
Ts

Ms
Ms

Ms
Ms

Ms
Ms

Ms
Ms

m

n

Ts
Ts

T
T

X
X

Y
Y

Z
Z

Ms
Ms

M
M

T
T

X
X

Y
Y

Figure 3-18 XGKICK Pipeline Operation
Figure 3-18 shows an XGKICK pipeline operation example. The XGKICK in Instruction 1 executes without
stalling. However, at the XGKICK in Instruction 2, since the transfer via PATH1 activated in Instruction 1 is
in process, the pipeline stalls until the PATH1 transfer caused by the preceding XGKICK instruction ends. At
this time, not Instruction 2, but the following Instruction 3 is delayed. Note that the Upper instruction of
Instruction 2 is executed without stalling.

© SCEI
-52-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.5. Micro Subroutine Execution
3.5.1. How to Execute a Micro Subroutine
There are three ways to execute a micro subroutine:
Macroinstruction VCALLMS / VCALLMSR instruction
Write the execution address to the control register
MSCAL/MSCALF of VIFcode

Executable in VU0
Executable in VU1
Executable in VU0/VU1

Operation is indeterminate if VCALLMS/VCALLMSR instruction from the EE Core and start-up from the VIF
are specified concurrently.

3.5.2. How to Terminate a Micro Subroutine
There are three ways to terminate a micro subroutine:
By a microinstruction that sets the E bit to 1.
By a microinstruction that sets the T bit or D bit to 1.
By a Force Break from an external source.
A micro subroutine is normally terminated by a microinstruction that sets the E bit to 1. Other termination
methods are only for debugging purposes. See "7.3. Micro Subroutine Debugging".

3.5.3. Operation of Execution and Termination
Figure 3-19 shows an example of executing a micro subroutine.
clock cycle

1

2

3

4

5

6

7

8

9

10

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

11

12

13

14

15

16

17

[Instruction address unit : bytes]

Instruction

1

address : 0x0100
MUL
NOP

2

address : 0x0108
MULA
NOP

3

address : 0x0110
MADDA
NOP

4

address : 0x0118
MADD
NOP

5

address : 0x0120
MUL
NOP

6

address : 0x0128
MUL
NOP

7

address : 0x0130
MUL
NOP

E bit

Address specified by callms

S
S

E bit delay Slot
(1 instruction)

Stop position according to E bit

Address stored in TPC

Figure 3-19 Execution by callms and Termination by the E bit

© SCEI
-53-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

In Figure 3-19, micro subroutine execution starts from Instruction 1. The E bit, which indicates the end of the
micro subroutine, is set in Instruction 5. There is a one-instruction E-bit delay slot, in which Instruction 6 is
executed; then the micro subroutine stops execution and returns to macro mode. The address of Instruction 7
is stored in the termination position program counter (TPC).
The following kinds of instructions cannot be placed in the E bit delay slot:
• Branch instructions
• Instructions that synchronize to external units, such as XTOP and XITOP
• XGKICK
• VU Mem load/store instruction
• Microinstructions that set the E bit to 1
If a micro subroutine is terminated during the execution of the FDIV or EFU instruction, the FDIV or EFU
process is continued, and the results are stored in the P or Q register at a given latency.

© SCEI
-54-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

3.6. Other Functions
3.6.1. Data Transfer with VU Mem/Micro Mem
VU Mem and Micro Mem are I/O-mapped to the main memory of the EE Core. When the VPU is not
operating, these memory locations are accessible directly from the EE Core.
The address map is shown in the following table.
Memory
MicroMem0
VUMem0
MicroMem1
VUMem1

Address
0x1100_0000 - 0x1100_0ff0
0x1100_4000 - 0x1100_4ff0
0x1100_8000 - 0x1100_bff0
0x1100_c000 - 0x1100_fff0

3.6.2. Debug Support Function
Execution of a micro subroutine can be suspended for debugging by setting the D bit in the operation code field
to 1. For further information, see "7.3. Micro Subroutine Debugging".

© SCEI
-55-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

(This page is left blank intentionally)

© SCEI
-56-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

4. Micro Mode Instruction Reference

© SCEI
-57-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

4.1. Micro Mode Instruction Set
4.1.1. Types of Upper Instruction
There are four types of Upper instructions that primarily execute floating-point calculations:
UpperOP field type 0
Specifies three registers (VF[fs],VF[ft],VF[fd]) and a broadcast field (e.g. ADDbc instruction), and performs scalar
calculations as follows:

Example: ADDx.xyzw VF10xyzw, VF20xyzw, VF30x
Operation:

VF10x = VF20x + VF30x
VF10y = VF20y + VF30x
VF10z = VF20z + VF30x
VF10w = VF20w + VF30x

Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

UpperOP field type 1
Specifies three registers (VF[fs],VF[ft],VF[fd] ), e.g. ADD instruction, and performs vector calculations as
follows:

Example: ADD.xyzw VF10xyzw, VF20xyzw, VF30xyzw
Operation:

VF10x = VF20x + VF30x
VF10y = VF20y + VF30y
VF10z = VF20z + VF30z
VF10w = VF20w + VF30w

Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

UpperOP field type 2
Specifies two registers (VF[fs],VF[ft]) and a broadcast field, e.g. ADDAbc instruction, and performs scalar
calculations as follows:

Example: ADDAx.xyzw ACCxyzw, VF20xyzw, VF30x
Operation:

ACCx = VF20x + VF30x
ACCy = VF20y + VF30x
ACCz = VF20z + VF30x
ACCw = VF20w + VF30x

© SCEI
-58-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Upper 32-bit word: UpperOP field type 2
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

OPCODE
----1111
9 bits

bc
-2 bits

UpperOP field type 3
Specifies two registers (VF[fs],VF[ft]), e.g. ADDA instruction. Performs vector calculations as follows:

Example: ADDA.xyzw ACCxyzw, VF20xyzw, VF30xyzw
Operation:

ACCx = VF20x + VF30x
ACCy = VF20y + VF30y
ACCz = VF20z + VF30z
ACCw = VF20w + VF30w

Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

-----

OPCODE
1111

--

11 bits

4.1.2. Types of Lower Instructions
There are 7 types of Lower instructions:
LowerOP field type 1
Specifies 3 registers, e.g. IADD instruction.
Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

LowerOP field type 3
Specifies up to two registers and a dest field, e.g. MOVE instruction.

Example: MOVE.xyzw VF10xyzw, VF20xyzw
Operation:

VF10x = VF20x
VF10y = VF20y
VF10z = VF20z
VF10w = VF20w

Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

7 bits

4 bits

5 bits

5 bits

-----

OPCODE
1111

--

11 bits

© SCEI
-59-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

LowerOP field type 4
Specifies 2 floating-point registers with a specific field for each, e.g. DIV instruction.

Example: DIV Q, VF10x, VF20y
Operation:

Q = VF10x ÷ VF20y

Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
--

fsf
--

2 bits 2 bits

ft reg
-----

fs reg
-----

5 bits

5 bits

-----

OPCODE
1111

--

11 bits

LowerOP field type 5
Specifies 2 registers and a 5-bit immediate value, e.g. IADDI instruction.
Lower 32-bit word: LowerOP field type 5
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
-----

Imm5
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

LowerOP field type 7
Specifies 2 registers and an 11-bit immediate value, e.g. ILW instruction.
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
0------

dest
----

it reg
-----

fs reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

LowerOP field type 8
Specifies 2 registers and a 15-bit immediate value.
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
0------

Imm15
----

it reg
-----

fs reg
-----

Imm15
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

LowerOP field type 9
Specifies a 24-bit immediate value, e.g. FCAND instruction.
Lower 32-bit word: LowerOP field type 9
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
0------

0

Imm24
------------------------

7 bits

1

24 bits

© SCEI
-60-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

4.1.3. Operation Fields for Micro Instructions
Various operation fields in operation codes are explained in this section.
dest field (Upper/Lower)
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

The dest field specifies the FMAC units to be operated in parallel; that is, either the x, y, z or w field of the 128bit data to be operated on.
The dest field is 4 bits: bits 56 through 53 for Upper instructions and bits 24 through 21 for Lower instructions.
Each of the 4 bits can be specified independently; when the bit is set to 1, the corresponding FMAC unit /field
becomes effective.
Bit
Upper
56
55
54
53

Corresponding FMAC /Field
Lower
24
23
22
21

x
y
z
w

bc field (Upper)
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

The bc field is bits 33 and 32, and specifies the broadcast field as below.
Specified value of bc field
00
01
10
11

Broadcast field
x
y
z
w

© SCEI
-61-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

fsf/ftf field
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

ftf
--

fsf
--

ft reg
-----

fs reg
-----

7 bits

2 bits

2 bits

5 bits

5 bits

-----

OPCODE
1111

--

11 bits

The combinations of the fsf field with the fs reg field and the ftf field with the ft reg field specify the field to be
calculated by the instruction. Bits 22 and 21 of the Lower instruction are used for the fsf field, and bits 24 and
23 are used for the ftf field.
Specified value for fsf/ftf field
00
01
10
11

Field to be operated
x
y
z
w

I bit (Upper)
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

The I bit is specified when loading an immediate value into the I register. When bit 63 of the Upper instruction
field is set to 1, the contents of the Lower instruction field are loaded into the I register as a single-precision
floating-point immediate value.

© SCEI
-62-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

E bit (Upper)
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

The E bit is bit 62 of the Upper instruction field; it is used when designating termination of a micro subroutine.
When the E bit is set to 1, the VU terminates execution of the micro subroutine after the next instruction and
returns to macro mode.
M bit (Upper)
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

The M bit is bit 61 of the Upper instruction field; it specifies QMTC2 / CTC2 instruction interlock. The
QMTC2 / CTC2 instruction is executed without interlocking when the M bit is set to 1. Refer to "5.4. Macro
Mode Pipeline".
D bit (Upper)
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

The D bit is bit 60 of the Upper instruction; it specifies a debug break instruction. When the D bit is set to 1
and the instruction is executed, the VU is halted and an interrupt signal is sent to the host processor. The
interrupt can be enabled/disabled by the DE bit (D bit Enable) of the control register FBRST.

© SCEI
-63-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

T bit (Upper)
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

The T bit is bit 59 of the Upper instruction; it specifies debug halt instruction. In the same manner as the D bit,
when the T bit is set to 1 and the instruction is executed, the VU is halted, and an interrupt signal is sent to the
host processor. The interrupt can be enabled/disabled by the TE bit (T bit Enable) of the control register
FBRST.

© SCEI
-64-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

4.2. Upper Instruction Reference
This section describes the function, operation code, mnemonic, operation, flag changes, and throughput/latency
of Upper instructions. They are listed in alphabetical order in mnemonic form. The descriptions also include
examples, programming notes, and reference information.

© SCEI
-65-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ABS : Absolute Value
Calculates the absolute value of VF[fs] and stores the result in VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00111

ABS
1111

01

11 bits

Mnemonic

ABS.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then VF[ft]x = |VF[fs]x|
if (y ⊆ dest) then VF[ft]y = |VF[fs]y|
if (z ⊆ dest) then VF[ft]z = |VF[fs]z|
if (w ⊆ dest) then VF[ft]w = |VF[fs]w|
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/latency
1/4
Example

ABS.xyzw VF10xyzw, VF20xyzw
VF10x = |VF20x|
VF10y = |VF20y|
VF10z = |VF20z|
VF10w = |VF20w|

© SCEI
-66-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADD : Add
Calculates the sum of VF[fs] and VF[ft], and stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

ADD
101000

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

ADD.dest VF[fd]dest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x + VF[ft]x
if (y ⊆ dest) then VF[fd]y = VF[fs]y + VF[ft]y
if (z ⊆ dest) then VF[fd]z = VF[fs]z + VF[ft]z
if (w ⊆ dest) then VF[fd]w = VF[fs]w + VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

ADD.xyzw VF10xyzw, VF20xyzw, VF30xyzw
VF10x = VF20x + VF30x
VF10y = VF20y + VF30y
VF10z = VF20z + VF30z
VF10w = VF20w + VF30w

© SCEI
-67-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADDi : Add to I Register
Adds each field of VF[fs] and the I register, and stores the sum in the corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

ADDi
100010

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

ADDi.dest VF[fd]dest, VF[fs]dest, I
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x + I
if (y ⊆ dest) then VF[fd]y = VF[fs]y + I
if (z ⊆ dest) then VF[fd]z = VF[fs]z + I
if (w ⊆ dest) then VF[fd]w = VF[fs]w + I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X
Throughput/Latency
1/4
Example

ADDi.xyzw VF10xyzw, VF20xyzw, I
VF10x = VF20x + I
VF10y = VF20y + I
VF10z = VF20z + I
VF10w = VF20w + I

© SCEI
-68-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADDq : Add to Q Register
Adds each field of VF[fs] and the Q register, and stores the sum in the corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

ADDq
100000

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

ADDq.dest VF[fd]dest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x + Q
if (y ⊆ dest) then VF[fd]y = VF[fs]y + Q
if (z ⊆ dest) then VF[fd]z = VF[fs]z + Q
if (w ⊆ dest) then VF[fd]w = VF[fs]w + Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

ADDq.xyzw VF10xyzw, VF20xyzw, Q
VF10x = VF20x + Q
VF10y = VF20y + Q
VF10z = VF20z + Q
VF10w = VF20w + Q

© SCEI
-69-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADDbc : Broadcast Add
Calculates the sum of each field of VF[fs] and the specified field of VF[ft], and stores the sum in the
corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

ADD?
0000

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Mnemonic

ADDbc.dest VF[fd]dest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x + VF[ft]bc
if (y ⊆ dest) then VF[fd]y = VF[fs]y + VF[ft]bc
if (z ⊆ dest) then VF[fd]z = VF[fs]z + VF[ft]bc
if (w ⊆ dest) then VF[fd]w = VF[fs]w + VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X
Throughput/Latency
1/4
Example

ADDx.xyzw VF10xyzw, VF20xyzw, VF30x
VF10x = VF20x + VF30x
VF10y = VF20y + VF30x
VF10z = VF20z + VF30x
VF10w = VF20w + VF30x

© SCEI
-70-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADDA : Add; to Accumulator
Adds VF[fs] and VF[ft], and stores the sum in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01010

ADDA
1111

00

11 bits

Mnemonic

ADDA.dest ACCdest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then ACCx = VF[fs]x + VF[ft]x
if (y ⊆ dest) then ACCy = VF[fs]y + VF[ft]y
if (z ⊆ dest) then ACCz = VF[fs]z + VF[ft]z
if (w ⊆ dest) then ACCw = VF[fs]w + VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

ADDA.xyzw ACCxyzw, VF20xyzw, VF30xyzw
ACCx = VF20x + VF30x
ACCy = VF20y + VF30y
ACCz = VF20z + VF30z
ACCw = VF20w + VF30w

© SCEI
-71-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADDAi : Add I Register; to Accumulator
Adds each field of VF[fs] and the I register, and stores the sum in the corresponding field of ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01000

ADDAi
1111

10

11 bits

Mnemonic

ADDAi.dest ACCdest, VF[fs]dest, I
Operation
if (x ⊆ dest) then ACCx = VF[fs]x + I
if (y ⊆ dest) then ACCy = VF[fs]y + I
if (z ⊆ dest) then ACCz = VF[fs]z + I
if (w ⊆ dest) then ACCw = VF[fs]w + I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X
Throughput/Latency
1/4
Example

ADDAi.xyzw ACCxyzw, VF20xyzw, I
ACCx = VF20x + I
ACCy = VF20y + I
ACCz = VF20z + I
ACCw = VF20w + I

© SCEI
-72-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADDAq : Add Q Register; to Accumulator
Adds each field of VF[fs] and the Q register, and stores the sum in the corresponding field of ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01000

ADDAq
1111

00

11 bits

Mnemonic

ADDAq.dest ACCdest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then ACCx = VF[fs]x + Q
if (y ⊆ dest) then ACCy = VF[fs]y + Q
if (z ⊆ dest) then ACCz = VF[fs]z + Q
if (w ⊆ dest) then ACCw = VF[fs]w + Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

ADDAq.xyzw ACCxyzw, VF20xyzw, Q
ACCx = VF20x + Q
ACCy = VF20y + Q
ACCz = VF20z + Q
ACCw = VF20w + Q

© SCEI
-73-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ADDAbc : Broadcast Add; to Accumulator
Adds each field of VF[fs] and the specified field of VF[ft], and stores the sum in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 2
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

ADDA?
00000
1111

bc
--

1

4 bits

5 bits

5 bits

9 bits

2 bits

1

1

1

1

1

1

Mnemonic

ADDAbc.dest ACCdest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then ACCx = VF[fs]x + VF[ft]bc
if (y ⊆ dest) then ACCy = VF[fs]y + VF[ft]bc
if (z ⊆ dest) then ACCz = VF[fs]z + VF[ft]bc
if (w ⊆ dest) then ACCw = VF[fs]w + VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X
Throughput/Latency
1/4
Example

ADDAx.xyzw ACCxyzw, VF20xyzw, VF30x
ACCx = VF20x + VF30x
ACCy = VF20y + VF30x
ACCz = VF20z + VF30x
ACCw = VF20w + VF30x

© SCEI
-74-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

CLIP : Clipping Judgment
Performs clipping judgment by the x,y,z field of VF[fs] and the w field of VF[ft] and sets the clipping flag (CF)
to the result.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
1110

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00111

CLIPw
1111

11

11 bits

Mnemonic

CLIPw.xyz VF[fs]xyz, VF[ft]w
Operation
CF = CF << 6
if (VF[fs]x > +|VF[ft]w|) then {set +x flag}
if (VF[fs]x < -|VF[ft]w|) then {set -x flag}
if (VF[fs]y > +|VF[ft]w|) then {set +y flag}
if (VF[fs]y < -|VF[ft]w|) then {set -y flag}
if (VF[fs]z > +|VF[ft]w|) then {set +z flag}
if (VF[fs]z < -|VF[ft]w|) then {set -z flag}
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - -

Z
-

clipping
flag
X

Throughput/Latency
1/4
Example

CLIPw.xyz VF10xyz, VF10w
Under the following condition:
VF10x > +|VF10w|,
-|VF10w| < VF10y < +|VF10w|,
VF10z < -|VF10w|

© SCEI
-75-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Before Execution
23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

3rd previous
2nd previous
Previous
Current
judgment
judgment
judgment
judgment
- + - + - + - + - + - + - + - + - + - + - + - +
z z y y x x z z y y x x z z y y x x z z y y x x
0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0
6 bits

6 bits

6 bits

6 bits

↓
After Execution
23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

3rd previous
2nd previous
Previous
Current
judgment
judgment
judgment
judgment
- + - + - + - + - + - + - + - + - + - + - + - +
z z y y x x z z y y x x z z y y x x z z y y x x
0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0 1 0 0 0 0 1
6 bits

6 bits

6 bits

6 bits

Remarks
In order to branch according to the results of clipping judgment, the Lower instructions
FCAND/FCEQ/FCGET/FCOR are available.

© SCEI
-76-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FTOI0 : Convert to Fixed Point
Converts the value of VF[fs] into a fixed-point number whose fractional part is 0 bit, and stores the result in
VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00101

FTOI0
1111

00

11 bits

Mnemonic

FTOI0.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then
VF[ft]x = float_to_integer0(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = float_to_integer0(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = float_to_integer0(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = float_to_integer0(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - -

Z
-

clipping
flag
-

Throughput/Latency
1/4
Example

FTOI0.xyzw VF10xyzw, VF20xyzw
VF10x = float_to_integer0(VF20x)
VF10y = float_to_integer0(VF20y)
VF10z = float_to_integer0(VF20z)
VF10w = float_to_integer0(VF20w)
Remarks
A few examples are shown in the following table:
x
-0.45
0.45
0.55
123.45

float_to_integer0(x)
0
0
0
123

© SCEI
-77-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FTOI4 : Convert to Fixed Point
Converts the value of VF[fs] into a fixed-point number whose fractional part is 4 bits, and stores the result in
VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00101

FTOI4
1111

01

11 bits

Mnemonic

FTOI4.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then
VF[ft]x = float_to_integer4(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = float_to_integer4(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = float_to_integer4(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = float_to_integer4(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - Throughput/Latency
1/4
Example

FTOI4.xyzw VF10xyzw, VF20xyzw
VF10x = float_to_integer4(VF20x)
VF10y = float_to_integer4(VF20y)
VF10z = float_to_integer4(VF20z)
VF10w = float_to_integer4(VF20w)
Remarks
A few examples are shown in the following table:
x
-0.45
0.45
0.55
123.45

float_to_integer4(x)
-7
7
8
1975

© SCEI
-78-

Z
-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FTOI12 : Convert to Fixed Point
Converts the value of VF[fs] into a fixed-point number whose fractional part is 12 bits, and stores the result in
VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

FTOI12
00101
1111

10

11 bits

Mnemonic

FTOI12.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then
VF[ft]x = float_to_integer12(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = float_to_integer12(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = float_to_integer12(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = float_to_integer12(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - -

Z
-

clipping
flag
-

Throughput/Latency
1/4
Example

FTOI12.xyzw VF10xyzw, VF20xyzw
VF10x = float_to_integer12(VF20x)
VF10y = float_to_integer12(VF20y)
VF10z = float_to_integer12(VF20z)
VF10w = float_to_integer12(VF20w)
Remarks
A few examples are shown in the following table:
x
-0.45
0.45
0.55
123.45

float_to_integer12(x)
-1843
1843
2252
505651

© SCEI
-79-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FTOI15 : Convert to Fixed Point
Converts the value of VF[fs] into a fixed-point number whose fractional part is 15 bits, and stores the result in
VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

FTOI15
00101
1111

11

11 bits

Mnemonic

FTOI15.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then
VF[ft]x = float_to_integer15(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = float_to_integer15(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = float_to_integer15(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = float_to_integer15(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - Throughput/Latency
1/4
Example

FTOI15.xyzw VF10xyzw, VF20xyzw
VF10x = float_to_integer15(VF20x)
VF10y = float_to_integer15(VF20y)
VF10z = float_to_integer15(VF20z)
VF10w = float_to_integer15(VF20w)
Remarks
A few examples are shown in the following table:
x
-0.45
0.45
0.55
123.45

float_to_integer15(x)
-14745
14745
18022
4045209

© SCEI
-80-

Z
-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ITOF0 : Convert to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional part is 0 bit, and converts it into floating
point and stores the result in VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00100

ITOF0
1111

00

11 bits

Mnemonic

ITOF0.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then
VF[ft]x = integer_to_float0(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = integer_to_float0(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = integer_to_float0(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = integer_to_float0(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - -

Z
-

clipping
flag
-

Throughput/Latency
1/4
Example

ITOF0.xyzw VF10xyzw, VF20xyzw
VF10x = integer_to_float0(VF20x)
VF10y = integer_to_float0(VF20y)
VF10z = integer_to_float0(VF20z)
VF10w = integer_to_float0(VF20w)
Remarks
A few examples are shown in the following table:
x
-12
1
123
1843

integer_to_float0 (x)
-12.0
1.0
123.0
1843.0

© SCEI
-81-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ITOF4 : Convert to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional part is 4 bits, and converts it into
floating point and stores the result in VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00100

ITOF4
1111

01

11 bits

Mnemonic

ITOF4.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then
VF[ft]x = integer_to_float4(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = integer_to_float4(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = integer_to_float4(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = integer_to_float4(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - Throughput/Latency
1/4
Example

ITOF4.xyzw VF10xyzw, VF20xyzw
VF10x = integer_to_float4(VF20x)
VF10y = integer_to_float4(VF20y)
VF10z = integer_to_float4(VF20z)
VF10w = integer_to_float4(VF20w)
Remarks
A few examples are shown in the following table:
x
-12
1
123
1843

integer_to_float4(x)
-0.750000
0.062500
7.687500
115.187500

© SCEI
-82-

Z
-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ITOF12 : Convert to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional part is 12 bits, and converts it into
floating point and stores the result in VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

ITOF12
00100
1111

10

11 bits

Mnemonic

ITOF12.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then
VF[ft]x = integer_to_float12(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = integer_to_float12(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = integer_to_float12(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = integer_to_float12(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - -

Z
-

clipping
flag
-

Throughput/Latency
1/4
Example

ITOF12.xyzw VF10xyzw, VF20xyzw
VF10x = integer_to_float12(VF20x)
VF10y = integer_to_float12(VF20y)
VF10z = integer_to_float12(VF20z)
VF10w = integer_to_float12(VF20w)
Remarks
A few examples are shown in the following table:
x
-12
1
123
1843

integer_to_float12(x)
-0.002930
0.000244
0.030029
0.449951

© SCEI
-83-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ITOF15 : Convert to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional part is 15 bits, and converts it into
floating point and stores the result in VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

ITOF15
00100
1111

11

11 bits

Mnemonic

ITOF15.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆dest) then
VF[ft]x = integer_to_float15(VF[fs]x)
if (y ⊆ dest) then
VF[ft]y = integer_to_float15(VF[fs]y)
if (z ⊆ dest) then
VF[ft]z = integer_to_float15(VF[fs]z)
if (w ⊆ dest) then
VF[ft]w = integer_to_float15(VF[fs]w)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S
------------- - - - - - - - - - Throughput/Latency
1/4
Example

ITOF15.xyzw VF10xyzw, VF20xyzw
VF10x = integer_to_float15(VF20x)
VF10y = integer_to_float15(VF20y)
VF10z = integer_to_float15(VF20z)
VF10w = integer_to_float15(VF20w)
Remarks
A few examples are shown in the following table:
x
-12
1
123
1843

integer_to_float15(x)
-0.000366
0.000031
0.003754
0.056244

© SCEI
-84-

Z
-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADD : Product Sum
Adds the value of ACC to the product of VF[fs] and VF[ft], and stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MADD
101001

1

4 bits

5 bits

5 bits

5 bits

6 bits

1 1

1

1

1

1

Mnemonic

MADD.dest VF[fd]dest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then VF[fd]x = ACCx + VF[fs]x × VF[ft]x
if (y ⊆ dest) then VF[fd]y = ACCy + VF[fs]y × VF[ft]y
if (z ⊆ dest) then VF[fd]z = ACCz + VF[fs]z × VF[ft]z
if (w ⊆ dest) then VF[fd]w = ACCw + VF[fs]w × VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADD.xyzw VF10xyzw, VF20xyzw, VF30xyzw
VF10x = ACCx + VF20x × VF30x
VF10y = ACCy + VF20y × VF30y
VF10z = ACCz + VF20z × VF30z
VF10w = ACCw + VF20w × VF30w
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-85-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDi : Product Sum; with I Register
Multiplies each field of VF[fs] by the I register, then adds the product to the corresponding field of ACC, and
stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MADDi
100011

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1 1

1

1

1

Mnemonic

MADDi.dest VF[fd]dest, VF[fs]dest, I
Operation
if (x ⊆ dest) then VF[fd]x = ACCx + VF[fs]x × I
if (y ⊆ dest) then VF[fd]y = ACCy + VF[fs]y × I
if (z ⊆ dest) then VF[fd]z = ACCz + VF[fs]z × I
if (w ⊆ dest) then VF[fd]w = ACCw + VF[fs]w × I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADDi.xyzw VF10xyzw, VF20xyzw, I
VF10x = ACCx + VF20x × I
VF10y = ACCy + VF20y × I
VF10z = ACCz + VF20z × I
VF10w = ACCw + VF20w × I
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-86-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDq : Product Sum; by Q Register
Multiplies each field of VF[fs] by the Q register, then adds the product to the corresponding field of ACC, and
stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MADDq
100001

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MADDq.dest VF[fd]dest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then VF[fd]x = ACCx + VF[fs]x × Q
if (y ⊆ dest) then VF[fd]y = ACCy + VF[fs]y × Q
if (z ⊆ dest) then VF[fd]z = ACCz + VF[fs]z × Q
if (w ⊆ dest) then VF[fd]w = ACCw + VF[fs]w × Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADDq.xyzw VF10xyzw, VF20xyzw, Q
VF10x = ACCx + VF20x × Q
VF10y = ACCy + VF20y × Q
VF10z = ACCz + VF20z × Q
VF10w = ACCw + VF20w × Q
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-87-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDbc : Broadcast Product Sum
Multiplies each field of VF[fs] by the specified field of VF[ft], then adds the product to the corresponding field
of ACC and stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MADD?
0010

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Mnemonic

MADDbc.dest VF[fd]dest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then VF[fd]x = ACCx + VF[fs]x × VF[ft]bc
if (y ⊆ dest) then VF[fd]y = ACCy + VF[fs]y × VF[ft]bc
if (z ⊆ dest) then VF[fd]z = ACCz + VF[fs]z × VF[ft]bc
if (w ⊆ dest) then VF[fd]w = ACCw + VF[fs]w × VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADDx.xyzw VF10xyzw, VF20xyzw, VF30x
VF10x = ACCx + VF20x × VF30x
VF10y = ACCy + VF20y × VF30x
VF10z = ACCz + VF20z × VF30x
VF10w = ACCw + VF20w × VF30x
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-88-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDA : Product Sum; to Accumulator
Multiplies VF[fs] by VF[ft], then adds the product to ACC and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01010

MADDA
1111

01

11 bits

Mnemonic

MADDA.dest ACCdest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then ACCx = ACCx + VF[fs]x × VF[ft]x
if (y ⊆ dest) then ACCy = ACCy + VF[fs]y × VF[ft]y
if (z ⊆ dest) then ACCz = ACCz + VF[fs]z × VF[ft]z
if (w ⊆ dest) then ACCw = ACCw + VF[fs]w × VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADDA.xyzw ACCxyzw, VF20xyzw, VF30xyzw
ACCx = ACCx + VF20x × VF30x
ACCy = ACCy + VF20y × VF30y
ACCz = ACCz + VF20z × VF30z
ACCw = ACCw + VF20w × VF30w
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-89-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDAi : Product Sum; by I register, to Accumulator
Multiplies each field of VF[fs] by the I register, then adds the product to the corresponding field of ACC and
stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

MADDAi
01000
1111

11

11 bits

Mnemonic

MADDAi.dest ACCdest, VF[fs]dest, I
Operation
if (x ⊆ dest) then ACCx = ACCx + VF[fs]x × I
if (y ⊆ dest) then ACCy = ACCy + VF[fs]y × I
if (z ⊆ dest) then ACCz = ACCz + VF[fs]z × I
if (w ⊆ dest) then ACCw = ACCw + VF[fs]w × I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADDAi.xyzw ACCxyzw, VF20xyzw, I
ACCx = ACCx + VF20x × I
ACCy = ACCy + VF20y × I
ACCz = ACCz + VF20z × I
ACCw = ACCw + VF20w × I
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-90-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDAq : Product Sum; by Q Register, to Accumulator
Multiplies each field of VF[fs] by the Q register, then adds the product to the corresponding field of ACC and
stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

MADDAq
01000
1111

01

11 bits

Mnemonic

MADDAq.dest ACCdest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then ACCx = ACCx + VF[fs]x × Q
if (y ⊆ dest) then ACCy = ACCy + VF[fs]y × Q
if (z ⊆ dest) then ACCz = ACCz + VF[fs]z × Q
if (w ⊆ dest) then ACCw = ACCw + VF[fs]w × Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADDAq.xyzw ACCxyzw, VF20xyzw, Q
ACCx = ACCx + VF20x × Q
ACCy = ACCy + VF20y × Q
ACCz = ACCz + VF20z × Q
ACCw = ACCw + VF20w × Q
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-91-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDAbc : Broadcast Product Sum; to Accumulator
Multiplies each field of VF[fs] by the specified field of VF[ft], then adds the product to the corresponding field
of ACC and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 2
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

MADDA?
00010
1111

bc
--

1

4 bits

5 bits

5 bits

9 bits

2 bits

1

1

1

1

1

1

Mnemonic

MADDAbc.dest ACCdest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then ACCx = ACCx + VF[fs]x × VF[ft]bc
if (y ⊆ dest) then ACCy = ACCy + VF[fs]y × VF[ft]bc
if (z ⊆ dest) then ACCz = ACCz + VF[fs]z × VF[ft]bc
if (w ⊆ dest) then ACCw = ACCw + VF[fs]w × VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MADDAx.xyzw ACCxyzw, VF20xyzw, VF30x
ACCx = ACCx + VF20x × VF30x
ACCy = ACCy + VF20y × VF30x
ACCz = ACCz + VF20z × VF30x
ACCw = ACCw + VF20w × VF30x
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-92-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MAX : Maximum Value
Compares VF[fs] with VF[ft] and stores the greater value in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MAX
101011

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MAX.dest VF[fd]dest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then
if (VF[fs]x > VF[ft]x)
{VF[fd]x = VF[fs]x}
else
{VF[fd]x = VF[ft]x}
(The same operation is performed for the y and z fields.)
if (w ⊆ dest) then
if (VF[fs]w > VF[ft]w)
{VF[fd]w = VF[fs]w}
else
{VF[fd]w = VF[ft]w}
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

MAX.xyzw VF10xyzw, VF20xyzw, VF30xyzw
if (VF20x > VF30x) then {VF10x = VF20x} else {VF10x = VF30x}
if (VF20y > VF30y) then {VF10y = VF20y} else {VF10y = VF30y}
if (VF20z > VF30z) then {VF10z = VF20z} else {VF10z = VF30z}
if (VF20w > VF30w) then {VF10w = VF20w} else {VF10w = VF30w}

© SCEI
-93-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MAXi : Maximum Value
Compares each field of VF[fs] with the I register and stores the greater field in the corresponding field of
VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MAXi
011101

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MAXi.dest VF[fd]dest, VF[fs]dest, I
Operation
if (x ⊆ dest) then
if (VF[fs]x > I) then
VF[fd]x = VF[fs]x
else
VF[fd]x = I
(The same operation is performed for the y and z fields.)
if (w ⊆ dest) then
if (VF[fs]w > I) then
VF[fd]w = VF[fs]w
else
VF[fd]w = I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/4
Example

MAXi.xyzw VF10xyzw, VF20xyzw, I
if (VF20x > I) then {VF10x = VF20x} else {VF10x = I}
if (VF20y > I) then {VF10y = VF20y} else {VF10y = I}
if (VF20z > I) then {VF10z = VF20z} else {VF10z = I}
if (VF20w > I) then {VF10w = VF20w} else {VF10w = I}

© SCEI
-94-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MAXbc : Maximum Value
Compares value of each field of VF[fs] with the specified field of VF[ft] and stores the greater value in the
corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MAX?
0100

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Mnemonic

MAXbc.dest VF[fd]dest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then
if (VF[fs]x > VF[ft]bc)
{VF[fd]x = VF[fs]x}
else
{VF[fd]x = VF[ft]bc}
(The same operation is performed for the y and z fields.)
if (w ⊆ dest) then
if (VF[fs]w > VF[ft]bc)
{VF[fd]w = VF[fs]w}
else
{VF[fd]w = VF[ft]bc}
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

MAXw.xyzw VF01xyzw, VF01xyzw, VF00w
A value of less than 1.0 in each field of VF01 is replaced with 1.0.

© SCEI
-95-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MINI : Minimum Value
Compares VF[fs] with VF[ft] and stores the smaller value in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MINI
101111

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MINI.dest VF[fd]dest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then
if (VF[fs]x < VF[ft]x)
{VF[fd]x = VF[fs]x}
else
{VF[fd]x = VF[ft]x}
(The same operation is performed for the y and z fields.)
if (w ⊆ dest) then
if (VF[fs]w < VF[ft]w)
{VF[fd]w = VF[fs]w}
else
{VF[fd]w = VF[ft]w}
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/4
Example

MINI.xyzw VF10xyzw, VF20xyzw, VF30xyzw
if (VF20x < VF30x) then {VF10x = VF20x} else {VF10x = VF30x}
if (VF20y < VF30y) then {VF10y = VF20y} else {VF10y = VF30y}
if (VF20z < VF30z) then {VF10z = VF20z} else {VF10z = VF30z}
if (VF20w < VF30w) then {VF10w = VF20w} else {VF10w = VF30w}

© SCEI
-96-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MINIi : Minimum Value
Compares each field of VF[fs] with the I register and stores the smaller value in the corresponding field of
VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MINIi
011111

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MINIi.dest VF[fd]dest, VF[fs]dest, I
Operation
if (x ⊆ dest) then
if (VF[fs]x < I) then
VF[fd]x = VF[fs]x
else
VF[fd]x = I
(The same operation is performed for the y and z fields.)
if (w ⊆ dest) then
if (VF[fs]w < I) then
VF[fd]w = VF[fs]w
else
VF[fd]w = I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

MINIi.xyzw VF10xyzw, VF20xyzw, I
if (VF20x < I) then {VF10x = VF20x} else {VF10x = I}
if (VF20y < I) then {VF10y = VF20y} else {VF10y = I}
if (VF20z < I) then {VF10z = VF20z} else {VF10z = I}
if (VF20w < I) then {VF10w = VF20w} else {VF10w = I}

© SCEI
-97-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MINIbc : Minimum Value
Compares each field of VF[fs] with the specified field of VF[ft] and stores the smaller value in the corresponding
field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MINI?
0101

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Mnemonic

MINIbc.dest VF[fd]dest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then
if (VF[fs]x < VF[ft]bc)
{VF[fd]x = VF[fs]x}
else
{VF[fd]x = VF[ft]bc}
(The same operation is performed for the y and z fields.)
if (w ⊆ dest) then
if (VF[fs]w < VF[ft]bc)
{VF[fd]w = VF[fs]w}
else
{VF[fd]w = VF[ft]bc}
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/4
Example

MINIx.xyzw VF10xyzw, VF20xyzw, VF30x
if (VF20x < VF30x) then {VF10x = VF20x} else {VF10x = VF30x}
if (VF20y < VF30x) then {VF10y = VF20y} else {VF10y = VF30x}
if (VF20z < VF30x) then {VF10z = VF20z} else {VF10z = VF30x}
if (VF20w < VF30x) then {VF10w = VF20w} else {VF10w = VF30x}

© SCEI
-98-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUB : Multiply and Subtract
Multiplies VF[fs] and VF[ft], then subracts the product obtained from the value of ACC, and stores the result in
VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MSUB
101101

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MSUB.dest VF[fd]dest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then VF[fd]x = ACCx - VF[fs]x × VF[ft]x
if (y ⊆ dest) then VF[fd]y = ACCy - VF[fs]y × VF[ft]y
if (z ⊆ dest) then VF[fd]z = ACCz - VF[fs]z × VF[ft]z
if (w ⊆ dest) then VF[fd]w = ACCw - VF[fs]w × VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUB.xyzw VF10xyzw, VF20xyzw, VF30xyzw
VF10x = ACCx - VF20x × VF30x
VF10y = ACCy - VF20y × VF30y
VF10z = ACCz - VF20z × VF30z
VF10w = ACCw - VF20w × VF30w
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-99-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUBi : Multiply and Subtract; with I Register
Multiplies each field of VF[fs] by the I register, then subtracts the product from the corresponding field of ACC
and stores the result in the corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MSUBi
100111

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MSUBi.dest VF[fd]dest, VF[fs]dest, I
Operation
if (x ⊆ dest) then VF[fd]x = ACCx - VF[fs]x × I
if (y ⊆ dest) then VF[fd]y = ACCy - VF[fs]y × I
if (z ⊆ dest) then VF[fd]z = ACCz - VF[fs]z × I
if (w ⊆ dest) then VF[fd]w = ACCw - VF[fs]w × I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUBi.xyzw VF10xyzw, VF20xyzw, I
VF10x = ACCx - VF20x × I
VF10y = ACCy - VF20y × I
VF10z = ACCz - VF20z × I
VF10w = ACCw - VF20w × I
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-100-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUBq : Multiply and Subtract; by Q Register
Multiplies each field of VF[fs] by the Q register, then subtracts the product from the corresponding field of
ACC and stores the result in the corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MSUBq
100101

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MSUBq.dest VF[fd]dest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then VF[fd]x = ACCx - VF[fs]x × Q
if (y ⊆ dest) then VF[fd]y = ACCy - VF[fs]y × Q
if (z ⊆ dest) then VF[fd]z = ACCz - VF[fs]z × Q
if (w ⊆ dest) then VF[fd]w = ACCw - VF[fs]w × Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUBq.xyzw VF10xyzw, VF20xyzw, Q
VF10x = ACCx - VF20x × Q
VF10y = ACCy - VF20y × Q
VF10z = ACCz - VF20z × Q
VF10w = ACCw - VF20w × Q
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-101-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUBbc : Broadcast Multiply and Subtract
Multiplies each field of VF[fs] by the specified field of VF[ft], then subtracts the product from the
corresponding field of ACC and stores the result in the corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MSUB?
0011

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Mnemonic

MSUBbc.dest VF[fd]dest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then VF[fd]x = ACCx - VF[fs]x × VF[ft]bc
if (y ⊆ dest) then VF[fd]y = ACCy - VF[fs]y × VF[ft]bc
if (z ⊆ dest) then VF[fd]z = ACCz - VF[fs]z × VF[ft]bc
if (w ⊆ dest) then VF[fd]w = ACCw - VF[fs]w × VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUBx.xyzw VF10xyzw, VF20xyzw, VF30x
VF10x = ACCx - VF20x × VF30x
VF10y = ACCy - VF20y × VF30x
VF10z = ACCz - VF20z × VF30x
VF10w = ACCw - VF20w × VF30x
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-102-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUBA : Multiply and Subtract; to Accumulator
Multiplies VF[fs] and VF[ft], then subtracts the product from ACC and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1 1

1

1

1

1

01011

MSUBA
1111

01

11 bits

Mnemonic

MSUBA.dest ACCdest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then ACCx = ACCx - VF[fs]x × VF[ft]x
if (y ⊆ dest) then ACCy = ACCy - VF[fs]y × VF[ft]y
if (z ⊆ dest) then ACCz = ACCz - VF[fs]z × VF[ft]z
if (w ⊆ dest) then ACCw = ACCw - VF[fs]w × VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUBA.xyzw ACCxyzw, VF20xyzw, VF30xyzw
ACCx = ACCx - VF20x × VF30x
ACCy = ACCy - VF20y × VF30y
ACCz = ACCz - VF20z × VF30z
ACCw = ACCw - VF20w × VF30w
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-103-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUBAi : Multiply and Subtract; with I Register, to Accumulator
Multiplies each field of VF[fs] and the I register, then subtracts the product from the corresponding field of
ACC and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

MSUBAi
01001
1111

11

11 bits

Mnemonic

MSUBAi.dest ACCdest, VF[fs]dest, I
Operation
if (x ⊆ dest) then ACCx = ACCx - VF[fs]x × I
if (y ⊆ dest) then ACCy = ACCy - VF[fs]y × I
if (z ⊆ dest) then ACCz = ACCz - VF[fs]z × I
if (w ⊆ dest) then ACCw = ACCw - VF[fs]w × I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUBAi.xyzw ACCxyzw, VF20xyzw, I
ACCx = ACCx - VF20x × I
ACCy = ACCy - VF20y × I
ACCz = ACCz - VF20z × I
ACCw = ACCw - VF20w × I
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-104-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUBAq : Multiply and Subtract; by Q Register, to Accumulator
Multiplies VF[fs] by the Q register, then subtracts the product from the corresponding field of ACC and stores
the result in the corresponding field of ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

MSUBAq
01001
1111

01

11 bits

Mnemonic

MSUBAq.dest ACCdest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then ACCx = ACCx - VF[fs]x × Q
if (y ⊆ dest) then ACCy = ACCy - VF[fs]y × Q
if (z ⊆ dest) then ACCz = ACCz - VF[fs]z × Q
if (w ⊆ dest) then ACCw = ACCw - VF[fs]w × Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUBAq.xyzw ACCxyzw, VF20xyzw, Q
ACCx = ACCx - VF20x × Q
ACCy = ACCy - VF20y × Q
ACCz = ACCz - VF20z × Q
ACCw = ACCw - VF20w × Q
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-105-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MSUBAbc : Broadcast Multiply and Subtract; to Accumulator
Multiplies each field of VF[fs] and the specified field of VF[ft], then subtracts the product from the
corresponding field of ACC and stores the result in the corresponding field of ACC.
Operation Code
Upper 32-bit word: UpperOP field type 2
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

MSUBA?
00011
1111

bc
--

1

4 bits

5 bits

5 bits

9 bits

2 bits

1

1

1

1

1

1

Mnemonic

MSUBAbc.dest ACCdest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then ACCx = ACCx - VF[fs]x × VF[ft]bc
if (y ⊆ dest) then ACCy = ACCy - VF[fs]y × VF[ft]bc
if (z ⊆ dest) then ACCz = ACCz - VF[fs]z × VF[ft]bc
if (w ⊆ dest) then ACCw = ACCw - VF[fs]w × VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MSUBAx.xyzw ACCxyzw, VF20xyzw, VF30x
ACCx = ACCx - VF20x × VF30x
ACCy = ACCy - VF20y × VF30x
ACCz = ACCz - VF20z × VF30x
ACCw = ACCw - VF20w × VF30x
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-106-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MUL : Multiply
Multiplies VF[fs] by VF[ft] and stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MUL
101010

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MUL.dest VF[fd]dest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x × VF[ft]x
if (y ⊆ dest) then VF[fd]y = VF[fs]y × VF[ft]y
if (z ⊆ dest) then VF[fd]z = VF[fs]z × VF[ft]z
if (w ⊆ dest) then VF[fd]w = VF[fs]w × VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MUL.xyzw VF10xyzw, VF20xyzw, VF30xyzw
VF10x = VF20x × VF30x
VF10y = VF20y × VF30y
VF10z = VF20z × VF30z
VF10w = VF20w × VF30w
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-107-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULi : Multiply by I Register
Multiplies each field of VF[fs] by the I register and stores the result in the corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MULi
011110

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MULi.dest VF[fd]dest, VF[fs]dest, I
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x × I
if (y ⊆ dest) then VF[fd]y = VF[fs]y × I
if (z ⊆ dest) then VF[fd]z = VF[fs]z × I
if (w ⊆ dest) then VF[fd]w = VF[fs]w × I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MULi.xyzw VF10xyzw, VF20xyzw, I
VF10x = VF20x × I
VF10y = VF20y × I
VF10z = VF20z × I
VF10w = VF20w × I
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-108-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULq : Multiply by Q Register
Multiplies each field of VF[fs] by the Q register and stores the result in the corresponding field of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

MULq
011100

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

MULq.dest VF[fd]dest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x × Q
if (y ⊆ dest) then VF[fd]y = VF[fs]y × Q
if (z ⊆ dest) then VF[fd]z = VF[fs]z × Q
if (w ⊆ dest) then VF[fd]w = VF[fs]w × Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MULq.xyzw VF10xyzw, VF20xyzw, Q
VF10x = VF20x × Q
VF10y = VF20y × Q
VF10z = VF20z × Q
VF10w = VF20w × Q
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-109-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULbc : Multiply by Broadcast
Multiplies each field of VF[fs] by the specified field of VF[ft] and stores the result in the corresponding field of
VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

MUL?
0110

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Mnemonic

MULbc.dest VF[fd]dest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x × VF[ft]bc
if (y ⊆ dest) then VF[fd]y = VF[fs]y × VF[ft]bc
if (z ⊆ dest) then VF[fd]z = VF[fs]z × VF[ft]bc
if (w ⊆ dest) then VF[fd]w = VF[fs]w × VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MULx.xyzw VF10xyzw, VF20xyzw, VF30x
VF10x = VF20x × VF30x
VF10y = VF20y × VF30x
VF10z = VF20z × VF30x
VF10w = VF20w × VF30x
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-110-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULA : Multiply; to Accumulator
Multiplies VF[fs] by VF[ft] and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01010

MULA
1111

10

11 bits

Mnemonic

MULA.dest ACCdest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then ACCx = VF[fs]x × VF[ft]x
if (y ⊆ dest) then ACCy = VF[fs]y × VF[ft]y
if (z ⊆ dest) then ACCz = VF[fs]z × VF[ft]z
if (w ⊆ dest) then ACCw = VF[fs]w × VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MULA.xyzw ACCxyzw, VF20xyzw, VF30xyzw
ACCx = VF20x × VF30x
ACCy = VF20y × VF30y
ACCz = VF20z × VF30z
ACCw = VF20w × VF30w
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-111-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULAi : Multiply by I Register, to Accumulator
Multiplies each field of VF[fs] by the value of the I register and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00111

MULAi
1111

10

11 bits

Mnemonic

MULAi.dest ACCdest, VF[fs]dest, I
Operation
if (x ⊆ dest) then ACCx = VF[fs]x × I
if (y ⊆ dest) then ACCy = VF[fs]y × I
if (z ⊆ dest) then ACCz = VF[fs]z × I
if (w ⊆ dest) then ACCw = VF[fs]w × I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MULAi.xyzw ACCxyzw, VF20xyzw, I
ACCx = VF20x × I
ACCy = VF20y × I
ACCz = VF20z × I
ACCw = VF20w × I
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-112-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULAq : Multiply by Q Register, to Accumulator
Multiplies each field of VF[fs] by the Q register, and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

00111

MULAq
1111

00

11 bits

Mnemonic

MULAq.dest ACCdest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then ACCx = VF[fs]x × Q
if (y ⊆ dest) then ACCy = VF[fs]y × Q
if (z ⊆ dest) then ACCz = VF[fs]z × Q
if (w ⊆ dest) then ACCw = VF[fs]w × Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MULAq.xyzw ACCxyzw, VF20xyzw, Q
ACCx = VF20x × Q
ACCy = VF20y × Q
ACCz = VF20z × Q
ACCw = VF20w × Q
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-113-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULAbc : Broadcast Multiply by broadcast, to Accumulator
Multiplies each field of VF[fs] by the specified field of VF[ft] and stores the result in the corresponding field of
ACC.
Operation Code
Upper 32-bit word: UpperOP field type 2
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

MULA?
00110
1111

bc
--

1

4 bits

5 bits

5 bits

9 bits

2 bits

1

1

1

1

1

1

Mnemonic

MULAbc.dest ACCdest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then ACCx = VF[fs]x × VF[ft]bc
if (y ⊆ dest) then ACCy = VF[fs]y × VF[ft]bc
if (z ⊆ dest) then ACCz = VF[fs]z × VF[ft]bc
if (w ⊆ dest) then ACCw = VF[fs]w × VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

MULAx.xyzw ACCxyzw, VF20xyzw, VF30x
ACCx = VF20x × VF30x
ACCy = VF20y × VF30x
ACCz = VF20z × VF30x
ACCw = VF20w × VF30x
Remarks
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value. By using VF[fs] as a multiplicand, the results of multiplication with 1 are guaranteed to be
accurate.

© SCEI
-114-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

NOP : No Operation
No operation is performed.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
0000

ft reg
00000

fs reg
00000

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01011

NOP
1111

11

11 bits

Mnemonic

NOP
Operation
None
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4

© SCEI
-115-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

OPMULA : Vector Outer Product
Calculates the first part of the vector outer product of VF[fs] and VF[ft] and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
1110

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01011

OPMULA
1111

10

11 bits

Mnemonic

OPMULA.xyz ACCxyz, VF[fs]xyz, VF[ft]xyz
Operation
ACCx = VF[fs]y × VF[ft]z
ACCy = VF[fs]z × VF[ft]x
ACCz = VF[fs]x × VF[ft]y
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example
Vector Outer Product: VF20 × VF30(Be careful to the description order of VF20 and VF30.)

OPMULA.xyz ACCxyz, VF20xyz, VF30xyz
OPMSUB.xyz VF10xyz, VF30xyz, VF20xyz
VF10x = VF20y × VF30z - VF30y × VF20z
VF10y = VF20z × VF30x - VF30z × VF20x
VF10z = VF20x × VF30y - VF30x × VF20y
Remarks
The fields subject to the operation are fixed to x,y,z.
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value.

© SCEI
-116-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

OPMSUB : Vector Outer Product
Calculates the last part of the vector outer product of VF[fs], VF[ft] and ACC and stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
1110

ft reg
-----

fs reg
-----

fd reg
-----

OPMSUB
101110

1

4 bits

5 bits

5 bits

5 bits

6 bits

1 1

1

1

1

1

Mnemonic

OPMSUB.xyz VF[fd]xyz, VF[fs]xyz, VF[ft]xyz
Operation
VF[fd]x = ACCx - VF[fs]y × VF[ft]z
VF[fd]y = ACCy - VF[fs]z × VF[ft]x
VF[fd]z = ACCz - VF[fs]x × VF[ft]y
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example
Vector Outer Product: VF20 × VF30(Be careful with the description order of VF20 and VF30.)

OPMULA.xyz ACCxyz, VF20xyz, VF30xyz
OPMSUB.xyz VF10xyz, VF30xyz, VF20xyz
VF10x = VF20y × VF30z - VF30y × VF20z
VF10y = VF20z × VF30x - VF30z × VF20x
VF10z = VF20x × VF30y - VF30x × VF20y
Remarks
The fields subject to the operation are fixed to x,y,z.
There is an operation error of 1 bit in multiplication, so the value multiplied by 1 may not be the same as the
original value.

© SCEI
-117-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUB : Subtract
Subtracts VF[ft] from VF[fs] and stores the result in VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

SUB
101100

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

SUB.dest VF[fd]dest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x - VF[ft]x
if (y ⊆ dest) then VF[fd]y = VF[fs]y - VF[ft]y
if (z ⊆ dest) then VF[fd]z = VF[fs]z - VF[ft]z
if (w ⊆ dest) then VF[fd]w = VF[fs]w - VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

SUB.xyzw VF01xyzw, VF00xyzw, VF00xyzw
The values of all fields of VF01 all become 0.0.
Remarks
When VF00 is specified as the destination, the instruction is used to compare VF[fs] with VF[ft].

© SCEI
-118-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUBi : Subtract I Register
Subtracts the I register from each field of VF[fs] and stores the result in the corresponding fields of VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

SUBi
100110

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

SUBi.dest VF[fd]dest, VF[fs]dest, I
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x - I
if (y ⊆ dest) then VF[fd]y = VF[fs]y - I
if (z ⊆ dest) then VF[fd]z = VF[fs]z - I
if (w ⊆ dest) then VF[fd]w = VF[fs]w - I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

SUBi.xyzw VF10xyzw, VF20xyzw, I
VF10x = VF20x - I
VF10y = VF20y - I
VF10z = VF20z - I
VF10w = VF20w - I
Remarks
When VF00 is specified as the destination, the instruction is used to compare each field of VF[fs] with the I
register.

© SCEI
-119-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUBq : Subtract Q Register
Subtracts the Q register from each field of VF[fs] and stores the result in the corresponding field of VF[ft].
Operation Code
Upper 32-bit word: UpperOP field type 1
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

SUBq
100100

1

4 bits

5 bits

5 bits

5 bits

6 bits

1

1

1

1

1

1

Mnemonic

SUBq.dest VF[fd]dest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x - Q
if (y ⊆ dest) then VF[fd]y = VF[fs]y - Q
if (z ⊆ dest) then VF[fd]z = VF[fs]z - Q
if (w ⊆ dest) then VF[fd]w = VF[fs]w - Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

SUBq.xyzw VF10xyzw, VF20xyzw, Q
VF10x = VF20x - Q
VF10y = VF20y - Q
VF10z = VF20z - Q
VF10w = VF20w - Q
Remarks
When VF00 is specified as the destination, the instruction is used to compare each field of VF[fs] with the Q
register .

© SCEI
-120-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUBbc : Broadcast Subtract
Subtracts the specified field of VF[ft] from each field of VF[fs] and stores the result in the corresponding field of
VF[fd].
Operation Code
Upper 32-bit word: UpperOP field type 0
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

SUB?
0001

bc
--

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

1

1

1

1

1

1

Mnemonic

SUBbc.dest VF[fd]dest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then VF[fd]x = VF[fs]x - VF[ft]bc
if (y ⊆ dest) then VF[fd]y = VF[fs]y - VF[ft]bc
if (z ⊆ dest) then VF[fd]z = VF[fs]z - VF[ft]bc
if (w ⊆ dest) then VF[fd]w = VF[fs]w - VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

SUBx.xyzw VF10xyzw, VF20xyzw, VF30x
VF10x = VF20x - VF30x
VF10y = VF20y - VF30x
VF10z = VF20z - VF30x
VF10w = VF20w - VF30x
Remarks
When VF00 is specified as the destination, the instruction is used to compare each field of VF[fs] with the
VF[ft]bc field.

© SCEI
-121-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUBA : Substract; to Accumulator
Subtracts VF[ft] from VF[fs] and stores the result in ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01011

SUBA
1111

00

11 bits

Mnemonic

SUBA.dest ACCdest, VF[fs]dest, VF[ft]dest
Operation
if (x ⊆ dest) then ACCx = VF[fs]x - VF[ft]x
if (y ⊆ dest) then ACCy = VF[fs]y - VF[ft]y
if (z ⊆ dest) then ACCz = VF[fs]z - VF[ft]z
if (w ⊆ dest) then ACCw = VF[fs]w - VF[ft]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X
Throughput/Latency
1/4
Example

SUBA.xyzw ACCxyzw, VF20xyzw, VF30xyzw
ACCx = VF20x - VF30x
ACCy = VF20y - VF30y
ACCz = VF20z - VF30z
ACCw = VF20w - VF30w

© SCEI
-122-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUBAi : Subtract I Register; to Accumulator
Subtracts the I register from each field of VF[fs] and stores the result in the corresponding field of ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01001

SUBAi
1111

10

11 bits

Mnemonic

SUBAi.dest ACCdest, VF[fs]dest, I
Operation
if (x ⊆ dest) then ACCx = VF[fs]x - I
if (y ⊆ dest) then ACCy = VF[fs]y - I
if (z ⊆ dest) then ACCz = VF[fs]z - I
if (w ⊆ dest) then ACCw = VF[fs]w - I
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

SUBAi.xyzw ACCxyzw, VF20xyzw, I
ACCx = VF20x - I
ACCy = VF20y - I
ACCz = VF20z - I
ACCw = VF20w - I

© SCEI
-123-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUBAq : Subtract Q Register; to Accumulator
Subtracts the Q register from each field of VF[fs] and stores the result in the corresponding field of ACC.
Operation Code
Upper 32-bit word: UpperOP field type 3
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E M D T - - - - - - 0 0

dest
----

ft reg
00000

fs reg
-----

1

4 bits

5 bits

5 bits

1

1

1

1

1

1

01001

SUBAq
1111

00

11 bits

Mnemonic

SUBAq.dest ACCdest, VF[fs]dest, Q
Operation
if (x ⊆ dest) then ACCx = VF[fs]x - Q
if (y ⊆ dest) then ACCy = VF[fs]y - Q
if (z ⊆ dest) then ACCz = VF[fs]z - Q
if (w ⊆ dest) then ACCw = VF[fs]w - Q
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X
Throughput/Latency
1/4
Example

SUBAq.xyzw ACCxyzw, VF20xyzw, Q
ACCx = VF20x - Q
ACCy = VF20y - Q
ACCz = VF20z - Q
ACCw = VF20w - Q

© SCEI
-124-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SUBAbc : Broadcast Subtract; to Accumulator
Subtracts the specified field of VF[ft] from each field of VF[fs] and stores the result in the corresponding field of
ACC.
Operation Code
Upper 32-bit word: UpperOP field type 2
63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32

I E MD T - - - - - - 0 0

dest
----

ft reg
-----

fs reg
-----

SUBA?
00001
1111

bc
--

1

4 bits

5 bits

5 bits

9 bits

2 bits

1

1

1

1

1

1

Mnemonic

SUBAbc.dest ACCdest, VF[fs]dest, VF[ft]bc
Operation
if (x ⊆ dest) then ACCx = VF[fs]x - VF[ft]bc
if (y ⊆ dest) then ACCy = VF[fs]y - VF[ft]bc
if (z ⊆ dest) then ACCz = VF[fs]z - VF[ft]bc
if (w ⊆ dest) then ACCw = VF[fs]w - VF[ft]bc
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
XXXX XXXX XXXX XXXX - - X X X X - - X X X X

clipping
flag
-

Throughput/Latency
1/4
Example

SUBAx.xyzw ACCxyzw, VF20xyzw, VF30x
ACCx = VF20x - VF30x
ACCy = VF20y - VF30x
ACCz = VF20z - VF30x
ACCw = VF20w - VF30x

© SCEI
-125-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

4.3. Lower Instruction Reference
This section describes the function, operation code, mnemonic, operation, flag changes, and throughput/latency
of Lower instructions. They are listed in alphabetical order in mnemonic form. The descriptions also include
examples, programming notes, and reference information.

© SCEI
-126-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

B : Unconditional Branch
Branches to the PC relative address specified with the immediate value.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

B
0100000

dest
0000

it reg
00000

is reg
00000

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

B Imm11
Imm11 is a signed interger of 11-bit long; specify the value obtained by dividing the offset to branch
destination by 8.
Operation
PC = PC + Imm11 × 8
The branch destination is determined by adding the value of Imm11, to the address of the instruction in the
branch delay slot (one instruction).
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example
In the example below, the branch destination varies when VI10 matches with either VI01, VI02, or VI03.
NOP
NOP
NOP
NOP
NOP
NOP
NOP

BEQ VI10, VI01, PROG1
NOP
BEQ VI10, VI02, PROG2
NOP
BEQ VI10, VI03, PROG3
NOP
B DEFAULT

Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-127-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

BAL : Unconditional Branch with Saving Address
Stores the address before branching in VI[it] and branches to the PC relative address specified with the
immediate value.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

BAL
0100001

dest
0000

it reg
-----

is reg
00000

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

BAL VI[it], Imm11
Imm11 is an 11-bit signed integer; specify the value obtained by dividing the offset to branch destination by
8.
Operation
VI[it] = PC + (2 × 8)
PC = PC + Imm11 × 8
The address of the instruction next to the branch delay slot (one instruction) is stored in VI[it].
The branch destination is determined by adding the value of Imm11 to the address of the instruction in the
slot.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
2/2
Example

BAL VI15, LABEL:
VI15 = PC+(2 × 8)
Branches to LABEL (PC relative address)
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-128-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

DIV : Divide
Divides fsf field of VF[fs] by ftf field of VF[ft] and stores the result in the Q register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
--

fsf
--

2 bits 2 bits

ft reg
-----

fs reg
-----

5 bits

5 bits

01110

DIV
1111

00

11 bits

Mnemonic

DIV Q, VF[fs]fsf, VF[ft]ftf
Operation
Q = VF[fs]fsf ÷ VF[ft]ftf
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------X X - - - - X X - - - -

clipping
flag
-

Throughput/Latency
7/7
Example

DIV Q, VF10x, VF20y
Q = VF10x ÷ VF20y
Remarks
A data dependency check is not performed with the Q register. To execute subsequent instructions after the
results of the DIV instruction are written to the Q register, use the WAITQ instruction for synchronization.

© SCEI
-129-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

EATAN : Arctangent
Calculates the arc tangent of fsf field of VF[fs] and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

2 bits 2 bits

ft reg
00000

fs reg
-----

5 bits

5 bits

11111

EATAN
1111

01

11 bits

Mnemonic

EATAN P, VF[fs]fsf
Operation
P = arctan( VF[fs]fsf )
The following approximation formula is used for calculating arctan.
Calculation is valid for: 0 <= x <= 1

arctan( x) = (T1 × t + T2 × t 3 + T3 × t 5 + T4 × t 7
+ T5 × t 9 + T6 × t 11 + T7 × t 13 + T8 × t 15 ) +

π
4

Provided t =
Constants
T1
T2
T3
T4
T5
T6
T7
T8
π/4

Decimal expressions
0.999999344348907
-0.333298563957214
0.199465364217758
-0.139085337519646
0.096420042216778
-0.055909886956215
0.021861229091883
-0.004054057877511
0.785398185253143

S
0
1
0
1
0
1
0
1
0

( x − 1)
( x + 1)

Single precision floating-point expressions
E
F
01111110
11111111111111111110101
01111101
01010101010011000011100
01111100
10011000100000010100110
01111100
00011100110110001100011
01111011
10001010111011111011111
01111010
11001010000000111000100
01111001
01100110001011001010010
01110111
00001001101011111100111
01111110
10010010000111111011011

Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
53 / 54
Example

EATAN P, VF10x
P = arctan( VF10x )

© SCEI
-130-

Hex.
Expressions
3f7ffff5
beaaa61c
3e4c40a6
be0e6c63
3dc577df
bd6501c4
3cb31652
bb84d7e7
3f490fdb

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the EATAN instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-131-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

EATANxy : Arctangent
Calculates the arctangent based on the x, y fields of VF[fs] and stores in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
1100

ft reg
00000

fs reg
-----

7 bits

4 bits

5 bits

5 bits

11101

EATANxy
1111

00

11 bits

Mnemonic

EATANxy P, VF[fs]
Operation
P = arctan( VF[fs]y / VF[fs]x )
The following approximation formula is used for calculating arctan.
Calculation is valid for: 0 <= y <= x (Excluding 0 = y =x)

arctan(x) = (T1 × t + T2 × t 3 + T3 × t 5 + T4 × t 7
+ T5 × t 9 + T6 × t 11 + T7 × t 13 + T8 × t 15 ) +

π
4

Provided t =
Constants
T1
T2
T3
T4
T5
T6
T7
T8
π/4

Decimal Expressions
0.999999344348907
-0.333298563957214
0.199465364217758
-0.139085337519646
0.096420042216778
-0.055909886956215
0.021861229091883
-0.004054057877511
0.785398185253143

S
0
1
0
1
0
1
0
1
0

( y − 1)
( y + x)

Single precision floating-point expressions
E
F
01111110
11111111111111111110101
01111101
01010101010011000011100
01111100
10011000100000010100110
01111100
00011100110110001100011
01111011
10001010111011111011111
01111010
11001010000000111000100
01111001
01100110001011001010010
01110111
00001001101011111100111
01111110
10010010000111111011011

Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
53 / 54
Example

EATANxy P, VF10
P = arctan( VF10y / VF10x )

© SCEI
-132-

Hex.
expressions
3f7ffff5
beaaa61c
3e4c40a6
be0e6c63
3dc577df
bd6501c4
3cb31652
bb84d7e7
3f490fdb

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the EATANxy instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-133-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

EATANxz : Arctangent
Calculates the arctangent based on the x, z fields of VF[fs] and stores in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
1010

ft reg
00000

fs reg
-----

7 bits

4 bits

5 bits

5 bits

11101

EATANxz
1111

01

11 bits

Mnemonic

EATANxz P, VF[fs]
Operation
P = arctan( VF[fs]z / VF[fs]x )
The following approximation formula is used for calculating arctan.
Calculation is valid for: 0 <= z <= x (Excluding 0 = z =x)

arctan( x) = (T1 × t + T2 × t 3 + T3 × t 5 + T4 × t 7
+ T5 × t 9 + T6 × t 11 + T7 × t 13 + T8 × t 15 ) +

π
4

Provided t =
Constants

Decimal Expressions

T1
T2
T3
T4
T5
T6
T7
T8
π/4

0.999999344348907
-0.333298563957214
0.199465364217758
-0.139085337519646
0.096420042216778
-0.055909886956215
0.021861229091883
-0.004054057877511
0.785398185253143

S
0
1
0
1
0
1
0
1
0

( z − x)
( z + x)

Single precision floating-point expressions
E
F
01111110
11111111111111111110101
01111101
01010101010011000011100
01111100
10011000100000010100110
01111100
00011100110110001100011
01111011
10001010111011111011111
01111010
11001010000000111000100
01111001
01100110001011001010010
01110111
00001001101011111100111
01111110
10010010000111111011011

Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
53 / 54
Example

EATANxz P, VF10
P = arctan( VF10z / VF10x )

© SCEI
-134-

Hex.
expressions
3f7ffff5
beaaa61c
3e4c40a6
be0e6c63
3dc577df
bd6501c4
3cb31652
bb84d7e7
3f490fdb

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the EATANxz instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-135-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

EEXP : Exponent
Calculates the exponent of fsf field of VF[fs] and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

ft reg
00000

fs reg
-----

5 bits

5 bits

2 bits 2 bits

11111

EEXP
1111

10

11 bits

Mnemonic

EEXP P, VF[fs]fsf
Operation
P = exp( - VF[fs]fsf )
The following approximation formula is used for calculating exp.
Calculation is valid for: 0 <= x <= +MAX

e−x =
Constants
E1
E2
E3
E4
E5
E6

1
(1 + E1 × x + E 2 × x + E 3 × x + E 4 × x 4 + E 5 × x 5 + E 6 × x 6 ) 4
2

Decimal Expressions
0.249998688697815
0.031257584691048
0.002591371303424
0.000171562001924
0.000005430199963
0.000000690600018

S
0
0
0
0
0
0

3

Single precision floating-point expressions
E
F
01111100
11111111111111110101000
01111010
00000000000011111110100
01110110
01010011101001111111111
01110010
01100111110010101010011
01101101
01101100011010100010000
01101010
01110010110000110101100

Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

Hex.
expressions
3e7fffa8
3d0007f4
3b29d3ff
3933e553
36b63510
353961ac

clipping
flag
-

Throughput/Latency
43 / 44
Example

EEXP P, VF10x
P = exp( - VF10x )
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the EEXP instruction are written to the P register, use the WAITP instruction for synchronization.

© SCEI
-136-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ELENG : Length
Calculates the length from the origin, with the x, y, z fields of VF[fs] as the three-dimensional coordinates, and
stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
1110

ft reg
00000

fs reg
-----

7 bits

4 bits

5 bits

5 bits

ELENG
11100
1111

10

11 bits

Mnemonic

ELENG P, VF[fs]
Operation
P = {(VF[fs]x)2+(VF[fs]y)2+(VF[fs]z)2}
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
17 / 18
Example

ELENG P, VF10
P = {(VF10x)2+(VF10y)2+(VF10z)2}
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ELENG instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-137-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ERCPR : Reciprocal Number
Calculates the reciprocal of the fsf field of VF[fs] and stores it in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

2 bits 2 bits

ft reg
00000

fs reg
-----

5 bits

5 bits

11110

ERCPR
1111

10

11 bits

Mnemonic

ERCPR P, VF[fs]fsf
Operation
P = 1 / VF[fs]fsf
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
11 / 12
Example

ERCPR P, VF10x
P = 1 / VF10x
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ERCPR instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-138-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ERLENG : Reciprocal Number of Length
Calculates the reciprocal of the length from the origin, using the x, y, z fields of VF[fs] as the three-dimensional
coordinates, and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
1110

ft reg
00000

fs reg
-----

7 bits

4 bits

5 bits

5 bits

ERLENG
11100
1111

11

11 bits

Mnemonic

ERLENG P, VF[fs]
Operation
P=

1
{(VF[fs]x)2+(VF[fs]y)2+(VF[fs]z)2}

Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
23 / 24
Example

ERLENG P, VF10
P=

1
{(VF10x)2+(VF10y)2+(VF10z)2}

Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ERLENG instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-139-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ERSADD : Reciprocal Number
Calculates the reciprocal of the sum of the squares of the x, y, z fields of VF[fs] and stores the result in the P
register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
1110

ft reg
00000

fs reg
-----

7 bits

4 bits

5 bits

5 bits

ERSADD
11100
1111

01

11 bits

Mnemonic

ERSADD P, VF[fs]
Operation
P = 1 / { (VF[fs]x)2 + (VF[fs]y)2 + (VF[fs]z)2 }
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
17 / 18
Example

ERSADD P, VF10
P = 1 / { (VF10x)2 + (VF10y)2 + (VF10z)2 }
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ERSADD instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-140-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ERSQRT : Reciprocal Number of Square Root
Calculates the reciprocal of the square root of the fsf field of VF[fs] and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

ftf
00

7 bits

fsf
--

2 bits 2 bits

ft reg
00000

fs reg
-----

5 bits

5 bits

11110

ERSQRT
1111

01

11 bits

Mnemonic

ERSQRT P, VF[fs]fsf
Operation
P=

1
VF[fs]fsf

Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
17 / 18
Example

ERSQRT P, VF10x
P=

1
VF10x

Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ERSQRT instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-141-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ESADD : Sum of Square Numbers
Calculates the sum of the squares of the x, y, z fields of VF[fs] and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
1110

ft reg
00000

fs reg
-----

7 bits

4 bits

5 bits

5 bits

11100

ESADD
1111

00

11 bits

Mnemonic

ESADD P, VF[fs]
Operation
P = (VF[fs]x)2 + (VF[fs]y)2 + (VF[fs]z)2
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
10 / 11
Example

ESADD P, VF10
P = (VF10x)2 + (VF10y)2 + (VF10z)2
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ESADD instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-142-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ESIN : Sine
Calculates the sine of the fsf field of VF[fs] and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

ft reg
00000

fs reg
-----

5 bits

5 bits

2 bits 2 bits

11111

ESIN
1111

00

11 bits

Mnemonic

ESIN P, VF[fs]fsf
Operation
P = sin( VF[fs]fsf )
The following approximation formula is used for calculating sin.
Calculation is valid for: -π/2 <= x <= +π/2

sin( x ) = S1 × x + S 2 × x 3 + S 3 × x 5 + S 4 × x 7 + S5 × x 9
Constants
S1
S2
S3
S4
S5

Decimal Expressions
1.00000000000000
-0.166666567325592
0.008333025500178
-0.000198074136279
0.000002601886990

S
0
1
0
1
0

Single precision floating-point expressions
E
F
01111111
00000000000000000000000
01111100
01010101010101010100100
01111000
00010001000011100111110
01110010
10011111011001000011111
01101100
01011101001110000010100

Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

Hex.
expressions
3f800000
be2aaaa4
3c08873e
b94fb21f
362e9c14

clipping
flag
-

Throughput/Latency
28 / 29
Example

ESIN P, VF10x
P = sin( VF10x )
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ESIN instruction are written to the P register, use the WAITP instruction for synchronization.

© SCEI
-143-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ESQRT : Square Root
Calculates the square root of the fsf field of VF[fs] and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

2 bits 2 bits

ft reg
00000

fs reg
-----

5 bits

5 bits

11110

ESQRT
1111

00

11 bits

Mnemonic

ESQRT P, VF[fs]fsf
Operation
P = VF[fs]fsf
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
11 / 12
Example

ESQRT P, VF10x
P = VF10x
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ESQRT instruction are written to the P register, use the WAITP instruction for
synchronization.

© SCEI
-144-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ESUM : Sum of Each Field
Calculates the total sum of the four fields of VF[fs] and stores the result in the P register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
1111

ft reg
00000

fs reg
-----

7 bits

4 bits

5 bits

5 bits

11101

ESUM
1111

10

11 bits

Mnemonic

ESUM P, VF[fs]
Operation
P = VF[fs]x + VF[fs]y + VF[fs]z + VF[fs]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
11 / 12
Example

ESUM P, VF10
P = VF10x + VF10y + VF10z + VF10w
Remarks
A data dependency check is not performed with the P register. To execute subsequent instructions after the
results of the ESUM instruction are written to the P register, use the WAITP instruction for synchronization.

© SCEI
-145-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FCAND : Test Clipping Flag
Tests the value of the clipping flag and stores the result in VI01.
Operation Code
Lower 32-bit word: LowerOP field type 9
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FCAND
0010010

−
0

Imm24
------------------------

7 bits

1

24 bits

Mnemonic

FCAND VI01, Imm24
Operation
Calculates the AND (logical product) of the clipping flag and Imm24 for every bit. A 1 is written to VI01 if
there is at least one 1 in the results; a 0 is written to VI01 if there are no 1's.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example
; Checks if there is "+x" even only one time in the 4 previous clipping tests or not.
NOP
NOP
NOP

FCAND VI01, 0x041041
IBEQ VI01, VI00, NEVER_CLIP_X
B
CLIP_X

© SCEI
-146-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FCEQ : Test Clipping Flag
Tests the value of clipping flag and stores the result in VI01.
Operation Code
Lower 32-bit word: LowerOP field type 9
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FCEQ
0010000

−
0

Imm24
------------------------

7 bits

1

24 bits

Mnemonic

FCEQ VI01, Imm24
Operation
Compares clipping flag and Imm24. A 1 is written to VI01 if they are the same; a 0 is written if they are not.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example
; Checks if all of the results of the 4 previous clipping tests are "+x" or not.
NOP
NOP
NOP

FCEQ VI01, 0x041041
IBEQ VI01, VI00, IN_RANGE_X
B
OUT_OF_RANGE_X

© SCEI
-147-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FCGET : Get Clipping Flag
Stores a part of the values of the clipping flag in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FCGET
0011100

Imm15
0000

it reg
-----

is reg
00000

Imm15
00000000000

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FCGET VI[it]
Operation
The lower 12 bits of the clipping flag (information from the two most recent checks) are written into VI[it].
0 is written to the upper 4 bits of VI[it].
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

FCGET VI01

© SCEI
-148-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FCOR : Test Clipping Flag
Tests the value of clipping flag and stores the result in VI01.
Operation Code
Lower 32-bit word: LowerOP field type 9
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FCOR
0010011

−
0

Imm24
------------------------

7 bits

1

24 bits

Mnemonic

FCOR VI01, Imm24
Operation
Calculates the OR (logical sum) of the clipping flag and Imm24 at every bit. A 1 is written into VI01 if all of
the results are 1, and 0 is written if they are not all 1's.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example

FCOR VI01, 0xffdf7d

;B1111,1111,1101,1111,0111,1101

VI01 becomes 1 if all of the results of the 3 previous clipping tests are "-x."

© SCEI
-149-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FCSET : Setting Clipping Flag
Sets the value of the clipping flag.
Operation Code
Lower 32-bit word: LowerOP field type 9
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FCSET
0010001

−
0

Imm24
------------------------

7 bits

1

24 bits

Mnemonic

FCSET Imm24
Operation
Sets Imm24 to the clipping flag. It is set is at the S stage of the basic pipeline.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
X

Throughput/Latency
1/4
Example

FCSET 0x000000

;B0000,0000,0000,0000,0000,0000

Clears the clipping flag.

© SCEI
-150-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FMAND : Test MAC Flag Check
Calculates the logical product of the MAC flag and VI[is] and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FMAND
0011010

Imm15
0000

it reg
-----

is reg
-----

Imm15
00000000000

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FMAND VI[it], VI[is]
Operation
Calculates the AND (logical product) of the MAC flag and VI[is] at every bit and stores the result in VI[it].
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example
; Branches if the Z field of the calculation result is negative.
NOP
NOP
NOP

IADDIU
FMAND
IBNE

VI10, VI00, 0x0020
VI01, VI10
VI01, VI00, Z_MINUS

© SCEI
-151-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FMEQ : Test MAC Flag Check
Compares the MAC flag and VI[is] and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FMEQ
0011000

Imm15
0000

it reg
-----

is reg
-----

Imm15
00000000000

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FMEQ VI[it], VI[is]
Operation
Writes 1 into VI[it] if the values of the MAC flag and VI[is] are the same, and writes 0 if they are not.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example
; Branches if only the Z field of the calculation result is negative.
NOP
NOP
NOP

IADDIU
FMEQ
IBNE

VI10, VI00, 0x0020
VI01, VI10
VI01, VI00, Z_MINUS

© SCEI
-152-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FMOR : Test MAC Flag Check
Calculates the logical sum of the MAC flag and VI[is] and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FMOR
0011011

Imm15
0000

it reg
-----

is reg
-----

Imm15
00000000000

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FMOR VI[it], VI[is]
Operation
Calculates the OR (logical sum) of the MAC flag and VI[is] at every bit and stores the result in VI[it] .
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example

FMOR VI01, VI10

© SCEI
-153-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FSAND : Test Status Flag Check
Calculates the logical product (AND) of the status flag and the immediate value and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FSAND
0010110

Imm12
000-

it reg
-----

is reg
00000

Imm12
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FSAND VI[it], Imm12
Operation
Calculates the logical product (AND) of the status flag and Imm12 at every bit and stores the result in VI[it].
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example
; Checks if there was an overflow in the past.
NOP
NOP

FSAND VI01, 0x0200
IBNE VI01, VI00, OVERFLOW

© SCEI
-154-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FSEQ : Test Status Flag Check
Compares the status flag and the immediate value and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FSEQ
0010100

Imm12
000-

it reg
-----

is reg
00000

Imm12
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FSEQ VI[it], Imm12
Operation
Writes 1 into VI[it] if the values of the status flag and Imm12 are the same, and writes 0 if they are not.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example
; Checks if there was an overflow immediately before.
NOP
NOP

FSEQ
IBNE

VI01, 0x0008
VI01, VI00, OVERFLOW

© SCEI
-155-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FSOR : Test Status Flag
Calculates the logical sum of the status flag and the immediate value and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FSOR
0010111

Imm12
000-

it reg
-----

is reg
00000

Imm12
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FSOR VI[it], Imm12
Operation
Calculates the logical sum (OR) of the status flag and Imm12 at every bit and stores the result in VI[it].
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

FSOR VI01, chk_flag

© SCEI
-156-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FSSET : Set Sticky Flags
Writes the immediate value to the sticky flags. (DS-ZS)
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

FSSET
0010101

Imm12
000-

it reg
00000

is reg
00000

Imm12
-----000000

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

FSSET Imm12
Operation
Writes the upper 6 bits of Imm 12 to the upper 6 bits (sticky flag) of the 12-bit status flag. The lower 6 bits
of Imm 12 are not set, and are therefore ignored. It is set at the S stage of the basic pipleline.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------X X X X X X - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

FSSET 0x0
Clears all the sticky flag.

© SCEI
-157-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IADD : ADD Integer
Adds VI[is] and VI[it] and stores the result in VI[id].
Operation Code
Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
-----

id reg
-----

IADD
110000

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

IADD VI[id], VI[is], VI[it]
Operation
VI[id] = VI[is] + VI[it]
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

IADD VI04, VI05, VI06
VI04 = VI05 + VI06

© SCEI
-158-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IADDI : Add Immediate Value Integer
Adds the immediate value to VI[is] and stores the result in VI[it].
Operation Code
Lower 32-bit word : LowerOP field type 5
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
-----

Imm5
-----

IADDI
110010

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

IADDI VI[it], VI[is], Imm5
Operation
VI[it] = VI[is] + Imm5
Imm5 is considered as a signed integer.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example

IADDI VI04, VI05, -1
VI04 = VI05 - 1

© SCEI
-159-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IADDIU : Add Immediate Integer
Adds the immediate value to VI[is] and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

IADDIU
0001000

Imm15
----

it reg
-----

is reg
-----

Imm15
---------------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

IADDIU VI[it], VI[is], Imm15
Operation
VI[it] = VI[is] + Imm15
Imm15 is considered as a unsigned integer.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

IADDIU VI04, VI05, 1
VI04 = VI05 + 1

© SCEI
-160-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IAND : Logical Product
Calculates the AND (logical product) of VI[is] and VI[it] at every bit and stores the result in VI[id].
Operation Code
Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
-----

id reg
-----

IAND
110100

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

IAND VI[id], VI[is], VI[it]
Operation
VI[id] = VI[is] AND VF[it]
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example

IAND VI04, VI05, VI06
VI04 = VI05 AND VI06

© SCEI
-161-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IBEQ : Conditional Branch
Branches to the relative address specified by the immediate value if the contents of VI[is] and VI[it] are equal.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

IBEQ
0101000

dest
0000

it reg
-----

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

IBEQ VI[it], VI[is], Imm11
Imm11 is an 11-bit signed integer and holds the value of offset to the branch destination by 8.
Operation
if (VI[it] == VI[is]) then jump (PC+Imm11 × 8)
An 11-bit signed offset Imm11 multiplied by 8 is added to the address of the instruction following the
branch (NOT the branch itself), in the branch delay slot, to form a PC-relative effective target address.
If the contents of VI[it] and VI[is] are equal, a branch to the PC-relative effective target address after the
instruction in the delay slot is executed.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example

IBEQ VI04, VI05, LABEL:
Remarks
This instruction cannot be placed in the E bit delay slot.
After the instruction for setting the value to VI[is] or VI[it], leave a slot for one instruction, then write IBEQ
instruction. Refer to "3.4.8. Conditional Branching and Pipeline."

© SCEI
-162-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IBGEZ : Conditional Branch
Branches to the relative address specified by the immediate value if the contents of VI[is] is greater than or equal
to 0.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

IBGEZ
0101111

dest
0000

it reg
00000

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

IBGEZ VI[is], Imm11
Imm11 is an 11-bit signed integer and holds the value of offset to the branch target address divided by 8.
Operation
if (VI[is] >= 0) then jump (PC+Imm11 × 8)
An 11-bit signed offset Imm11 multiplied by 8 is added to the address of the instruction following the
branch (NOT the branch itself), in the branch delay slot, to form a PC-relative effective target address.
If the contents of VI[is] are greater than or equal to 0, a branch to the PC-relative effective target address
after the instruction in the delay slot is executed.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example

IBGEZ VI04, LABEL:
Remarks
This instruction cannot be placed in the E bit delay slot.
After an instruction that writes a value to VI[is], leave a slot for one instruction before an IBGEZ
instruction. Refer to "3.4.8. Conditional Branching and Pipeline."

© SCEI
-163-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IBGTZ : Conditional Branch
Branches to the relative address specified by the immediate value if the value of VI[is] is greater than 0.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

IBGTZ
0101101

dest
0000

it reg
00000

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

IBGTZ VI[is], Imm11
Imm11 is an 11-bit signed integer and holds the value of offset to the branch target address divided by 8.
Operation
if (VI[is] > 0) then jump (PC+Imm11 × 8)
An 11-bit signed offset Imm11 multiplied by 8 is added to the address of the instruction following the
branch (NOT the branch itself), in the branch delay slot, to form a PC-relative effective target address.
If the contents of VI[is] are greater than 0, a branch to the PC-relative effective target address after the
instruction in the delay slot is executed.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example

IBGTZ VI04, LABEL:
Remarks
This instruction cannot be placed in the E bit delay slot.
After an instruction that writes a value to VI[is], leave a slot for one instruction before the IBGTZ
instruction. Refer to "3.4.8. Conditional Branching and Pipeline."

© SCEI
-164-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IBLEZ : Conditional Branch
Branches to the relative address specified by the immediate value if the value of VI[is] is less than or equal to 0.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

IBLEZ
0101110

dest
0000

it reg
00000

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

IBLEZ VI[is], Imm11
Imm11 is an 11-bit signed integer and holds the value of offset to branch destination by 8.
Operation
if (VI[is] <= 0) then jump (PC+Imm11 × 8)
An 11-bit signed offset Imm11 multiplied by 8 is added to the address of the instruction following the
branch (NOT the branch itself), in the branch delay slot, to form a PC-relative effective target address.
If the contents of VI[is] are less than or equal to 0, a branch to the PC-relative effective target address after
the instruction in the delay slot is executed.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example

IBLEZ VI04, LABEL:
Remarks
This instruction cannot be placed in the E bit delay slot.
After an instruction that writes a value to VI[is], leave a slot for one instruction before the IBLEZ
instruction. Refer to "3.4.8. Conditional Branching and Pipeline."

© SCEI
-165-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IBLTZ : Conditional Branch
Branches to the relative address specified by the immediate value, if the contents of VI[is] are less than 0.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

IBLTZ
0101100

dest
0000

it reg
00000

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

IBLTZ VI[is], Imm11
Imm11 is an 11-bit signed integer and holds the value of offset to the branch target address divided by 8.
Operation
if (VI[is] < 0) then jump (PC+Imm11 × 8)
An 11-bit signed offset Imm11 multiplied by 8 is added to the address of the instruction following the
branch (NOT the branch itself), in the branch delay slot, to form a PC-relative effective target address.
If the contents of VI[is] are less than 0, a branch to the PC-relative effective target address after the
instruction in the delay slot is executed.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example

IBLTZ VI04, LABEL:
Remarks
This instruction cannot be placed in the E bit delay slot.
After an instruction that writes a value to VI[is], leave a slot for one instruction before the IBLTZ
instruction. Refer to "3.4.8. Conditional Branching and Pipeline."

© SCEI
-166-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IBNE : Conditional Branch
Branches to the PC relative address determined by the immediate value, if the contents of VI[is] and VI[it] are
not equal.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

IBNE
0101001

dest
0000

it reg
-----

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

IBNE VI[it], VI[is], Imm11
Imm11 is an 11-bit signed integer and holds the value of offset to the branch target address divided by 8.
Operation
if (VI[it] != VI[is]) then jump (PC+Imm11 × 8)
An 11-bit signed offset Imm11 multiplied by 8 is added to the address of the instruction following the
branch (NOT the branch itself), in the branch delay slot, to form a PC-relative effective target address.
If the contents of VI[is] and VI[it] are not equal, a branch to the PC-relative effective target address after the
instruction in the delay slot is executed.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example

IBNE VI04, VI05, LABEL:
Remarks
This instruction cannot be placed in the E bit delay slot.
After an instruction that writes values to VI[is] and VI[it], leave a slot for one instruction before the IBNE
instruction. Refer to "3.4.8. Conditional Branching and Pipeline."

© SCEI
-167-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ILW : Integer Load with Offset Specification
Loads the specific field of data, whose address is specified by VI[is] and the immediate value, to VI[it] from VU
Mem.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

ILW
0000100

dest
----

it reg
-----

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

ILW.dest VI[it], Imm11(VI[is])dest
Specifies any one of x, y, z or w for dest. Operation is undefined when multiple specifications are made.
Imm11 is an 11-bit signed integer. For Imm11 and VI[is], specifies the address divided by 16.
Operation
VI[it] = VU_MEMdest((Imm11 + VI[is]) × 16)
Loads the lower 16 bits of dest field of data located in VU Mem, whose address is determined with Imm11
and VI[is] to VI[it].
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/4
Example

ILW.y VI10, offset(VI04)y
VI10 = VU_MEMy(offset + VI04)
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-168-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ILWR : Integer Load
Loads the specific field of data, whose address is specified by VI[is], to VI[it] from VU Mem.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

it reg
-----

is reg
-----

7 bits

4 bits

5 bits

5 bits

01111

ILWR
1111

10

11 bits

Mnemonic

ILWR.dest VI[it], (VI[is])dest
Specifies any one of x, y, z or w for dest. Operation is undefined when multiple specifications are made. For
VI[is], specifies the address divided by 16.
Operation
VI[it] = VU_MEMdest(VI[is] × 16)
Loads lower 16 bits of dest field of data located in VU Mem, whose address is determined by VI[is] to VI[it].
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

ILWR.y VI10, (VI04) y
VI10 = VU_MEMy(VI04)
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-169-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

IOR : Logical Sum
Calculates the logical sum of VI[is] and VI[it] at every bit and stores the result in VI[id].
Operation Code
Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
-----

id reg
-----

IOR
110101

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

IOR VI[id], VI[is], VI[it]
Operation
VI[id] = VI[is] OR VI[it]
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

IOR VI04, VI05, VI06
VI04 = VI05 OR VI06

© SCEI
-170-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ISUB : Integer Subtract
Subtracts VI[it] from VI[is] and stores the result in VI[id].
Operation Code
Lower 32-bit word: LowerOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
-----

id reg
-----

ISUB
110001

7 bits

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

ISUB VI[id], VI[is], VI[it]
Operation
VI[id] = VI[is] - VI[it]
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example

ISUB VI04, VI05, VI06
VI04 = VI05 - VI06

© SCEI
-171-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ISUBIU : Immediate Value Integer Subtract
Subtracts the immediate value from VI[is] and stores the result in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 8
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

ISUBIU
0001001

Imm15
----

it reg
-----

is reg
-----

Imm15
---------------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

ISUBIU VI[it], VI[is], Imm15
Operation
VI[it] = VI[is] - Imm15
Imm15 is considered as unsigned integer.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

ISUBIU VI04, VI05, 1
VI04 = VI05 - 1

© SCEI
-172-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ISW : Integer Store with Offset
Stores data from VI[it] in VU Mem. The destination is the field specified with dest of the address determined by
VI[is] and the immediate value.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

ISW
0000101

dest
----

it reg
-----

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

ISW.dest VI[it], Imm11(VI[is])dest
Imm11 is a 11-bit signed integer. For Imm11 and VI[is], specifies the value obtained by dividing the address
by 16.
Operation
if (x ⊆ dest) then VU_MEMx(Imm11 + VI[is]) = VI[it]
if (y ⊆ dest) then VU_MEMy(Imm11 + VI[is]) = VI[it]
if (z ⊆ dest) then VU_MEMz(Imm11 + VI[is]) = VI[it]
if (w ⊆ dest) then VU_MEMw(Imm11 + VI[is]) = VI[it]
The value of VI[it] is stored in the lower 16 bits of dest field of data located in VU Mem, whose address is
determined by Imm11 and VI[is]. The upper 16 bits are filled with 0.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

ISW.y VI10, offset(VI04)y
VU_MEMy(offset + VI04) = VI10
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-173-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ISWR : Integer Store
Stores data from VI[it] in VU Mem. The destination is the field specified with dest of the address determined by
VI[is].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

it reg
-----

is reg
-----

7 bits

4 bits

5 bits

5 bits

01111

ISWR
1111

11

11 bits

Mnemonic

ISWR.dest VI[it], (VI[is])dest
For VI[is], specifies the value obtained by dividing the address by 16.
Operation
if (x ⊆ dest) then VU_MEMx(VI[is]) = VI[it]
if (y ⊆ dest) then VU_MEMy(VI[is]) = VI[it]
if (z ⊆ dest) then VU_MEMz(VI[is]) = VI[it]
if (w ⊆ dest) then VU_MEMw(VI[is]) = VI[it]
The value of VI[it] is stored in lower 16 bits of dest field of data located in VU Mem, whose address is
determined by VI[is]. The upper 16 bits are filled with 0.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/4
Example

ISWR.y VI10, (VI04)y
VU_MEMy(VI04) = VI10
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-174-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

JALR : Unconditional Jump with Address Saving
Stores the instruction address in VI[it] and jumps to the address specified with VI[is].
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

JALR
0100101

dest
0000

it reg
-----

is reg
-----

Imm11
00000000000

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

JALR VI[it], VI[is]
For VI[is], specifies the branch destination address aligned on a dword boundary.
Operation
VI[it] = PC + (2 × 8)
PC = VI[is]
The branch delay slot is one instruction.
The addresss of the next instruction after the branch delay slot is stored in VI[it].
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
2/2
Example
; Calls a subroutine indicated with the label SYMBOL

IADDIU VI04, VI00, SYMBOL
NOP
JALR VI15, VI04
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-175-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

JR : Unconditional Jump
Jumps to the address specified with VI[is].
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

JR
0100100

dest
0000

it reg
00000

is reg
-----

Imm11
00000000000

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

JR VI[is]
As the branch destination, specifies the address in dword alignment for VI[is].
Operation
It jumps to the address specified with VI[is]. The branch delay slot is one instruction.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
2/2
Example

JR VI04
Branches to the address indicated by VI04.
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-176-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

LQ : Load Qword
Loads data from a qword, whose address is specified with VI[is] and the immediate value, to VF[ft] from VU
Mem.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

LQ
0000000

dest
----

ft reg
-----

is reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

LQ.dest VF[ft]dest, Imm11(VI[is])
Imm11 is a 11-bit signed integer.
For Imm11 and VI[is], specifies the value obtained by dividing the address by 16.
Operation
if (x ⊆ dest) then VF[ft]x = VU_MEMx((Imm11 + VI[is]) × 16)
if (y ⊆ dest) then VF[ft]y = VU_MEMy((Imm11 + VI[is]) × 16)
if (z ⊆ dest) then VF[ft]z = VU_MEMz((Imm11 + VI[is]) × 16)
if (w ⊆ dest) then VF[ft]w = VU_MEMw((Imm11 + VI[is]) × 16)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

LQ.xyzw VF10xyzw, offset(VI04)
VF10x = VU_MEMx((offset + VI04) × 16)
VF10y = VU_MEMy((offset + VI04) × 16)
VF10z = VU_MEMz((offset + VI04) × 16)
VF10w = VU_MEMw((offset + VI04) × 16)
Remarks
This instruction cannot be placed in the E bit delay slot.
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-177-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

LQD : Load Qword with Pre-Decrement
Subtracts 1 from VI[is] and loads data from a qword specified by VI[is] from VU Mem to VF[ft].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

is reg
-----

7 bits

4 bits

5 bits

5 bits

01101

LQD
1111

10

11 bits

Mnemonic

LQD.dest VF[ft]dest, (--VI[is])
For VI[is], specifies the value obtained by dividing the address by 16.
Operation
VI[is] = VI[is] - 1
if (x ⊆ dest) then VF[ft]x = VU_MEMx(VI[is] × 16)
if (y ⊆ dest) then VF[ft]y = VU_MEMy(VI[is] × 16)
if (z ⊆ dest) then VF[ft]z = VU_MEMz(VI[is] × 16)
if (w ⊆ dest) then VF[ft]w = VU_MEMw(VI[is] × 16)
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

LQD.xyzw VF10xyzw, (--VI04)
VI04 = VI04 - 1
VF10x = VU_MEMx(VI04 × 16)
VF10y = VU_MEMy(VI04 × 16)
VF10z = VU_MEMz(VI04 × 16)
VF10w = VU_MEMw(VI04 × 16)
Remarks
This instruction cannot be placed in the E bit delay slot.
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-178-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

LQI : Load with Post-Increment
Loads data from VI[is] from VU Mem to VF[ft] and adds 1 to VI[is].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

is reg
-----

7 bits

4 bits

5 bits

5 bits

01101

LQI
1111

00

11 bits

Mnemonic

LQI.dest VF[ft]dest, (VI[is]++)
For VI[is], specifies the value obtained by dividing the address by 16.
Operation
if (x ⊆ dest) then VF[ft]x = VU_MEMx(VI[is] × 16)
if (y ⊆ dest) then VF[ft]y = VU_MEMy(VI[is] × 16)
if (z ⊆ dest) then VF[ft]z = VU_MEMz(VI[is] × 16)
if (w ⊆ dest) then VF[ft]w = VU_MEMw(VI[is] × 16)
VI[is] = VI[is] + 1
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

LQI.xyzw VF10xyzw, (VI04++)
VF10x = VU_MEMx(VI04 × 16)
VF10y = VU_MEMy(VI04 × 16)
VF10z = VU_MEMz(VI04 × 16)
VF10w = VU_MEMw(VI04 × 16)
VI04 = VI04 + 1
Remarks
This instruction cannot be placed in the E bit delay slot.
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-179-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MFIR : Move from Integer Register to Floating-Point Register
Moves the data of VI[is] to VF[ft].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

is reg
-----

7 bits

4 bits

5 bits

5 bits

01111

MFIR
1111

01

11 bits

Mnemonic

MFIR.dest VF[ft]dest, VI[is]
Operation
if (x ⊆ dest) then VF[ft]x = VI[is]
if (y ⊆ dest) then VF[ft]y = VI[is]
if (z ⊆ dest) then VF[ft]z = VI[is]
if (w ⊆ dest) then VF[ft]w = VI[is]
The data of VI[is] is sign extended from 16 bits to 32 bits.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

MFIR.y VF10y, VI12
VF10y = VI12
Remarks
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-180-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MFP : Move from P Register to Floating-Point Register
Moves the value of the P register to VF[ft].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
00000

7 bits

4 bits

5 bits

5 bits

11001

MFP
1111

00

11 bits

Mnemonic

MFP.dest VF[ft]dest, P
Operation
if (x ⊆ dest) then VF[ft]x = P
if (y ⊆ dest) then VF[ft]y = P
if (z ⊆ dest) then VF[ft]z = P
if (w ⊆ dest) then VF[ft]w = P
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

MFP.x VF10x, P
VF10x = P
Remarks
A data dependency check is not performed with the P register. To read the appropriate operation results,
use the WAITP instruction for synchronization.
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-181-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MOVE : Transfer between Floating-Point Registers
Transfers the value of VF[fs] to VF[ft].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

7 bits

4 bits

5 bits

5 bits

01100

MOVE
1111

00

11 bits

Mnemonic

MOVE.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then VF[ft]x = VF[fs]x
if (y ⊆ dest) then VF[ft]y = VF[fs]y
if (z ⊆ dest) then VF[ft]z = VF[fs]z
if (w ⊆ dest) then VF[ft]w = VF[fs]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

MOVE.xyzw VF10xyzw, VF20xyzw
VF10x = VF20x
VF10y = VF20y
VF10z = VF20z
VF10w = VF20w
Remarks
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-182-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MR32 : Move with Rotate
Moves the value of VF[fs] to VF[ft], rotating field by field.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
-----

7 bits

4 bits

5 bits

5 bits

01100

MR32
1111

01

11 bits

Mnemonic

MR32.dest VF[ft]dest, VF[fs]dest
Operation
if (x ⊆ dest) then VF[ft]x = VF[fs]y
if (y ⊆ dest) then VF[ft]y = VF[fs]z
if (z ⊆ dest) then VF[ft]z = VF[fs]w
if (w ⊆ dest) then VF[ft]w = VF[fs]x
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

MR32.xyzw VF10xyzw, VF20xyzw
VF10x = VF20y
VF10y = VF20z
VF10z = VF20w
VF10w = VF20x
Remarks
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-183-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MTIR : Move from Floating-Point Register to Integer Register
Moves the lower 16 bits of the fsf field of VF[fs] to VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

2 bits 2 bits

it reg
-----

fs reg
-----

5 bits

5 bits

01111

MTIR
1111

00

11 bits

Mnemonic

MTIR VI[it], VF[fs]fsf
Operation
VI[it] = VF[fs]fsf
Only the lower 16 bits of VF[fs]fsf are transferred.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

MTIR VI12, VF10y
VI12 = VF10y

© SCEI
-184-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

RGET : Get Random Number
Gets a random number and stores it in VF[ft].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
00000

7 bits

4 bits

5 bits

5 bits

10000

RGET
1111

01

11 bits

Mnemonic

RGET.dest VF[ft]dest, R
Operation
if (x ⊆ dest) then VF[ft]x = R*
if (y ⊆ dest) then VF[ft]y = R*
if (z ⊆ dest) then VF[ft]z = R*
if (w ⊆ dest) then VF[ft]w = R*
The random number obtained here (R*) consists of the 23-bit R register as mantissa and 001111111 as sign
and exponent. The value range is: +1.0 < R* < +2.0
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

RGET.x VF10x, R
RNEXT.y VF10y, R
RNEXT.z VF10z, R
Remarks
When multiple fields are specified as destinations, the same value is stored in each field.
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-185-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

RINIT : Random Number Intialize
Sets the R register to the field specified by the fsf field of VF[fs].
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

2 bits 2 bits

ft reg
00000

fs reg
-----

5 bits

5 bits

10000

RINIT
1111

10

11 bits

Mnemonic

RINIT R, VF[fs]fsf
Operation
R = VF[fs]fsf
The lower 23 bits of VF[fs]fsf are transferred to the R register.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

RINIT R, VF10x

© SCEI
-186-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

RNEXT : Next Random Number
Generates a new random number and stores it in VF[ft].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

ft reg
-----

fs reg
00000

7 bits

4 bits

5 bits

5 bits

10000

RNEXT
1111

00

11 bits

Mnemonic

RNEXT.dest VF[ft]dest, R
Operation
R = next( R )
if (x ⊆ dest) then VF[ft]x = R*
if (y ⊆ dest) then VF[ft]y = R*
if (z ⊆ dest) then VF[ft]z = R*
if (w ⊆ dest) then VF[ft]w = R*
Random numbers are generated in M series.
The random number obtained here (R*) consists of the 23-bit R register as mantissa and 001111111 as sign
and exponent. The value range is: +1.0 < R* < +2.0
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

RGET.x VF10x, R
RNEXT.y VF10y, R
RNEXT.z VF10z, R
Remarks
When multiple fields are specified as destinations, the same value is stored in each field.
When an Upper instruction in the same cycle writes data to the VF[ft] register, the result of this instruction is
discarded with priority given to the Upper instruction, regardless of whether the data is written to the same
field or not.

© SCEI
-187-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

RSQRT : Square Root Division
Divides the fsf field of VF[fs] by the square root of the ftf field of VF[ft] and stores the result in the Q register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
--

fsf
--

2 bits 2 bits

ft reg
-----

fs reg
-----

5 bits

5 bits

01110

RSQRT
1111

10

11 bits

Mnemonic

RSQRT Q, VF[fs]fsf, VF[ft]ftf
Operation
Q = VF[fs]fsf ÷ VF[ft]ftf
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------X X - - - - X X - - - -

clipping
flag
-

Throughput/Latency
13 / 13
Example

RSQRT Q, VF10x, VF20y
Q = VF10x ÷ |VF20y|
Remarks
A data dependency check is not performed with the Q register. To execute subsequent instructions after the
results of the RSQRT instruction are written to the Q register, use the WAITQ instruction for
synchronization.

© SCEI
-188-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

RXOR : Random Number Set
Sets the exclusive logical sum of the R register and the fsf field of VF[fs] to the R register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
00

fsf
--

ft reg
00000

fs reg
-----

5 bits

5 bits

2 bits 2 bits

10000

RXOR
1111

11

11 bits

Mnemonic

RXOR R, VF[fs]fsf
Operation
R = xor( R, VF[fs]fsf )
Only the lower 23 bits of VF[fs]fsf are used.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example

RXOR R, VF10x

© SCEI
-189-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SQ : Store Qword with Offset
Stores the data of VF[fs] in the VU Mem address specified by VI[it] and the immediate value.
Operation Code
Lower 32-bit word: LowerOP field type 7
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

SQ
0000001

dest
----

it reg
-----

fs reg
-----

Imm11
-----------

7 bits

4 bits

5 bits

5 bits

11 bits

Mnemonic

SQ.dest VF[fs]dest, Imm11(VI[it])
Imm11 is a 11-bit signed integer.
For Imm11 and VI[it], specify the value obtained by dividing the address by 16.
Operation
if (x ⊆ dest) then VU_MEMx((Imm11 + VI[it]) × 16) = VF[fs]x
if (y ⊆ dest) then VU_MEMy((Imm11 + VI[it]) × 16) = VF[fs]y
if (z ⊆ dest) then VU_MEMz((Imm11 + VI[it]) × 16) = VF[fs]z
if (w ⊆ dest) then VU_MEMw((Imm11 + VI[it]) × 16) = VF[fs]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/4
Example

SQ.xyzw VF10xyzw, offset(VI04)
VU_MEMx((offset + VI04) × 16) = VF10x
VU_MEMy((offset + VI04) × 16) = VF10y
VU_MEMz((offset + VI04) × 16) = VF10z
VU_MEMw((offset + VI04) × 16) = VF10w
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-190-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SQD : Store Qword with Pre-Decrement
Subtracts 1 from VI[it] and stores the data of VF[fs] to the address of VU Mem.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

it reg
-----

fs reg
-----

7 bits

4 bits

5 bits

5 bits

01101

SQD
1111

11

11 bits

Mnemonic

SQD.dest VF[fs]dest, (--VI[it])
For VI[it], specifies the value obtained by dividing the address by 16.
Operation
VI[is] = VI[is] – 1
if (x ⊆ dest) then VU_MEMx(VI[it] × 16) = VF[fs]x
if (y ⊆ dest) then VU_MEMy(VI[it] × 16) = VF[fs]y
if (z ⊆ dest) then VU_MEMz(VI[it] × 16) = VF[fs]z
if (w ⊆ dest) then VU_MEMw(VI[it] × 16) = VF[fs]w
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/4
Example

SQD.xyzw VF10xyzw, (--VI04)
VI04 = VI04 – 1
VU_MEMx(VI04 × 16) = VF10x
VU_MEMy(VI04 × 16) = VF10y
VU_MEMz(VI04 × 16) = VF10z
VU_MEMw(VI04 × 16) = VF10w
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-191-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SQI : Store with Post-Increment
Stores the data of VF[fs] to the VU Mem address specified by VI[it] and adds 1 to VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
----

it reg
-----

fs reg
-----

7 bits

4 bits

5 bits

5 bits

SQI
01101

1111

01

11 bits

Mnemonic

SQI.dest VF[fs]dest, (VI[it]++)
For VI[it], specifies the value obtained by dividing the address by 16.
Operation
if (x ⊆ dest) then VU_MEMx(VI[it] × 16) = VF[fs]x
if (y ⊆ dest) then VU_MEMy(VI[it] × 16) = VF[fs]y
if (z ⊆ dest) then VU_MEMz(VI[it] × 16) = VF[fs]z
if (w ⊆ dest) then VU_MEMw(VI[it] × 16) = VF[fs]w
VI[is] = VI[is] + 1
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/4
Example

SQI.xyzw VF10xyzw, (VI04++)
VU_MEMx(VI04 × 16) = VF10x
VU_MEMy(VI04 × 16) = VF10y
VU_MEMz(VI04 × 16) = VF10z
VU_MEMw(VI04 × 16) = VF10w
VI04 = VI04 + 1
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-192-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SQRT : Square Root
Obtains the square root of the ftf field of VF[ft] and stores the result in the Q register.
Operation Code
Lower 32-bit word: LowerOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000
7 bits

ftf
--

fsf
00

2 bits 2 bits

ft reg
-----

fs reg
00000

5 bits

5 bits

01110

SQRT
1111

01

11 bits

Mnemonic

SQRT Q, VF[ft]ftf
Operation
Q = |VF[ft]ftf|
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------X X - - - - X X - - - -

clipping
flag
-

Throughput/Latency
7/7
Example

SQRT Q, VF10x
Q = |VF10x|
Remarks
A data dependency check is not performed with the Q register. To execute subsequent instructions after the
results of the SQRT instruction are written to the Q register, use the WAITQ instruction for
synchronization.

© SCEI
-193-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

WAITP : P Register Syncronize
Waits until the calculation result is written to the P register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

ft reg
00000

fs reg
00000

7 bits

4 bits

5 bits

5 bits

11110

WAITP
1111

11

11 bits

Mnemonic

WAITP
Operation
Interlocks until the calculation result of the previous EFU instruction (e.g. ESIN) is written to the P register.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1~54 / 1~54
Example

NOP ELENG P, VF10
NOP WAITP
NOP MFP.x VF20.x, P
Remarks
A data dependency check is not performed with the P register. Use the WAITP instruction to synchronize
after an EFU instruction, if the subsequent instruction uses the result.

© SCEI
-194-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

WAITQ : Q Register Syncronize
Waits until the calculation result is written into Q register.
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

ft reg
00000

fs reg
00000

7 bits

4 bits

5 bits

5 bits

01110

WAITQ
1111

11

11 bits

Mnemonic

WAITQ
Operation
Interlocks until the calculation result of the previous FDIV instruction (DIV/SQRT/RSQRT) is written into
Q register.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1~13 / 1~13
Example

NOP SQRT Q, VF10x
NOP WAITQ
ADDq.xyzw VF20xyzw, VF11xyzw, Q NOP
Remarks
A data dependency check is not performed with the Q register. Use the WAITQ instruction to synchronize
after the FDIV instruction, if the subsequent instruction uses the result.

© SCEI
-195-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

XGKICK : GIF Control
Activates GIF transfer via PATH1 using the value of VI[is].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

ft reg
00000

is reg
-----

7 bits

4 bits

5 bits

5 bits

11011

XGKICK
1111

00

11 bits

Mnemonic

XGKICK VI[is]
Operation
Provides the value of VI[is] for the external unit (GIF) and gives an instruction to start transfer via PATH1.
If the preceding transfer via PATH1 is still in process, stalls until it ends.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1~ / 1~
Example

XGKICK VI10
Remarks
XGKICK instruction may be used only on VU1.
It cannot be placed in the E bit delay slot. If the E bit is set by an Upper instruction in the same cycle, the
VIFcode FLUSH may not wait for the end of the XGKICK instruction properly. For details, see "FLUSH"
in the "EE User’s Manual".
For operations with stalls, refer to "3.4.9. XGKICK Pipeline".
GIF considers the provided value of VI[is] as a GS packet (leading GIFtag) address, and transfers data.
Since the XGKICK instruction itself ends when the GIF starts transferring, caution should be taken against
rewriting data in the VU Mem during data transfer.
In addition, give attention to the latency of a store instruction so that the transfer starts after the data stored
immediately before the XGKICK instruction is written in the VU Mem.

© SCEI
-196-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

XITOP : VIF Control
Reads the value of the ITOP register of VIF and stores it in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
00000

7 bits

4 bits

5 bits

5 bits

11010

XITOP
1111

01

11 bits

Mnemonic

XITOP VI[it]
Operation
VI[it] = ITOP
Reads the value of the ITOP register of the external unit (VIF). The value is read at the T stage.
In VU0, whose data memory capacity is 4 Kbytes, the lower 8 bits of the VIF0_ITOP register are zeroextended and read. In VU1, the lower 10 bits of the VIF1_ITOP register are zero-extended and read.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - -

clipping
flag
-

Throughput/Latency
1/1
Example

XITOP VI10
Remarks
This instruction cannot be placed in the E bit delay slot.

© SCEI
-197-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

XTOP : VIF Control
Reads the value of the TOP register of VIF and stores it in VI[it].
Operation Code
Lower 32-bit word: LowerOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

Lower OP.
1000000

dest
0000

it reg
-----

is reg
00000

7 bits

4 bits

5 bits

5 bits

11010

XTOP
1111

00

11 bits

Mnemonic

XTOP VI[it]
Operation
VI[it] = TOP
Reads the value of the TOP register from the external unit (VIF). The value is read at the T stage.
Flag Changes
MAC flag
status flag
Oxyzw Uxyzw Sxyzw Zxyzw DS IS OS US SS ZS D I O U S Z
------------- - - - - - - - - - - Throughput/Latency
1/1
Example

XTOP VI10
Remarks
XTOP can be used only on VU1.
This instruction cannot be placed in the E bit delay slot.

© SCEI
-198-

clipping
flag
-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5. Macro Mode
Macro mode is an operation mode that exists only for VU0. From the point of view of the EE Core, the VU
operates as a coprocessor (COP2) in this mode. Operation of the VU can be specified by coprocessor
instructions.
The VU can be used as an FPU by using coprocessor calculation instructions in macro mode. Basic floatingpoint instructions such as add, subtract, multiply, and divide are defined.
In addition, it is possible to transfer data to VU registers by using coprocessor transfer instructions, and to make
the VU execute micro subroutines by using micro subroutine execution instructions. Since the VU then enters
micro mode and operates independently, high performance can be realized.
This chapter describes the method of controlling VU1 from the EE Core as well as macro mode in VU0.

© SCEI
-199-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5.1. Macro Mode Register Set
5.1.1. Floating-Point Registers
The VU0 floating-point registers are allocated to the COP2 data register. The register fields etc. are the same as
those in the micro mode. For details, see "3.1.1. Floating-Point Registers".

CPR[2,00]
CPR[2,01]
CPR[2,02]
CPR[2,03]
CPR[2,31]

32 bits
32 bits
32 bits
32 bits
127
96 95
64 63
32 31
0
VF00w
VF00z
VF00y
VF00x
VF01w
VF01z
VF01y
VF01x
VF02w
VF02z
VF02y
VF02x
VF03w
VF03z
VF03y
VF03x
:
:
VF31w
VF31z
VF31y
VF31x

5.1.2. Integer Registers
The VU0 integer registers are allocated to the lower 16 bits of the first 16 COP2 control registers. For details of
the integer registers, see "3.1.2. Integer Registers".
31
CCR[2,00]
CCR[2,01]
CCR[2,02]
CCR[2,03]
CCR[2,04]
CCR[2,05]
CCR[2,06]
CCR[2,07]
CCR[2,08]
CCR[2,09]
CCR[2,10]
CCR[2,11]
CCR[2,12]
CCR[2,13]
CCR[2,14]
CCR[2,15]

0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000
0000000000000000

15
0
VI00 (0 register)
VI01
VI02
VI03
VI04
VI05
VI06
VI07
VI08
VI09
VI10
VI11
VI12
VI13
VI14 (Stack pointer: recommended)
VI15 (Link register: recommended)

5.1.3. Control Registers
Control registers such as flag registers are allocated to the last 16 COP2 control registers.
These control registers can be accessed only by the CTC2/CFC2 instruction, and cannot be accessed by other
macroinstructions and microinstructions.

© SCEI
-200-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

31

23

15
Status flag
MAC flag
clipping flag
reserved
R register
I register
Q register
reserved
reserved
reserved
TPC register
CMSAR0 register
FBRST register
VPU-STAT register
reserved
CMSAR1 register

CCR[2,16]
CCR[2,17]
CCR[2,18]
CCR[2,19]
CCR[2,20]
CCR[2,21]
CCR[2,22]
CCR[2,23]
CCR[2,24]
CCR[2,25]
CCR[2,26]
CCR[2,27]
CCR[2,28]
CCR[2,29]
CCR[2,30]
CCR[2,31]

7

0

The initial values and access limits of the control registers are shown in the table below. r* and w* indicate the
necessity of inserting a VNOP instruction considering the access timing.
Control register

Initial value

VU0 is stopped

status flag
MAC flag
clipping flag
R register
I register
Q register
TPC register
CMSAR0 register
FBRST register
VPU-STAT register
CMSAR1 register

all 0
all 0
all 0
indeterminate
indeterminate
indeterminate
indeterminate
indeterminate
all 0
all 0
indeterminate

r/w
r/r/w
r/w
r/w
r/w
r/r/w
r/w
r/-/w

VU0 is operating
(Running state)
r*/w*
r*/r*/w*
-/-/-/-/r/w
r/w
r/-/w

Writes to the CMSAR1 register are ignored while VU1 is operating.
Status flag
31

24

00000000

23

16

00000000

15

8

0000

7

0

D I O U S Z D I O U S Z
S S S S S S

The 12-bit status flag, which reflects VU0 calculation results, is allocated to the lower 12 bits of the control
register, CCR[2,16]. For details of the flag, see "3.3.2. Status Flags (SF)".
When accessing the status flag during VU0 operation, it is necessary to insert a VNOP instruction etc.
considering the timing. When performing a write operation, values written to the lower 6 bits (D, I, O, U, S, Z)
are ignored.

© SCEI
-201-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MAC flag
31

24

23

00000000

16

15

8

7

0

O O O O U U U U S S S S Z Z Z Z
x y z w x y z w x y z w x y z w

00000000

The 16-bit MAC flag, which reflects VU0 calculation results, is allocated to the lower 16 bits of the control
register, CCR[2,17]. For details of the flag, see "3.3.1. MAC Flags".
When accessing the MAC flag during VU0 operation, an instruction such as VNOP should be inserted in
consideration of access timing.
Clipping flag
31

24

23

16

15

8

7

0

- + - + - + - + - + - + - + - + - + - + - + - +
z z y y x x z z y y x x z z y y x x z z y y x x

00000000

The 24-bit clipping flag, which reflects VU0 clipping judgment results, is allocated to the lower 24 bits of the
control register, CCR[2,18]. For details of the flag, see "3.3.3. Clipping Flags (CF)".
When accessing the clipping flag during VU0 operation, an instruction such as VNOP should be inserted in
consideration of access timing.
R, I, Q register
31

24

00000000

23

16

15

0

8

7

0

R
I
Q

These registers are used for accessing the R, I, Q registers in VU0 during debugging.
TPC register
31

24

23

00000000

16

15

00000000

8

7

0

TPC

This register indicates the PC (Program Counter) at which the micro subroutine has stopped. The cause of the
stop can be checked with the VPU-STAT register.
CMSAR0 register
31

24

00000000

23

16

15

00000000

8

7

0

CMSAR0

This register specifies the starting address when attempting to make VU0 execute a micro subroutine by means
of the VCALLMSR instruction.

© SCEI
-202-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

FBRST register
31

24

23

16

15

8

7

0

T D R F
T D R F
E E S B
E E S B
00000000
00000000
0000
0000
1 1 1 1
0 0 0 0
This register controls the operation status of VU0 and VU1. The function of each bit is shown in the table
below.
Bit
FB0

Function
VU0 Force Break

RS0

VU0 Reset

DE0

VU0 D bit enable

TE0

VU0 T bit enable

FB1

VU1 Force Break

RS1

VU1 Reset

DE1

VU1 D bit enable

TE1

VU1 T bit enable

Write
0: 1: Force Break
0: 1: Reset
0: Disable
1: Enable
0: Disable
1: Enable
0: 1: Force Break
0: 1: Reset
0: Disable
1: Enable
0: Disable
1: Enable

Read
Always 0
Always 0
Current setting
Current setting
Always 0
Always 0
Current setting
Current setting

For details of the status of VU operations such as Force Break and Reset, see "1.3. VU Operation Status".
VPU-STAT register
31

24

23

16

15

8

7

0

E D V V V V V I
D
V V V V
I G F T D B B 0 I 0 F T D B
00000000
00000000
0 F
U V W S S S S S
V
S S S S
1 1 1 1 1 1 1 0
0
0 0 0 0
This is a read-only register, which reflects the operation status of VU0 and VU1. The bits are described in the
table below.

© SCEI
-203-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Bit
VBS0

Definition
VU0 operation status
0: idle (Stopped by E bit or Reset)
1: busy (Executing micro subroutine)
VU0 operation status
0: Operating or stopped by a factor other than D bit
1: Stopped by D bit
VU0 operation status
0: Operating or stopped by a factor other than T bit
1: Stopped by T bit
VU0 operation status
0: Operating or stopped by a factor other than Force Break
1: Stopped by Force Break
VU0 DIV unit operation status
0: idle
1: busy
VIF0 operation status
0: idle
1: busy
VU1 operation status
0: idle (Stopped by E bit or Reset)
1: busy (Executing micro subroutine)
VU1 operation status
0: Not D bit stop
1: Stopped by D bit
VU1 operation status
0: Not T bit stop
1: Stopped by T bit
VU1 operation status
0: Not Force Break stop
1: Stopped by Force Break
VU1 XGKICK related status
0: Not-waiting
1: Waiting for sync with external unit
VU1 DIV unit operation status
0: idle
1: busy
VU1 EFU unit operation status
0: idle
1: busy

VDS0
VTS0
VFS0
DIV0
IBS0
VBS1
VDS1
VTS1
VFS1
VGW1
DIV1
EFU1

CMSAR1 register
31

24

00000000

23

16

15

00000000

8

7

0

CMSAR1

This is the control register to activate a VU1 micro subroutine. By writing a 16-bit address (an address in byte
units in Micro Mem) to this register when VU1 is in the stopped state, VU1 starts micro subroutine execution
from this address. (In contrast to CMSAR0, a VU1 micro subroutine can be activated just with a write
operation.)

© SCEI
-204-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5.1.4. Special Registers
ACC, I, Q, and R registers in the macro mode function in the same way as those in the micro mode. See "3.1.
Micro Mode Register Set".
Since the macro instruction set does not have the I bit and the Lower instruction field, a coprocessor transfer
instruction is used to set the I register to a floating-point immediate value.

© SCEI
-205-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5.2. Macro Instruction Set Overview
5.2.1. MIPS COP2 Instructions
Macro instructions conform to the MIPS COP2 Instruction format. They are broadly divided into four
classifications:
• Coprocessor transfer instructions
• Coprocessor branch instructions
• Macro (primitive) calculation instructions
• Micro subroutine execution instructions
The VU macroinstructions start with the letter "V" to differentiate them from other MIPS instructions.

5.2.2. Coprocessor Transfer Instructions
Coprocessor transfer instructions are used to transfer data between VU0 registers and the EE Core.
When VU0 is executing a micro subroutine, the user can specify whether or not the EE Core operation
interlocks. For details, see the description of each instruction.
Category
Coprocessor
transfer
instruction

Instruction
QMFC2
QMTC2
LQC2
SQC2
CFC2
CTC2

Function
Floating-point data transfer from VU to EE Core
Floating-point data transfer from EE Core to VU
Floating-point data transfer from EE Core to VU
Floating-point data transfer from VU to EE Core
Integer data transfer from VU to EE Core
Integer data transfer from EE Core to VU

5.2.3. Coprocessor Branch Instructions
Coprocessor branch instructions branch by means of the MIPS COP2 condition signal. They are used to judge
VU1 status. For details, see the description of each instruction.
Category
Coprocessor
branch
instruction

Instruction
BC2F
BC2FL
BC2T
BC2TL

Function
Branch on COP2 condition signal
Branch on COP2 condition signal
Branch on COP2 condition signal
Branch on COP2 condition signal

5.2.4. Coprocessor Calculation Instructions
Coprocessor calculation instructions are the instructions to make VU0 execute floating-point calculations such
as add, subtract, multiply, and divide. Since they are a subset of the micro mode instructions, refer to Chapter 3
if necessary.

© SCEI
-206-

SCE CONFIDENTIAL

Category
Floating-point
calculation
instruction

Instruction
VABS
VADD
VADDi
VADDq
VADDbc
VADDA
VADDAi
VADDAq
VADDAbc
VSUB
VSUBi
VSUBq
VSUBbc
VSUBA
VSUBAi
VSUBAq
VSUBAbc
VMU
VMULi
VMULq
VMULbc
VMULA
VMULAi
VMULAq
VMULAbc
VMADD
VMADDi
VMADDq
VMADDbc
VMADDA
VMADDAi
VMADDAq
VMADDAbc
VMSUB
VMSUBi
VMSUBq
VMSUBbc
VMSUBA
VMSUBAi
VMSUBAq
VMSUBAbc
VMAX
VMAXi
VMAXbc
VMINI
VMINIi
VMINIbc
VOPMULA
VOPMSUB
VNOP

VU User's Manual Version 6.0

Function
absolute
addition
ADD broadcast I register
ADD broadcast Q register
ADD broadcast bc field
ADD output to ACC
ADD output to ACC broadcast I register
ADD output to ACC broadcast Q register
ADD output to ACC broadcast bc field
subtraction
SUB broadcast I register
SUB broadcast Q register
SUB broadcast bc field
SUB output to ACC
SUB output to ACC broadcast I register
SUB output to ACC broadcast Q register
SUB output to ACC broadcast bc field
multiply
MUL broadcast I register
MUL broadcast Q register
MUL broadcast bc field
MUL output to ACC
MUL output to ACC broadcast I register
MUL output to ACC broadcast Q register
MUL output to ACC broadcast bc field
MUL and ADD (SUB)
MUL and ADD (SUB) broadcast I register
MUL and ADD (SUB) broadcast Q register
MUL and ADD (SUB) broadcast bc field
MUL and ADD (SUB) output to ACC
MUL and ADD (SUB) output to ACC broadcast I register
MUL and ADD (SUB) output to ACC broadcast Q register
MUL and ADD (SUB) output to ACC broadcast bc field
Multiply and SUB
Multiply and SUB broadcast I register
Multiply and SUB broadcast Q register
Multiply and SUB broadcast bc field
Multiply and SUB output to ACC
Multiply and SUB output to ACC broadcast I register
Multiply and SUB output to ACC broadcast Q register
Multiply and SUB output to ACC broadcast bc field
maximum
maximum broadcast I register
maximum broadcast bc field
minimum
minimum broadcast I register
minimum broadcast bc field
outer product MULA
outer product MSUB
no operation

© SCEI
-207-

SCE CONFIDENTIAL

Category
Floating-point/
fixed point
conversion
instructions

Clipping
judgment
instruction
Floating-point
divider
instructions
Integer calculation
instructions

Register-register
transfer
instructions

VU User's Manual Version 6.0

Instruction
VFTOI0
VFTOI4
VFTOI12
VFTOI15
VITOF0
VITOF4
VITOF12
VITOF15
VCLIP

Function
float to integer, fixed point 0 bit
float to integer, fixed point 4 bits
float to integer, fixed point 12 bits
float to integer, fixed point 15 bits
integer to float, fixed point 0 bit
integer to float, fixed point 4 bits
integer to float, fixed point 12 bits
integer to float, fixed point 15 bits
clipping

VDIV
VSQRT
VRSQRT
VIADD
VIADDI
VIAND
VIOR
VISUB
VMOVE
VMFIR

floating divide
floating square-root
floating reciprocal square-root
integer ADD
integer ADD immediate
integer AND
integer OR
integer SUB
move floating register
move from integer register

VMTIR
VMR32
Load/Store
VLQD
instructions
VLQI
VSQD
VSQI
VILWR
VISWR
Random numbers VRINIT
VRGET
VRNEXT
VRXOR
Synchronization VWAITQ
instruction

move to integer register
rotate right 32 bits
Load Quadword with pre-decrement
Load Quadword with post-increment
Store Quadword with pre-decrement
Store Quadword with post-increment
integer load word register
integer store word register
random-unit init R register
random-unit get R register
random-unit next M sequence
random-unit XOR R register
wait Q register

5.2.5. Micro Subroutine Execution Instructions
Micro subroutine execution instructions execute microinstruction programs in the Micro Mem. For details, see
the description of each instruction.
Category
Instruction
Micro subroutine VCALLMS
execution
VCALLMSR
instruction

Function
Call micro subroutine
Call micro subroutine register

© SCEI
-208-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5.3. Flags
Functions and operations of MAC flags, status flags, and clipping flags in the macro mode are the same as those
in the micro mode. See "3.3.1. MAC Flags", "3.3.2. Status Flags (SF)", and "3.3.3. Clipping Flags (CF)"
respectively.
Since there is no flag set instruction in the macro mode, the CFC2 instruction is used to set a value to the flag.

© SCEI
-209-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5.4. Macro Mode Pipeline
5.4.1. Pipeline Structure of Macroinstructions
Figure 5-1 illustrates the pipeline structure of macroinstructions.
General macroinstruction

I

Q

R

A

T

X

Y

Z

S

VDIV/VSQRT

I

Q

R

A

T

D1

D2

D3

D4

D5

D6

F

VRSQRT

I

Q

R

A

T

D1

D2

D3

D4

D5

D6

D7

VCALLMS
/VCALLMSR

I

Q

R

A

D

M

T

X

Y

Z

S

T

X

Y

Z

S

M

T

X

Y

Z

Ms Ms Ms Ms

W

S

M

EE core instruction

I

Q

R

A

D

W

QMFC2/CFC2

I

Q

R

T

D

W

SQC2

I

Q

R

T

D

W

QMTC2/CTC2

I

Q

R

A

D

W

S

LQC2

I

Q

R

A

D

W

S

LQC2 (cache miss)

I

Q

R

A

D

I, Q
R
A
W
T
D
M
X -Z
S
D1-

D8

D9

DA

DB

DC

F

(E bit)
S

: instruction fetch(EE Core)
: instruction decode, read GPR(EE Core)
: execute(EE Core)
: write(EE Core)
: read VGPR
: data transaction between EE Core and VPU
: read micro-Mem
: execution stage
: store VGPR
: FDIV execution stage

Figure 5-1 Macroinstruction Pipeline List

5.4.2. Hazards in Macro Mode
The following situations can generate stalls in macro mode:
DIV resource hazard
When a macroinstruction that uses the floating-point divider unit (VDIV/VSQRT/VRSQRT) is being
executed, and another macroinstruction of this type is executed concurrently.

VDIV
VDIV

I

Q

R

A

D

W

I

Q

R

A

T

D1 D2 D3 D4 D5 D6

I

Q

R

As As As As As As

A

T

X

I

Q

R

As As As As As As

A

D

W

F
Y

Z

S

D cache miss
When a D cache miss occurs, the LQC2/SQC2 instruction stalls at the M stage. Since this is a blocking load,
the next instruction stalls at the same time.
© SCEI
-210-

SCE CONFIDENTIAL

LQC2

VU User's Manual Version 6.0

I

Q

R

A

D

W

I

Q

R

A

D

M

M

M

M

M

M

W

S

(core)

pipe 1
pipe 0

Data hazard
An instruction that uses the value in a register is executed before instruction results are written to the register
(RAW: Read After Write Hazard), or the next instruction writes the results to the same register before
instruction results are written to the register (WAW: Write After Write Hazard).
In macro mode, register data dependency checks are performed in units of register numbers. Unlike micro
mode, no distinction is made between the respective fields of the floating-point register. Moreover, the
floating-point register and integer register of the same number are regarded as the same register, and may
cause hazards occasionally.
VF00 and VI00 are not subject to hazard checks.
Data hazards do not occur to the special registers (ACC/I/Q/R). For the Q register, however,
synchronization is enabled by means of the VWAITQ instruction.

5.4.3. Macroinstruction Operation
The following figure illustrates the normal macroinstruction pipeline operation.

COP2M
COP2M

I

Q
I

R
Q

A
A

T
R

X
T

Y
X

Z
Y

S
Z

S

When there is a data hazard between macroinstructions, stalls occur as below.

COP2M
COP2M

I

Q
I

R
Q

A
R

T X Y
As As As

Z
A

S
T

X

Y

Z

S

5.4.4. Operation when Transferring Data with EE Core
After a QMTC2/CTC2/LQC2 instruction which transfers data from the EE Core to the VU, the next
macroinstruction always stalls for one cycle. This allows the next macroinstruction to use the transferred data.

QMTC2 VF1, r4
COP2M VF3, VF2, VF1

I

Q
I

R
Q

A
R

D
A

W
Ts

S
T

X

Y

Z

S

When a D cache miss is generated with the LQC2 instruction, stalls occur as illustrated below.
© SCEI
-211-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

I

LQC2 VF1, 0(r4)
COP2M VF3, VF2, VF1

Q
I

R
Q

A
R

D
A

M M M W
Ts Ts Ts Ts

S
T

X

Y

Z

S

If the output register is the same when transferring data to the VU in the QMTC2/CTC2/LQC2 instruction
following a macroinstruction, a WAW hazard occurs and results in stalls as illustrated below.
No WAW hazard

COP2M
QMTC2

I
I

Q
Q
I
I

R
R
Q
Q

A
A
R
R

D
T
A
A

W
X
D
D

I
I

Q
Q
I

R
R
Q

A
A
R

D W
T X Y Z
A Ds Ds D

S
W

I

Q

R

A

W

Y
W
W

Z
S

pipe 1
pipe 0
pipe 1
pipe 0

S

WAW hazard

COP2M
QMTC2

Ds Ds D

S

When transferring data successively with the QMTC2/CTC2/LQC2 instruction, if there is not a D cache miss,
stalls do not occur.
No D cache miss

QMTC2
QMTC2

I

Q

R

A

D

W

I

Q

R

A

D

W

S

I

Q

R

T

D

W

I

Q

R

A

D

W

Q
Q
I
I

R
R
Q
Q

A
A
R
R

D W
D M M M M W
A Ds Ds Ds Ds D
A Ds Ds Ds Ds D

S

D cache miss

LQC2
QMTC2

I
I

S
W
W

S

When data is transferred from the VU to the EE Core with the QMFC2/CFC2/SQC2 instruction, stalls do not
occur normally as illustrated below.
© SCEI
-212-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

I
I

QMFC2
COP2M

Q
Q
I

R
R
Q

A
T
R

D
D
A

W
W
T

X

I

Q

R

A

D

W

Y

Z

S

Even when executing QMFC2/CFC2/SQC2 instructions successively, no stalls occur as long as the SQC2
instruction does not overflow the store buffer.
Normal conditions

QMFC2
QMFC2

I

Q

R

T

D

W

I

Q

R

A

D

W

I
I

Q
Q

R
R

T
A

D
D

Q
Q
I
I

R
R
Q
Q

T
A
R
R

D W
D W
T Ds Ds Ds Ds D
A Ds Ds Ds Ds D

W
W

Store buffer full

QMFC2
SQC2

I
I

W
W

When transferring macroinstruction results to the EE Core with the QMFC2/CFC2/SQC2 instruction, stalls
occur as illustrated below.

COP2M
QMFC2

I

Q
I

R
Q

A
R

T X Y Z
Ts Ts Ts Ts

S
T

D

W

The flow of transferring data from the EE Core, processing it with a macroinstruction, and returning it to the
EE Core is summarized in the figure below:
QMTC2 r4, VF1
COP2M VF3, VF4, VF1
QMFC2 r9, VF3

I

Q

R

A

D

W

I

Q

R

A

D

W

S

I

Q

R

As

A

T

Z

S

I

Q
I

R Ts Ts Ts Ts Ts Ts
Q Rs Rs Rs Rs Rs Rs

T
R

X

Y

D
A

W
D

To read a VI register that is incremented/decremented by a VLQI/VLQD/VSQI/VSQD instruction, using the
CFC2 instruction, the program should be written to enable the following pipeline operation:

© SCEI
-213-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VLQI VF1, (VI1++)
VNOP
VNOP
VNOP
VNOP
VNOP
CFC2 r2, VI1

I

Q
I

R
Q
I

A
R
Q
I

T
A
R
Q
I

X
T
A
R
Q
I
I

Y
X
T
A
R
Q
Q

Z
Y
X
T
A
R
R

S
Z
Y
X
T
A
T

S
Z
Y
X
T
D

S
Z
Y
X
W

S
Z
Y
W

S
Z

S

pipe 0
pipe 0
pipe 0
pipe 0
pipe 0
pipe 0
pipe 1

Furthermore, if there is a RAW hazard when reading data with QMFC2/CFC2/SQC2 immediately after the data
is transferred from the EE Core with QMTC2/CTC2/LQC2, stalls occur for two cycles.
No RAW hazard
I
I

QMTC2
QMFC2

Q
Q

R
R

A
A

D
D

W
W

S

I

Q

R

T

D

W

I

Q

R

A

D

W

Q
Q

R
R

A
A

D
D

W
W

S

I

Q

R

Ts Ts

T

D

W

I

Q

R

As As

A

D

W

RAW hazard

I
I

QMTC2
QMFC2

Conversely, QMTC2/CTC2/LQC2 after QMFC2/CFC2/SQC2 does not stall as long as a D cache miss is not
generated or the store buffer is not full.
Normal conditions
QMFC2

I
I

QMTC2

Q
Q
I
I

R
R
Q
Q

T
A
R
R

D
D
A
A

W
W
D
D

Q
Q
I
I

R
R
Q
Q
I
I

T
A
R
R
Q
Q

D
D
A
A
R
R

W
W
D M M M M Z
D W
A Ds Ds Ds Ds D
A Ds Ds Ds Ds D

W
W

S

D cache miss
QMFC2
LQC2

I
I

© SCEI
-214-

S
W
W

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Status flag read/write
The following figure shows an example in which the VU0 status flag (CCR[2,16] register) is read with the
CFC2 instruction.

COP2M
VNOP
VNOP
VNOP
VNOP
VNOP
CFC2 r0, VI16

I

Q
I

R
Q
I

A
R
Q
I

T
A
R
Q
I

X
T
A
R
Q
I
I

Y
X
T
A
R
Q
Q

Z
Y
X
T
A
R
R

S
Z
Y
X
T
A
Ts

S
Z
Y
X
T
T

S
Z
Y
X
D

S
Z
Y
W

S
Z

pipe 0
pipe 0
pipe 0
pipe 0
pipe 0
pipe 0
pipe 1

S

The following figure shows an example in which the status flag is written by means of the CTC2 instruction.

COP2M
VNOP
VNOP
CTC2 r0, VI16

I

Q
I

R
Q
I
I

A
R
Q
Q

T
A
R
R

X
T
A
As

Y
X
T
A

Z
Y
X
D

S
Z
Y
W

S
Z
S

pipe 0
pipe 0
pipe 0
pipe 1

S

5.4.5. Operation when Executing a Micro Subroutine
A VCALLMS instruction executed following a macroinstruction always stalls for one cycle. This enables the
micro subroutine to use the results of the preceding macroinstruction.

COP2M

I

VCALLMS

Q

R

A

T

X

Y

Z

S

I

Q

R

A

Ds D

M

T
M

X
T

Y
X

Z
Y

S
Z

S

A macroinstruction executed following VCALLMS stalls at the T stage overlapping with the S stage of the last
microinstruction. This enables the subsequent macroinstruction to use the results of the micro subroutine.
VCALLMS

COP2M

I

Q

I

R

Q

A

R

D

M

T
M

X
T

Y
X

Z
Y

S
Z

S

M

T

X

Y

Z

S

As As As As As As As

A

T

E bit

X

Y

...

When two VCALLMS instructions are executed successively, the second VCALLMS stalls while the first
VCALLMS is executing the microinstruction. If there is a data hazard between the called microinstructions,
stalls are generated in the microinstruction.

© SCEI
-215-

SCE CONFIDENTIAL

VCALLMS

VU User's Manual Version 6.0

I

Q

VCALLMS

VCALLMS

I

I

Q

VCALLMS

I

R

Q

R

Q

A

R

A

R

D

A

D

A

M

Z
Y
X

S
Z
Y

Ds Ds Ds Ds D

M

M

T
M

T
M

X
T
M

X
T
M

Y
X
T

Y
X
T

Z
Y
X

Ds Ds Ds Ds D

S
Z
Y

S
Z
T
M

S
Z

E bit
S
X
T
M

Y
X
T

Z
Y
X

S
Z
Y

S
Z

S

Y
X
T

Z
Y
X

S
Z
Y

S
Z

E bit

E bit
S

M Ts T
ls M
‚

X
T
M

E bit
S

5.4.6. Micro Subroutine and Data Transfer Operations
The VCALLMS instruction does not stall after transferring data from the EE Core with QMTC2/CTC2/LQC2,
if there is no D cache miss. However, the transferred data can be used for the micro subroutine as illustrated
below.
Normal conditions

QMTC2
VCALLMS

I

Q
I

R
Q

A
R

D
A

W
D

S
M

T
M

X
T
M

Y
X
T

Z
Y
X

S
Z
Y

S
Z

S

X
T
M

Y
X
T

D cache miss with LQC2

LQC2
VCALLMS

I

Q
I

R
Q

A
R

D
A

M M M M W
Ds Ds Ds Ds D

S
M

I

Q

R

A

Ds Ds Ds Ds

W

D

T
M

Z
Y
X

S
Z
Y

When transferring the micro subroutine results to the EE Core, a QMFC2.i/CFC2.i instruction with interlocks
is used. QMFC2.i/CFC2.i stalls in the same way as the macroinstruction that follows VCALLMS, and can read
the calculation results of the final instruction of the micro subroutine.

© SCEI
-216-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VCALLMS

I

Q

QMFC2.i

R

I

A

Q

R

D

M

T
M

X
T
M

Y
X
T

Z
Y
X

S
Z
Y

E bit

S
Z

S

Ts Ts Ts Ts Ts Ts Ts Ts

T

D

W

Examples of transferring data from the EE Core, executing a micro subroutine, and reading the results are
shown below.
QMTC2 VF1, r4
VCALLMS

I

Q

R

A

D

W

I

Q
I

R
Q

A
R

D
A

W
D

QMFC2.i

I

S
M

T

X

Y

Z

S

M

T

X

Y

Z

E bit

S

Q

R

M T X Y Z
Ts Ts Ts Ts Ts Ts Ts Ts Ts

I

Q

Rs Rs Rs Rs Rs Rs Rs Rs Rs

S
T

D

W

R

A

D

The interlock-free QMFC2.ni/CFC2.ni/SQC2 instruction is executed independently of the preceding
VCALLMS instruction and does not stall. This method can be used to read the data, which is irrelevant to the
micro subroutine.
VCALLMS

I

Q

QMFC2.ni

I

R

Q

A

R

D

T

M

D

T
M

X
T
M

Y
X
T

Z
Y
X

S
Z
Y

S
Z

E bit
S

W

When transferring data to the VU with QMTC2/CTC2/LQC2 during execution of a micro subroutine,
operation is determined by the specification of interlock and the M bit of the microinstruction.
When transferring data with LQC2 and QMTC2.ni/CTC2.ni without interlocks, no stalls are generated, as
shown below. This method can be used to write registers irrelevant to the micro subroutine.
VCALLMS

I

QMTC2.ni

Q

I

R

Q

A

R

D

A

M

D

T

X

Y

Z

M

T

X

Y

Z

S

M

T

X

Y

Z

W

S

E bit
S

S

When transferring data with QMTC2.i/CTC2.i with interlocks, stalls occur and continue until a
microinstruction, which sets the M bit to 1, is executed in the micro subroutine. This means that prohibition
and cancellation of data transfer to the VU can be controlled on the micro subroutine side. In the following
example, stalls are cancelled by setting the M bit in the second microinstruction to 1.

VCALLMS

QMTC2.i

I

Q

I

R

Q

A

R

D

M

T
M

X
T
M

Y
X
T

Z
Y
X

S
Z
Y

S
Z

As As As As

A

D

W

S

M:0
E bit M:1
S
M:1

© SCEI
-217-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Using this method, the next input data can be set during execution of the micro subroutine and performance can
be improved.
Once stalls are cancelled, set the M bit in all the instructions to 1 until the completion of the micro subroutine.
Operations that repeat prohibition and cancellation of data transfer are undefined.
VCALLMSR instruction execution timing
The VCALLMSR instruction is executed after writing the micro subroutine address in the CMSAR0 register
with the CTC2 instruction. Execute this with the following timing:
CTC2 CMSAR0
VNOP
VNOP
VCALLMSR

I

Q

R

A

D

S

I

Q

R

T

X

Y

Z

S

I

Q
I

R
Q

T
R

X
A

Y
M

Z
T

S
X

Y

Z

S

M

T
M

X
T

Y
X

Z
Y

S
Z

S

5.4.7. Q Register Synchronization
The VWAITQ instruction stalls until the VDIV/VSQRT/VRSQRT instruction executed most recently
completes. Insertion of this VWAITQ enables the next macroinstruction to use the results of
VDIV/VSQRT/VRSQRT.
VDIV Q, VF2, VF1
VWAITQ
VMULq VF5, VF4, Q

I

Q

R

A

T D1 D2 D3 D4 D5 D6

F

I

Q
I

R
Q

As As As As As As
Rs Rs Rs Rs Rs Rs

T
A

A
R

X
T

Y
X

Z
Y

S
Z

S

There are many factors that stop the instruction stream, such as cache miss and interrupt in micro mode. So if
the Q register is referred to following an instruction such as VDIV, calculation results are not guaranteed to be
correct. Therefore, perform synchronization with VWAITQ before accessing the Q register.

5.4.8. Notes on Other Pipeline Operations
W stage kill
W stage kill occurs when EE Core exceptions are generated. All the instructions operating concurrently are
stopped by the W stage kill. However, COP2 divide instructions (VDIV, VSQRT, and VRSQRT) continue
operation. For this reason, if a COP2 divide instruction follows an instruction to read the Q register within
two cycles, it may overwrite the Q register value to be read at the occurrence of an exception. To avoid this,
place two VNOP or five NOP instructions between the instruction to read the Q register and a COP2 divide
instruction.

© SCEI
-218-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

W stage kill
ADD r1, r4, r5
VDIV
Pipe1 any
Pipe0 any

I
I

Q
Q

R
R
I
I

A
A
Q
Q

D
nullify
T D1 D2 D3 D4 D5 D6
R
nullify
R
nullify

pipe 1
pipe 0
pipe 1
pipe 0

F

D stage stall
D stage stall occurs as a cache resource hazard when D cache misses are overlapped. The pipeline operation
is shown below. The VDIV instruction is executed regardless of the D stage stall.
LW r1, 0(r2)
COP2M
COP2M

I
I

LW r1, 0(r2)
VDIV

I
I

Q
Q
I

R
R
Q

A
A
R

Ds Ds Ds Ds D
Ts Ts Ts Ts T
As As As As A

W
X
T

Q
Q

R
R

A
A

Ds Ds Ds Ds D W
T D1 D2 D3 D4 D5 D6 F

Y
X

Z
Y

S
Z

S

pipe 1
pipe 0
pipe 0

pipe 1
pipe 0

© SCEI
-219-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5.5. VU1 Control
Since VU1 is not connected to the EE Core via a coprocessor connection, it does not have macro mode. VU1
is controlled by the registers, which are mapped to VU0 data memory (VU Mem0), and some other control
registers.

5.5.1. MIPS COP2 Condition Signal
The COP2 condition signal becomes TRUE when a VU1 micro subroutine is operating, and FALSE when not
operating. By performing polling with instructions such as BC2T, VU1 operation status can be checked.

5.5.2. MIPS COP2 Control Register
As described in "5.1.3. Control Registers", the VU control registers are allocated to the last 16 COP2 control
registers. The registers related to VU1 are shown below.
FBRST register
31

24

23

00000000

16

15

00000000

8

0000

7

T D R F
E E S B
1 1 1 1

0

0000

T D R F
E E S B
0 0 0 0

This register controls the operation status of VU0 and VU1. The function of each bit related to VU1 is shown
in the table below.
Bit
FB1

Function
VU1 Force Break

Write
Read
0: Always 0
1: Force Break
RS1
VU1 Reset
0: Always 0
1: Reset
DE1
VU1 D bit enable
0: Disable
Current setting
1: Enable
TE1
VU1 T bit enable
0: Disable
Current setting
1: Enable
For details of the status of VU operations such as Force Break and Reset, see "1.3. VU Operation Status".
VPU-STAT register
31

24

23

16

15

8

7

0

V V V V
D
E D V V V V V I
F T D B
I
F I G F T D B B
00000000
00000000
0 U V W S S S S S 0 V 0 S S S S
0 0 0 0
0
1 1 1 1 1 1 1 0
This is a read only register, which reflects the operation status of VU0 and VU1. The bits related to VU1 are
described in the table below.

© SCEI
-220-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Bit
VBS1

Definition
VU1 operation status
0: idle (Stopped by E bit or Reset)
1: busy (Executing micro subroutine)
VU1 operation status
0: Operating or stopped by a factor other than D bit
1: Stopped by D bit
VU1 operation status
0: Operating or stopped by a factor other than T bit
1: Stopped by T bit
VU1 operation status
0: Operating or stopped by a factor other than Force Break
1: Stopped by Force Break
VU1 XGKICK-related status
0: Not-waiting
1: Waiting for sync with external unit
VU1 DIV unit operation status
0: idle
1: busy
VU1 EFU unit operation status
0: idle
1: busy

VDS1
VTS1
VFS1
VGW1
DIV1
EFU1

CMSAR1 register
31

24

23

00000000

16

15

00000000

8

7

0

CMSAR1

This is the control register to activate a VU1 micro subroutine. By writing a 16-bit address (an address in byte
units in Micro Mem) to this register when VU1 is in the stopped state, VU1 starts micro subroutine execution
from this adress.

5.5.3. Floating-Point Registers
VU1 floating-point registers are mapped to VU Mem0 as shown below. These registers are readable/writable
only when VU1 is stopped. Access while VU1 is operating causes an indeterminate result.
VU Mem0
Address
0x4000
0x4010
0x4020
0x4030
0x41F0

32 bits
32 bits
32 bits
32 bits
127
96 95
64 63
32 31
0
VF00w
VF00z
VF00y
VF00x
VF01w
VF01z
VF01y
VF01x
VF02w
VF02z
VF02y
VF02x
VF03w
VF03z
VF03y
VF03x
:
:
VF31w
VF31z
VF31y
VF31x

© SCEI
-221-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

5.5.4. Integer Registers
VU1 integer registers are mapped to the VU Mem0 as below. These registers are readable/writable only when
VU1 is stopped. Access while VU1 is operating causes an indeterminate result.
VU Mem0
Address
0x4200
0x4210
0x4220
0x4230
0x4240
0x4250
0x4260
0x4270
0x4280
0x4290
0x42A0
0x42B0
0x42C0
0x42D0
0x42E0
0x42F0

127

15

all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0

0

VI00 ( 0 register)
VI01
VI02
VI03
VI04
VI05
VI06
VI07
VI08
VI09
VI10
VI11
VI12
VI13
VI14 (Stack pointer recommended)
VI15 (Link register recommended)

5.5.5. Control Registers
VU1 flag registers and special registers are mapped to the VU Mem0 as below. These registers are
readable/writable only when VU1 and the GIF are both stopped. Access while either of them is operating
causes an indeterminate result.
VU Mem0
address
0x4300
0x4310
0x4320
0x4330
0x4340
0x4350
0x4360
0x4370
0x4380
0x4390
0x43A0
0x43B0
0x43C0
0x43D0
0x43E0
0x43F0

127

31

all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0
all 0

0

Status flag
MAC flag
Clipping flag
Reserved
R register
I register
Q register
P register
Reserved
Reserved
TPC
Reserved
Reserved
Reserved
Reserved
Reserved

© SCEI
-222-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

The initial value and access limit of each register are shown in the table below.
Control register
Status flag
MAC flag
Clipping flag
R register
I register
Q register
P register
TPC register

Initial Value
all 0
all 0
all 0
Indeterminate
Indeterminate
Indeterminate
Indeterminate
Indeterminate

VU1/GIF is stopped
r/w
r/r/w
r/w
r/w
r/w
r/w
r/-

VU1/GIF is operating
-/-/-/-/-/-/-/-/-

Status flag: VU Mem0 0x4300
127

16

15

all 0

8

0000

7

0

D I O U S Z
D I O U S Z
S S S S S S

The status flag (12-bit), which reflects VU1 calculation results, is allocated to the lower 12 bits. When
performing a write operation, values written to the lower 6 bits are ignored.
MAC flag: VU Mem0 0x4310
127

16

15

8

7

0

O O O O U U U U S S S S Z Z Z Z
x y z w x y z w x y z w x y z w

all 0

The MAC flag (16-bit), which reflects VU1 calculation results, is allocated to the lower 16 bits.
Clipping flag: VU Mem0 0x4320
127

24

23

16

15

8

7

0

- + - + - + - + - + - + - + - + - + - + - + - +
z z y y x x z z y y x x z z y y x x z z y y x x

all 0

The clipping flag (24-bit), which reflects VU1 clipping judgment results, is allocated to the lower 24 bits.
R, I, Q, P register: VU Mem0 0x4340, 0x4350, 0x4360, 0x4370
127

32

31

24

all 0

23

0

0

R register

all 0

I register

all 0

Q register

all 0

P register

These registers are used for accessing the R, I, Q, P registers during debugging.
TPC register: VU Mem0 0x43A0
127

16

15

all 0

0

TPC

This register indicates the PC (Program Counter) in which the micro subroutine has stopped.
© SCEI
-223-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

(This page is left blank intentionally)

© SCEI
-224-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

6. Macro Mode Instruction Reference

© SCEI
-225-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

6.1. Macro Instruction Operation Code
6.1.1. Macro Instruction Operation Type
There are seven types of macro instructions:
MacroOP field type 0
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Specifies three registers and the broadcast field.
MacroOP field type 1
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
------

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Specifies three registers.
MacroOP field type 2
32-bit word: MacroOP field type 2
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

OPCODE?
----1111
9 bits

bc
-2 bits

Specifies two registers and the broadcast field.
MacroOP field type 3
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

-----

OPCODE
1111

--

11 bits

Specifies two registers.
MacroOP field type 4
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 -1

fsf
--

ft reg
-----

fs reg
-----

5 bits

5 bits

2 bits 2 bits

Specifies one field each of two registers.

© SCEI
-226-

OPCODE
----1111
11 bits

--

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MacroOP field type 5
32-bit word: MacroOP field type 5
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

it reg
-----

is reg
-----

Imm5
-----

OPCODE
------

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Specifies two registers and a 5-bit immediate value.
MacroOP field type 6
32-bit word: MacroOP field type 6
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

Imm15
---------------

OPCODE
------

6 bits

1

4 bits

15 bits

6 bits

Specifies a 15-bit immediate value.
MacroOP field type 11
32-bit word: MacroOP field type 11
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

opcode
0----

rt reg
-----

id reg
-----

---------00 0000 0000

I
-

6 bits

5 bits

5 bits

5 bits

10 bits

1

Coprocessor transfer instructions, e.g. CFC2. The I bit (bit 0) specifies whether or not to wait for the
completion of the preceding VCALLMS instruction. If the I bit is set to1, the EE Core stalls, and starts
transferring upon completion of the micro sub-routine.
MacroOP field type 12
32-bit word: MacroOP field type 12
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

opcode
-----

base
-----

ft reg
-----

offset
----------------

6 bits

5 bits

5 bits

16 bits

Branch instruction using COP2 conditional signal and transfer instruction for main memory and coprocessor
register.

6.1.2. Macro Instruction Operation Field
This section explains the operation fields used in operation codes.
dest field
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE?
----

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

The dest field specifies the FMAC units to be operated in parallel: that is, the x, y, z or w fields of the 128-bit
data to be operated on. Bits 24 through 21 can be specified independently. When the bit is set to 1, the
corresponding FMAC unit/field becomes effective.

© SCEI
-227-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Bit
24
23
22
21

Corresponding FMAC/field
x
y
z
w

bc field
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

OPCODE
----

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

The bc field specifies the field to be broadcasted for broadcast-series instructions, as follows:
Specified value of bc field
00
01
10
11

Broadcast field
x
y
z
w

fsf/ftf field
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 -1

fsf
--

2 bits 2 bits

ft reg
-----

fs reg
-----

5 bits

5 bits

-----

OPCODE
1111

--

11 bits

The combinations of the fsf field with the fs reg field and the ftf field with the ft reg field specify the field to be
calculated by the instruction. Bits 22 and 21 are used for the fsf field, and bits 24 and 23 for the ftf field.
Specified value for fsf/ftf field
00
01
10
11

Field to be an arithmetic object
x
y
z
w

© SCEI
-228-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

6.2. Macro Instruction Set
This section describes the function, operation code, mnemonic, operation, flag changes, and throughput/latency
of macroinstructions. They are listed in alphabetical order in mnemonic form. The descriptions also include
examples, programming notes, and reference information.

© SCEI
-229-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

BC2F : Branch on COP2 Conditional Signal
Branches if COP2 conditional signal is FALSE, i.e., branches if VU1 is not active.
Operation Code
32-bit word: MacroOP field type 12
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

BC
01000

BCF
00000

offset
----------------

6 bits

5 bits

5 bits

16 bits

Mnemonic

BC2F offset
Operation
If the COP2 condition signal (CPCOND) sampled during execution of the previous instruction is FALSE,
the program branches to the specified PC relative address. The branch delay slot is one instruction.
The branch target address is obtained by adding a signed 16-bit offset multiplied by 4 to the branch delay
slot.
Throughput/Latency
2/2
Example
// no operation during VU execution

vu_run:
bc2f

vu_idle

nop
j vu_run
nop
vu_idle:
(The rest is omitted.)

© SCEI
-230-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

BC2FL : Branch on COP2 Conditional Signal
Branches if COP2 conditional signal is FALSE, i.e. ,VU1 is not activated.
Operation Code
32-bit word: MacroOP field type 12
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

BC
01000

BCFL
00010

offset
----------------

6 bits

5 bits

5 bits

16 bits

Mnemonic

BC2FL offset
Operation
Branches to the specified PC-relative address, and one instruction delay occurs, if the COP2 condition signal
(CPCOND) sampled during execution of the previous instruction is FALSE. If a conditional branch does
not occur, the instruction in the branch delay slot is nullified.
The branch target address is obtained by adding the signed 16-bit offset multiplied by 4 to the address of the
instruction in the branch delay slot.
Throughput/Latency
2/2
Example
// no operation during VU execution

vu_run:
bc2fl vu_idle
nop
j vu_run
nop
vu_idle:
(The rest is omitted.)

© SCEI
-231-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

BC2T : Branch on COP2 Conditional Signal
Branches if COP2 conditional signal is TRUE, i.e., VU1 is activated.
Operation Code
32-bit word: MacroOP field type 12
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

BC
01000

BCT
00001

offset
----------------

6 bits

5 bits

5 bits

16 bits

Mnemonic

BC2T offset
Operation
Branches to the specified PC-relative address, and one instruction delay occurs, if the COP2 condition signal
(CPCOND) sampled during execution of the previous instruction is TRUE.
The branch target address is obtained by adding the signed 16-bit offset multiplied by 4 to the address of the
instruction in the branch delay slot.
Throughput/Latency
2/2
Example
// no operation while VU is stopped

vu_idle:
bc2t

vu_run

nop
j vu_idle
nop
vu_run:
(The rest is omitted.)

© SCEI
-232-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

BC2TL : Branch on COP2 Conditional signal
Branches if COP2 conditional signal is TRUE, i.e., VU1 is activated.
Operation Code
32-bit word: MacroOP field type 12
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

BC
01000

BCTL
00011

offset
----------------

6 bits

5 bits

5 bits

16 bits

Mnemonic

BC2TL offset
Operation
Branches to the specified PC-relative address, and one instruction delay occurs, if the COP2 condition signal
(CPCOND) sampled during execution of the previous instruction is TRUE. If a conditional branch does
not occur, the instruction in the branch delay slot is nullified.
The branch target address is obtained by adding the signed 16-bit offset multiplied by 4 to the address of the
instruction in the branch delay slot.
Throughput/Latency
2/2
Example
// no operation while VU is stopped

vu_idle:
bc2tl vu_run
nop
j vu_idle
nop
vu_run:
(The rest is omitted.)

© SCEI
-233-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

CFC2 : Transfer Integer Data from VU to EE Core
Transfers the content of CCR[2,id] to GPR[rt].
Operation Code
32-bit word: MacroOP field type 11
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

CFC2
00010

rt reg
-----

id reg
-----

---------00 0000 0000

I
-

6 bits

5 bits

5 bits

5 bits

10 bits

1

Mnemonic

CFC2
rt, id
CFC2.NI rt, id
CFC2.I rt, id

(I=0: Without interlock)
(I=0: Without interlock)
(I=1: With interlock)

Operation
Transfers the sign-extended 32-bit data of CCR[2,id], the COP2 control register in which the VU0
integer/control register is mapped, to GPR[rt], a CPU general purpose register. (However, the integer
registers VI00 through VI15 are 16 bits and are mapped to the lower 16 bits of CCR[2,0] through CCR[2,15],
therefore, these registers are in fact not sign-extended.)
When interlocking is specified, the CFC2 instruction stalls until the previously executed VCALLMS (micro
subroutine) instruction completes, and therefore the results of the micro subroutine instruction can be read.
When interlocking is not specified, the CFC2 instruction does not stall, and it immediately reads the contents
of the CCR[2,id] register.
Throughput/Latency
1/1
Example
// Copies VI01 to t0.

cfc2

t0, vi1

sw

t0, 0(t1)

Remarks
The interlock specification does not affect synchronization with macro instructions. When the preceding
macro instruction writes to the CCR[2,id] register or the CPR[2,id] register, the CFC2 instruction stalls until
the macro instruction completes, regardless of the interlock specification.

© SCEI
-234-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

CTC2 : Transfer Integer Data from EE Core to VU
Transfers the contents of GPR[rt] to CCR[2,id].
Operation Code
32-bit word: MacroOP field type 11
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

CTC2
00110

rt reg
-----

id reg
-----

---------00 0000 0000

I
-

6 bits

5 bits

5 bits

5 bits

10 bits

1

Mnemonic

CTC2
rt, id
CTC2.NI rt, id
CTC2.I rt, id

(I=0: Without interlock)
(I=0: Without interlock)
(I=1: With interlock)

Operation
Transfers the lower 32-bit data of GPR[rt], the CPU general purpose register, to CCR[2,id], the COP2
control register in which the integer/control register of VU0 is mapped.
When interlocking is specified, the CTC2 instruction stalls until the preceding VCALLMS (micro subroutine) instruction completes or until a micro instruction with the M bit set to 0 is executed in that micro
sub-routine. When interlocking is not specified, the CTC2 instruction immediately writes the data into the
CCR[2,id] register even if the micro sub-routine is currently being executed.
Throughput/Latency
1/1
Example
// Copies t0 to VI01.

lq

t0, 0(t1)

ctc2

t0, vi1

Remarks
The interlock specification does not affect synchronization with macro instructions. When the preceding
macro instruction writes to the CCR[2,id] register or the CPR[2,id] register, the CTC2 instruction stalls until
the macro instruction completes, regardless of the interlock specification.

© SCEI
-235-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

LQC2 : Floating-Point Data Transfer from EE Core to VU
Transfers the 128-bit data specified with GPR[base] and the immediate value offset to CPR[2,ft].
Operation Code
32-bit word: MacroOP field type 12
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

LQC2
110110

base
-----

ft reg
-----

offset
----------------

6 bits

5 bits

5 bits

16 bits

Mnemonic

LQC2

ft, offset(base)

Operation
Transfers the 128-bit data located at the address obtained by adding the sign-extended offset to the
GPR[base] register, to the CPR[2,ft] register (VF[ft] floating-point register of VU0).
Throughput/Latency
1/1
Example

lqc2

vf1, 0(t0)

;VF01 = t0[0]

qmtc2 a0, vf2

;VF02 = a0

vmul

;VF03 = VF02 * VF01

vf3, vf2, vf1

qmfc2 a1, vf3

;a1 = VF03

sqc2

;t1[0] = VF03

vf3, 0(t1)

Remarks
Synchronization with micro instructions is not taken into consideration. LQC2 writes to the CPR[2,ft]
register (VF[ft] register) even when a micro sub-routine is currently being executed.
If a D-cache error occurs, LQC2 stalls as well as the subsequent instructions.
If the effective address (GPR[base]+offset) is not on a 128-bit boundary, that is, when the lower 4 bits of the
effective address are not all 0, an address error exception is generated on the EE Core side.

© SCEI
-236-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

QMFC2 : Floating-Point Data Transfer from VU to EE Core
Transfers data from CPR[2,fd] (VF[fd] register of VU0) to GPR[rt].
Operation Code
32-bit word: MacroOP field type 11
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

QMFC2
00001

rt reg
-----

fd reg
-----

---------00 0000 0000

I
-

6 bits

5 bits

5 bits

5 bits

10 bits

1

Mnemonic

QMFC2
QMFC2.NI
QMFC2.I

rt, fd (I=0: Without Interlock)
rt, fd (I=0: Without Interlock)
rt, fd (I=1: With Interlock)

Operation
Transfers the contents of the CPR[2,fd] register (the floating-point register VF[fd] of VU0) to GPR[rt].
When interlocking is specified, the QMFC2 instruction stalls until the preceding VCALLMS instruction
(micro sub-routine) completes, and QMFC2 reads the results of the micro sub-routine. When interlocking is
not specified, QMFC2 reads the contents of CPR[2,fd] immediately even if the micro sub-routine is currently
being executed.
Throughput/Latency
1/1
Example

lqc2

vf1, 0(t0)

;VF01 = t0[0]

qmtc2 a0, vf2

;VF02 = a0

vmul

;VF03 = VF02 * VF01

vf3, vf2, vf1

qmfc2 a1, vf3

;a1 = VF03

sqc2

;t1[0] = VF03

vf3, 0(t1)

Remarks
The interlock specification does not affect synchronization with macro-instructions. When the preceding
macro-instruction writes to CCR[2,fd] or CPR[2,fd], QMFC2 stalls until the instruction completes, regardless
of the interlock specification.

© SCEI
-237-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

QMTC2 : Floating-Point Data Transfer from EE Core to VU
Transfers data from GPR[rt] to CPR[2,fd] (VF[fd] register).
Operation Code
32-bit word: MacroOP field type 11
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

QMTC2
00101

rt reg
-----

fd reg
-----

---------00 0000 0000

I
-

6 bits

5 bits

5 bits

5 bits

10 bits

1

Mnemonic

QMTC2
QMTC2.NI
QMTC2.I

rt, fd (I=0: Without Interlock)
rt, fd (I=0: Without Interlock)
rt, fd (I=1: With Interlock)

Operation
Transfers the contents of GPR[rt] to CPR[2,fd] (the VF[fd] register of VU0).
When interlocking is specified, QMTC2 stalls until the preceding VCALLMS instruction (micro sub-routine)
completes or until a micro instruction with the M bit set to 0 is executed in that micro sub-routine. When
interlocking is not specified, QMTC2 immediately writes data to CPR[2,fd] (the VF[fd] register) even if the
micro sub-routine is currently being executed.
Throughput/Latency
1/1
Example

lqc2

vf1, 0(t0)

;VF01 = t0[0]

qmtc2 a0, vf2

;VF02 = a0

vmul

;VF03 = VF02 * VF01

vf3, vf2, vf1

qmfc2 a1, vf3

;a1 = VF03

sqc2

;t1[0] = VF03

vf3, 0(t1)

Remarks
The interlock specification does not affect synchronization with macro-instructions. When the preceding
macro-instruction writes to CCR[2,fd] register or CPR[2,fd] register, QMTC2 stalls until the instruction
terminates, regardless of the interlock specification.

© SCEI
-238-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

SQC2 : Floating-Point Data Transfer from VU to EE Core
Stores data from CPR[2,ft] (VF[ft] register) in the memory address specified with GPR[base] and offset.
Operation Code
32-bit word: MacroOP field type 12
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

SQC2
111110

base
-----

ft reg
-----

offset
----------------

6 bits

5 bits

5 bits

16 bits

Mnemonic

SQC2

ft, offset(base)

Operation
Stores data from the CPR[2,ft] register (VF[ft] register of VU0) in the address obtained by adding the signextended offset to GPR[base].
Throughput/Latency
1/1
Example

lqc2

vf1, 0(t0)

;VF01 = t0[0]

qmtc2 a0, vf2

;VF02 = a0

vmul

;VF03 = VF02 * VF01

vf3, vf2, vf1

qmfc2 a1, vf3

;a1 = VF03

sqc2

;t1[0] = VF03

vf3, 0(t1)

Remarks
Synchronization with micro instructions is not taken into consideration. SQC2 immediately reads the
contents of CPR[2,ft] (VF[ft] register of VU0) even if the micro sub-routine program is currently being
executed.
When the preceding macro-instruction writes to CPR[2,ft] or CCR[2,ft], the SQC2 instruction stalls until the
macro-instruction completes.
If the effective address (GPR[base]+offset) is not on a 128-bit boundary, that is, when the lower 4 bits of the
execution address are not all 0, an address error exception is generated on the EE Core side.

© SCEI
-239-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VABS : Absolute Value
Calculates the absolute value of VF[fs] and stores the result in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

00111

VABS
1111

01

11 bits

Mnemonic

VABS.dest

ftdest, fsdest

Operation
Same as the micro instruction ABS. For details, refer to the appropriate pages in "4.2. Upper Instruction
Reference".

© SCEI
-240-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADD : Add
Calculates the sum of VF[fs] and VF[ft] and stores it in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VADD
101000

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VADD.dest fddest, fsdest, ftdest
Operation
Same as the micro instruction ADD. For details, refer to "4.2. Upper Instruction Reference".

© SCEI
-241-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADDi : Add to I Register
Calculates the sum of VF[fs] and the I register and stores the sum in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VADDi
100010

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VADDi.dest fddest, fsdest, I
Operation
Same as the micro instruction ADDi. For details, refer to "4.2. Upper Instruction Reference".

© SCEI
-242-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADDq : Add to Q Register
Calculates the sum of VF[fs] and the Q register and stores the sum in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VADDq
100000

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VADDq.dest fddest, fsdest, Q
Operation
Same as the micro instruction ADDq. For details, refer to "4.2. Upper Instruction Reference".

© SCEI
-243-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADDbc : Broadcast Add
Calculates the sum of each field of VF[fs] and the specified field of VF[ft] and stores the sum in the
corresponding field of VF[fd].
Operation Code
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VADD?
0000

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Mnemonic

VADDx.dest
VADDy.dest
VADDz.dest
VADDw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction ADDbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-244-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADDA : Add to Accumulator
Calculates the sum of VF[fs] and VF[ft] and stores the sum in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01010

VADDA
1111

00

11 bits

Mnemonic

VADDA.dest ACCdest, fsdest, ftdest
Operation
Same as the micro instruction ADDA. Refer to "4.2. Upper Instruction Reference".

© SCEI
-245-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADDAi : Add I Register to Accumulator
Calculates the sum of VF[fs] and the I register and stores the sum in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01000

VADDAi
1111
11 bits

Mnemonic

VADDAi.dest ACCdest, fsdest, I
Operation
Same as the micro instruction ADDAi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-246-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADDAq : Add Q Register to Accumulator
Calculates the sum of VF[fs] and the Q register and stores the sum in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01000

VADDAq
1111

00

11 bits

Mnemonic

VADDAq.dest ACCdest, fsdest, Q
Operation
Same as the micro instruction ADDAq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-247-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VADDAbc : Broadcast Add to Accumulator
Calculates sum of each field of VF[fs] and the specified field of VF[ft] and stores the sum in ACC.
Operation Code
32-bit word: MacroOP field type 2
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

VADDA?
00000
1111

bc
--

6 bits

1

4 bits

5 bits

5 bits

9 bits

2 bits

Mnemonic

VADDAx.dest
VADDAy.dest
VADDAz.dest
VADDAw.dest

ACCdest,
ACCdest,
ACCdest,
ACCdest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction ADDAbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-248-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VCALLMS : Start Micro Sub-Routine
Starts the micro sub-routine at the address specified by the immediate value.
Operation Code
32-bit word: MacroOP field type 6
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

Imm15
---------------

VCALLMS
111000

6 bits

1

4 bits

15 bits

6 bits

Mnemonic

VCALLMS Imm15
For Imm15, specify the address divided by 8.
Operation
Starts the sub-routine in MicroMem0 at the address specified with Imm15.
Example

qmtc2

vf1, a0

Micro Program

qmtc2

vf2, a1

MulMatrix:

qmtc2

vf3, a2

NOP

MULAx

ACC, VF02, VF01x

qmtc2

vf4, a3

NOP

MADDAy

ACC, VF02, VF01y

qmtc2

vf5, a4

NOP

MADDAz

ACC, VF02, VF01z

vcallms MulMatrix

NOP

MADDAw

VF01, VF02, VF01w

qmfc2

NOP

NOP @ E

vf1, a0

Remarks
When executed at the same time as other external micro sub-routine calls such as MSCAL of VIFcode, the
operation is undefined.

© SCEI
-249-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VCALLMSR : Start Micro Sub-Routine by Register
Starts the micro sub-routine at the address specified by the CMSAR0 register.
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

ft reg
00000

fs reg
11011

fd reg
00000

VCALLMSR
111001

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VCALLMSR VI27
Operation
Starts the micro sub-routine in MicroMem0. The address of the micro sub-routine must be set to CMSAR0
(CCR[2,27] register) in advance using the CTC2 instruction. (Specify the address divided by 8.)
Example

CTC2

rt, id

VNOP

;requires two VNOP instructions in order to adjust
timing.

VNOP
VCALLMSR VI27
Remarks
When executed at the same time as other external micro sub-routine calls such as MSCAL of VIFcode, the
operation is undefined.

© SCEI
-250-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VCLIP : Clipping Judgment
Performs clipping judgment with the x, y, z fields of VF[fs] and the w field of VF[ft] and puts the result in the
clipping flag (CF).
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
1110

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

00111

VCLIP
1111

11

11 bits

Mnemonic

VCLIPw.xyz fsxyz, ftw
Operation
Same as the micro instruction CLIP. Refer to "4.2. Upper Instruction Reference".

© SCEI
-251-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VDIV : Divide
Divides the fsf field of VF[fs] by the ftf field of VF[ft] and stores the quotient in the Q register.
Operation Code
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 --

fsf
--

1 2 bits 2 bits

ft reg
-----

fs reg
-----

5 bits

5 bits

01110

VDIV
1111
11 bits

Mnemonic

VDIV Q, fsfsf, ftftf
Operation
Same as the micro instruction DIV. Refer to "4.3. Lower Instruction Reference".

© SCEI
-252-

00

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VFTOI0 : Conversion to Fixed Point
Converts the contents of VF[fs] into a fixed-point number whose fractional portion is 0 bit, and stores the result
in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VFTOI0
00101
1111

00

11 bits

Mnemonic

VFTOI0.dest ftdest, fsdest
Operation
Same as the micro instruction FTOI0. Refer to "4.2. Upper Instruction Reference".

© SCEI
-253-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VFTOI4 : Conversion to Fixed Point
Converts the contents of VF[fs] into a fixed-point number whose fractional portion is 4 bits and stores the result
in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VFTOI4
00101
1111
11 bits

Mnemonic

VFTOI4.dest ftdest, fsdest
Operation
Same as the micro instruction FTOI4. Refer to "4.2. Upper Instruction Reference".

© SCEI
-254-

01

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VFTOI12 : Conversion to Fixed Point
Converts the contents of VF[fs] into a fixed-point number whose fractional portion is 12 bits and stores the
result in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VFTOI12
00101
1111

10

11 bits

Mnemonic

VFTOI12.dest ftdest, fsdest
Operation
Same as the micro instruction FTOI12. Refer to "4.2. Upper Instruction Reference".

© SCEI
-255-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VFTOI15 : Conversion to Fixed Point
Converts the contents of VF[fs] into a fixed-point number whose fractional portion is 15 bits and stores the
result in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VFTOI15
00101
1111
11 bits

Mnemonic

VFTOI15.dest ftdest, fsdest
Operation
Same as the micro instruction FTOI15. Refer to "4.2. Upper Instruction Reference".

© SCEI
-256-

11

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VIADD : Add Integer
Adds VI[is] and VI[it] and stores the result in VI[id].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

it reg
-----

is reg
-----

id reg
-----

VIADD
110000

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VIADD id, is, it
Operation
Same as the micro instruction IADD. Refer to "4.3. Lower Instruction Reference".

© SCEI
-257-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VIADDI : Add Immediate Value Integer
Adds the immediate value to VI[is] and stores the sum in VI[it].
Operation Code
32-bit word: MacroOP field type 5
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

it reg
-----

is reg
-----

Imm5
-----

VIADDI
110010

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VIADDI it, is, Imm5
Operation
Same as the micro instruction IADDI. Refer to the appropriate page in "4.3. Lower Instruction
Reference".

© SCEI
-258-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VIAND : Logical Product
Calculates the AND (logical product) of VI[is] and VI[it] at every bit and stores the result in VI[id].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

it reg
-----

is reg
-----

id reg
-----

VIAND
110100

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VIAND id, is, it
Operation
Same as the micro instruction IAND. Refer to "4.3. Lower Instruction Reference".

© SCEI
-259-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VILWR : Integer Load
Loads the specific field of the data, whose address is specified with VI[is], to VI[it] from VU Mem.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

it reg
-----

is reg
-----

6 bits

1

4 bits

5 bits

5 bits

01111

VILWR
1111
11 bits

Mnemonic

VILWR.dest it, (is)dest
Operation
Same as the micro instruction ILWR. Refer to "4.3. Lower Instruction Reference".

© SCEI
-260-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VIOR : Logical Sum
Calculates the logical sum of VI[is] and VI[it] at every bit and stores the result in VI[id].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

it reg
-----

is reg
-----

id reg
-----

VIOR
110101

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VIOR id, is, it
Operation
Same as the micro instruction IOR. Refer to "4.3. Lower Instruction Reference".

© SCEI
-261-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VISUB : Integer Subtract
Subtracts VI[it] from VI[is] and stores the result in VI[id].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

it reg
-----

is reg
-----

id reg
-----

VISUB
110001

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VISUB id, is, it
Operation
Same as the micro instruction ISUB. Refer to "4.3. Lower Instruction Reference".

© SCEI
-262-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VISWR : Integer Store
Stores data from VI[it] in VU Mem. The destination address is the dest field specified by VI[is].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

it reg
-----

is reg
-----

6 bits

1

4 bits

5 bits

5 bits

01111

VISWR
1111

11

11 bits

Mnemonic

VISWR.dest it, (is)dest
Operation
Same as the micro instruction ISWR. Refer to "4.3. Lower Instruction Reference".

© SCEI
-263-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VITOF0 : Conversion to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional portion is 0 bits, converts it to floatingpoint, and stores the result in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VITOF0
00100
1111
11 bits

Mnemonic

VITOF0.dest ftdest, fsdest
Operation
Same as the micro instruction ITOF0. Refer to "4.2. Upper Instruction Reference".

© SCEI
-264-

00

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VITOF4 : Conversion to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional portion is 4 bits, converts it to floatingpoint, and stores the result in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VITOF4
00100
1111

01

11 bits

Mnemonic

VITOF4.dest ftdest, fsdest
Operation
Same as the micro instruction ITOF4. Refer to "4.2. Upper Instruction Reference".

© SCEI
-265-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VITOF12 : Conversion to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional portion is 12 bits, converts it to floatingpoint, and stores the result in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VITOF12
00100
1111
11 bits

Mnemonic

VITOF12.dest ftdest, fsdest
Operation
Same as the micro instruction ITOF12. Refer to "4.2. Upper Instruction Reference".

© SCEI
-266-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VITOF15 : Conversion to Floating-Point Number
Considers the value of VF[fs] as a fixed-point number whose fractional portion is 15 bits, converts it to floatingpoint, and stores the result in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VITOF15
00100
1111

11

11 bits

Mnemonic

VITOF15.dest ftdest, fsdest
Operation
Same as the micro instruction ITOF15. Refer to "4.2. Upper Instruction Reference".

© SCEI
-267-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VLQD : Load with Pre-Decrement
Loads data from VU Mem at the address VI[is] - 1 into VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

is reg
-----

6 bits

1

4 bits

5 bits

5 bits

01101

VLQD
1111
11 bits

Mnemonic

VLQD.dest ftdest, (--is)dest
Operation
Same as the micro instruction LQD. Refer to "4.3. Lower Instruction Reference".

© SCEI
-268-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VLQI : Load with Post-Increment
Loads the data specified by VI[is] from VU Mem to VF[ft] and adds 1 to VI[is].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

is reg
-----

6 bits

1

4 bits

5 bits

5 bits

01101

VLQI
1111

00

11 bits

Mnemonic

VLQI.dest ftdest, (is++)dest
Operation
Same as the micro instruction LQI. Refer to "4.3. Lower Instruction Reference".

© SCEI
-269-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADD : Product Sum
Adds the value of ACC to the product of VF[fs] and VF[ft], and stores the result in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMADD
101001

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMADD.dest fddest, fsdest, ftdest
Operation
Same as the micro instruction MADD. Refer to "4.2. Upper Instruction Reference".

© SCEI
-270-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDi : Product Sum; with I Register
Adds the product of each field of VF[fs] and the I register to the corresponding value of ACC, and stores the
result in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMADDi
100011

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMADDi.dest fddest, fsdest, I
Operation
Same as the micro instruction MADDi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-271-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDq : Product Sum; with Q Register
Adds the product of each field of VF[fs] and the Q register to the corresponding value of ACC and stores the
result in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMADDq
100001

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMADDq.dest fddest, fsdest, Q
Operation
Same as the micro instruction MADDq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-272-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDbc : Broadcast Product Sum
Adds the product of each field of VF[fs] and the specified field of VF[ft] to the corresponding value of ACC and
stores the result in VF[fd].
Operation Code
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMADD?
0010

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Mnemonic

VMADDx.dest
VMADDy.dest
VMADDz.dest
VMADDw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MADDbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-273-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDA : Product Sum; to Accumulator
Adds the product of VF[fs] and VF[ft] to the value of ACC and stores the result in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01010

VMADDA
1111
11 bits

Mnemonic

VMADDA.dest ACCdest, fsdest, ftdest
Operation
Same as the micro instruction MADDA. Refer to "4.2. Upper Instruction Reference".

© SCEI
-274-

01

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDAi : Product Sum; with I Register, to Accumulator
Adds the product of each field of VF[fs] and the I register to the corresponding value of ACC and stores the
result in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VMADDAi
01000
1111

11

11 bits

Mnemonic

VMADDAi.dest ACCdest, fsdest, I
Operation
Same as the micro instruction MADDAi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-275-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDAq : Product Sum; with Q Register, to Accumulator
Adds the product of each field of VF[fs] and the Q register to the corresponding field of ACC and stores the
result in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VMADDAq
01000
1111
11 bits

Mnemonic

VMADDAq.dest ACCdest, fsdest, Q
Operation
Same as the micro instruction MADDAq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-276-

01

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMADDAbc : Broadcast Product Sum; to Accumulator
Adds the product of each field of VF[fs] and the specified field of VF[ft] to the corresponding field of ACC and
stores the result in ACC.
Operation Code
32-bit word: MacroOP field type 2
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

VMADDA?
00010
1111

bc
--

6 bits

1

4 bits

5 bits

5 bits

9 bits

2 bits

Mnemonic

VMADDAx.dest
VMADDAy.dest
VMADDAz.dest
VMADDAw.dest

ACCdest,
ACCdest,
ACCdest,
ACCdest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MADDAbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-277-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMAX : Maximum Value
Compares the value of VF[fs] with that of VF[ft] and stores the greater value in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMAX
101011

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMAX.dest fddest, fsdest, ftdest
Operation
Same as the micro instruction MAX. Refer to "4.2. Upper Instruction Reference".

© SCEI
-278-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMAXi : Maximum Value
Compares the value of each field of VF[fs] with that of the I register and stores the greater value in the
corresponding field of VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMAXi
011101

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMAXi.dest fddest, fsdest, I
Operation
Same as the micro instruction MAXi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-279-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMAXbc : Maximum Value
Compares each field of VF[fs] with the specified field of VF[ft] and stores the greater value in the corresponding
field of VF[fd].
Operation Code
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMAX?
0100

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Mnemonic

VMAXx.dest
VMAXy.dest
VMAXz.dest
VMAXw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MAXbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-280-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMFIR : Transfer from Integer Register to Floating-Point Register
Transfers the contents of VI[is] to VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

is reg
-----

6 bits

1

4 bits

5 bits

5 bits

01111

VMFIR
1111

01

11 bits

Mnemonic

VMFIR.dest ftdest, is
Operation
Same as the micro instruction MFIR. Refer to "4.2. Upper Instruction Reference".

© SCEI
-281-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMINI : Minimum Value
Compares the contents of VF[fs] with VF[ft] and stores the smaller value in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMINI
101111

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMINI.dest fddest, fsdest, ftdest
Operation
Same as the micro instruction MINI. Refer to "4.2. Upper Instruction Reference".

© SCEI
-282-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMINIi : Minimum Value
Compares each field of VF[fs] with the I register and stores the smaller value in the corresponding field of
VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMINIi
011111

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMINIi.dest fddest, fsdest, I
Operation
Same as the micro instruction MINIi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-283-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMINIbc : Minimum Value
Compares each field of VF[fs] with the specified field of VF[ft] and stores the smaller value in the corresponding
field of VF[fd].
Operation Code
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMINI?
0101

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Mnemonic

VMINIx.dest
VMINIy.dest
VMINIz.dest
VMINIw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MINIbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-284-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMOVE : Transfer between Floating-Point Registers
Transfers the value of VF[fs] to VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01100

VMOVE
1111

00

11 bits

Mnemonic

VMOVE.dest ftdest, fsdest
Operation
Same as the micro instruction MOVE. Refer to "4.3. Lower Instruction Reference".

© SCEI
-285-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMR32 : Vector Rotate
Rotates the fields of VF[fs] to the right and transfers them to VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01100

VMR32
1111
11 bits

Mnemonic

VMR32.dest ftdest, fsdest
Operation
Same as the micro instruction MR32. Refer to "4.3. Lower Instruction Reference".

© SCEI
-286-

01

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUB : Multiply and Subtract
Subtracts the product of VF[fs] and VF[ft] from the value of ACC and stores the result in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMSUB
101101

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMSUB.dest fddest, fsdest, ftdest
Operation
Same as the micro instruction MSUB. Refer to "4.2. Upper Instruction Reference".

© SCEI
-287-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUBi : Multiply and Subtract with I Register
Subtracts the product of the values of each field of VF[fs] and the I register from the corresponding field of
ACC, and stores the result in the corresponding field of VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMSUBi
100111

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMSUBi.dest fddest, fsdest, I
Operation
Same as the micro instruction MSUBi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-288-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUBq : Multiply and Subtract; Q Register
Subtracts the product of the values of each field of VF[fs] and the Q register from the corresponding field of
ACC, and stores the result in the corresponding field of VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMSUBq
100101

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMSUBq.dest fddest, fsdest, Q
Operation
Same as the micro instruction MSUBq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-289-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUBbc : Broadcast Multiply and Subtract
Subtracts the product of each field of VF[fs] and the specified field of VF[ft] from the corresponding field of
ACC, and stores the result in the corresponding field of VF[fd].
Operation Code
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMSUB?
0011

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Mnemonic

VMSUBx.dest
VMSUBy.dest
VMSUBz.dest
VMSUBw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MSUBbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-290-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUBA : Multiply and Subtract; to Accumulator
Subtracts the product of VF[fs] and VF[ft] from ACC, and stores the result in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01011

VMSUBA
1111

01

11 bits

Mnemonic

VMSUBA.dest ACCdest, fsdest, ftdest
Operation
Same as the micro instruction MSUBA. Refer to "4.2. Upper Instruction Reference".

© SCEI
-291-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUBAi : Multiply and Subtract; with I Register, to Accumulator
Subtracts the product of each field of VF[fs] and the I register from the corresponding field of ACC, and stores
the result in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VMSUBAi
01001
1111
11 bits

Mnemonic

VMSUBAi.dest ACCdest, fsdest, I
Operation
Same as the micro instruction MSUBAi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-292-

11

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUBAq : Multiply and Subtract; with Q Register, to Accumulator
Subtracts the product of each field of VF[fs] and the Q register from the corresponding field of ACC, and stores
the result in the corresponding field of ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

VMSUBAq
01001
1111

01

11 bits

Mnemonic

VMSUBAq.dest ACCdest, fsdest, Q
Operation
Same as the micro instruction MSUBAq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-293-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMSUBAbc : Broadcast Multiply and Subtract; to Accumulator
Subtracts the product of each field of VF[fs] and the specified field of VF[ft] from the corresponding field of the
ACC, and stores the result in the corresponding field of ACC.
Operation Code
32-bit word: MacroOP field type 2
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

VMSUBA?
00011
1111

bc
--

6 bits

1

4 bits

5 bits

5 bits

9 bits

2 bits

Mnemonic

VMSUBAx.dest
VMSUBAy.dest
VMSUBAz.dest
VMSUBAw.dest

ACCdest,
ACCdest,
ACCdest,
ACCdest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MSUBAbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-294-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMTIR : Transfer from Floating-Point Register to Integer Register
Transfers the lower 16 bits of the field specified by fsf of VF[fs] to VI[it].
Operation Code
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 00

fsf
--

1 2 bits 2 bits

it reg
-----

fs reg
-----

5 bits

5 bits

01111

VMTIR
1111

00

11 bits

Mnemonic

VMTIR it, fsfsf
Operation
Same as the micro instruction MTIR. Refer to "4.3. Lower Instruction Reference".

© SCEI
-295-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMUL : Multiply
Multiplies VF[fs] and VF[ft], and stores the product in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMUL
101010

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMUL.dest fddest, fsdest, ftdest
Operation
Same as the micro instruction MUL. Refer to "4.2. Upper Instruction Reference".

© SCEI
-296-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMULi : Multiply; by I Register
Multiplies the values of each field of VF[fs] by the I register, and stores the product in the corresponding field of
VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMULi
011110

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMULi.dest fddest, fsdest, I
Operation
Same as the micro instruction MULi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-297-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMULq : Multiply; by Q Register
Multiplies the values of each field of VF[fs] by the value of the Q register, and stores the product in the
corresponding field of VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VMULq
011100

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VMULq.dest fddest, fsdest, Q
Operation
Same as the micro instruction MULq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-298-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMULbc : Broadcast Multiply
Multiplies each field of VF[fs] by the specified field of VF[ft], and stores the product in the corresponding field
of VF[fd].
Operation Code
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VMUL?
0110

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Mnemonic

VMULx.dest
VMULy.dest
VMULz.dest
VMULw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MULbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-299-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMULA : Multiply; to Accumulator
Multiplies VF[fs] by VF[ft] and stores the product in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01010

VMULA
1111
11 bits

Mnemonic

VMULA.dest ACCdest, fsdest, ftdest
Operation
Same as the micro instruction MULA. Refer to "4.2. Upper Instruction Reference".

© SCEI
-300-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMULAi : Multiply by I Register; to Accumulator
Multiplies each field of VF[fs] by the value of the I register, and stores the product in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

00111

VMULAi
1111

10

11 bits

Mnemonic

VMULAi.dest ACCdest, fsdest, I
Operation
Same as the micro instruction MULAi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-301-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMULAq : Multiply by Q Register; to Accumulator
Multiplies each field of VF[fs] by value of the Q register, and stores the product in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

00111

VMULAq
1111
11 bits

Mnemonic

VMULAq.dest ACCdest, fsdest, Q
Operation
Same as the micro instruction MULAq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-302-

00

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VMULAbc : Broadcast Multiply; to Accumulator
Multiplies each field of VF[fs] by the specified field of VF[ft], and stores the product in the corresponding field
of ACC.
Operation Code
32-bit word: MacroOP field type 2
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

VMULA?
00110
1111

bc
--

6 bits

1

4 bits

5 bits

5 bits

9 bits

2 bits

Mnemonic

VMULAx.dest
VMULAy.dest
VMULAz.dest
VMULAw.dest

ACCdest,
ACCdest,
ACCdest,
ACCdest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction MULAbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-303-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VNOP : No Operation
No operation is performed.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

ft reg
00000

fs reg
00000

6 bits

1

4 bits

5 bits

5 bits

Mnemonic

VNOP
Operation
No operation is performed. The status flag does not change.
Throughput/Latency
1/4

© SCEI
-304-

01011

VNOP
1111
11 bits

11

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VOPMULA : Vector Outer Product
Calculates the first part of the vector outer product of VF[fs] and VF[ft] and stores the result in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
1110

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01011

VOPMULA
1111

10

11 bits

Mnemonic

VOPMULA.xyz ACCxyz, fsxyz, ftxyz
Operation
Same as the micro instruction OPMULA. Refer to "4.2. Upper Instruction Reference".

© SCEI
-305-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VOPMSUB : Vector Outer Product
Calculates the last part of the vector outer product using VF[fs], VF[ft] and ACC, and stores the result in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
1110

ft reg
-----

fs reg
-----

fd reg
-----

VOPMSUB
101110

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VOPMSUB.xyz fdxyz, fsxyz, ftxyz
Operation
Same as the micro instruction OPMSUB. Refer to "4.2. Upper Instruction Reference".

© SCEI
-306-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VRGET : Get Random Numbers
Obtains random numbers and stores them in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
00000

6 bits

1

4 bits

5 bits

5 bits

10000

VRGET
1111

01

11 bits

Mnemonic

VRGET.dest ftdest, R
Operation
Same as the micro instruction RGET. Refer to "4.3. Lower Instruction Reference".

© SCEI
-307-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VRINIT : Random Number Initial Set
Sets the R register to the field specified by fsf of VF[fs].
Operation Code
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 00

fsf
--

1 2 bits 2 bits

ft reg
00000

fs reg
-----

5 bits

5 bits

10000

VRINIT
1111
11 bits

Mnemonic

VRINIT R, fsfsf
Operation
Same as the micro instruction RINIT. Refer to "4.3. Lower Instruction Reference".

© SCEI
-308-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VRNEXT : New Random Numbers
Generates new random numbers and stores them in VF[ft].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
00000

6 bits

1

4 bits

5 bits

5 bits

10000

VRNEXT
1111

00

11 bits

Mnemonic

VRNEXT.dest ftdest, R
Operation
Same as the micro instruction RNEXT. Refer to "4.3. Lower Instruction Reference".

© SCEI
-309-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VRSQRT : Square Root Division
Divides the fsf field of VF[fs] by the square root of the ftf field of VF[ft], and stores the result in the Q register.
Operation Code
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 --

fsf
--

1 2 bits 2 bits

ft reg
-----

fs reg
-----

5 bits

5 bits

01110

VRSQRT
1111
11 bits

Mnemonic

VRSQRT Q, fsfsf, ftftf
Operation
Same as the micro instruction RSQRT. Refer to "4.3. Lower Instruction Reference".

© SCEI
-310-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VRXOR : Random Number Set
Takes the exclusive OR of the field specified with fsf of VF[fs] and the R register, and sets the R register to the
result.
Operation Code
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 00

fsf
--

1 2 bits 2 bits

ft reg
00000

fs reg
-----

5 bits

5 bits

VRXOR
10000
1111

11

11 bits

Mnemonic

VRXOR R, fsfsf
Operation
Same as the micro instruction RXOR. Refer to "4.3. Lower Instruction Reference".

© SCEI
-311-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSQD : Store with Pre-Decrement
Stores the contents of VF[fs] at the VU Mem0 address VI[it] - 1.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

it reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01101

VSQD
1111
11 bits

Mnemonic

VSQD.dest fsdest, (--it)dest
Operation
Same as the micro instruction SQD. Refer to "4.3. Lower Instruction Reference".

© SCEI
-312-

11

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSQI : Store with Post-Increment
Stores the contents of VF[fs] in the VU Mem0 address specified by VI[it] and adds 1 to VI[it].
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

it reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01101

VSQI
1111

01

11 bits

Mnemonic

VSQI.dest fsdest, (it++)dest
Operation
Same as the micro instruction SQI. Refer to "4.3. Lower Instruction Reference".

© SCEI
-313-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSQRT : Square Root
Obtains the square root of the field specified by the ftf field of VF[ft], and stores the result in the Q register.
Operation Code
32-bit word: MacroOP field type 4
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010
6 bits

co ftf
1 --

fsf
00

1 2 bits 2 bits

ft reg
-----

fs reg
00000

5 bits

5 bits

01110

VSQRT
1111
11 bits

Mnemonic

VSQRT Q, ftftf
Operation
Same as the micro instruction SQRT. Refer to "4.3. Lower Instruction Reference".

© SCEI
-314-

01

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUB : Subtract
Subtracts VF[ft] from VF[fs] and stores the result in VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VSUB
101100

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VSUB.dest fddest, fsdest, ftdest
Operation
Same as the micro instruction SUB. Refer to "4.2. Upper Instruction Reference".

© SCEI
-315-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBi : Subtract I Register
Subtracts the I register from each field of VF[fs], and stores the result in the corresponding field of VF[fd].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VSUBi
100110

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VSUBi.dest fddest, fsdest, I
Operation
Same as the micro instruction SUBi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-316-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBq : Subtract Q Register
Subtracts the Q register from each field of VF[fs], and stores the result in the corresponding field of VF[ft].
Operation Code
32-bit word: MacroOP field type 1
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

fd reg
-----

VSUBq
100100

6 bits

1

4 bits

5 bits

5 bits

5 bits

6 bits

Mnemonic

VSUBq.dest fddest, fsdest, Q
Operation
Same as the micro instruction SUBq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-317-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBbc : Broadcast Subtract
Subtracts the specified field of VF[ft] from each field of VF[fs] and stores the result in the corresponding field of
VF[fd].
Operation Code
32-bit word: MacroOP field type 0
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

fd reg
-----

VSUB?
0001

bc
--

6 bits

1

4 bits

5 bits

5 bits

5 bits

4 bits

2 bits

Mnemonic

VSUBx.dest
VSUBy.dest
VSUBz.dest
VSUBw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction SUBbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-318-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBA : Subtract; to Accumulator
Subtracts VF[ft] from VF[fs], and stores the result in ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01011

VSUBA
1111

00

11 bits

Mnemonic

VSUBA.dest ACCdest, fsdest, ftdest
Operation
Same as the micro instruction SUBA. Refer to "4.2. Upper Instruction Reference".

© SCEI
-319-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBAi : Subtract I Register; to Accumulator
Subtracts the I register from each field of VF[fs], and stores the result in the corresponding field of ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01001

VSUBAi
1111
11 bits

Mnemonic

VSUBAi.dest ACCdest, fsdest, I
Operation
Same as the micro instruction SUBAi. Refer to "4.2. Upper Instruction Reference".

© SCEI
-320-

10

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBAq : Subtract Q Register; to Accumulator
Subtracts the Q register from each field of VF[fs], and stores the result in the corresponding field of ACC.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
00000

fs reg
-----

6 bits

1

4 bits

5 bits

5 bits

01001

VSUBAq
1111

00

11 bits

Mnemonic

VSUBAq.dest ACCdest, fsdest, Q
Operation
Same as the micro instruction SUBAq. Refer to "4.2. Upper Instruction Reference".

© SCEI
-321-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VSUBAbc : Broadcast Subtract; to Accumulator
Subtracts the specified field of VF[ft] from each field of VF[fs], and stores the result in the corresponding field
of ACC.
Operation Code
32-bit word: MacroOP field type 2
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
----

ft reg
-----

fs reg
-----

VSUBA?
00001
1111

bc
--

6 bits

1

4 bits

5 bits

5 bits

9 bits

2 bits

Mnemonic

VSUBAx.dest
VSUBAy.dest
VSUBAz.dest
VSUBAw.dest

fddest,
fddest,
fddest,
fddest,

fsdest,
fsdest,
fsdest,
fsdest,

ftx
fty
ftz
ftw

Operation
Same as the micro instruction SUBAbc. Refer to "4.2. Upper Instruction Reference".

© SCEI
-322-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

VWAITQ : Q Register Synchronize
Stops the VU until the result is written to the Q register.
Operation Code
32-bit word: MacroOP field type 3
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 09 08 07 06 05 04 03 02 01 00

COP2
010010

co
1

dest
0000

ft reg
00000

fs reg
00000

6 bits

1

4 bits

5 bits

5 bits

01110

VWAITQ
1111

11

11 bits

Mnemonic

VWAITQ
Operation
Same as the micro instruction WAITQ. Refer to "4.3. Lower Instruction Reference".

© SCEI
-323-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

(This page is left blank intentionally)

© SCEI
-324-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

7. Appendix

© SCEI
-325-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

7.1. Sample Micro Programs
Examples of micro programs are shown below. Note that these programs are insufficient as practical programs,
since they have been created as samples.
NOP* indicates a NOP that can be deleted. (When deleted, however, the subsequent instruction stalls). There
is no NOP in the Lower instruction, so substitute a meaningless instruction such as MOVE VF00, VF00 for
NOP.
Register allocation
;Floating point register
;
x
y
z
;VF00: 0
0
0
;VF01: 1
1
1
;VF02: L0x
L1x
L2x
;VF03: L0y
L1y
L2y
;VF04: L0z
L1z
L2z
;VF05: R0
G0
B0
;VF06: R1
G1
B1
;VF07: R2
G2
B2
;VF08: R3
G3
B3
;VF09: Ra
Ga
Ba
;VF10: M00
M01
M02
;VF11: M10
M11
M12
;VF12: M20
M21
M22
;VF13: M30
M31
M32
;VF14: MINI
MAX
SEED
;VF15: ;VF16: Locally used by each subroutine
;VF17: Locally used by each subroutine
;VF18: Locally used by each subroutine
;VF19: Locally used by each subroutine
;VF20: Locally used by each subroutine
;VF21: Locally used by each subroutine
;VF22: Locally used by each subroutine
;VF23: Locally used by each subroutine
;VF24: Locally used by each subroutine
;VF25: Locally used by each subroutine
;VF26: Locally used by each subroutine
;VF27: Locally used by each subroutine
;VF28: ;VF29: u_0
v_0
;VF30: Delta u
Delta v
;VF31: 1/32
1/16
1/8
;
; Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: ;VI04: ;VI05: ;VI06: ;VI07: ;VI08: -

w
1
1
L3x
L3y
L3z
r0
M03
M13
M23
M33
MR
-

; Special constant register
; All 1 register (would be helpful to have)
; Normalized light source vector (X)
; Normalized light source vector (Y)
; Normalized light source vector (Z)
; Light source color RGB 0
; Light source color RGB 1
; Light source color RGB 2
; Light source color RGB 3
; Peripheral light
; Coordinate conversion matrix (row 0)
; Coordinate conversion matrix (row 1)
; Coordinate conversion matrix (row 2)
; Coordinate conversion matrix (row 3)
; MINI=0, MAX=255

- ; Texture coordinate temporary memory
- ; Texture coordinate Delta
1/4

© SCEI
-326-

SCE CONFIDENTIAL

;VI09:
;VI10:
;VI11:
;VI12:
;VI13:
;VI14:
;VI15:

VU User's Manual Version 6.0

Base register for Buffer 0
Base register for Buffer 1
Buffer 0, 1 swap temporary, and texture output pointer
Stack pointer recommended
Link register

Curve generation
This program generates a Bezier curve. A Bezier curve represented by 16 control points is replaced with a plane
polygon divided into 32 x 32 two-dimensional meshes and is output as 33 x 33 vertex columns.
;-------------------------------------------------------------------; SURF subroutine temporary register
; Floating point register
;VF16:
; BTv load temporary (BTu temp when optimized)
;VF17:
; BTu load temporary
;VF18:
; Output register
;VF19:
; Control point temporary 1
;VF20:
; Control point temporary 2
;VF21:
; Control point temporary 3
;VF22:
; Control point temporary 4
;VF23:
; SUM (BTu X CP) temporary 1 tmp1
;VF24:
; SUM (BTu X CP) temporary 2 tmp2
;VF25:
; SUM (BTu X CP) temporary 3 tmp3
;VF26:
; SUM (BTu X CP) temporary 4 tmp4
;VF27:
; Output register (use when optimized)
;Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: pBTu (pointer to u blending table)
;VI05: pBTv (pointer to v blending table)
;VI06: pCP (pointer to control point)
;VI07: Output vertex column pointer (dummy is output right before the pointer when optimized)
; :
;VI14: Stack pointer recommended
;VI15: Link register
; Curve generation subroutine (u one-column 33-vertex)
;-------------------------------------; Non-optimized version (for understanding the algorithms)
;-------------------------------------SURF:
NOP
LQI VF16, (VI05++) ; load BTv
NOP
LQI VF19, (VI06++) ; load CP
NOP
LQI VF20, (VI06++) ; load CP
NOP
LQI VF21, (VI06++) ; load CP
NOP
LQI VF22, (VI06++) ; load CP
MULAx.xyzw ACC, VF19, VF16x LQI VF19, (VI06++) ; tmp1=CP*BTv
MADDAy.xyzw ACC, VF20, VF16y LQI VF20, (VI06++) ; +=CP*BTv
MADDAz.xyzw ACC, VF21, VF16z LQI VF21, (VI06++) ; +=CP*BTv
MADDw.xyzw VF23, VF22, VF16w LQI VF22, (VI06++) ; +=CP*BTv
MULAx.xyzw ACC, VF19, VF16x LQI VF19, (VI06++) ; tmp2=CP*BTv
MADDAy.xyzw ACC, VF20, VF16y LQI VF20, (VI06++) ; +=CP*BTv

© SCEI
-327-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDAz.xyzw ACC, VF21, VF16z LQI VF21, (VI06++) ; +=CP*BTv
MADDw.xyzw VF24, VF22, VF16w LQI VF22, (VI06++) ; +=CP*BTv
MULAx.xyzw ACC, VF19, VF16x LQI VF19, (VI06++) ; tmp3=CP*BTv
MADDAy.xyzw ACC, VF20, VF16y LQI VF20, (VI06++) ; +=CP*BTv
MADDAz.xyzw ACC, VF21, VF16z LQI VF21, (VI06++) ; +=CP*BTv
MADDw.xyzw VF25, VF22, VF16w LQI VF22, (VI06++) ; +=CP*BTv
MULAx.xyzw ACC, VF19, VF16x LQI VF19, (VI06++) ; tmp4=CP*BTv
MADDAy.xyzw ACC, VF20, VF16y LQI VF20, (VI06++) ; +=CP*BTv
MADDAz.xyzw ACC, VF21, VF16z LQI VF21, (VI06++) ; +=CP*BTv
MADDw.xyzw VF26, VF22, VF16w LQI VF22, (VI06++) ; +=CP*BTv
NOP
IADDI VI01, VI00, 33
Lu:
NOP
LQI VF17, (VI04++) ; load BTu
NOP
NOP
NOP
NOP
NOP
NOP
MULAx.xyzw ACC, VF23, VF17x NOP
; = tmp1*BTu
MADDAy.xyzw ACC, VF24, VF17y NOP
; += tmp2*BTu
MADDAz.xyzw ACC, VF25, VF17z NOP
; += tmp3*BTu
MADDw.xyzw VF18, VF26, VF17w NOP
; += tmp4*BTu
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI VF18, (VI07++) ; store Data
NOP
NOP
NOP
IBNE VI01, VI00, Lu: ; loop
NOP
NOP
; BDSlot
NOP
NOP

JR (VI15)
NOP

; return
; BDSlot

;-------------------------------------; Optimized version
;-------------------------------------; Lu: Loop unrolls twice and makes the load delay no longer visible.
; Therefore, the looping count is not 33 loops, but is instead 16 loops + epilog (for 1 loop).
SURF_O:
NOP
LQI VF16, (VI05++) ; load BTv
NOP
LQI VF19, (VI06++) ; load CP
NOP
LQI VF20, (VI06++) ; load CP
NOP
LQI VF21, (VI06++) ; load CP
NOP
LQI VF22, (VI06++) ; load CP
MULAx.xyzw ACC, VF19, VF16x LQI VF19, (VI06++) ; tmp1=CP*BTv
MADDAy.xyzw ACC, VF20, VF16y LQI VF20, (VI06++) ; +=CP*BTv
MADDAz.xyzw ACC, VF21, VF16z LQI VF21, (VI06++) ; +=CP*BTv
MADDw.xyzw VF23, VF22, VF16w LQI VF22, (VI06++) ; +=CP*BTv
MULAx.xyzw ACC, VF19, VF16x LQI VF19, (VI06++) ; tmp2=CP*BTv
MADDAy.xyzw ACC, VF20, VF16y LQI VF20, (VI06++) ; +=CP*BTv
MADDAz.xyzw ACC, VF21, VF16z LQI VF21, (VI06++) ; +=CP*BTv
MADDw.xyzw VF24, VF22, VF16w LQI VF22, (VI06++) ; +=CP*BTv
MULAx.xyzw ACC, VF19, VF16x LQI VF19, (VI06++) ; tmp3=CP*BTv
MADDAy.xyzw ACC, VF20, VF16y LQI VF20, (VI06++) ; +=CP*BTv

© SCEI
-328-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDAz.xyzw ACC, VF21, VF16z LQI VF21, (VI06++) ; +=CP*BTv
MADDw.xyzw VF25, VF22, VF16w LQI VF22, (VI06++) ; +=CP*BTv
MULAx.xyzw ACC, VF19, VF16x
MADDAy.xyzw ACC, VF20, VF16y
MADDAz.xyzw ACC, VF21, VF16z
MADDw.xyzw VF26, VF22, VF16w

LQI VF17, (VI04++) ; tmp4.. ,ld BTu
NOP
; +=CP*BTv
IADDI VI07,VI07,-1 ; +=CP*BTv
IADDI VI01,VI00,16 ; +=CP*BTv

Lu:
MULAx.xyzw ACC, VF23, VF17x
MADDAy.xyzw ACC, VF24, VF17y
MADDAz.xyzw ACC, VF25, VF17z
MADDw.xyzw VF18, VF26, VF17w
MULAx.xyzw ACC, VF23, VF16x
MADDAy.xyzw ACC, VF24, VF16y
MADDAz.xyzw ACC, VF25, VF16z
MADDw.xyzw VF27, VF26, VF16w

LQI VF16, (VI04++) ; = tmp1*BTu
NOP
; += tmp2*BTu
IADDI VI01,VI01,-1 ; += tmp3*BTu
SQI VF27, (VI07++) ; += tmp4*BTu
LQI VF17, (VI04++) ; = tmp1*BTu
NOP
; += tmp2*BTu
IBNE VI01, VI00, Lu: ; loop
SQI VF18, (VI07++) ; BDSlot

MULAx.xyzw ACC, VF23, VF17x
MADDAy.xyzw ACC, VF24, VF17y
MADDAz.xyzw ACC, VF25, VF17z
MADDw.xyzw VF18, VF26, VF17w

NOP
; = tmp1*BTu
NOP
; += tmp2*BTu
NOP
; += tmp3*BTu
SQI VF27, (VI07++) ; += tmp4*BTu

NOP*
NOP*
NOP
NOP

NOP*
NOP*
JR (VI15)
; return
SQI VF28, (VI07++) ; last store

Perspective conversion
This program performs perspective conversion to each vertex of the mesh.
;-------------------------------------------------------------------; PRSP subroutine temporary register
; Floating point register
;VF16:
; Input vertex
;VF17:
; Vertex after coordinate conversion
;VF18:
; Output vertex
;VF19:
; Temporary for optimized version
; Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: Input vertex pointer
;VI05: Output vertex pointer (Two dummies are written before this pointer in optimized version)
; :
;VI14: Stack pointer recommended
;VI15: Link register
; Coordinate conversion and perspective conversion (for 33 vertices)
;-------------------------------------; Non-optimized version (for understanding the algorithms)
;-------------------------------------PRSP:
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00, 33 ; VF23w=1
Lp: NOP
LQI VF16, (VI04++) ; load V
NOP
NOP
NOP
NOP

© SCEI
-329-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

NOP
NOP
MULAx.xyzw ACC, VF10, VF16x
NOP
MADDAy.xyzw ACC, VF11, VF16y
NOP
MADDAz.xyzw ACC, VF12, VF16z
NOP
MADDw.xyzw VF17, VF13, VF16w
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
DIV Q, VF00w, VF17w
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MULq.xyz VF18, VF17, Q
NOP
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI VF18, (VI05++)
NOP
NOP
NOP
IBNE VI01, VI00, Lp:
NOP
NOP
; BDSlot
NOP
JR (VI15)
; return
NOP
NOP
; BDSlot
;-------------------------------------; Optimized version
;-------------------------------------PRSP_O:
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00, 33 ; VF23w=1
NOP
LQI VF16, (VI04++) ; load V
NOP
IADDI VI05, VI05, -2
NOP
NOP
Lp: MULq.xyz VF18, VF19, Q
DIV Q, VF00w, VF17w
MULAx.xyzw ACC, VF10, VF16x
MOVE.xyzw VF19, VF17
MADDAy.xyzw ACC, VF11, VF16y
NOP
MADDAz.xyzw ACC, VF12, VF16z
IADDI VI01, VI01, -1
MADDw.xyzw VF17, VF13, VF16w
LQI VF16, (VI04++) ; load V
NOP
IBNE VI01, VI00, Lp:
NOP
SQI VF18, (VI05++) ; BDSlot
MULq.xyz VF18, VF19, Q
DIV Q, VF00w, VF17w
NOP
MOVE.xyzw VF19, VF17
NOP*
NOP*
NOP*
NOP*
NOP
SQI VF18, (VI05++)
NOP*
NOP*
NOP
WAITQ
MULq.xyz VF18, VF19, Q
NOP
NOP*
NOP*
NOP*
NOP*
NOP
JR (VI15)
; return
NOP
SQI VF18, (VI05++) ; BDSlot

© SCEI
-330-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Normalization
This program obtains the normal for each vertex according to the vector product of the vector to two adjoining
vertices, and normalizes it so that the length becomes 1.
;-------------------------------------------------------------------; NORM subroutine temporary register
; Floating point register
;VF16:
; Input vertex 1
;VF17:
; Input vertex 2
;VF18:
; Input vertex 3
;VF19:
; Input vector 1
;VF20:
; Input vector 2
;VF21:
; Vector product results
;VF22:
; Vector product calculation temporary
;VF23:
; Output normal
;VF24:
; Temporary during optimization
;VF25:
; Temporary during optimization
;VF26:
; Temporary during optimization
; Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: Input vertex pointer (previous column)
;VI05: Input vertex pointer (current column)
;VI06: Output normal pointer (Two dummies are written before this pointer in optimized version)
; :
;VI14: Stack pointer recommended
;VI15: Link register
; Normal calculation (for 33 vertices)
;-------------------------------------; Non-optimized version (for understanding the algorithms)
;-------------------------------------NORM:
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00, 33 ; VF23w=1
NOP
IADDI VI01, VI01, -1
Ln: NOP
LQ VF17, (VI04 + 1)
NOP
LQI VF16, (VI04++)
NOP
LQI VF18, (VI05++)
NOP
NOP
NOP
NOP
SUB.xyz VF19xyz, VF17xyz, VF16xyz
NOP
SUB.xyz VF20xyz, VF18xyz, VF16xyz
NOP
NOP
NOP
NOP
NOP
NOP
NOP
OPMULA.xyz ACCxyz,VF19xyz,VF20xyz
NOP
OPMSUB.xyz VF21xyz,VF19xyz,VF20xyz
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MUL.xyz VF22xyz, VF21xyz, VF21xyz
NOP
NOP
NOP
NOP
NOP
NOP
NOP
ADDy.x VF22x, VF22x, VF22y
NOP
NOP
NOP

© SCEI
-331-

SCE CONFIDENTIAL

NOP
NOP
NOP
NOP
ADDz.x VF22x, VF22x, VF22z
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
RSQRT Q, VF00w, VF22x
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MULq.xyz VF23xyz, VF21xyz, Q
NOP
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI VF23, (VI06++)
NOP
NOP
NOP
IBNE VI01, VI00, Ln:
NOP
NOP
; BDSlot
NOP
JR (VI15)
; return
NOP
NOP
; BDSlot
; Vertex that obtains the differential varies in the final normal.
NOP
LQ VF17, (VI04 - 1) ;***
NOP
LQI VF16, (VI04++)
NOP
LQI VF18, (VI05++)
NOP
NOP
NOP
NOP
SUB.xyz VF19xyz, VF17xyz, VF16xyz
NOP
SUB.xyz VF20xyz, VF18xyz, VF16xyz
NOP
NOP
NOP
NOP
NOP
NOP
NOP
OPMULA.xyz ACCxyz,VF20xyz,VF19xyz
NOP
OPMSUB.xyz VF21xyz,VF20xyz,VF19xyz
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MUL.xyz VF22xyz, VF21xyz, VF21xyz
NOP
NOP
NOP
NOP
NOP
NOP
NOP
ADDy.x VF22x, VF22x, VF22y
NOP
NOP
NOP
NOP
NOP
NOP
NOP
ADDz.x VF22x, VF22x, VF22z
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
RSQRT Q, VF00w, VF22x
NOP
NOP

VU User's Manual Version 6.0

;***
;***

© SCEI
-332-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MULq.xyz VF23xyz, VF21xyz, Q
NOP
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI VF23, (VI06++)
NOP
NOP
NOP
IBNE VI01, VI00, Ln:
NOP
NOP
; BDSlot
NOP
JR (VI15)
; return
NOP
NOP
; BDSlot
;-------------------------------------; Optimized version (also performs texture coordinate calculations)
;-------------------------------------NORM_TEXD:
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00,(33-2); VF23w=1
NOP
LQ.xyzw VF17, (VI04 + 1)
NOP
LQI.xyzw VF16, (VI04++)
NOP
LQI.xyzw VF18, (VI05++)
NOP
IADDI VI05, VI05, -3
NOP
SQI XY, VF29, VI13
Ln: MULq.xyz VF23xyz, VF24xyz, Q
RSQRT Q, VF00w, VF25x
SUB.xyz VF19xyz, VF17xyz, VF16xyz MOVE.xyzw VF24, VF26
SUB.xyz VF20xyz, VF18xyz, VF16xyz MOVE.xyzw VF26, VF21
ADDy.x VF22x, VF22x, VF22y
IADDI VI01, VI01, -1
ADD.x VF29x, VF29x, VF30x
SQI.xyzw VF23, (VI06++) ; TEXD
NOP
NOP
OPMULA.xyz ACCxyz,VF19xyz,VF20xyz LQ.xyzw VF17, (VI04 + 1)
OPMSUB.xyz VF21xyz,VF19xyz,VF20xyz LQI.xyzw VF16, (VI04++)
ADDz.x VF25x, VF22x, VF22z
SQI.xy VF29xy, (VI13++) ; TEXD
NOP
LQI.xyzw VF18, (VI05++)
NOP
NOP
MUL.xyz VF22xyz, VF21xyz, VF21xyz IBNE VI01, VI00, Ln:
NOP
WAITQ
; BDSlot
; This is a gigantic epilog. The vertex that obtains the differential varies in the final normal.
; Expels 29 vertices, Q of 30 vertices, normalize of 31 vertices
; Vector product of 32 vertices, texture coordinates of 33 vertices
MULq.xyz VF23xyz, VF24xyz, Q
RSQRT Q, VF00w, VF25x
SUB.xyz VF19xyz, VF17xyz, VF16xyz MOVE.xyzw VF24, VF26
SUB.xyz VF20xyz, VF18xyz, VF16xyz MOVE.xyzw VF26, VF21
ADDy.x VF22x, VF22x, VF22y
NOP
ADD.x VF29x, VF29x, VF30x
SQI.xyzw VF23, (VI06++) ; TEXD
NOP*
NOP*
OPMULA.xyz ACCxyz,VF19xyz,VF20xyz LQ.xyzw VF17, (VI04 - 1)
OPMSUB.xyz VF21xyz,VF19xyz,VF20xyz LQI.xyzw VF16, (VI04++)
ADDz.x VF25x, VF22x, VF22z
SQI.xy VF29xy, (VI13++) ; TEXD

© SCEI
-333-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

NOP
LQI.xyzw VF18, (VI05++)
NOP*
NOP*
MUL.xyz VF22xyz, VF21xyz, VF21xyz IADDI VI01, VI01, -1
NOP
WAITQ
; Expels 30 vertices, Q of 31 vertices, normalize of 32 vertices
; Vector product of 33 vertices
MULq.xyz VF23xyz, VF24xyz, Q
RSQRT Q, VF00w, VF25x
SUB.xyz VF19xyz, VF17xyz, VF16xyz MOVE.xyzw VF24, VF26
SUB.xyz VF20xyz, VF18xyz, VF16xyz MOVE.xyzw VF26, VF21
ADDy.x VF22x, VF22x, VF22y
NOP
ADD.x VF29x, VF29x, VF30x
SQI.xyzw VF23, (VI06++) ; TEXD
NOP*
NOP*
OPMULA.xyz ACCxyz,VF19xyz,VF20xyz NOP
OPMSUB.xyz VF21xyz,VF19xyz,VF20xyz NOP
ADDz.x VF25x, VF22x, VF22z
NOP
NOP
NOP
NOP*
NOP*
NOP*
NOP*
MUL.xyz VF22xyz, VF21xyz, VF21xyz NOP
NOP
WAITQ
; Expels 31 vertices, Q of 32 vertices, normalize of 33 vertices
MULq.xyz VF23xyz, VF24xyz, Q
RSQRT Q, VF00w, VF25x
NOP
MOVE.xyzw VF24, VF26
ADDy.x VF22x, VF22x, VF22y
MOVE.xyzw VF26, VF21
NOP*
NOP*
NOP
SQI.xyzw VF23, (VI06++)
NOP*
NOP*
ADDz.x VF25x, VF22x, VF22z
NOP
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP
WAITQ
; Expels 32 vertices, Q of 33 vertices
MULq.xyz VF23xyz, VF24xyz, Q
RSQRT Q, VF00w, VF25x
NOP
MOVE.xyzw VF24, VF26
NOP*
NOP*
NOP*
NOP*
NOP
SQI.xyzw VF23, (VI06++)
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
NOP
WAIT
; Expels 33 vertices
MULq.xyz VF23xyz, VF24xyz, Q
NOP
NOP*
NOP*
NOP*
NOP*
NOP
JR (VI15)
; return
NOP
SQI.xyzw VF23, (VI06++) ; BDSlot

© SCEI
-334-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Parallel light source
This program obtains the color of each vertex from the light source color, the direction vector of the light
source, and the normal.
;-------------------------------------------------------------------; LIGHT subroutine temporary register
; Floating point register
;VF16:
; Input normal
;VF17:
; cos (theta) 0, cos (theta) 1, cos (theta) 2, cos (theta) 3
;VF18:
; cos (theta)^ns temporary
;VF19:
; Output R, G, B
; Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: Input normal pointer
;VI05: Output RGB pointer
;VI06: NS (=0, 1, 2, 4, ....)
; :
;VI13 :
;VI14: Stack pointer recommended
;VI15: Link register
; Illumination processing of parallel light source (for 33 vertices)
;-------------------------------------; Non-optimized version (for understanding algorithms)
;-------------------------------------LIGHT:
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00, 33 ; VF23w=1
Ll: NOP
LQI VF16, (VI04++)
NOP
IADDI VI06, VI00, NS
NOP
NOP
NOP
NOP
MULAx.xyzw ACC, VF02, VF16x
NOP ; lighting
MADDAy.xyzw ACC, VF03, VF16y
NOP ; calculate cos (theta)
MADDz.xyzw VF17, VF04, VF16z
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MAXx.xyzw VF17, VF17, VF14x
NOP ; clipping MINI=0
NOP
NOP
NOP
NOP
MUL.xyzw VF18, VF01, VF01
NOP ; VF18=(1,1,1,1)
Lns:
MUL.xyzw VF18, VF18, VF17
IADDI VI06, VI06, -1 ; cos(theta)^(1,2,4,8,...)
NOP
NOP
NOP
IBNE VI06, VI00, Lns:
NOP
NOP ; BDSlot
MULAy.xyzw ACC, VF09, VF00w
MADDAx.xyzw ACC, VF05, VF18x
MADDAy.xyzw ACC, VF06, VF18y
MADDAz.xyzw ACC, VF07, VF18z
MADDw.xyzw VF19, VF08, VF18w
NOP
NOP
NOP
NOP
NOP
NOP

NOP ; Ra, Ga, Ba
NOP ; R0,G0,B0*cos(theta)^NS
NOP ; R1,G1,B1*cos(theta)^NS
NOP ; R2,G2,B2*cos(theta)^NS
NOP ; R3,G3,B3*cos(theta)^NS

© SCEI
-335-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MINIy.xyzw VF19, VF19, VF14y
NOP ; clipping MAX=255
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI VF19, (VI05++)
NOP
NOP
NOP
IBNE VI01, VI00, Ll:
NOP
NOP
; BDSlot
NOP
JR (VI15)
; return
NOP
NOP
; BDSlot
;-------------------------------------; Optimized version
;-------------------------------------LIGHT_O:
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00, 33 ; VF23w=1
NOP
IADDI VI05 VI05 -1
Ll:
MULAy.xyzw ACC, VF09, VF00w
LQI VF16, (VI04++) ; Ra, Ga, Ba
MADDAx.xyzw ACC, VF05, VF18x
NOP ; R0,G0,B0*cos(theta)^NS
MADDAy.xyzw ACC, VF06, VF18y
NOP ; R1,G1,B1*cos(theta)^NS
MADDAz.xyzw ACC, VF07, VF18z
NOP ; R2,G2,B2*cos(theta)^NS
MADDw.xyzw VF19, VF08, VF18w
NOP ; R3,G3,B3*cos(theta)^NS
MULAx.xyzw ACC, VF02, VF16x
MADDAy.xyzw ACC, VF03, VF16y
MADDz.xyzw VF17, VF04, VF16z

NOP ; lighting
NOP ; calculate cos (theta)
NOP

MINIy.xyzw VF19, VF19, VF14y
NOP ; clipping MAX=255
NOP*
NOP*
NOP*
NOP*
MAXx.xyzw VF17, VF17, VF14x
NOP ; clipping MINI=0
NOP
SQI VF19, (VI05++)
NOP*
NOP*
MUL.xyzw VF18, VF01, VF01
IADDI VI01, VI01, -1 ; VF18=(1,1,1,1)
Lns:
MUL.xyzw VF18, VF18, VF17
IADDI VI06, VI06, -1 ; cos(theta)^(1,2,4,8,...)
NOP
NOP
NOP
IBNE VI06, VI00, Lns:
NOP
NOP ; BDSlot
NOP
NOP

IBNE VI01, VI00, Ll:
NOP
; BDSlot

MULAy.xyzw ACC, VF09, VF00w
NOP ; Ra, Ga, Ba
MADDAx.xyzw ACC, VF05, VF18x
NOP ; R0,G0,B0*cos (theta)^NS
MADDAy.xyzw ACC, VF06, VF18y
NOP ; R1,G1,B1*cos (theta)^NS
MADDAz.xyzw ACC, VF07, VF18z
NOP ; R2,G2,B2*cos (theta)^NS
MADDw.xyzw VF19, VF08, VF18w
NOP ; R3,G3,B3*cos (theta)^NS
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MINIy.xyzw VF19, VF19, VF14y
NOP ; clipping MAX=255
NOP*
NOP*
NOP*
NOP*
NOP
JR (VI15)
; return
NOP
SQI VF19, (VI05++) ; BDSlot

© SCEI
-336-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Point light source
This program obtains the color of each vertex from the light source color, light source position, and the normal.
;-------------------------------------------------------------------; LIGHTP subroutine temporary register
In the case of point light source, there are coordinates for four point light sources:
VF02: L0x L0y L0z ; Light source coordinate 0
VF03: L1x L1y L1z ; Light source coordinate 1
VF04: L2x L2y L2z ; Light source coordinate 2
VF15: L3x L3y L3z ; Light source coordinate 3
; Floating point register
;VF16:
;VF17:
;VF18:
;VF19:
;VF20:
;VF21:
;VF22:
;VF23:
;VF24:
;VF25:
;VF26:
;VF27:
;VF28:
;VF29:

; Input normal
; cos (theta) 0, cos (theta) 1, cos (theta) 2, cos (theta) 3
; cos(theta)^ns temporary
; Output R, G, B
; Input vetex coordinates
; Light source vector temporary
; Light source vector temporary
; Light source vector temporary
; Light source vector temporary
; Light source vector temporary 2
; Light source vector temporary 2
; Light source vector temporary 2
; Light source vector temporary 2
; 1/ri

; Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: Input normal pointer
;VI05: Output RGB pointer
;VI06: NS (=0, 1, 2, 4, ....)
;VI07: Input vertex pointer
; :
;VI13:
;VI14: Stack pointer recommended
;VI15: Link register
; Illumination processing of point light source (for 33 vertices)
;-------------------------------------; Non-optimized version (for understanding algorithms)
;-------------------------------------LIGHTP:
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00, 33 ; VF23w=1
Ll:
NOP
LQI VF20, (VI07++)
NOP
IADDI, VI06, VI00, NS
NOP
NOP
NOP
NOP
SUB.xyzw VF21, VF02, VF20
NOP ; L0: calc lx, ly, lz
SUB.xyzw VF22, VF03, VF20
NOP ; L1: calc lx, ly, lz
SUB.xyzw VF23, VF04, VF20
NOP ; L2: calc lx, ly, lz
SUB.xyzw VF24, VF15, VF20
NOP ; L3: calc lx, ly, lz
MUL.xyzw VF25, VF21, VF21
NOP ; L0: lx^2, ly^2, lz^2
MUL.xyzw VF26, VF22, VF22
NOP ; L1: lx^2, ly^2, lz^2

© SCEI
-337-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MUL.xyzw VF27, VF23, VF23
MUL.xyzw VF28, VF24, VF24

NOP ; L2: lx^2, ly^2, lz^2
NOP ; L3: lx^2, ly^2, lz^2

MULAx.x ACCx, VF21x, VF21x
MADDAy.x ACCx, VF01x, VF25y
MADDz.x VF25x, VF01x, VF25z

NOP ; L0: lx^2
NOP ; L0: lx^2+ly^2
NOP ; L0: lx^2+ly^2+lz^2

MULAx.x ACCx, VF22x, VF22x
MADDAy.x ACCx, VF01x, VF26y
MADDz.x VF26x, VF01x, VF26z

NOP ; L1: lx^2
NOP ; L1: lx^2+ly^2
NOP ; L1: lx^2+ly^2+lz^2

MULAx.x ACCx, VF23x, VF23x
MADDAy.x ACCx, VF01x, VF27y
MADDz.x VF27x, VF01x, VF27z

NOP ; L2: lx^2
NOP ; L2: lx^2+ly^2
NOP ; L2: lx^2+ly^2+lz^2

MULAx.x ACCx, VF24x, VF24x
MADDAy.x ACCx, VF01x, VF28y
MADDz.x VF28x, VF01x, VF28z

NOP ; L3: lx^2
NOP ; L3: lx^2+ly^2
NOP ; L3: lx^2+ly^2+lz^2

NOP
RSQRT Q, VF25x ; L0
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MULq.x VF29x, VF01x, Q
NOP ; L0: move Q
NOP
NOP
NOP
NOP
NOP
NOP
MULz.xyz VF23xyz, VF23xyz, VF29z NOP ; L2: normalize
NOP
NOP
NOP
NOP
NOP
NOP
MULx.z VF25z, VF01z, VF23x
NOP ; L2: transpose
MULy.z VF26z, VF01z, VF23y
NOP ; L2: transpose
MULz.z VF27z, VF01z, VF23z
NOP ; L2: transpose
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP

RSQRT Q, VF26x ; L1
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP

© SCEI
-338-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULq.y VF29y, VF01y, Q
NOP ; L1: move Q
NOP
NOP
NOP
NOP
NOP
NOP
MULy.xyz VF22xyz, VF22xyz, VF29y NOP ; L1: normalize
NOP
NOP
NOP
NOP
NOP
NOP
MULx.y VF25y, VF01y, VF22x
NOP ; L1: transpose
MULy.y VF26y, VF01y, VF22y
NOP ; L1: transpose
MULz.y VF27y, VF01y, VF22z
NOP ; L1: transpose
NOP
NOP
NOP
NOP
NOP
RSQRT Q, VF27x ; L2
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MULq.z VF29z, VF01z, Q
NOP ; L2: move Q
NOP
NOP
NOP
NOP
NOP
NOP
MULx.xyz VF21xyz, VF21xyz, VF29x NOP ; L0: normalize
NOP
NOP
NOP
NOP
NOP
NOP
MULx.x VF25x, VF01x, VF21x
NOP ; L0: transpose
MULy.x VF26x, VF01x, VF21y
NOP ; L0: transpose
MULz.x VF27x, VF01x, VF21z
NOP ; L0: transpose
NOP
NOP
NOP
NOP
NOP
RSQRT Q, VF28x ; L3
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MULq.w VF29w, VF01w, Q
NOP ; L3: move Q
NOP
NOP
NOP
NOP
NOP
NOP
MULw.xyz VF24xyz, VF24xyz, VF29w NOP ; L3: normalize

© SCEI
-339-

SCE CONFIDENTIAL

NOP
NOP
NOP
NOP
NOP
NOP
MULx.w VF25w, VF01w, VF24x
MULy.w VF26w, VF01w, VF24y
MULz.w VF27w, VF01w, VF24z
MULw.xyzw VF29, VF29, VF09w
NOP
NOP
NOP
NOP
NOP
NOP
MULw.xyzw VF29, VF29, VF29
NOP
NOP
NOP
NOP
NOP
NOP
MINIy.xyzw VF29, VF29, VF00w
NOP
NOP
NOP
NOP
NOP
NOP
MULx.xyzw VF21, VF05, VF29x
MULx.xyzw VF22, VF06, VF29y
MULx.xyzw VF23, VF07, VF29z
MULx.xyzw VF24, VF08, VF29w

VU User's Manual Version 6.0

NOP ; L3: transpose
NOP ; L3: transpose
NOP ; L3: transpose
NOP ; 1.ri * r0

NOP; (r0/ri)^2

NOP; (r0/ri)^2>1 ? 1: (r0/ri)^2

NOP; R0,G0,B0*(r0/ri)^2
NOP ;R1,G1,B1*(r0/ri)^2
NOP; R2,G2,B2*(r0/ri)^2
NOP; R3,G3,B3*(r0/ri)^2

NOP
LQI VF16, (VI04++)
NOP
NOP
NOP
NOP
NOP
NOP
MULAx.xyzw ACC, VF25, VF16x
NOP ; lighting
MADDAy.xyzw ACC, VF26, VF16y
NOP ; calculate cos(theta)
MADDz.xyzw VF17, VF27, VF16z
NOP
NOP
NOP
NOP
NOP
NOP
NOP
MAXx.xyzw VF17, VF17, VF14y
NOP ; clipping MINI=0
NOP
NOP
NOP
NOP
MUL.xyzw VF18, VF01, VF01
NOP ; VF18=(1,1,1,1)
Lns:
MUL.xyzw VF18, VF18, VF17
IADDI VI06, VI06, -1 ; cos(theta)^(1,2,4,8,...)
NOP
NOP
NOP
IBNE VI06, VI00, Lns:
NOP
NOP ; BDSlot
MULAy.xyzw ACC, VF09, VF00w
NOP ; Ra, Ga, Ba
MADDAx.xyzw ACC, VF21, VF18x
NOP ; R0,G0,B0*cos(theta)^NS
MADDAy.xyzw ACC, VF22, VF18y
NOP ; R1,G1,B1*cos(theta)^NS
MADDAz.xyzw ACC, VF23, VF18z
NOP ; R2,G2,B2*cos(theta)^NS
MADDw.xyzw VF19, VF24, VF18w
NOP ; R3,G3,B3*cos(theta)^NS
NOP
NOP
NOP
NOP
NOP
NOP
MINIy.xyzw VF19, VF19, VF14x
NOP ; clipping MAX=255
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI VF19, (VI05++)
NOP
NOP

© SCEI
-340-

SCE CONFIDENTIAL

NOP
NOP
NOP
NOP

VU User's Manual Version 6.0

IBNE VI01, VI00, Ll:
NOP
; BDSlot
JR (VI15)
; return
NOP
; BDSlot

;Illumination processing of point light source (for 33 vertices)
;---------------;Optimized version
;---------------LIGHTP_O:
NOP
LQI VF20, (VI07++)
MULw.w VF23w, VF01w, VF01w
IADDI VI01, VI00, 33; VF23w=1
NOP
IADDI VI05, VI05, -1
NOP
NOP
Ll:
SUB.xyzw VF21, VF02, VF20
IADDI VI06,VI00,NS; L0: calc lx,ly,lz
SUB.xyzw VF22, VF03, VF20
NOP; L1: calc lx, ly, lz
SUB.xyzw VF23, VF04, VF20
NOP; L2: calc lx, ly, lz
SUB.xyzw VF24, VF15, VF20
NOP; L3: calc lx, ly, lz
MUL.xyzw VF25, VF21, VF21
NOP; L0: lx^2, ly^2, lz^2
MULAx.x ACCx, VF21x, VF21x
NOP; L0: lx^2
MADDAy.x ACCx, VF01x, VF25y
NOP; L0: lx^2+ly^2
MADDz.x VF25x, VF01x, VF25z
NOP; L0: lx^2+ly^2+lz^2
MUL.xyzw VF26, VF22, VF22
NOP; L1: lx^2, ly^2, lz^2
MUL.xyzw VF27, VF23, VF23
NOP; L2: lx^2, ly^2, lz^2
MUL.xyzw VF28, VF24, VF24
NOP; L3: lx^2, ly^2, lz^2
MULAx.x ACCx, VF22x, VF22x
MADDAy.x ACCx, VF01x, VF26y
MADDz.x VF26x, VF01x, VF26z

RSQRT Q, VF25x; L1: lx^2,L0: 1/square root
NOP; L1: lx^2+ly^2
NOP; L1: lx^2+ly^2+lz^2

MULAx.x ACCx, VF23x, VF23x
MADDAy.x ACCx, VF01x, VF27y
MADDz.x VF27x, VF01x, VF27z

NOP; L2: lx^2
NOP; L2: lx^2+ly^2
NOP; L2: lx^2+ly^2+lz^2

MULAx.x ACCx, VF24x, VF24x
MADDAy.x ACCx, VF01x, VF28y
MADDz.x VF28x, VF01x, VF28z

NOP; L3: lx^2
NOP; L3: lx^2+ly^2
NOP; L3: lx^2+ly^2+lz^2

MINIy.xyzw VF19, VF19, VF14x
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*

NOP; result: clipping MAX=255

MULq.x VF29x, VF01x, Q
RSQRT Q, VF26x; L0: moveQ,L1: 1/square root
NOP
SQI VF19, (VI05++); store result***
NOP*
NOP*
NOP*
NOP*
MULx.xyz VF21xyz, VF21xyz, VF29x NOP; L0 normalize
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MULx.x VF25x, VF01x, VF21x
NOP; L0: transpose
MULy.x VF26x, VF01x, VF21y
NOP; L0: transpose
MULz.x VF27x, VF01x, VF21z
NOP; L0: transpose
NOP*
NOP*
NOP*
NOP*
MULq.y VF29y, VF01y, Q

RSQRT Q, VF27x; L1: moveQ,L2: 1/squre root

© SCEI
-341-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MULy.xyz VF22xyz, VF22xyz, VF29y NOP; L1: normalize
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MULx.y VF25y, VF01y, VF22x
NOP; L1: transpose
MULy.y VF26y, VF01y, VF22y
NOP; L1: transpose
MULz.y VF27y, VF01y, VF22z
NOP; L1: transpose
NOP*
NOP*
NOP*
NOP*
MULq.z VF29z, VF01z, Q
RSQRT Q, VF28x; L2: moveQ,L3: 1/square root
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MULz.xyz VF23xyz, VF23xyz, VF29z NOP; L2: normalize
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MULx.z VF25z, VF01z, VF23x
NOP; L2: transpose
MULy.z VF26z, VF01z, VF23y
NOP; L2: transpose
MULz.z VF27z, VF01z, VF23z
NOP; L2: transpose
NOP*
NOP*
NOP*
NOP*
MULq.w VF29w, VF01w, Q
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MULw.xyzw VF29, VF29, VF09w
MULq.xyz VF24xyz, VF24xyz, Q
NOP*
NOP*
NOP*
NOP*
MULw.xyzw VF29, VF29, VF29
MULx.w VF25w, VF01w, VF24x
MULy.w VF26w, VF01w, VF24y
MULz.w VF27w, VF01w, VF24z
MINIy.xyzw VF29, VF29, VF00y
NOP*
NOP*
NOP*
NOP*
MUL.xyzw VF18, VF01, VF01

NOP; L3: move Q

NOP; 1/ri * r0
NOP; L3: normalize

NOP; (r0/ri)^2
NOP; L3: transpose
NOP; L3: transpose
LQI VF16, (VI04++); L3: transpose
NOP; (r0/ri)^2>1 ? 1: (r0/ri)^2

NOP

; VF18=(1,1,1,1)

MULAx.xyzw ACC, VF25, VF16x
NOP ; lighting
MADDAy.xyzw ACC, VF26, VF16y
NOP ; calculate cos (theta)
MADDz.xyzw VF17, VF27, VF16z
NOP
MULx.xyzw VF21, VF05, VF29x
NOP; R0,G0,B0*(r0/ri)^2
MULx.xyzw VF22, VF06, VF29y
NOP; R1,G1,B1*(r0/ri)^2
MULx.xyzw VF23, VF07, VF29z
NOP; R2,G2,B2*(r0/ri)^2
MAXx.xyzw VF17, VF17, VF14y
NOP ; clipping MINI=0
MULx.xyzw VF24, VF08, VF29w
NOP; R3,G3,B3*(r0/ri)^2
NOP
LQI VF20, (VI07++)
NOP
IADDI VI01,VI01,-1
Lns:
MUL.xyzw VF18, VF18, VF17
IADDI VI06,VI06,-1;
NOP
NOP
NOP
IBNE VI06,VI00,Lns;

© SCEI
-342-

SCE CONFIDENTIAL

MULAw.xyzw ACC, VF09, VF00w
MADDAx.xyzw ACC, VF21, VF18x
MADDAy.xyzw ACC, VF22, VF18y
MADDAz.xyzw ACC, VF23, VF18z
MADDw.xyzw VF19, VF24, VF18w

VU User's Manual Version 6.0

NOP
NOP
NOP
IBNE VI01,VI00,Ll;
NOP

NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MINIy.xyzw VF19, VF19, VF14x
NOP ; clipping MAX=255
NOP*
NOP*
NOP*
NOP*
NOP
JR (VI15)
; return
NOP
SQI VF19, (VI05++) ; BDSlot

Texture coordinates partitioning
This program obtains the coordinate values of the texture for each vertex.
;-------------------------------------------------------------------;TEXD subroutine temporary register
;Floating point register
VF29: u_0 v_0 ; Texture coordinates temporary memory
VF30: Delta u Delta v -; Texture coordinates Delta
;Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
; :
;VI13: Output RGB pointer
;VI14: Stack pointer recommended
;VI15: Link register
;Texture coordinates partitioning (for 33 vertices)
;-------------------------------------;Non-optimized version (for understanding algorithms)
;-------------------------------------TEXD:
NOP
IADDI VI01, VI00, 33
Lt: ADD.x VF29x, VF29x, VF30x
NOP
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI.xy VF29xy, (VI13++)
NOP
NOP
NOP
IBNE VI01, VI00, Lt:
NOP
NOP
; BDSlot
ADD.y VF29y, VF29y, VF30y
LQ.x VF29x, (VI00, #pTEX)
NOP
JR (VI15)
; return
NOP
NOP
; BDSlot
;-------------------------------------;Optimized version
;-------------------------------------;Is positioned so there is no conflict with the register and is embedded inside the NORM routine.

© SCEI
-343-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

;-------------------------------------------------------------------;OUTP subroutine temporary register
;Floating point register
;VF16: ;VF17: ;VF18: ;VF19: ;VF20: ;VF21: ;VF22: ;VF23: ;VF24:
; Input vertex coordinate 0
;VF25:
; Input texture coordinate 0
;VF26:
; Input RGB 0
;VF27:
; Input vertex coordinate 1
;VF28:
; Input texture coordinate 1
;VF29:
; Input RGB 1
;Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: Input vertex coordinate 0 pointer
;VI05: Input vertex coordinate 1 pointer
;VI06: Input texture coordinate 0 pointer
;VI07: Input texture coordinate 1 pointer
;VI08: Input RGB0 pointer
;VI09: Input RGB1 pointer
;VI10: Output pointer (output by strip triangle)
;VI11: ;VI12: ;VI13: ;VI14: Stack pointer recommended
;VI15: Link register
;Consecutive polygon/strip triangle output (33 vertices)
;-------------------------------------;Non-optimized version (for understanding algorithms)
;-------------------------------------OUTP:
NOP
IADDI VI01, VI00, 33
Lo: NOP
IADDI VI01, VI01, -1
NOP
LQI VF24, (VI04++)
NOP
LQI VF25, (VI06++)
NOP
LQI VF26, (VI08++)
NOP
LQI VF27, (VI05++)
NOP
LQI VF28, (VI07++)
NOP
LQI VF29, (VI09++)
NOP
SQI VF24, (VI10++)
NOP
SQI VF25, (VI10++)
NOP
SQI VF26, (VI10++)
NOP
SQI VF27, (VI10++)
NOP
SQI VF28, (VI10++)
NOP
IBNE VI01, VI00, Lo:
NOP
SQI VF29, (VI10++) ; BDSlot
NOP
JR (VI15)
; return
NOP
NOP
; BDSlot

© SCEI
-344-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

;-------------------------------------;Optimized version
;-------------------------------------; Same as above

Displacement Mapping
This program adds offsets, which vary from vertex to vertex, to the coordinate values of respective vertices.
The arrangement of the above offset values is called displacement, and is given by input displacement table.
;-------------------------------------------------------------------;DISP subroutine temporary register
;Floating point register
;VF16:
; Input displacement data
;VF17:
; Input vertex coordinate 0
;VF18:
; Input normal vector 0
;VF19:
; Input vertex coordinate 1
;VF20:
; Input normal vector 1
;VF21:
; Output vertex coordinate 0
;VF22:
; Output vertex coordinate 1
;Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: Input vertex coordinate pointer
;VI05: Input normal pointer
;VI06: Input displacement table pointer
;VI07: Output vertex coordinate pointer
; :
;VI13 :
;VI14: Stack pointer recommended
;VI15: Link register
;Table displacement mapping (33 vertices)
; It is assumed that the displacement data is continuous in the x, y, z, w fields.
; Eight loops are invoked by 32/4. The final vertex is calculated outside the loop.
;-------------------------------------;Non-optimized version (for understanding the algorithms)
;-------------------------------------DISP:
NOP
IADDI VI01, VI00, 8
Ld: NOP
LQI VF16, (VI06++)
NOP
LQI VF17, (VI04++)
NOP
LQI VF18, (VI05++)
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF17
NOP
MADDx.xyzw VF21, VF18, VF16x
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF21, (VI07++)
NOP
LQI VF19, (VI04++)
NOP
LQI VF20, (VI05++)
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF19
NOP

© SCEI
-345-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDy.xyzw VF22, VF20, VF16y
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF22, (VI07++)
NOP
LQI VF17, (VI04++)
NOP
LQI VF18, (VI05++)
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF17
NOP
MADDz.xyzw VF21, VF18, VF16z
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF21, (VI07++)
NOP
LQI VF19, (VI04++)
NOP
LQI VF20, (VI05++)
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF19
NOP
MADDw.xyzw VF22, VF20, VF16w
NOP
NOP
NOP
NOP
NOP
NOP
IADDI VI01, VI01, -1
NOP
SQI VF22, (VI07++)
NOP
NOP
NOP

NOP
IBNE VI01, VI00, Ld:
NOP
; BDSlot

NOP
LQI VF16, (VI06++)
NOP
LQI VF17, (VI04++)
NOP
LQI VF18, (VI05++)
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF17
NOP
MADDx.xyzw VF21, VF18, VF16x
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF21, (VI07++)
NOP
NOP

JR (VI15)
NOP

; return
; BDSlot

;-------------------------------------;Optimized version
;-------------------------------------DISP_O:
NOP
IADDI VI01, VI00, 8
NOP
LQI VF16, (VI06++)
Ld: NOP
LQI VF17, (VI04++)
NOP
LQI VF18, (VI05++)
NOP
LQI VF19, (VI04++)
NOP
LQI VF20, (VI05++)
MULA.xyzw ACC, VF01, VF17
LQI VF17, (VI04++)

© SCEI
-346-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MADDx.xyzw VF21, VF18, VF16x
LQI VF18, (VI05++)
MULA.xyzw ACC, VF01, VF19
LQI VF19, (VI04++)
MADDy.xyzw VF22, VF20, VF16y
LQI VF20, (VI05++)
MULA.xyzw ACC, VF01, VF17
NOP
MADDz.xyzw VF21, VF18, VF16z
SQI VF21, (VI07++)
MULA.xyzw ACC, VF01, VF19
IADDI VI01, VI01, -1
MADDw.xyzw VF22, VF20, VF16w
SQI VF22, (VI07++)
NOP
LQI VF16, (VI06++)
NOP
SQI VF21, (VI07++)
NOP
IBNE VI01, VI00, Ld:
NOP
SQI VF22, (VI07++) ; BDSlot
NOP
LQI VF17, (VI04++)
NOP
LQI VF18, (VI05++)
NOP*
NOP*
NOP*
NOP*
MULA.xyzw ACC, VF01, VF17
NOP
MADDx.xyzw VF21, VF18, VF16x
NOP
NOP*
NOP*
NOP*
NOP*
NOP
JR (VI15)
; return
NOP
SQI VF21, (VI07++) ; BDSlot

Random number displacement mapping
This subroutine obtains the displacement value from random numbers, not from the table.
;-------------------------------------------------------------------;DISPR subroutine temporary register
;Floating point register
;VF16:
; Input displacement data
;VF17:
; Input vertex coordinate 0
;VF18:
; Input normal vector 0
;VF19:
; Input vertex coordinate 1
;VF20:
; Input normal vector 1
;VF21:
; Output vertex coordinate 0
;VF22:
; Output vertex coordinate 1
;Integer register
;VI00: 0 register
;VI01: u counter (0 - 32)
;VI02: v counter (0 - 32)
;VI03: i counter (0 - 3)
;VI04: Input vertex coordinate pointer
;VI05: Input normal pointer
;VI06: Input displacement table pointer
;VI07: Output vertex coordinate pointer
; :
;VI13:
;VI14: Stack pointer recommended
;VI15: Link register
;Random number displacement mapping (33 vertices)
; Random numbers are created by the RNEXT instruction within the range +1.0 to +2.0.
; Scaling to 0 - MR by RNEXT(1 - 2) x MR - MR.
; Eight loops are invoked by 32/4 and the final vertex is calculated outside the loop.
;Non-optimized version (for understanding algorithms)
;-------------------------------------DISPR:

© SCEI
-347-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

;-------------------------------------NOP
IADDI VI01, VI00, 8
Lr:
NOP
RNEXT VF16x
NOP
NOP
NOP
NOP
NOP
NOP
MULAw.x ACCx, VF16x, VF14w
LQI VF17, (VI04++); Rnad x MR
MSUBw.x VF16x, VF01x, VF14w
LQI VF18, (VI05++); RNEXT x MR – MR
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF17
NOP
MADDx.xyzw VF21, VF18, VF16x
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF21, (VI07++)
NOP
RNEXT VF16y
NOP
NOP
NOP
NOP
NOP
NOP
MULAw.y ACCy, VF16y, VF14w
LQI VF19, (VI04++); Rnad x MR
MSUBw.y VF16y, VF01y, VF14w
LQI VF20, (VI05++); RNEXT x MR – MR
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF19
NOP
MADDy.xyzw VF22, VF20, VF16y
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF22, (VI07++)
NOP
RNEXT VF16z
NOP
NOP
NOP
NOP
NOP
NOP
MULAw.z ACCz, VF16z, VF14w
LQI VF17, (VI04++); Rnad x MR
MSUBw.z VF16z, VF01z, VF14w
LQI VF18, (VI05++); RNEXT x MR – MR
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF17
NOP
MADDz.xyzw VF21, VF18, VF16z
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF21, (VI07++)
NOP
RNEXT VF16w
NOP
NOP
NOP
NOP
NOP
NOP
MULAw.w ACCw, VF16w, VF14w
LQI VF19, (VI04++); Rnad x MR
MSUBw.w VF16w, VF01w, VF14w LQI VF20, (VI05++); RNEXT x MR – MR
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF19
NOP
MADDw.xyzw VF22, VF20, VF16w
NOP

© SCEI
-348-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

NOP
NOP
NOP
NOP

NOP
NOP
IADDI VI01, VI01, -1
SQI VF22, (VI07++)

NOP
NOP
NOP

NOP
IBNE VI01, VI00, Lr:
NOP
; BDSlot

NOP
RNEXT VF16x
NOP
NOP
NOP
NOP
NOP
NOP
MULAw.x ACCx, VF16x, VF14w
LQI VF17, (VI04++)
MSUBw.x VF16x, VF01x, VF14w
LQI VF18, (VI05++)
NOP
NOP
NOP
NOP
MULA.xyzw ACC, VF01, VF17
NOP
MADDx.xyzw VF21, VF18, VF16x
NOP
NOP
NOP
NOP
NOP
NOP
NOP
NOP
SQI VF21, (VI07++)
NOP
NOP

JR (VI15)
NOP

; return
; BDSlot

;-------------------------------------;Optimized version
;-------------------------------------DISPR_O:
NOP
IADDI VI01, VI00, 8
NOP
LQI VF16, (VI06++)
NOP
NOP*
NOP
NOP*

RNEXT VF16x
NOP*
RNEXT VF16y
NOP*

lr: MULAw.x ACCx, VF16x, VF14w
LQI VF17, (VI04++)
MSUBw.x VF16x, VF01x, VF14w
LQI VF18, (VI05++)
MULAw.y ACCy, VF16y, VF14w
LQI VF19, (VI04++)
MSUBw.y VF16y, VF01y, VF14w
LQI VF20, (VI05++)
MULA.xyzw ACC, VF01, VF17
RNEXT VF16z
MADDx.xyzw VF21, VF18, VF16x
NOP
MULA.xyzw ACC, VF01, VF19
RNEXT VF16w
MADDy.xyzw VF22, VF20, VF16y
NOP
MULAw.z ACCz, VF16z, VF14w
LQI VF17, (VI04++)
MSUBw.z VF16z, VF01z, VF14w
LQI VF18, (VI05++)
MULAw.w ACCw, VF16w, VF14w
LQI VF19, (VI04++)
MSUBw.w VF16w, VF01w, VF14w
LQI VF20, (VI05++)
MULA.xyzw ACC, VF01, VF17
SQI VF21, (VI07++)
MADDz.xyzw VF21, VF18, VF16z
SQI VF22, (VI07++)
MULA.xyzw ACC, VF01, VF19
RNEXT VF16x
MADDw.xyzw VF22, VF20, VF16w
RNEXT VF16y
NOP
IADDI VI01, VI01, -1
NOP
SQI VF21, (VI07++)
NOP
IBNE VI01, VI00, Lr:
NOP
SQI VF22, (VI07++) ; BDSlot

© SCEI
-349-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

NOP
RNEXT VF16x
NOP*
NOP*
NOP*
NOP*
NOP*
NOP*
MULAw.x ACCx, VF16x, VF14w
LQI VF17, (VI04++)
MSUBw.x VF16x, VF01x, VF14w
LQI VF18, (VI05++)
NOP*
NOP*
NOP*
NOP*
MULA.xyzw ACC, VF01, VF17
NOP
MADDx.xyzw VF21, VF18, VF16x
NOP
NOP*
NOP*
NOP*
NOP*
NOP
JR (VI15)
; return
NOP
SQI VF21, (VI07++) ; BDSlot

Main routine
Processing such as initialization or polygon generation should be performed by the EE Core originally. This
program, however, is described as a sample using micro mode instructions.
;-------------------------------------------------------------------;main routine
;-------------------------------------------------------------------; This was originally a EE Core program.
; pBASE0 Base address 0 for temporary buffer
; pBASE1 Base address 1 for temporary buffer
; pTEX(VI00) is the address in which the texture coordinates reside.
; pBT32(VI00) is the address in which the 32-part partition blending table resides.
; pCO_P0(VI00) is the address in which the patch 0 control point resides.
; The following buffers take each of VI11 and VI12 as the base address.
; pQP(VI11) and pQP(VI12) are the output buffer addresses for partition coordinates of curved
surface partitioning
; pOP(VI11) and pOP(VI12) are the output buffer addresses for coordinate conversion +
perspective converted coordinates
; pOtex(VI11) and pOtex(VI11) are the output buffer addresses for partitioned texture coordinates
; pN(VI11) and pN(VI12) are normal output buffer addresses
; pRGB(VI11) and pRGB(VI12) are RGB output buffer addresses

; Main routine
; Main routine initialization processing
NOP
IADDI VI01, VI00, SEED
NOP
RINIT VI01
; SEED set
NOP
IADDI VI11, VI00, #pBASE0
NOP
IADDI VI12, VI00, #pBASE1
; Texture partitioning initialization processing
NOP
IADDI VI04, VI00, #pTEX
NOP
LQI.xy VF29xy, (VI04++)
NOP
LQI.xy VF30xy, (VI04++)
NOP
NOP
NOP
NOP
NOP
NOP
SUB.xy VF30xy, VF30xy, VF29xy NOP
NOP
NOP
NOP
NOP
NOP
NOP

© SCEI
-350-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

MULx.xy VF30xy, VF30xy, VF31x NOP
NOP

; (x,y) * 1/32

IADDI VI02, VI00, 33

Lmain:
; Curved surface generation (Generates nth column, 33 vertices)
NOP
IADDI VI04, VI00, #pBT32
NOP
IADDI VI05, VI00, #pBT32+1
NOP
IADDI VI06, VI00, #pCP_P0
NOP
BAL VI15, SURF_OUTP: ; subroutine call
NOP
IADDI VI07, VI12, #pQP; BDSlot
; Coordinate conversion + perspective conversion
NOP
IADDI VI04, VI12, #pQP
NOP
BAL VI15, PRSP_O: ; subroutine call
NOP
IADDI VI05, VI12, #pOP; BDSlot
; Using the vertices for two columns, the vector product is calculated, normal is derived, and the light source is calculated.
; Illumination processing is performed by the vertex in the modeling coordinate system prior to coordinate conversion + perspective
conversion.
;Normal calculation + texture coordinate partitioning
; Setting parameters for texture coordinate partitioning
NOP
IADDI VI13, VI12, #pOtex; BDSlot
;Setting parameters for normal (vector product) calculation
NOP
IADDI VI04, VI11, #pQP
NOP
IADDI VI05, VI12, #pQP
NOP
BAL VI15, NORM_TEXD: ; subroutine call
NOP
IADDI VI06, VI12, #pN; BDSlot
;Processing illumination using four parallel light sources
NOP
IADDI VI04, VI12, #pN
NOP
IADDI VI06, VI00, NS
NOP
BAL VI15, LIGHT_O: ; subroutine call
NOP
IADDI VI05, VI12, #pRGB; BDSlot
; Outputting strip triangles
NOP
IADDI VI04, VI12, #pOP
NOP
IADDI VI05, VI11, #pOP
NOP
IADDI VI06, VI12, #pOtex
NOP
IADDI VI07, VI11, #pOtex
NOP
IADDI VI08, VI12, #pRGB
NOP
BAL VI15, OUTP: ; subroutine call
NOP
IADDI VI09, VI11, #pRGB
;* * * * * * * * * * * * * * * * * * *
;swap VI11<-->VI12
NOP
NOP
NOP
; v loop
NOP
NOP
NOP

IADD VI13, VI00, VI11
IADD VI11, VI00, VI12
IADD VI12, VI00, VI13
IADDI VI02, VI02, -1
IBNE VI02, VI00, Lmain:
NOP
; BDSlot

; Outputting final strip triangles
NOP
IADDI VI08, VI11, #pOP
NOP
IADDI VI09, VI12, #pOP
NOP
IADDI VI10, VI11, #pOtex
NOP
IADDI VI11, VI12, #pOtex
NOP
IADDI VI12, VI11, #pRGB
NOP
BAL VI15, OUTP:
; subroutine call

© SCEI
-351-

SCE CONFIDENTIAL

NOP

VU User's Manual Version 6.0

IADDI VI13, VI12, #pRGB; BDSlot

© SCEI
-352-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

7.2. EFU Processing
The processing of the elementary function calculation unit, EFU, built into VU1, is shown as pseudo
microinstructions for reference purpose.
Registers
Input registers
VN00: 32 bits: x reg
VN01: 32 bits: y reg (Can be used with tmp3)
VN02: 32 bits: z reg (Can be used with tmp2)
VN03: 32 bits: w reg (Can be used with tmp1)
Output register
VN04: 32 bits: p reg
tmp registers
VN05: 32 bits: tmp1
VN06: 32 bits: tmp2
VN07: 32 bits: tmp3
Constant registers
VN08: 32 bits: 1
VN09: 32 bits: S1
VN10: 32 bits: S2
VN11: 32 bits: S3
VN12: 32 bits: S4
VN13: 32 bits: S5
VN14: 32 bits: T1
VN15: 32 bits: T2
VN16: 32 bits: T3
VN17: 32 bits: T4
VN18: 32 bits: T5
VN19: 32 bits: T6
VN20: 32 bits: T7
VN21: 32 bits: T8
VN22: 32 bits: PI/4
VN23: 32 bits: E1
VN24: 32 bits: E2
VN25: 32 bits: E3
VN26: 32 bits: E4
VN27: 32 bits: E5
VN28: 32 bits: E6
VN29: 32 bits: ---VN30: 32 bits: ---VN31: 32 bits: ----

© SCEI
-353-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Others
ACC
The input copies the four fields (128 bits) of the source register to the x, y, z, and w registers or copies an arbitrary field (32 bits) of the
source register to the x register.
The output register is p.

© SCEI
-354-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

EATAN
[description]
x <- VF[fs]fsf
p = arctan(x) (0 =< x =< 1)
p = ( T1 x Y + T2 x Y^3 + T3 x Y^5 + T4 x Y^7
+ T5 x Y^9 + T6 x Y^11 + T7 x Y^13 + T8 x Y^15 )+ PI/4
Y = (X - 1) / (X + 1)

[nano code]
ADD tmp1, x, 1
SUB tmp2, x, 1
nop
; Arctans are the same after this.
nop
nop
DIV x, tmp2, tmp1
nop
nop
nop
nop
nop
nop
nop
MUL tmp3, x, x
; x^2
MULA ACC, T1, x
nop
nop
MUL tmp1, tmp3, x ; x^3
nop
nop
nop
MUL tmp2, tmp1, tmp3 ; x^5
MADDA ACC, tmp1, T2
nop
nop
MUL tmp1, tmp2, tmp3 ; x^7
MADDA ACC, tmp2, T3
nop
nop
MUL tmp2, tmp1, tmp3 ; x^9
MADDA ACC, tmp1, T4
nop
nop
MUL tmp1, tmp2, tmp3 ; x^11
MADDA ACC, tmp2, T5
nop
nop
MUL tmp2, tmp1, tmp3 ; x^13
MADDA ACC, tmp1, T6
nop
nop
MUL tmp1, tmp2, tmp3 ; x^15
MADDA ACC, tmp2, T7
MADDA ACC, 1, PI/4
nop
MADD p, tmp1, T8
nop
nop
nop
Outputting p.

© SCEI
-355-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

EATANxy
[description]
x <- VF[ft]x
y <- VF[ft]y
z <- VF[ft]z
w <- VF[ft]w
p = arctan(y/x)

[nano code]
ADD tmp1, y, x
SUB tmp2, y, x
nop
; Same processing as arctan follows.
......
(0 =< y =< x)
[nano code]
ADD tmp1, z, x
SUB tmp2, z, x
nop
; Same processing as arctan follows.

EATANxz
[description]
x <- VF[ft]x
y <- VF[ft]y
z <- VF[ft]z
w <- VF[ft]w
p = arctan(z/x)

......
(0 =< z =< x)

© SCEI
-356-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

EEXP
[description]

[nano code]
MUL tmp1, x, x ; x^2
MULA ACC, x, E1 ; E1 * x
nop
nop
MUL tmp2, tmp1, x ; x^3
MADDA ACC, tmp1, E2 ; E2 * x^2
nop
nop
MUL tmp1, tmp2, x ; x^4
MADDA ACC, tmp2, E3 ; E3 * x^3
nop
nop
MUL tmp2, tmp1, x ; x^5
MADDA ACC, tmp1, E4 ; E4 * x^4
nop
nop
MUL tmp1, tmp2, x ; x^6
MADDA ACC, tmp2, E5 ; E5 * x^5
MADDA ACC, 1, 1
;+1
nop
MADD p, tmp1, E6 ; E6 * x^6
nop
nop
nop
MUL p, p, p
; p^2
nop
nop
nop
MUL p, p, p
; p^4
nop
nop
nop
DIV p, 1, p
nop
nop
nop
nop
nop
nop
nop
Outputting p.

x <- VF[fs]fsf
p = exp(-x)

(0 =< x =< +MAX)

© SCEI
-357-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ELENG
[description]
x <- VF[ft]x
y <- VF[ft]y
z <- VF[ft]z
w <- VF[ft]w
p = (x*x+y*y+z*z)

[nano code]
MULA ACC, x, x
MADDA ACC, y, y
MADD p, z, z
nop
nop
nop
SQRT p, 1, p
nop
nop
nop
nop
nop
nop
nop
Outputting p.

ERCPR
[description]
x <- VF[fs]fsf
p=1/x

[nano code]
DIV p, 1, x
nop
nop
nop
nop
nop
nop
nop
Outputting p.

ERLENG
[description]
x <- VF[ft]x
y <- VF[ft]y
z <- VF[ft]z
w <- VF[ft]w
1
p=
(x*x+y*y+z*z)

[nano code]
MULA ACC, x, x
MADDA ACC, y, y
MADD p, z, z
nop
nop
nop
RSQRT p, 1, p
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
Outputting p.

© SCEI
-358-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ERSADD
[description]
x <- VF[ft]x
y <- VF[ft]y
z <- VF[ft]z
w <- VF[ft]w
p = 1 / (x*x + y*y + z*z)

[nano code]
MULA ACC, x, x
MADDA ACC, y, y
MADD p, z, z
nop
nop
nop
DIV p, 1, p
nop
nop
nop
nop
nop
nop
nop
Outputting p.

ERSQRT
[description]
x <- VF[fs]dest
1
p=
x

[nano code]
RSQRT p, 1, x
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
Outputting p.

ESADD
[description]
x <- VF[ft]x
y <- VF[ft]y
z <- VF[ft]z
w <- VF[ft]w
p = x*x + y*y + z*z

[nano code]
MULA ACC, x, x
MADDA ACC, y, y
MADD p, z, z
nop
nop
nop
Outputting p.

© SCEI
-359-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ESIN
[nano code]
MUL tmp3, x, x
; x^2
MULA ACC, x, S1
nop
nop
MUL tmp1, tmp3, x ; x^3
nop
nop
nop
MUL tmp2, tmp1, tmp3 ; x^5
MADDA ACC, tmp1, S2
nop
nop
MUL tmp1, tmp2, tmp3 ; x^7
MADDA ACC, tmp2, S3
nop
nop
MUL tmp2, tmp1, tmp3 ; x^9
MADDA ACC, tmp1, S4
nop
nop
MADD p, tmp2, S5
nop
nop
nop
Outputting p.

[description]
x <- VF[fs]fsf
p = sin(x) (-π/2 =< x =< +π/2)
p = S1 x X + S2 x X^3 + S3 x X^5 + S4 x X^7 + S5 x X^9

© SCEI
-360-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

ESQRT
[description]
x <- VF[fs]fsf
p= x

[nano code]
SQRT p, 1, x
nop
nop
nop
nop
nop
nop
nop
Outputting p.

ESQUR
[description]
x <- VF[fs]fsf
p = x*x

[nano code]
MUL p, x, x
nop
nop
nop
Outputting p.

ESUM
[description]
x <- VF[ft]x
y <- VF[ft]y
z <- VF[ft]z
w <- VF[ft]w
p=x+y+z+w

[nano code]
MULA ACC, x, 1
MADDA ACC, y, 1
MADDA ACC, z, 1
MADD p, w, 1
nop
nop
nop
Outputting p.

© SCEI
-361-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

7.3. Micro Subroutine Debugging
7.3.1. Debug Flow
Micro subroutines can be debugged as follows: Stop the micro subroutine with the D bit or the T bit, examine
the VU status by using CFC2 or other instructions, and restart execution referring to the TPC.
Stop of Micro Subroutine
When the D bit or the T bit is set to 1 in a microinstruction, the VU enters the Stop state after executing the
instruction, an interrupt is sent to the EE Core, and the address after the last instruction executed is stored in the
TPC. Unlike when setting the E bit, there is no delay slot, and the following microinstruction is canceled during
its execution.
However, if the D bit or T bit is set in a branch instruction, the following instruction is the last instruction
executed, and the branch destination address is stored in the TPC.
The D bit and the T bit function similarly, but use the T bit in the application since the debugger uses the D bit.
Forcible Stop of Micro Subroutine
To forcibly stop execution of a micro subroutine externally, write 1 to the FB bit of the FBRST register using
the CTC2 instruction. This generates a ForceBreak, and the VU enters the Stop state.
However, correct re-execution of the micro subroutine after the forcible stop is not guaranteed.
Access to VU Resources
The methods of accessing the VU registers and other resources are shown in the table below. For details, refer
to "5. Macro Mode", and other corresponding macro instruction descriptions.
VU0 Resource
Access Method
Floating-point register
COP2 data register (CPR[2,00] - CPR[2,31])
Integer register
COP2 control register (CCR[2,00] - CCR[2,15])
Special register
COP2 control register (CCR[2,20] - CCR[2,22])
Flag
COP2 control register (CCR[2,16] - CCR[2,18])
VU Mem0
Main memory 0x1100_4000 - 0x1100_4ff0
(or VLQD/VLQI/VSQD/VSQI instruction)
MicroMem0
Main memory 0x1100_0000 - 0x1100_0ff0
Accesses to VU Mem0 and MicroMem0 are possible only when VU0 is stopped.

VU1 Resource
Floating-point register
Integer register
Special register
Flag
VU Mem1
MicroMem1

Access Method
VU Mem0 addresses 0x4000 - 0x41ff
VU Mem0 addresses 0x4200 - 0x42ff
VU Mem0 addresses 0x4340 - 0x437f
VU Mem0 addresses 0x4300 - 0x432f
Main memory 0x1100_c000 - 0x1100_fff0
Main memory 0x1100_8000 - 0x1100_bff0

Accesses to VU Mem1 and MicroMem1 are possible only when VU1 is stopped. Since the above methods of accessing VU Mem1 and
MicroMem1 are strictly for the purpose of debugging with a considerable speed penalty involved, transfer data via VIF in ordinary cases.

Execution Restart
To restart execution of a VU0 micro subroutine, read the TPC with an instruction such as CFC2, transfer the
value to CMSAR0 with the CTC2 instruction, and restart execution with VCALLMSR.
In VU1, read the TPC with an instruction such as VILWR or CFC2 and transfer the value to CMSAR1 with the
CTC2 instruction to restart execution.
© SCEI
-362-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

Control of Debugging Function
Stopping micro programs with the D bit/T bit can be controlled with the DE bit/TE bit of the control register.
Stopping a microprogram with the D bit is permitted only when the DE bit is set to 1; similarly for the T bit and
TE bit.
Debugging efficiency can be improved by setting the DE bit only when a routine that becomes a problem is
executed while debugging a large-scale program.

7.3.2. Notes on Re-execution
Since the VU performs pipeline operation, continuous execution of a micro subroutine and re-execution after
stopping during its execution may sometimes cause different results. It is recommended to stop at a place where
the re-execution result is guaranteed by paying attention to the instructions before and after the stop by the D
bit.
The figures below show how results can differ between continuous execution of a micro subroutine and reexecution after stopping during its execution with the D bit.

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

16

17

Instruction clock cycle

1

MUL
NOP

2

MULA
NOP

3

MADDA
NOP

4

MADD
NOP

5

MUL
FSAND

6

MUL
NOP

D bit

2

MUL
NOP
MULA
NOP

3

MADDA
NOP

4

MADD
NOP

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

7

8

9

10

11

D bit

Instruction clock cycle

1

M
M

D bit

FSAND of Instruction 5 refers to
MUL calculation result flag of Instruction 1

1

2

3

4

5

6

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

12

13

14

15

FSAND of Instruction 5 refers to
MADD calculation result flag of Instruction 4

S
S

Stop by D bit

5
6

M
M

MUL
FSAND
MUL
NOP

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

© SCEI
-363-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

A flag is referred to by the FSAND instruction in Instruction 5. Although it is the calculation result of
Instruction 1 in continuous execution, the calculation result of Instruction 4 is referred to in re-execution after
stopping.
The following figure illustrates an example in which the value of the Q register (to be referred to when the micro
subroutine is re-executed after the stop) is changed.

Instruction clock cycle

1

MUL
DIV

2

MULA
NOP

3

MADDA
NOP

4

MADD
NOP

1

2

M
M

T
T
M
M

D bit

8

9

X Y
Z S
D1 D2 D3 D4 D5 D6

3

4

5

6

7

F

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

10

11

12

13

14

15

16

DIV of Instruction 1 is referred to
for re-execution after the stop.

S
S

Stop by D bit

5
6

MULq
NOP

M
M

MUL
NOP

© SCEI
-364-

T
T

X
X

Y
Y

Z
Z

S
S

M
M

T
T

X
X

Y
Y

Z
Z

S
S

17

SCE CONFIDENTIAL

VU User's Manual Version 6.0

7.4. Throughput / Latency List
Micro Instruction
ABS
ADD
ADDi ADDq
ADDbc
ADDA
ADDAi ADDAq
ADDAbc
B
BAL
------------------------CLIP
--------DIV
EATAN
EATANxy
EATANxz
EEXP
ELENG
ERCPR
ERLENG
ERSADD
ERSQRT
ESADD
ESIN
ESQRT
ESUM
FCAND
FCEQ
FCGET
FCOR
FCSET
FMAND
FMEQ
FMOR
FSAND
FSEQ
FSOR
FSSET
FTOI0 FTOI4
FTOI12 FTOI15
IADD
IADDI
IAND

Macro Instruction
VABS
VADD
VADDi VADDq
VADDbc
VADDA
VADDAi VADDAq
VADDAbc
--------BC2F
BC2FL
BC2T
BC2TL
VCALLMS
VCALLMSR
VCLIP
CFC2
CTC2
VDIV
----------------------------------------------------------------------------------------------------VFTOI0 VFTOI4
VFTOI12 VFTOI15
VIADD
VIADDI
VIAND

Throughput
1
1
1
1
1
1
1
2
2
2
2
2
2
--------1
1
1
7
53
53
53
43
17
11
23
17
17
10
28
11
11
1
1
1
1
1
1
1
1
1
1
1
1
1

Latency
4
4
4
4
4
4
4
2
2
2
2
2
2
--------4
1
1
7
54
54
54
44
18
12
24
18
18
11
29
12
12
1
1
1
1
4
1
1
1
1
1
1
4
4

1
1
1

1
1
1
© SCEI

-365-

SCE CONFIDENTIAL

Micro Instruction
IADDIU
IBEQ
IBGEZ
IBGTZ
IBLEZ
IBLTZ
IBNE
ILW
ILWR
IOR
ISUB
ISUBIU
ISW
ISWR
ITOF0 ITOF4
ITOF12 ITOF15
JALR
JR
LQ
----LQD
LQI
MADD
MADDi MADDq
MADDbc
MADDA
MADDAi MADDAq
MADDAbc
MAX
MAXi
MAXbc
MFIR
MFP
MINI
MINIi
MINIbc
MOVE
MR32
MSUB
MSUBi MSUBq
MSUBbc
MSUBA
MSUBAi MSUBAq
MSUBAbc
MTIR
MUL
MULi MULq
MULbc
MULA
MULAi MULAq
MULAbc
NOP

VU User's Manual Version 6.0

Macro Instruction
--------------------------------VILWR
VIOR
VISUB
--------VISWR
VITOF0 VITOF4
VITOF12 VITOF15
------------LQC2
VLQD
VLQI
VMADD
VMADDi VMADDq
VMADDbc
VMADDA
VMADDAi
VMADDAq
VMADDAbc
VMAX
VMAXi
VMAXbc
VMFIR
----VMINI
VMINIi
VMINIbc
VMOVE
VMR32
VMSUB
VMSUBi VMSUBq
VMSUBbc
VMSUBA
VMSUBAi MSUBAq
VMSUBAbc
VMTIR
VMUL
VMULi VMULq
VMULbc
VMULA
VMULAi VMULAq
VMULAbc
VNOP

© SCEI
-366-

Throughput
1
2
2
2
2
2
2
1
1
1
1
1
1
1
1

Latency
1
2
2
2
2
2
2
4
4
1
1
1
4
4
4

2
2
1
1
1
1
1
1
1
1
1

2
2
4
1
4
4
4
4
4
4
4

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
1
4
4
4
4
4
4
4

SCE CONFIDENTIAL

Micro Instruction
OPMULA
OPMSUB
--------RGET
RINIT
RNEXT
RSQRT
RXOR
SUB
SUBi SUBq
SUBbc
SUBA
SUBAi SUBAq
SUBAbc
SQ
----SQD
SQI
SQRT
WAITP
WAITQ
XGKICK
XITOP
XTOP

VU User's Manual Version 6.0

Macro Instruction
VOPMULA
VOPMSUB
QMFC2
QMTC2
VRGET
VRINIT
VRNEXT
VRSQRT
VRXOR
VSUB
VSUBi VSUBq
VSUBbc
VSUBA
VSUBAi VSUBAq
VSUBAbc
----SQC2
VSQD
VSQI
VSQRT
----VWAITQ
-------------

Throughput
1
1
1
1
1
1
1
13
1
1
1
1
1
1
1
1
1
1
1
7
1~54
1~13
1~
1
1

Latency
4
4
1
1
4
1
4
13
1
4
4
4
4
4
4
4
1
4
4
7
1~54
1~13
1~
1
1

© SCEI
-367-

SCE CONFIDENTIAL

VU User's Manual Version 6.0

(This page is left blank intentionally)

© SCEI
-368-

Source Exif Data:

File Type                       : PDF
File Type Extension             : pdf
MIME Type                       : application/pdf
PDF Version                     : 1.4
Linearized                      : Yes
Encryption                      : Standard V1.2 (40-bit)
User Access                     : Print, Fill forms, Extract, Assemble, Print high-res
Creator                         : Acrobat PDFMaker 5.0 for Word
Modify Date                     : 2002:06:10 14:27:01-07:00
Create Date                     : 2002:05:31 21:04:07Z
Page Count                      : 368
Creation Date                   : 2002:05:31 21:04:07Z
Mod Date                        : 2002:06:10 14:27:01-07:00
Producer                        : Acrobat Distiller 5.0.5 (Windows)
Author                          : Sony Computer Entertainment
Metadata Date                   : 2002:06:10 14:27:01-07:00
Title                           : VU User’s Manual
Page Mode                       : UseOutlines
Page Layout                     : SinglePage

EXIF Metadata provided by EXIF.tools

VU User’s Manual Users

Navigation menu

Versions of this User Manual:

Views

Navigation