Section 2. CPU 70204 70204C

User Manual: 70204

Open the PDF directly: View PDF .
Page Count: 46

Section 2. CPU

CPU

Section 2. CPU

HIGHLIGHTS

This section of the manual contains the following topics:

2.1 Introduction ....................................................................................................................2-2

2.2 Programmer’s Model......................................................................................................2-5

2.3 Software Stack Pointer...................................................................................................2-9

2.4 CPU Register Descriptions ..........................................................................................2-12

2.5 Arithmetic Logic Unit....................................................................................................2-17

2.6 DSP Engine .................................................................................................................2-18

2.7 Divide Support .............................................................................................................2-27

2.8 Instruction Flow Types .................................................................................................2-28

2.9 Loop Constructs...........................................................................................................2-31

2.10 Address Register Dependencies .................................................................................2-37

2.11 Register Maps..............................................................................................................2-40

2.12 Related Application Notes............................................................................................2-42

2.13 Revision History...........................................................................................................2-43

dsPIC33F/PIC24H Family Reference Manual

2.1 INTRODUCTION

The dsPIC33F/PIC24H CPU has a 16-bit (data) modified Harvard architecture with an

enhanced instruction set, including significant support for digital signal processing. The CPU

has a 24-bit instruction word, with a variable length opcode field. The Program Counter (PC) is

24 bits wide and addresses up to 4M x 24 bits of user program memory space.

A single-cycle instruction prefetch mechanism helps maintain throughput and provides

predictable execution. All instructions execute in a single cycle, except the instructions that

change the program flow, double-word move (MOV.D) instruction, table instructions and also

instructions accessing Program Space Visibility (PSV) take more than one cycle. Overhead-free

program loop constructs are supported using the DO and REPEAT instructions, both of which are

interruptible at any point.

2.1.1 Registers

The dsPIC33F/PIC24H devices have sixteen 16-bit working registers in the programmer’s

model. Each working register can operate as data, address, or address offset register. The

sixteenth working register (W15) operates as a Software Stack Pointer (SSP) for interrupts and

calls.

2.1.2 Instruction Set

The dsPIC33F/PIC24H instruction set has two classes of instructions:

• MCU class instructions

• DSP class instructions

These two instruction classes are seamlessly integrated into the architecture and execute from

a single execution unit. The instruction set includes many Addressing modes and is designed

for optimum C compiler efficiency.

2.1.3 Data Space Addressing

The data space can be addressed as 32K words or 64 Kbytes and is split into two blocks,

referred to as X and Y data memory. Each memory block has its own independent

Address Generation Unit (AGU). The MCU class of instructions operate solely through the X

memory AGU, which accesses the entire memory map as one linear data space. A few DSP

instructions operate through the X and Y AGU to support dual operand read, which splits the

data address space into two parts. The X and Y data space boundary is device specific.

The upper 32 Kbytes of the data space memory map can optionally be mapped into program space

at any 16K program word boundary defined by the 8-bit Program Space Visibility Page (PSVPAG)

space as if it were data space. Furthermore, RAM can be connected to the program memory bus

on devices with an external bus, and can be used to extend the internal data RAM.

Overhead-free circular buffers (modulo addressing) are supported in both X and Y address

spaces. The modulo addressing removes the software boundary-checking overhead for DSP

algorithms. The X AGU circular addressing can be used with any MCU class of instructions. The

X AGU also supports bit-reverse addressing to greatly simplify input or output data reordering for

radix-2 FFT algorithms.

Note: This family reference manual section is meant to serve as a complement to device

data sheets. Depending on the device variant, this manual section may not apply to

all dsPIC33F/PIC24H devices.

Please consult the note at the beginning of the “CPU” chapter in the current device

data sheet to check whether this document supports the device you are using.

Device data sheets and family reference manual sections are available for

download from the Microchip Worldwide Web site at: http://www.microchip.com

Section 2. CPU

CPU

2.1.4 Addressing Modes

The CPU supports the following addressing modes:

• Inherent (no operand)

• Relative

•Literal

• Memory Direct

• Register Direct

• Register Indirect

Each instruction is associated with a predefined addressing mode group, depending upon its

functional requirements. As many as six addressing modes are supported for each instruction.

For most instructions, the dsPIC33F/PIC24H CPU can execute the following functions in a single

instruction cycle:

• Data (or program data) memory read

• Working register (data) read

• Data memory write

• Program (instruction) memory read

Therefore, three operand instructions can be supported allowing A + B = C operations to be

executed in a single cycle.

2.1.5 DSP Engine and Instructions

Following are the DSP Engine features:

• A high-speed, 17-bit by 17-bit multiplier (for Signed 16-bit multiplication)

• A 40-bit Arithmetic Logic Unit (ALU)

• Two 40-bit saturating accumulators

• A 40-bit bidirectional barrel shifter that is capable of shifting a 40-bit value up to 16 bits right

or left, in a single cycle

The DSP instructions operate seamlessly with all other instructions and are designed for optimal

real-time performance. The MAC instruction and other associated instructions can concurrently

fetch two data operands from memory while multiplying two Working (W) registers. This requires

that the data space be split for these instructions and linear for all others. This is achieved in a

transparent and flexible manner by dedicating certain working registers to each address space.

dsPIC33F/PIC24H Family Reference Manual

2.1.6 Exception Processing

The dsPIC33F/PIC24H CPU has a vectored exception scheme with up to eight sources of

non-maskable traps and 118 interrupt sources. Each interrupt source can be assigned to one of

the seven priority levels. Figure 2-1 shows the CPU block diagram.

Figure 2-1: dsPIC33F/PIC24H CPU Block Diagram

Power-up

Timer

Oscillator

Start-up Timer

Instruction

Decode and

Control

OSC1/CLKI

MCLR

VDD, VSS

UART1,

ECAN2

Timing

Generation

ECAN1,

PCH PCL

Program Counter (PC)

16-bit ALU

X Data Bus

I2C™

DCI

PCU

10-bit or

Timers

Input

Capture

Module

Output

Compare

Module

16 16

16 x 16

W Reg Array

Divide

Support

Engine

DSP

ROM Latch

Y Data Bus

EA MUX

X RAGU

X WAGU

Y AGU

AVDD, AVSS

UART2

SPI2

Interrupt

Controller PSV and Table

Data Access

Control Block

Stack

Control

Logic

Loop

Control

Logic

Data LatchData Latch

Y Data

RAM X Data

RAM

Address

Latch

Address

Latch

Control Signals

to Various Blocks

SPI1,

Data Latch

I/O

X Address Bus

Y Address Bus

Literal Data

12-bit ADC

Program Memory

Watchdog

Timer

Reset

POR/BOR

Address Latch

Ports

Section 2. CPU

CPU

2.2 PROGRAMMER’S MODEL

The programmer’s model for the dsPIC33F/PIC24H CPU is shown in Figure 2-2. All registers in

the programmer’s model are memory mapped and can be manipulated directly using

instructions. Table 2-1 provides a description of each register.

In addition to the registers contained in the programmer’s model, the dsPIC33F/PIC24H contains

control registers for modulo addressing, bit-reversed addressing and interrupts. These registers

are described in the subsequent sections in this document. All registers associated with the

programmer’s model are memory mapped as shown in Table 2-3.

Table 2-1: Programmer’s Model Register Descriptions

W0 through W15 Working register array

ACCA, ACCB 40-bit DSP Accumulators

PC 23-bit PC

SR ALU and DSP Engine Status register

SPLIM Stack Pointer Limit Value register

TBLPAG Table Memory Page Address register

PSVPAG Program Space Visibility Page Address register

RCOUNT REPEAT Loop Count register

DCOUNT DO Loop Count register

DOSTART DO Loop Start Address register

DOEND DO Loop End Address register

CORCON Contains DSP Engine and DO Loop control bits

dsPIC33F/PIC24H Family Reference Manual

Figure 2-2: Programmer’s Model

NOVZ C

TBLPAG

22 0

015

Program Counter

Data Table Page Address

STATUS Register

Working/Address

Registers

DSP Operand

Registers

W0 (WREG)

W10

W11

W12

W13

Frame Pointer/W14

Stack Pointer/W15

DSP Address

Registers

39 031

DSP

Accumulators

PSVPAG

70Program Space Visibility

OA OB SA SB

RCOUNT

15 0

DCOUNT(1)

15 0

DO Loop Counter

DOSTART(1)

22 0

DO Loop Start Address

DOEND(1) DO Loop End Address

IPL<2:0>

SPLIM Stack Pointer Limit

22 0

SRL

PUSH.S and POP.S Shadows

OAB SAB

Page Address

DA DC

CORCON

15 0

Core Control Register

ACCAH ACCAL

ACCAU

ACCBU ACCBH ACCBL

ACCA

ACCB

SRH

Note 1: DCOUNT, DOSTART and DOEND have one level of shadow registers (not shown) for nested DO loops.

Section 2. CPU

CPU

2.2.1 Working Register Array

The 16 working (W) registers can function as data, address or address offset registers. The

function of a W register is determined by the addressing mode of the instruction that accesses it.

The dsPIC33F/PIC24H instruction set can be divided into two instruction types:

• Register instructions

• File register instructions

2.2.1.1 REGISTER INSTRUCTIONS

shown in Example 2-1.

Example 2-1: Register Instructions

2.2.1.2 FILE REGISTER INSTRUCTIONS

File register instructions operate on a specific memory address contained in the instruction

opcode and W0 register. W0 is a special working register used in file register instructions.

Working registers W1-W15 cannot be specified as target registers in file register instructions.

The file register instructions provide backward compatibility with existing PIC® MCU devices that

have only one W register. The label “WREG” is used in the assembler syntax to denote W0 in a

file register instruction, as shown in Example 2-2.

Example 2-2: File Register Instructions

2.2.1.3 W REGISTER MEMORY MAPPING

Since the W registers are memory mapped, it is possible to access a W register in a file register

instruction, as shown in Example 2-3.

Example 2-3: Accessing W Register in File Register Instruction

Further, it is also possible to execute an instruction that uses a W register as both an address

pointer and operand destination, as shown in Example 2-4.

Example 2-4: W Register Used as Address Pointer and Operand Destination

In Example 2-4, the contents of W2 are 0x0004. As W2 is used as an address pointer, it points

to location 0x0004 in memory. W2 is also mapped to this address in memory. Even though this

is an unlikely event, it is impossible to detect until run time. The dsPIC33F/PIC24H CPU ensures

that the data write dominates, resulting in W2 = 0x1234 in this example.

Note: Refer to the “16-bit MCU and DSC Programmer’s Reference Manual” (DS70157),

for complete descriptions of addressing modes and instruction syntax.

MOV W0, W1 ; move contents of W0 to W1

MOV W0, [W1] ; move W0 to address contained in W1

ADD W0, [W4], W5 ; add contents of W0 to contents pointed

; to by W4. Place result in W5.

MOV WREG, 0x0100 ; move contents of W0 to address 0x0100

ADD 0x0100, WREG ; add W0 to address 0x0100, store in W0

MOV 0x0004, W10 ; equivalent to MOV W2, W10

where:

0x0004 is the memory address of W2

MOV W1, [W2++]

where:

W1 = 0x1234

W2 = 0x0004 ;[W2] addresses W2

dsPIC33F/PIC24H Family Reference Manual

2.2.1.4 W REGISTERS AND BYTE MODE INSTRUCTIONS

The byte instructions, which target the W register array, only affect the

Least Significant Byte (LSB) of the target register. As the working registers are memory mapped,

the LSB and Most Significant Byte (MSB) can be manipulated through byte wide data memory

space access.

2.2.2 Shadow Registers

Many of the registers in the programmer’s model have an associated shadow register, as shown

in Figure 2-2. None of the shadow registers are accessible directly. There are two types of

shadow registers:

• Used by PUSH.S and POP.S instructions

• Used by DO instruction

2.2.2.1 PUSH.S AND POP.S SHADOW REGISTERS

The PUSH.S and POP.S instructions are useful for fast context save/restore during a function call

or Interrupt Service Routine (ISR). The PUSH.S instruction transfers the following register values

into their respective shadow registers:

•W0...W3

• SR (N, OV, Z, C, DC bits only)

The POP.S instruction restores the values from the shadow registers into these register

locations. Example 2-5 shows a code example using the PUSH.S and POP.S instructions.

Example 2-5: PUSH.S and POP.S Instructions

The PUSH.S instruction overwrites the contents that are previously saved in the shadow

registers. The shadow registers are only one level in depth. Therefore, care must be taken if the

shadow registers are to be used for multiple software tasks.

The user-assigned application must ensure that any task using the shadow registers are not

interrupted by a higher priority task that also uses the shadow registers. If the higher priority task

is allowed to interrupt the lower priority task, the contents of the shadow registers saved in the

lower priority task are overwritten by the higher priority task.

2.2.2.2 DO LOOP SHADOW REGISTERS

The following registers are automatically saved in shadow registers when a DO instruction is

executed:

•DOSTART

•DOEND

• DCOUNT

The DO shadow registers are one level in depth, permitting two loops to be automatically nested.

Refer to 2.9.2.2 “DO Loop Nesting”, for more details on DO loop nesting.

2.2.3 Uninitialized W Register Reset

The W register array (with the exception of W15) is cleared during all resets and is considered

uninitialized until written to. An attempt to use an uninitialized register as an address pointer will

reset the device.

A word write must be performed to initialize a W register. A byte write will not affect the initialization

detection logic.

MyFunction:

PUSH.S ; Save W registers, MCU status

MOV #0x03,W0 ; load a literal value into W0

ADD RAM100 ; add W0 to contents of RAM100

BTSC SR,#Z ; is the result 0?

BSET Flags,#IsZero ; Yes, set a flag

POP.S ; Restore W regs, MCU status

RETURN

Section 2. CPU

CPU

2.3 SOFTWARE STACK POINTER

The W15 register serves as a dedicated Software Stack Pointer (SSP) and is automatically

modified by exception processing, subroutine calls and returns. However, W15 can be

referenced by any instruction in the same manner as all other W registers. This simplifies

reading, writing and manipulating of the Stack Pointer (SP). For example, creating stack frames.

W15 is initialized to 0x0800 during all resets. This address ensures that the SP points to valid

RAM in all dsPIC33F/PIC24H devices and permits stack availability for non-maskable trap

exceptions. These can occur before the SP is initialized by the user-assigned application

software. The user can reprogram the SP during initialization to any location within data space.

The SP always points to the first available free word and fills the software stack, working from

lower toward higher addresses. Figure 2-3 shows how it pre-decrements for a stack pop (read)

and post-increments for a stack push (writes).

When the PC is pushed onto the stack, the PC<15:0> bits are pushed onto the first available

stack word, and then PC<22:16> bits are pushed onto the second available stack location. For

a PC push during any CALL instruction, the MSB of the PC is zero-extended before the push, as

shown in Figure 2-3. During exception processing, the MSB of the PC is concatenated with the

lower eight bits of the CPU status register, SR. This allows the contents of SRL to be preserved

automatically during interrupt processing.

Figure 2-3: Stack Operation for a CALL Instruction

2.3.1 Software Stack Examples

The software stack is manipulated using the PUSH and POP instructions. The PUSH and POP

instructions are the equivalent of a MOV instruction with W15 used as the destination pointer. The

contents of W0 can be pushed onto the stack as shown in Example 2-6.

Example 2-6: Software Stack Example

Figure 2-4 through Figure 2-7 show examples of how the software stack is used. Figure 2-4

shows the software stack at device initialization. W15 has been initialized to 0x0800. This

example assumes the values 0x5A5A and 0x3636 have been written to W0 and W1, respectively.

The stack is pushed for the first time in Figure 2-5 and the value contained in W0 is copied to the

stack. W15 is automatically updated to point to the next available stack location (0x0802). In

Figure 2-6, the contents of W1 are pushed onto the stack. Figure 2-7 shows how the stack is

popped and the Top-of-Stack (TOS) value (previously pushed from W1) is written to W3.

Note: To protect against misaligned stack access, W15<0> is set to ‘0’ by the hardware.

PC<15:0>

PC<22:16>

015

W15 (before CALL)

W15 (after CALL)

Stack Grows Towards

Higher Address

B‘000000000’

CALL SUBR

PUSH W0

This syntax is equivalent to:

MOV W0,[W15++]

The contents of the top-of-stack can be returned to W0 by:

POP W0

This syntax is equivalent to:

MOV [--W15],W0

dsPIC33F/PIC24H Family Reference Manual

Figure 2-4: Stack Pointer at Device Reset

Figure 2-5: Stack Pointer After the First PUSH Instruction

Figure 2-6: Stack Pointer After the Second PUSH Instruction

Figure 2-7: Stack Pointer After a POP Instruction

2.3.2 W14 Software Stack Frame Pointer

A frame is a user-defined section of memory in the stack that is used by a single subroutine.

Working register W14 can be used as a stack frame pointer with the LNK (link) and ULNK (unlink)

instructions. W14 can be used in a normal working register by instructions when it is not used as

a frame pointer.

For software examples that use W14 as a stack frame pointer, refer to the “16-bit MCU and DSC

Programmer’s Reference Manual” (DS70157).

0x0000

0xFFFE

0x0800

W15

W15 = 0x0800

W0 = 0x5A5A

W1 = 0x3636

0x0000

0xFFFE

0x5A5A

W15 = 0x0802

W0 = 0x5A5A

W1 = 0x3636

0x0800 PUSH W0

0x0802

W15

0x0000

0xFFFE

0x5A5A

0x3636

W15 = 0x0804

W0 = 0x5A5A

W1 = 0x3636

0x0800 PUSH W1

0x0802

0x0804

W15

0x0000

0xFFFE

0x05A5A

0x03636

0x3636 → W3

W15 = 0x0802

POP W3

0x0802

0x0800

W15

Section 2. CPU

CPU

2.3.3 Stack Pointer Overflow

The Stack Pointer Limit register (SPLIM) specifies the size of the stack buffer. The SPLIM register

is a 16-bit register, but SPLIM<0> is set to ‘0’ because all stack operations must be word-aligned.

The stack overflow check is not enabled until a word write to the SPLIM register occurs. After this

time it can only be disabled by a device Reset. All effective addresses generated using W15 as

a source or destination are compared against the value in SPLIM. If the contents of the

Stack Pointer (W15) exceed the contents of the SPLIM register by 2, and a Push operation is

performed, a stack error trap occurs on a subsequent Push operation. Thus, for example, if it is

desirable to cause a stack error trap when the stack grows beyond address 0x2000 in RAM,

initialize the SPLIM register with the value 0x1FFE.

If stack overflow checking is enabled, a stack error trap also occurs, if the W15 effective address

calculation wraps over the end of data space (0xFFFF).

Refer to Section 6. “Interrupts” (DS70184), for more details on stack error traps.

2.3.4 Stack Pointer Underflow

The stack is initialized to 0x0800 during a Reset. A stack error trap is initiated, if the Stack Pointer

address is less than 0x0800.

Note: A stack error trap can be caused by any instruction that uses the contents of the

W15 register to generate an Effective Address (EA). Thus, if the contents of W15

are greater than the contents of the SPLIM register by 2, and a CALL instruction is

executed, or if an interrupt occurs, a stack error trap is generated.

Note: A write to the SPLIM register should not be followed by an indirect read operation

using W15.

Note: Locations in data space between 0x0000 and 0x07FF are, in general, reserved for

core and peripheral Special Function Registers (SFRs).

dsPIC33F/PIC24H Family Reference Manual

2.4 CPU REGISTER DESCRIPTIONS

2.4.1 CPU Status Register: SR

The dsPIC33F/PIC24H CPU has a 16-bit Status Register (SR). A detailed description of the CPU

Status Register is shown in Register 2-1. The LSB of this register is referred to as the SRL (Status

The SRL register contains all of the MCU ALU operation Status flags, the CPU Interrupt Priority

Level Status bits (IPL<2:0>), and the REPEAT Loop Active Status bit, RA (SR<4>). During

exception processing, SRL is concatenated with the MSB of the PC, to form a complete word

value, which is then stacked.

The SRH register contains:

• The DSP Adder/Subtractor status bits

•The DO loop active bit, DA (SR<9>)

• The Digit Carry bit, DC (SR<8>)

The SR register bits are readable/writable with the following exceptions:

• The DA bit (SR<8>) is read-only

• The RA bit (SR<4>) is read-only

• The OA, OB (SR<15:14>) and OAB (SR<11>) bits are read-only and can only be modified

by the DSP engine hardware

• The SA, SB (SR<13:12>) and SAB (SR<10>) bits are read- and clear-only, and can only be

set by the DSP engine hardware. Once set, they remain set until cleared by the

user-assigned application, irrespective of the results from any subsequent DSP operations.

2.4.2 Core Control Register: CORCON

The CORCON register contains bits that control the operation of the DSP multiplier and DO loop

hardware. The CORCON register contains the IPL3 status bit, which is concatenated with

IPL<2:0> (SR<7:5>), to form the CPU Interrupt Priority Level (IPL). A detailed description of the

CORCON register is provided in Register 2-2.

Note 1: Clearing the SAB bit also clears both the SA and SB bits.

2: A description of the SR bits affected by each instruction is provided in the “16-bit

MCU and DSC Programmer’s Reference Manual” (DS70157).

Section 2. CPU

CPU

R-0 R-0 R/C-0 R/C-0 R-0 R/C-0 R -0 R/W-0

OA OB SA(1) SB(2) OAB SAB(3) DA DC

bit 15 bit 8

R/W-0(4,5) R/W-0(4,5) R/W-0(4,5) R-0 R/W-0 R/W-0 R/W-0 R/W-0

IPL<2:0> RA N OV Z C

bit 7 bit 0

Legend: C = Clearable bit

R = Readable bit W = Writable bit U = Unimplemented bit, read as ‘0’

-n = Value at POR ‘1’ = Bit is set ‘0’ = Bit is cleared x = Bit is unknown

bit 15 OA: Accumulator A Overflow Status bit

1 = Accumulator A has overflowed

0 = Accumulator A has not overflowed

bit 14 OB: Accumulator B Overflow Status bit

1 = Accumulator B has overflowed

0 = Accumulator B has not overflowed

bit 13 SA: Accumulator A Saturation ‘Sticky’ Status bit(1)

1 = Accumulator A is saturated or has been saturated at some time

0 = Accumulator A is not saturated

bit 12 SB: Accumulator B Saturation ‘Sticky’ Status bit(2)

1 = Accumulator B is saturated or has been saturated at some time

0 = Accumulator B is not saturated

bit 11 OAB: OA || OB Combined Accumulator Overflow Status bit

1 = Accumulators A or B have overflowed

0 = Neither Accumulators A or B have overflowed

bit 10 SAB: SA || SB Combined Accumulator ‘Sticky’ Status bit(3)

1 = Accumulators A or B are saturated or have been saturated at some time

0 = Neither Accumulator A or B are saturated

bit 9 DA: DO Loop Active bit

1 = DO loop in progress

0 = DO loop not in progress

bit 8 DC: MCU ALU Half Carry/Borrow bit

1 = A carry-out from the 4th low order bit (for byte-sized data) or 8th low order bit (for word-sized data)

of the result occurred

0 = No carry-out from the 4th low order bit (for byte-sized data) or 8th low order bit (for word-sized data)

of the result occurred

Note 1: The SA bit can be read or cleared (not set).

2: The SB bit can be read or cleared (not set).

3: The SAB bit can be read or cleared (not set). Clearing this bit clears SA and SB.

4: The IPL<2:0> bits are concatenated with the IPL<3> bit (CORCON<3>) to form the CPU Interrupt Priority

Level. The value in parentheses indicates the IPL, if IPL<3> = 1. User interrupts are disabled when

IPL<3> = 1.

5: The IPL<2:0> status bits are read only when NSTDIS = 1 (INTCON1<15>).

dsPIC33F/PIC24H Family Reference Manual

bit 7-5 IPL<2:0>: CPU Interrupt Priority Level Status bits(4,5)

111 = CPU Interrupt Priority Level is 7 (15). User interrupts disabled

110 = CPU Interrupt Priority Level is 6 (14)

101 = CPU Interrupt Priority Level is 5 (13)

100 = CPU Interrupt Priority Level is 4 (12)

011 = CPU Interrupt Priority Level is 3 (11)

010 = CPU Interrupt Priority Level is 2 (10)

001 = CPU Interrupt Priority Level is 1 (9)

000 = CPU Interrupt Priority Level is 0 (8)

bit 4 RA: REPEAT Loop Active bit

1 = REPEAT loop in progress

0 = REPEAT loop not in progress

bit 3 N: MCU ALU Negative bit

1 = Result was negative

0 = Result was non-negative (zero or positive)

bit 2 OV: MCU ALU Overflow bit

This bit is used for signed arithmetic (2’s complement). It indicates an overflow of the magnitude that

causes the sign bit to change state.

1 = Overflow occurred for signed arithmetic (in this arithmetic operation)

0 = No overflow occurred

bit 1 Z: MCU ALU Zero bit

1 = An operation that affects the Z bit has set it at some time in the past

0 = The most recent operation that affects the Z bit has cleared it (i.e., a non-zero result)

bit 0 C: MCU ALU Carry/Borrow bit

1 = A carry-out from the Most Significant bit (MSb) of the result occurred

0 = No carry-out from the Most Significant bit (MSb) of the result occurred

Note 1: The SA bit can be read or cleared (not set).

2: The SB bit can be read or cleared (not set).

3: The SAB bit can be read or cleared (not set). Clearing this bit clears SA and SB.

4: The IPL<2:0> bits are concatenated with the IPL<3> bit (CORCON<3>) to form the CPU Interrupt Priority

Level. The value in parentheses indicates the IPL, if IPL<3> = 1. User interrupts are disabled when

IPL<3> = 1.

5: The IPL<2:0> status bits are read only when NSTDIS = 1 (INTCON1<15>).

Section 2. CPU

CPU

U-0 U-0 U-0 R/W-0 R/W-0 R-0 R-0 R-0

— — —USEDT

(1) DL<2:0>

bit 15 bit 8

R/W-0 R/W-0 R/W-1 R/W-0 R/C-0 R/W-0 R/W-0 R/W-0

SATA SATB SATDW ACCSAT IPL3(2) PSV RND IF

bit 7 bit 0

Legend: C = Clearable bit

R = Readable bit W = Writable bit U = Unimplemented bit, read as ‘0’

-n = Value at POR ‘1’ = Bit is set ‘0’ = Bit is cleared x = Bit is unknown

bit 15-13 Unimplemented: Read as '0’

bit 12 US: DSP Multiply Unsigned/Signed Control bit

1 = DSP engine multiplies are unsigned

0 = DSP engine multiplies are signed

bit 11 EDT: Early DO Loop Termination Control bit(1)

1 = Terminate executing DO loop at end of current loop iteration

0 = No effect

bit 10-8 DL<2:0>: DO Loop Nesting Level Status bits

111 = 7 DO loops active

•

001 = 1 DO loop active

000 = 0 DO loop active

bit 7 SATA: ACCA Saturation Enable bit

1 = Accumulator A saturation enabled

0 = Accumulator A saturation disabled

bit 6 SATB: ACCB Saturation Enable bit

1 = Accumulator B saturation enabled

0 = Accumulator B saturation disabled

bit 5 SATDW: Data Space Write from DSP Engine Saturation Enable bit

1 = Data space write saturation enabled

0 = Data space write saturation disabled

bit 4 ACCSAT: Accumulator Saturation Mode Select bit

1 = 9.31 saturation (super saturation)

0 = 1.31 saturation (normal saturation)

bit 3 IPL3: CPU Interrupt Priority Level Status bit 3(2)

1 = CPU interrupt priority level is greater than 7

0 = CPU interrupt priority level is 7 or less

bit 2 PSV: Program Space Visibility in Data Space Enable bit

1 = Program space visible in data space

0 = Program space not visible in data space

bit 1 RND: Rounding Mode Select bit

1 = Biased (conventional) rounding enabled

0 = Unbiased (convergent) rounding enabled

bit 0 IF: Integer or Fractional Multiplier Mode Select bit

1 = Integer mode enabled for DSP multiply

0 = Fractional mode enabled for DSP multiply

Note 1: The EDT bit always reads as ‘0’.

2: The IPL3 bit is concatenated with the IPL<2:0> bits (SR<7:5>) to form the CPU interrupt priority level.

dsPIC33F/PIC24H Family Reference Manual

2.4.3 Other dsPIC33F/PIC24H CPU Control Registers

This section provides brief descriptions of additional registers that are associated with the

dsPIC33F/PIC24H CPU. These registers are described in greater detail in other sections of the

“dsPIC33F/PIC24H Family Reference Manual”.

•TBLPAG: Table Page Register

The TBLPAG register holds the upper eight bits of a program memory address during the

table read and write operations. Table instructions are used to transfer data between

program memory space and data memory space. Refer to Section 4. “Program

Memory” (DS70203), for more details.

• PSVPAG: Program Space Visibility Page Register

The PSVPAG register allows the user-assigned application to map a 32 Kbytes section of

the program memory space into the upper 32 Kbytes of data address space. This feature

allows transparent access to constant data through dsPIC33F/PIC24H instructions that

operate on data memory. The PSVPAG register selects the 32 Kbytes region of program

memory space that is mapped to the data address space. Refer to Section 4. “Program

Memory” (DS70203), for more details on the PSVPAG register.

• MODCON: Modulo Control Register

The MODCON register enables and configures modulo addressing (circular buffers). Refer

to Section 3. “Data Memory” (DS70202), for more details on modulo addressing.

• XMODSRT, XMODEND: X Modulo Start and End Address Registers

The XMODSRT and XMODEND registers hold the start and end addresses for

modulo (circular) buffers implemented in the X data memory address space. Refer to

Section 3. “Data Memory” (DS70202), for more details on modulo addressing.

• YMODSRT, YMODEND: Y Modulo Start and End Address Registers

The YMODSRT and YMODEND registers hold the start and end addresses for

modulo (circular) buffers implemented in the Y data memory address space. Refer to

Section 3. “Data Memory” (DS70202), for more details on modulo addressing.

• XBREV: X Modulo Bit-Reverse Register

The XBREV register sets the buffer size used for bit-reversed addressing. Refer to

Section 3. “Data Memory” (DS70202), for more details on modulo addressing.

• DISICNT: Disable Interrupts Count Register

The DISICNT register is used by the DISI instruction to disable interrupts of priority 1-6 for

a specified number of cycles. Refer to Section 6. “Interrupts” (DS70184), for more details

on interrupts.

Section 2. CPU

CPU

2.5 ARITHMETIC LOGIC UNIT

The dsPIC33F/PIC24H Arithmetic Logic Unit (ALU) is 16 bits wide and is capable of addition,

subtraction, single bit shifts, and logic operations. Unless otherwise mentioned, arithmetic

operations are 2’s complement in nature. Depending on the operation, the ALU can affect the

values of these status bits in the SR register:

• Carry (C)

•Zero (Z)

• Negative (N)

• Overflow (OV)

• Digit Carry (DC)

The C and DC status bits operate as Borrow and Digit Borrow bits, respectively, for subtraction

operations.

The ALU can perform 8-bit or 16-bit operations, depending on the mode of the instruction used.

Data for the ALU operation can come from the W register array, or data memory, depending on

the addressing mode of the instruction. Likewise, output data from the ALU can be written to the

W register array or a data memory location.

Refer to the “16-bit MCU and DSC Programmer’s Reference Manual” (DS70157), for more

details on the SR bits affected by each instruction, addressing modes and 8-bit/16-bit instruction

modes.

2.5.1 Byte to Word Conversion

The dsPIC33F/PIC24H has two instructions that are helpful when mixing 8-bit and 16-bit ALU

operations:

• The Sign-Extend (SE) instruction takes a byte value in a W register or data memory and

creates a sign-extended word value that is stored in a W register.

• The Zero-Extend (ZE) instruction clears the 8 MSb of a word value in a W register or data

memory and places the result in a destination W register.

Note 1: Byte operations use the 16-bit ALU and can produce results in excess of eight bits.

However, to maintain backward compatibility with PIC MCU devices, the ALU result

from all of the byte operations is written back as a byte (i.e., MSB is not modified),

and the SR register is updated based only on the LSB state of the result.

2: All register instructions performed in Byte mode affect only the LSB of the

W registers. The MSB of any W register can be modified by using file register

instructions that access the memory mapped contents of the W registers.

dsPIC33F/PIC24H Family Reference Manual

2.6 DSP ENGINE

The DSP engine is a block of hardware that is fed with data from the W register array, but

contains its own specialized result registers. The DSP engine is driven from the same instruction

decoder that directs the MCU ALU. In addition, all operand Effective Addresses (EAs) are

generated in the W register array. Concurrent operation with MCU instruction flow is not possible,

though both the MCU ALU and DSP engine resources can be shared by all instructions in the

instruction set.

The DSP engine consists of the following components:

• High-speed 17-bit by 17-bit multiplier

• Barrel shifter

• 40-bit adder/subtractor

• Two target accumulator registers

• Rounding logic with selectable modes

• Saturation logic with selectable modes

Data input to the DSP engine is derived from one of the following sources:

• Directly from the W array (registers W4, W5, W6 or W7) for dual source operand DSP

instructions. Data values for the W4, W5, W6 and W7 registers are prefetched via the

X and Y memory data buses

• From the X memory data bus for all other DSP instructions

Data output from the DSP engine is written to one of the following destinations:

• The target accumulator, as defined by the DSP instruction being executed

• The X memory data bus to any location in the data memory address space

The DSP engine can perform inherent accumulator-to-accumulator operations that require no

additional data.

The MCU shift and multiply instructions use the DSP engine hardware to obtain their results. The

X memory data bus is used for data reads and writes in these operations.

Figure 2-8 shows the block diagram of the DSP engine.

Note: Refer to the “16-bit MCU and DSC Programmer’s Reference Manual” (DS70157),

for detailed code examples and instruction syntax related to this section.

Section 2. CPU

CPU

Figure 2-8: DSP Engine Block Diagram

Zero Backfill

Sign-Extend

Barrel

Shifter

40-bit Accumulator A

40-bit Accumulator B

Round Logic

X Data Bus

To/From W Array

Adder

Saturate

Negate

16 16

40 40

Y Data Bus

Multiplier/Scaler

17-bit x 17-bit

16-bit to 17-bit

Conversion

Saturation Logic

dsPIC33F/PIC24H Family Reference Manual

2.6.1 Data Accumulators

Two 40-bit data accumulators, ACCA and ACCB, are the result registers for the DSP instructions

listed in Table 2-3. Each accumulator is memory mapped to these three registers, where ‘x’

denotes the particular accumulator:

• ACCxL: ACCx<15:0>

• ACCxH: ACCx<31:16>

• ACCxU: ACCx<39:32>

For fractional operations that use the accumulators, the radix point is located to the right of

bit 31. The range of fractional values that can be stored in each accumulator is -256.0 to

(256.0 – 2-31).

For integer operations that use the accumulators, the radix point is located to the right of bit 0.

The range of integer values that can be stored in each accumulator is -549,755,813,888 to

549,755,813,887.

2.6.2 Multiplier

The dsPIC33F/PIC24H CPU features a 17-bit by 17-bit multiplier (for 16-bit Signed

multiplication) shared by both the MCU ALU and the DSP engine. The multiplier is capable of

signed or unsigned operation and supports either 1.31 fractional (Q.31) or 32-bit integer results.

The multiplier takes in 16-bit input data and converts the data to 17 bits. Signed operands to the

multiplier are sign-extended. Unsigned input operands are zero-extended. The internal 17-bit

representation of data in the multiplier allows correct execution of mixed-sign and unsigned

16-bit by 16-bit multiplication operations.

The representation of data in hardware for Integer and Fractional Multiplier modes is as follows:

• Integer data is inherently represented as a signed 2’s complement value, where the Most

Significant bit is defined as a sign bit. Generally, the range of an N-bit 2’s complement

integer is -2N-1 to 2N-1 – 1

• Fractional data is represented as a 2’s complement fraction where the MSb is defined as a

sign bit and the radix point is implied to lie just after the sign bit (Q.X format). The range of

an N-bit 2’s complement fraction with this implied radix point is -1.0 to (1 – 21-N).

The range of data in both Integer and Fractional modes is listed in Table 2-2. Figure 2-9 and

Figure 2-10 illustrate how the multiplier hardware interprets data in Integer and Fractional modes.

The Integer or Fractional Multiplier Mode Select bit (IF) in the Core Control register

(CORCON<0>) determines integer/fractional operation for the instructions listed in Table 2-3.

The IF bit does not affect MCU multiply instructions listed in Table 2-4, which are always integer

operations. The multiplier scales the result one bit to the left for fractional operation. The Least

Significant bits (LSbs) of the result are always cleared. The multiplier defaults to Fractional mode

for DSP operations at a device Reset.

Table 2-2: dsPIC33F/PIC24H Data Ranges

Size Integer Range Fraction Range Fraction

Resolution

16-bit -32768 to 32767 -1.0 to (1.0 – 2-15)

(Q.15 Format) 3.052 x 10-5

32-bit -2,147,483,648 to

2,147,483,647 -1.0 to (1.0 – 2-31)

(Q.31 Format) 4.657 x 10-10

40-bit -549,755,813,888 to

549,755,813,887 -256.0 to (256.0 – 2-31)

(Q.31 Format with eight guard bits) 4.657 x 10-10

Section 2. CPU

CPU

Figure 2-9: Integer and Fractional Representation of 0x4001

Figure 2-10: Integer and Fractional Representation of 0xC002

Different Representations of 0x4001

Integer:

-215 214 213 212 . . . .

0x4001 = 214 + 20 = 16385

1.15 Fractional:

2-15

2-1 2-2 2-3 . . .

-20

0x4001 = 2-1 + 2-15 = 0.500030518

Implied Radix Point

1 10000000000000

01 10000000000000

Different Representations of 0xC002

Integer:

-215 214 213 212 . . . .

0xC002 = -215 + 214 + 20 = -32768 + 16384 + 2 = -16382

1.15 Fractional:

2-15

. 2-1 2-2 2-3 . . .

-20

0xC002 = -20 + 2-1 + 2-14 = -1 + 0.5 + 0.000061035 = -0.499938965

Implied Radix Point

11 00000000000001

dsPIC33F/PIC24H Family Reference Manual

2.6.2.1 DSP MULTIPLY INSTRUCTIONS

The DSP instructions that use the multiplier are summarized in Table 2-3.

Table 2-3: DSP Instructions that Use the Multiplier

The DSP Multiplier Unsigned/Signed Control bit (US) in the CORCON register determines

whether DSP multiply instructions are signed (default) or unsigned. The US bit does not influence

the MCU multiply instructions, which have specific instructions for signed or unsigned operation.

If the US bit is set, the input operands for instructions shown in Table 2-3 are considered as

unsigned values, which are always zero-extended into the seventeenth bit of the multiplier value.

2.6.2.2 MCU MULTIPLY INSTRUCTIONS

The same multiplier supports the MCU multiply instructions, which include integer 16-bit signed,

unsigned, and mixed sign multiplies as shown in Table 2-4. All multiplications performed by the

MUL instruction produce integer results. The MUL instruction can be directed to use byte-sized or

word-sized operands. Byte input operands produce a 16-bit result and word input operands

produce a 32-bit result to the specified register(s) in the W array.

Table 2-4: MCU Instructions that Utilize the Multiplier

Table 2-5: Multiplication Options

DSP Instruction Description Algebraic Equivalent

MAC Multiply and Add to Accumulator or

Square and Add to Accumulator

a = a + b * c

a = a + b2

MSC Multiply and Subtract from Accumulator a = a – b * c

MPY Multiply a = b * c

MPY.N Multiply and Negate Result a = -b * c

ED Partial Euclidean Distance a = (b – c)2

EDAC Add Partial Euclidean Distance to the

Accumulator a = a + (b – c)2

Note: DSP instructions using the multiplier can operate in Fractional (1.15) or Integer

modes.

MCU Instruction Description

MUL/MUL.UU Multiply two unsigned integers

MUL.SS Multiply two signed integers

MUL.SU/MUL.US Multiply a signed integer with an unsigned integer

Note 1: MCU instructions using the multiplier operate only in Integer mode.

2: Result of an MCU multiply is 32 bits long and is stored in a pair of W registers.

Instruction Class Signed x

Signed Unsigned x

Unsigned Unsigned x

Signed Signed x

Unsigned

MAC

(DSP Multiplication) Yes

(Integer or

Fractional)

Yes

(Integer or

Fractional)

No No

MUL

(MCU Multiplication) Yes

(Integer only) Yes

(Integer only)

Section 2. CPU

CPU

2.6.3 Data Accumulator Adder/Subtractor

The data accumulators have a 40-bit adder/subtractor with automatic sign extension logic for the

multiplier result (if signed). It can select one of two accumulators (A or B) as its pre-accumulation

source and post-accumulation destination. For the ADD (accumulator) and LAC instructions, the

data to be accumulated or loaded can optionally be scaled via the barrel shifter prior to

accumulation.

The 40-bit adder/subtractor can optionally negate one of its operand inputs to change the sign of

the result (without changing the operands). The negate is used during multiply and subtract

(MSC), or multiply and negate (MPY.N) operations.

The 40-bit adder/subtractor has an additional saturation block that controls accumulator data

saturation, if enabled.

2.6.3.1 ACCUMULATOR STATUS BITS

Six STATUS register bits that support saturation and overflow are located in the CPU STATUS

Table 2-6: Accumulator Overflow and Saturation Status Bits

The OA and OB bits are read-only and are modified each time the data is passed through the

accumulator add/subtract logic. When set, they indicate that the most recent operation has

overflowed into the accumulator guard bits (bit 32 through bit 39). This type of overflow is not

catastrophic; the guard bits preserve the accumulator data. The OAB status bit is the logically

OR value of OA and OB.

The OA and OB bits, when set, can optionally generate an arithmetic error trap. The trap is

enabled by setting the corresponding Overflow Trap Flag Enable bit (OVATE or OVBTE) in

Interrupt Control Register 1 (INTCON1<10> or <9>) in the Interrupt controller. The trap event

allows the user to perform immediate corrective action, if desired.

The SA and SB bits can be set each time data passes through the accumulator saturation logic.

Once set, these bits remain set until cleared by the user-assigned application. The SAB status

bit indicates the logical OR value of SA and SB. The SA and SB bits are cleared when SAB is

cleared. When set, these bits indicate that the accumulator has overflowed its maximum range

(bit 31 for 32-bit saturation or bit 39 for 40-bit saturation) and are saturated (if saturation is

enabled).

When saturation is not enabled, the SA and SB bits indicate that a catastrophic overflow has

occurred (the sign of the accumulator has been destroyed). If the Catastrophic Overflow Trap

Enable bit (COVTE) in the INTCON1 register is set, SA and SB bits will generate an arithmetic

error trap when saturation is disabled.

Status Bit Location Description

OA SR<15> Accumulator A overflowed into guard bits (ACCA<39:32>)

OB SR<14> Accumulator B overflowed into guard bits (ACCB<39:32>)

SA SR<13> ACCA saturated (bit 31 overflow and saturation)

ACCA overflowed into guard bits and saturated

(bit 39 overflow and saturation)

SB SR<12> ACCB saturated (bit 31 overflow and saturation)

ACCB overflowed into guard bits and saturated

(bit 39 overflow and saturation)

OAB SR<11> OA logically ORed with OB

SAB SR<10> SA logically ORed with SB

Clearing SAB also clears SA and SB

Note 1: Refer to Section 6. “Interrupts” (DS70184), for more details on arithmetic warning

traps.

2: The SA, SB and SAB status bits can have different meaning depending on whether

accumulator saturation is enabled. The Accumulator Saturation mode is controlled

via the CORCON register.

dsPIC33F/PIC24H Family Reference Manual

2.6.3.2 SATURATION AND OVERFLOW MODES

The dsPIC33F/PIC24H CPU supports three saturation and overflow modes: Accumulator 39-bit

Saturation, Accumulator 31-bit Saturation and Accumulator Catastrophic Overflow.

• Accumulator 39-bit Saturation

In this mode, the saturation logic loads the maximally positive 9.31 value (0x7FFFFFFFFF),

or maximally negative 9.31 value (0x8000000000), into the target accumulator. The SA or

SB bit is set and remains set until cleared by the user-assigned application. This saturation

mode is useful for extending the dynamic range of the accumulator.

To configure for this mode of saturation, set the Accumulator Saturation Mode Select bit

(ACCSAT) in the CORCON register. Additionally, set the ACCA Saturation Enable bit (SATA)

in the CORCON register, and/or the ACCB Saturation Enable bit (SATB) in the CORCON

• Accumulator 31-bit Saturation

In this mode, the saturation logic loads the maximally positive 1.31 value (0x007FFFFFFF)

or maximally negative 1.31 value (0xFF80000000) into the target accumulator. The SA or

SB bit is set and remains set until cleared by the user. When this saturation mode is in effect,

the guard bits 32 through 39 are not used except for sign-extension of the accumulator

value. Consequently, the OA, OB or OAB bits in SR are never set.

To configure for this mode of overflow and saturation, the ACCSAT bit (CORCON<4>) must

be cleared. Additionally, the SATA (CORCON<7>) and/or SATB (CORCON<6>) bits must be

set to enable accumulator saturation.

• Accumulator Catastrophic Overflow

If the SATA (CORCON<7>) and/or SATB (CORCON<6>) bits are not set, then no saturation

operation is performed on the accumulator, and the accumulator is allowed to overflow all

the way up to bit 39 (destroying its sign). If the Catastrophic Overflow Trap Enable bit

(COVTE) (INTCON1<8> in the interrupt controller) is set, a catastrophic overflow initiates an

arithmetic error trap.

Accumulator saturation and overflow detection can only result from the execution of a DSP

instruction that modifies one of the two accumulators via the 40-bit DSP ALU. Saturation and

overflow detection does not take place when the accumulators are accessed as memory mapped

registers through the MCU class instructions. Furthermore, the accumulator status bits shown in

Table 2-6 are not modified. However, the MCU status bits (Z, N, C, OV, DC) will be modified,

depending on the MCU instruction that accesses the accumulator.

2.6.3.3 DATA SPACE WRITE SATURATION

In addition to adder/subtractor saturation, writes to data space can be saturated without affecting

the contents of the source accumulator. This feature allows data to be limited, while not

sacrificing the dynamic range of the accumulator during intermediate calculation stages. Data

space write saturation is enabled by setting the Data Space Write from DSP Engine Saturation

Enable bit (SATDW) in the CORCON register. Data space write saturation is enabled by default

at a device Reset.

The data space write saturation feature works with the SAC and SAC.R instructions. The value

held in the accumulator is never modified when these instructions are executed. The hardware

performs the following steps to obtain the saturated write result:

1. The read data is scaled based upon the arithmetic shift value specified in the instruction.

2. The scaled data is rounded (SAC.R only).

3. The scaled/rounded value is saturated to a 16-bit result based on the value of the guard

bits. For data values greater than 0x007FFF, the data written to memory is saturated to

the maximum positive 1.15 value, 0x7FFF. For input data less than 0xFF8000, data written

to memory is saturated to the maximum negative 1.15 value, 0x8000.

Note: Refer to Section 6. “Interrupts” (DS70184), for more details on arithmetic error

traps.

Section 2. CPU

CPU

2.6.3.4 ACCUMULATOR ‘WRITE BACK’

The MAC and MSC instructions can optionally write a rounded version of the accumulator that is

not the target of the current operation into data space memory. The write is performed across the

X bus into combined X and Y address space. This accumulator write-back feature is beneficial

in certain FFT and LMS algorithms.

The following addressing modes are supported by the accumulator write back hardware:

• W13, register direct: The rounded contents of the non-target accumulator are written into

W13 as a 1.15 fractional result

• [W13]+=2, register indirect with post-increment: The rounded contents of the non-target

accumulator are written into the address pointed to by W13 as a 1.15 fraction. W13 is then

incremented by 2.

2.6.4 Round Logic

The round logic can perform a conventional (biased) or convergent (unbiased) round function

during an accumulator write (store). The Round mode is determined by the state of the Rounding

Mode Select bit (RND) in the CORCON register. It generates a 16-bit, 1.15 data value, which is

passed to the data space write saturation logic. If rounding is not indicated by the instruction, a

truncated 1.15 data value is stored.

The two rounding modes are shown in Figure 2-11. Conventional rounding takes bit 15 of the

accumulator, zero-extends it and adds it to the most significant word (msw), excluding the guard

or overflow bits (bits 16 through 31). If the least significant word (lsw) of the accumulator is

between 0x8000 and 0xFFFF (0x8000 included), the msw is incremented. If the lsw of the

accumulator is between 0x0000 and 0x7FFF, the msw does not change. A consequence of this

algorithm is that over a succession of random rounding operations, the value tends to be biased

slightly positive.

Convergent (or unbiased) rounding operates in the same manner as conventional rounding

except when the lsw equals 0x8000. If this is the case, the LSb of the msw (bit 16 of the

accumulator) is examined. If it is ‘1’, the msw is incremented. If it is ‘0’, the msw is not modified.

Assuming that bit 16 is effectively random in nature, this scheme removes any rounding bias that

may accumulate.

The SAC and SAC.R instructions store either a truncated (SAC) or rounded (SAC.R) version of

the contents of the target accumulator to data memory via the X-bus (subject to data saturation).

Refer to 2.6.3.3 “Data Space Write Saturation”, for more details.

For the MAC class of instructions, the accumulator write-back data path is always subject to

rounding.

Figure 2-11: Conventional and Convergent Rounding Modes

01516

1000 0000 0000 00001XXX XXXX XXXX XXXX

0XXX XXXX XXXX XXXX 1000 0000 0000 0000

Conventional (Biased) Convergent (Unbiased)

Round Up (add 1 to msw) when:

Round Down (add nothing) when:

Round Up (add 1 to msw) when:

1. lsw = 0x8000 and bit 16 = 1

2. lsw > 0x8000

lsw >= 0x8000

lsw < 0x8000

Round Down (add nothing) when:

1. lsw = 0x8000 and bit 16 = 0

2. lsw < 0x8000

msw

dsPIC33F/PIC24H Family Reference Manual

2.6.5 Barrel Shifter

The barrel shifter can perform up to a 16-bit arithmetic right shift, or up to a 16-bit left shift, in a

single cycle. DSP or MSU instructions can use the barrel shifter for multi-bit shifts.

The shifter requires a signed binary value to determine both the magnitude (number of bits) and

direction of the shift operation:

• A positive value shifts the operand right

• A negative value shifts the operand left

• A value of ‘0’ does not modify the operand

The barrel shifter is 40 bits wide to accommodate the width of the accumulators. A 40-bit output

result is provided for DSP shift operations, and a 16-bit result is provided for MCU shift

operations.

Table 2-7 provides a summary of instructions that use the barrel shifter.

Table 2-7: Instructions that Use the DSP Engine Barrel Shifter

2.6.6 DSP Engine Mode Selection

The operational characteristics of the DSP engine discussed in previous sections can be

selected through the CPU Core Configuration register (CORCON):

• Fractional or integer multiply operation

• Conventional or convergent rounding

• Automatic saturation on/off for ACCA

• Automatic saturation on/off for ACCB

• Automatic saturation on/off for writes to data memory

• Accumulator saturation mode selection

2.6.7 DSP Engine Trap Events

The following arithmetic error traps can be generated for handling exceptions in the DSP engine

and are selected through the Interrupt Control register (INTCON1):

• Trap on ACCA overflow enable, using OVATE (INTCON1<10>)

• Trap on ACCB overflow enable, using OVBTE (INTCON1<9>)

• Trap on catastrophic ACCA and/or ACCB overflow enable, using COVTE (INTCON1<8>)

Occurrence of the traps is indicated by the following error status bits:

• OVAERR (INTCON1<14>)

• OVBERR (INTCON1<13>)

• COVAERR (INTCON1<12>)

• COVBERR (INTCON1<11>)

An arithmetic error trap is also generated when the user-assigned application attempts to shift a

value beyond the maximum allowable range (±16 bits) using the SFTAC instruction. This trap

source cannot be disabled, and is indicated by the Shift Accumulator Error Status bit

(SFTACERR) (INTCON1<7> in the Interrupt controller). The instruction will execute, but the

results of the shift are not written to the target accumulator. Refer to Section 6. “Interrupts”

(DS70184), for more details on bits in the INTCON1 register and arithmetic error traps.

Instruction Description

ASR Arithmetic multi-bit right shift of data memory location

LSR Logical multi-bit right shift of data memory location

SL Multi-bit shift left of data memory location

SAC Store DSP accumulator with optional shift

SFTAC Shift DSP accumulator

Section 2. CPU

CPU

2.7 DIVIDE SUPPORT

The dsPIC33F/PIC24H supports the following types of division operations:

•DIVF: 16/16 signed fractional divide

•DIV.SD: 32/16 signed divide

•DIV.UD: 32/16 unsigned divide

•DIV.SW: 16/16 signed divide

•DIV.UW: 16/16 unsigned divide

The quotient for all divide instructions is placed in working register W0. The remainder is placed

in W1. The 16-bit divisor can be located in any W register. A 16-bit dividend can be located in

any W register and a 32-bit dividend must be located in an adjacent pair of W registers.

All divide instructions are iterative operations and must be executed 18 times within a REPEAT

loop. The developer is responsible for programming the REPEAT instruction. A complete divide

operation takes 19 instruction cycles to execute.

The divide flow is interruptible, just like any other REPEAT loop. All data is restored into the

respective data registers after each iteration of the loop. Therefore, the user-assigned application

is responsible for saving the appropriate W registers in the ISR. Although they are important to

the divide hardware, the intermediate values in the W registers have no meaning to the

user-assigned application. The divide instructions must be executed 18 times in a REPEAT loop

to produce a meaningful result.

A Divide-by-Zero error generates a math error trap. This condition is indicated by the Math Error

Status bit (DIV0ERR) (INTCON1<6> in the Interrupt controller). Refer to the “16-bit MCU and

DSC Programmer’s Reference Manual” (DS70157), for more details and programming examples

for the divide instructions.

dsPIC33F/PIC24H Family Reference Manual

2.8 INSTRUCTION FLOW TYPES

Most instructions in the dsPIC33F/PIC24H architecture occupy a single word of program memory

and execute in a single cycle. An instruction prefetch mechanism facilitates single cycle (1 TCY)

execution. However, some instructions take two or three instruction cycles to execute.

Consequently, there are seven different types of instruction flow in the dsPIC® DSC architecture,

which are listed below and are described in this section:

• 1 Instruction Word, 1 Instruction Cycle

• 1 Instruction Word, 2 Instruction Cycles

• 1 Instruction Word, 2 or 3 Instruction Cycles (Program Flow Changes)

• 1 Instruction Word, 3 Instruction Cycles (RETFIE, RETURN, RETLW)

• Table Read and Table Write Instructions

• 2 Instruction Words, 2 Instruction Cycles

• Address Register Dependencies

2.8.1 1 Instruction Word, 1 Instruction Cycle

These instructions take one instruction cycle to execute as shown in Figure 2-12. Most

instructions are one-word, one-cycle instructions.

Figure 2-12: Instruction Flow: One-Word, One-Cycle

2.8.2 1 Instruction Word, 2 Instruction Cycles

In these instructions, there is no prefetch flush. The only instructions of this type are the MOV.D

instructions (load and store double-word). Two cycles are required to complete these

instructions, as shown in Figure 2-13.

Figure 2-13: Instruction Flow: One-Word, Two-Cycle (MOV.D Operation)

TCY0TCY1TCY2TCY3TCY4TCY5

1. MOV #0x55AA,W0 Fetch 1 Execute 1

2. MOV W0,PORTA Fetch 2 Execute 2

3. MOV W0,PORTB Fetch 3 Execute 3

TCY0TCY1TCY2TCY3TCY4TCY5

1. MOV #0x1234,W0 Fetch 1 Execute 1

2. MOV.D [W0++],W1 Fetch 2 Execute 2

R/W Cycle 1

3. MOV #0x00AA,W1 Fetch 3 Execute 2

R/W Cycle 2

No Fetch Execute 3

4. MOV #0x00CC,W0 Fetch 4 Execute 4

Section 2. CPU

CPU

2.8.3 1 Instruction Word, 2 or 3 Instruction Cycles (Program Flow

Changes)

These instructions include relative call and branch instructions, and skip instructions. When an

instruction changes the PC (other than to increment it), the program memory prefetch data must

be discarded. This makes the instruction take two effective cycles to execute, as shown in

Figure 2-14.

Figure 2-14: Instruction Flow: One-Word, Two-Cycle (Program Flow Change)

Three cycles are required when a two-word instruction is skipped. In this case, the program

memory prefetch data is discarded and the second word of the two-word instruction is fetched.

Figure 2-15 shows the second word of the instruction is executed as a NOP.

Figure 2-15: Instruction Flow: One-Word, Three-Cycle (Two-Word Instruction Skipped)

2.8.4 1 Instruction Word, 3 Instruction Cycles (RETFIE, RETURN,

RETLW)

Figure 2-16 shows the RETFIE, RETURN and RETLW instructions, used to return from a

subroutine call or an ISR, take three instruction cycles to execute.

Figure 2-16: Instruction Flow: One-Word, Three-Cycle (RETURN, RETFIE, RETLW)

TCY0TCY1TCY2TCY3TCY4TCY5

1. MOV.B #0x55,W0 Fetch 1 Execute 1

2. BTSC PORTA,#3 Fetch 2 Execute 2

Skip Taken

3. ADD.B PORTA (executed as NOP) Fetch 3 Forced NOP

4. BRA SUB_1 Fetch 4 Execute 4

5. ADD.B PORTB (executed as NOP) Fetch 5 Forced NOP

6. SUB_1: Instruction @ address SUB_1 Fetch SUB_1

TCY0TCY1TCY2TCY3TCY4TCY5

1. BTSC SR,#Z Fetch 1 Execute 1,

Skip Taken

2. GOTO LABEL Fetch 2 Forced NOP

(GOTO 2nd word) Fetch 2nd

word of

GOTO

2nd word

executed as

a NOP

3. BCLR PORTB,#3 Fetch 3 Execute 3

4. MOV W0,W1 Fetch 4 Execute 4

TCY0TCY1TCY2TCY3TCY4TCY5

1. MOV #0x55AA,W0 Fetch 1 Execute 1

2. RETURN Fetch 2 Execute 2

3. (instruction in old program flow) Fetch 3 Execute 2

4. MOV W0, W3 (instruction in new program flow) No Fetch Execute 2

5. MOV W3, W5 Fetch 4 Execute 4

Fetch 5

dsPIC33F/PIC24H Family Reference Manual

2.8.5 Table Read/Write Instructions

These instructions suspend fetching to insert a read or write cycle to the program memory.

Figure 2-17 shows the instruction fetched while executing the table operation is saved for one

cycle and executed in the cycle immediately after the table operation.

Figure 2-17: Instruction Pipeline Flow: Table Operations

2.8.6 2 Instruction Words, 2 Instruction Cycles

In these instructions, the fetch after the instruction contains data. This results in a two-cycle

instruction, as shown in Figure 2-18. The second word of a two-word instruction is encoded so

that it executes as a NOP if it is fetched by the CPU, when the CPU did not first fetch the first word

of the instruction. This is important when a two-word instruction is skipped by a skip instruction

(refer to Figure 2-15).

Figure 2-18: Instruction Pipeline Flow: Two-Word, Two-Cycle

2.8.7 Address Register Dependencies

These are instructions that are subject to a stall due to data address dependency between the

X-data space read and write operations. An additional cycle is inserted to resolve the resource

conflict, and is discussed in 2.10 “Address Register Dependencies”.

Figure 2-19: Instruction Pipeline Flow: One-Word, One-Cycle (With Instruction Stall)

TCY0TCY1TCY2TCY3TCY4TCY5

1. MOV #0x1234,W0 Fetch 1 Execute 1

2. TBLRDL.w [W0++],W1 Fetch 2 Execute 2

3. MOV #0x00AA,W1 Fetch 3 PM Data

Read Cycle

Bus Read Execute 3

4. MOV #0x00CC,W0 Fetch 4 Execute 4

TCY0TCY1TCY2TCY3TCY4TCY5

1. MOV #0xAA55,W0 Fetch 1 Execute 1

2. GOTO LABEL Fetch 2L Update PC

Fetch 2H Forced NOP

3. LABEL: MOV W0,W2 Fetch 3 Execute 3

4. BSET PORTA, #3 Fetch 4 Execute 4

TCY0TCY1TCY2TCY3TCY4TCY5

1. MOV W0,W1 Fetch 1 Execute 1

2. MOV [W1],[W4] Fetch 2 Execute 1

Stall Execute 2

3. MOV W2,W1 Fetch 3 Execute 3

Note: If the RETURN instruction is placed at the end of the program memory, the illegal

address error trap will be generated by the device during the run-time. This is due

to the prefetch operation that will try to preload the next two instructions from the

memory location, which in this case do not exist. The solution is to leave two extra

instruction words available after the RETURN instruction, so that the compiler can

place NOP and RESET instructions at the end of the program memory.

Section 2. CPU

CPU

2.9 LOOP CONSTRUCTS

The dsPIC33F/PIC24H CPU supports both REPEAT and DO instruction constructs to provide

unconditional automatic program loop control. The REPEAT instruction implements a single

instruction program loop. The DO instruction implements a multiple instruction program loop. Both

instructions use control bits within the CPU Status register (SR) to temporarily modify the CPU

operation.

2.9.1 REPEAT Loop Construct

The REPEAT instruction causes the instruction that follows it to be repeated a specified number

of times. A literal value contained in the instruction or a value in one of the W registers can be

used to specify the REPEAT count value. The W register option enables the loop count to be a

software variable.

An instruction in a REPEAT loop is executed at least once. The number of iterations for a REPEAT

loop is the 14-bit literal value + 1, or Wn + 1.

The syntax for the two forms of the REPEAT instruction is as follows:

Example 2-7: REPEAT Instruction Syntax

2.9.1.1 REPEAT OPERATION

The loop count for REPEAT operations is held in the 14-bit REPEAT Loop Counter register

(RCOUNT), which is memory mapped. The RCOUNT register is initialized by the REPEAT

instruction. The REPEAT instruction sets the REPEAT Loop Active bit (RA) in the SR register to

‘1’, if the RCOUNT is a non-zero value.

The RA bit is read-only and cannot be modified through software. For REPEAT loop count values

greater than ‘0’, the PC is not incremented. Further, PC increments are inhibited until

RCOUNT = 0. For an instruction flow example of a REPEAT loop, refer to Figure 2-20.

For a loop count value equal to ‘0’, REPEAT has the effect of a NOP and the RA bit (SR<4>) is not

set. The REPEAT loop is essentially disabled before it begins, allowing the target instruction to

execute only once while prefetching the subsequent instruction (i.e., normal execution flow).

Figure 2-20: REPEAT Instruction Pipeline Flow

Note: The instruction immediately following the REPEAT instruction (i.e., the target

instruction) is always executed at least one time, and it is always executed one time

more than the value specified in the 14-bit literal or the W register operand.

REPEAT #lit14 ; RCOUNT <-- lit14

(Valid target Instruction)

REPEAT Wn ; RCOUNT <-- Wn

(Valid target Instruction)

TCY0TCY1TCY2TCY3TCY4TCY5

1.REPEAT #0x2 Fetch 1 Execute 1

2.MAC W4*W5,A,[W8]+=2,W4 Fetch 2 Execute 2

No Fetch Execute 2

3.BSET PORTA,#3 Fetch 3 Execute 3

PC (at end of

instruction)

PC PC+2 PC+2 PC+2 PC+4 PC+6

RCOUNT

(at end of

instruction)

X210 0 0

RA (at end of

instruction)

0110 0 0

dsPIC33F/PIC24H Family Reference Manual

2.9.1.2 INTERRUPTING A REPEAT LOOP

A REPEAT instruction loop can be interrupted any time. The state of the RA bit is preserved on

the stack during exception processing to enable the user-assigned application to execute

further REPEAT loops from within any number of nested interrupts. After the SRL register is

stacked, the RA status bit is cleared to restore normal execution flow within the ISR.

Returning into a REPEAT loop from an ISR using the RETFIE instruction requires no special

handling. Interrupts prefetch the repeated instruction during the third cycle of the RETFIE

instruction. The stacked RA bit is restored when the SRL register is popped and, if set, the

interrupted REPEAT loop is resumed.

2.9.1.2.1 Early Termination of a REPEAT Loop

An interrupted REPEAT loop can be terminated earlier than normal in the ISR by clearing the

RCOUNT register in software.

2.9.1.3 RESTRICTIONS ON THE REPEAT INSTRUCTION

Any instruction can immediately follow a REPEAT except for the following:

• Program Flow Control instructions (any branch, compare and skip, subroutine calls,

returns, etc.)

• Another REPEAT or DO instruction

•DISI, ULNK, LNK, PWRSAV or RESET

•MOV.D instruction

Note 1: If a REPEAT loop has been interrupted, and an ISR is being processed, the

user-assigned application must stack the REPEAT Count register (RCOUNT)

before it executes another REPEAT instruction within an ISR.

2: If a REPEAT instruction is used within an ISR, the user-assigned application must

unstack the RCOUNT register before it executes the RETFIE instruction.

Note: When the repeated instruction (target instruction in the REPEAT loop) accesses the

data from Program Space (PS) using PSV, it will require two instruction cycles the

first time it is executed after a return from an exception. Similar to the first iteration

of a loop, the timing limitations do not allow the first instruction to access data

residing in Program Space in a single instruction cycle.

Note: Some instructions and/or instruction addressing modes can be executed within a

REPEAT loop, but it is not necessary to repeat all instructions.

Section 2. CPU

CPU

2.9.2 DO Loop Construct

The DO instruction can execute a group of instructions that follow a specified number of times

without the software overhead. The set of instructions up to and including the end address are

repeated. The DO iteration count value for the DO instruction can be specified by a 14-bit literal or

by the contents of a W register declared within the instruction. The syntax for the 14-bit literal

form of the DO instruction is as shown in Example 2-8.

Example 2-8: Syntax for the 14-bit Literal Form of the DO Instruction

The syntax for the W register declared form of the DO instruction is as shown in Example 2-9:

Example 2-9: Syntax for the W Register Declared Form of the DO Instruction

The following features are provided in the DO loop construct:

• A W register can be used to specify the loop count, which allows the loop count to be

defined at run-time

• The instruction execution order need not be sequential (i.e., there can be branches,

subroutine calls, etc.)

• The loop end address need not be greater than the start address

2.9.2.1 DO LOOP REGISTERS AND OPERATION

The number of iterations executed by a DO loop will be the 14-bit literal value + 1 or the

Wn value + 1. If a W register is used to specify the number of iterations, the two Most Significant

bits are not used to specify the loop count. The operation of a DO loop is similar to the DO-WHILE

construct in the C programming language, because the instructions in the loop will always be

executed at least once.

The dsPIC33F/PIC24H CPU consists of the following three registers that are associated with DO

loops:

•The DO Loop Start Address register (DOSTART) is a 22-bit register that holds the starting

address of the DO loop

•The DO Loop End Address register (DOEND) is a 16-bit register that holds the end address

of the DO loop

•The DO Loop Counter register (DCOUNT) is a 16-bit register that holds the number of

iterations to be executed by the loop

These registers are memory mapped and are automatically loaded by the hardware when the DO

instruction is executed. The MSb and LSb of these registers are set to ‘0’. The LSb is not stored

in these registers because PC<0> is always forced to ‘0’.

The DO Loop Active bit (DA) in the SR register indicates that a single DO loop (or nested DO loops)

is active. When a DO instruction is executed, the DA bit is set, which enables the PC address to

be compared with the DOEND register on each subsequent instruction cycle. When the PC

matches the value in DOEND, DCOUNT is decremented.

DO #lit14,LOOP_END ; DCOUNT <-- lit14

Instruction1

Instruction2

LOOP_END: Instruction n

DO Wn,LOOP_END ; DCOUNT <-- Wn<13:0>

Instruction1

Instruction2

LOOP_END: Instruction n

dsPIC33F/PIC24H Family Reference Manual

If the DCOUNT register is not zero, the PC is loaded with the address contained in the DOSTART

terminates.

If no other nested DO loops are in progress, the DA bit is also cleared.

2.9.2.2 DO LOOP NESTING

The DOSTART, DOEND and DCOUNT registers each have an associated shadow register that

allows the DO loop hardware to support one level of automatic nesting. The DOSTART, DOEND

and DCOUNT registers are user accessible. They can be manually saved to permit additional

nesting, where required.

The DO Loop Nesting Level bits (DL<2:0>) in the CORCON register indicate the nesting level of

the DO loop currently being executed. When the first DO instruction is executed, DL<2:0> is set

to 'b001 to indicate that one level of the DO loop is underway. The DO Loop Active bit (DA) in the

SR register is also set.

When another DO instruction is executed within the first DO loop, the DOSTART, DOEND and

DCOUNT registers are transferred into the shadow registers before they are updated with the

new loop values. The DL<2:0> bits are set to 'b010 to indicate that a second, nested DO loop is

in progress. The DA bit (SR<9>) also remains set.

If no more than one level of DO loop nesting is required in the application, no special attention is

required. However, if the user-assigned application requires more than one level of DO loop

nesting, it can be achieved by saving the DOSTART, DOEND and DCOUNT registers before

executing the next DO instruction. These registers should be saved whenever DL<2:0> is 'b010

or greater.

The DOSTART, DOEND and DCOUNT registers are automatically restored from their shadow

registers when a DO loop terminates and DL<2:0> = 'b010.

2.9.2.3 INTERRUPTING A DO LOOP

DO loops can be interrupted at any time. If another DO loop is to be executed during the ISR, the

user-assigned application must check the DL<2:0> status bits and save the DOSTART, DOEND

and DCOUNT registers as required.

No special handling is required if the user-assigned application can ensure that only one level of

DO loop will be executed in:

• Both background and any one ISR handler (if interrupt nesting is enabled) or

• Both background and any ISR (if interrupt nesting is disabled)

Alternatively, up to two (nested) DO loops can be executed in either background or within:

•One ISR handler (if interrupt nesting is enabled) or

• Any ISR (if interrupt nesting is disabled)

It is assumed that no DO loops are used within any trap handlers.

Returning to a DO loop from an ISR using the RETFIE instruction requires no special handling.

Note: The group of instructions in a DO loop construct is always executed at least one

time. The DO loop is always executed one time more than the value specified in the

literal or W register operand.

Note: The DL<2:0> (CORCON<10:8>) bits are combined (logically ORed) to form the DA

bit (SR<9>). If nested DO loops are being executed, the DA bit is cleared only when

the loop count associated with the outermost loop expires.

Section 2. CPU

CPU

2.9.2.4 EARLY TERMINATION OF THE DO LOOP

There are two ways to terminate a DO loop earlier than normal:

•The Early DO Loop Termination Control bit (EDT) in the CORCON register provides a

means for the user-assigned application to terminate a DO loop before it completes all

loops. Writing a ‘1’ to the EDT bit forces the loop to complete the iteration underway and

then terminate. If EDT is set during the next-to-last (pen-ultimate) or last instruction of the

loop, one more iteration of the loop occurs. EDT always reads as a ‘0’; clearing it has no

effect. After the EDT bit is set, the user can optionally branch out of the DO loop.

• Alternatively, the code can branch out of the loop at any point except from the last

instruction, which cannot be a flow-control instruction. Although the DA bit (SR<9>)

enables the DO loop hardware, it has no effect unless the address of the pen-ultimate

instruction is encountered during an instruction prefetch. This is not a recommended

method for terminating a DO loop.

2.9.2.5 DO LOOP RESTRICTIONS

The use of DO loops imposes restrictions such as:

• When the DOEND register can be read.

• A few instructions must not be used as the last instruction in the loop.

• Certain small loop lengths are prohibited, as listed in Table 2-8 (loop length refers to the

size of the block of instructions that is being repeated in the loop).

2.9.2.5.1 DOEND Register Restrictions

All DO loops must contain at least two instructions because the loop termination tests are

performed in the pen-ultimate instruction. REPEAT should be used for single instruction loops.

The DOEND Special Function Register (SFR), cannot be read by user software in the instruction

that immediately follows either a DO instruction or a file register write operation to the DOEND

SFR.

The instruction before the pen-ultimate instruction in a DO loop should not modify:

• CPU priority level governed by the CPU IPL status bits (SR<7:5>).

• Peripheral Interrupt Enable bits governed by Interrupt Enable Control registers IEC0, IEC1

and IEC2.

• Peripheral Interrupt Priority bits governed by Interrupt Priority Control registers IPC0

through IPC11.

If these restrictions are not observed, the DO loop may execute incorrectly.

2.9.2.5.2 Last Instruction Restrictions

The last instruction in a DO loop should not be any of the following:

• Flow control instruction (e.g., any branch, compare and skip, GOTO, CALL, RCALL, TRAP).

• Another REPEAT or DO instruction.

• Target instruction within a REPEAT loop. This restriction implies that the one before the last

instruction also cannot be a REPEAT.

• Any instruction that occupies two words in program space.

•DISI instruction.

RETURN, RETFIE and RETLW work correctly as the last instruction of a DO loop, but the

user-assigned application is responsible for returning to the loop to complete it.

Note: Exiting a DO loop without using EDT is not recommended, because the hardware

will continue to check for DOEND addresses.

dsPIC33F/PIC24H Family Reference Manual

2.9.2.5.3 Loop Length Restrictions

Loop length is defined as the signed offset of the last instruction from the first instruction in the

DO loop. The loop length, when added to the address of the first instruction in the loop, forms the

address of the last instruction of the loop. Table 2-8 lists the loop lengths to avoid.

Table 2-8: Loop Lengths to Avoid

Loop Length Reason to Avoid

-2 Execution starts at the first instruction in the loop (i.e., at the PC address) and continues until the

loop-end address (in this case [PC – 4]) is prefetched. As this is the first word of the DO instruction, it

executes the DO instruction again, re-initializing the DCOUNT and prefetching [PC]. This continues

forever as long as the loop end address [PC – 4] is prefetched. This value of n has the potential of

creating an infinite loop (subject to a Watchdog Timer Reset).

For example:

end_loop: DO #33, end_loop ;DO is a two-word instruction

NOP ;2nd word of DO executes as a NOP

ADD W2,W3,W4 ;First instruction in DO loop([PC])

-1 Execution starts at the first instruction in the loop (i.e., at [PC]) and continues until the loop end

address ([PC – 2]) is prefetched. Since the loop end address is the second word of the

DO instruction, it executes as a NOP but will still prefetch [PC]. The loop will then execute again. This

will continue as long as the loop end address [PC – 2] is prefetched and the loop does not terminate.

If the value in the DCOUNT register reaches zero and on a subsequent decrement generates a

borrow, the loop will terminate. However, in such a case the initial instruction outside the loop is

once again the first loop instruction.

For example:

DO #33, end_loop ;DO is a two-word instruction

end_loop: NOP ;2nd word of DO executes as a NOP

ADD W2,W3,W4 ;First instruction in DO loop([PC])

0Execution starts at the first instruction in the loop (i.e., at [PC]) and continues until the loop end

address ([PC]) is prefetched. If the loop continues, this prefetch causes the DO loop hardware to load

the DOEND address ([PC]) into the PC for the next fetch (which will be [PC] again). After the first true

iteration of the loop, the first instruction in the loop executes repeatedly until the loop count

underflows and the loop terminates. When this occurs, the initial instruction outside the loop is the

instruction after [PC].

For example:

DO #33, end_loop ;DO is a two-word instruction

NOP ;2nd word of DO executes as a NOP

end_loop: ADD W2,W3,W4 ;First instruction in DO loop([PC])

Section 2. CPU

CPU

2.10 ADDRESS REGISTER DEPENDENCIES

The dsPIC33F/PIC24H architecture supports a data space read (source) and a data space write

(destination) for most MCU class instructions. The Effective Address (EA) calculation by the

AGU, and subsequent data space read or write, each take one instruction cycle to complete. This

timing causes the data space read and write operations for each instruction to partially overlap,

as shown in Figure 2-21. Because of this overlap, a Read-After-Write (RAW) data dependency

can occur across instruction boundaries. RAW data dependencies are detected and handled at

run-time by the dsPIC33F/PIC24H CPU.

Figure 2-21: Data Space Access Timing

2.10.1 Read-After-Write (RAW) Dependency Rules

If the W register is used as a write operation destination in the current instruction, and the

W register being read in the prefetched instruction are the same, the following rules apply:

• If the destination write (current instruction) does not modify the contents of Wn, no stalls will

occur

• If the source read (prefetched instruction) does not calculate an EA using Wn, no stalls will

occur

During each instruction cycle, the dsPIC33F/PIC24H hardware automatically checks to see if a

RAW data dependency is about to occur. If the conditions specified above are not satisfied, the

CPU automatically adds a one-instruction-cycle delay before executing the prefetched

instruction. The instruction stall provides enough time for the destination W register write to take

place before the next (prefetched) instruction uses the written data. Table 2-9 provides the RAW

dependency summary.

ADD MOV

[W7]

[W10] [W9]++

X-Space Address W7 W10 W8 W9

Add W0, [W7], [W10]

MOV [W8], [W9]++

[W8]

X-Space RAGU

Instruction Register

Contents

X-Space WAGU

1 Instruction Cycle (TCY)

TCY0TCY1TCY2

dsPIC33F/PIC24H Family Reference Manual

Table 2-9: Read-After-Write Dependency Summary

2.10.2 Instruction Stall Cycles

An instruction stall is essentially a one instruction cycle wait period appended in front of the read

phase of an instruction to allow the prior write to complete before the next read operation. For

the purposes of interrupt latency, the stall cycle is associated with the instruction following the

instruction where it was detected (i.e., stall cycles always precede instruction execution cycles).

If a RAW data dependency is detected, the dsPIC33F/PIC24H CPU begins an instruction stall.

During an instruction stall, the following events occur:

• The write operation underway (for the previous instruction) is allowed to complete as normal.

• Data space is not addressed until after the instruction stall.

• PC increment is inhibited until after the instruction stall.

• Further instruction fetches are inhibited until after the instruction stall.

2.10.2.1 INSTRUCTION STALL CYCLES AND INTERRUPTS

When an interrupt event coincides with two adjacent instructions that will cause an instruction

stall, one of two possible outcomes can occur:

• If the interrupt coincides with the first instruction, the first instruction is allowed to complete

while the second instruction is executed after the ISR completes. In this case, the stall

cycle is eliminated from the second instruction because the exception process provides

time for the first instruction to complete the write phase.

• If the interrupt coincides with the second instruction, the second instruction and the

appended stall cycle are allowed to execute before to the ISR. In this case, the stall cycle

associated with the second instruction executes normally. However, the stall cycle is

effectively absorbed into the exception process timing. The exception process proceeds as

if an ordinary one-cycle instruction or two-cycle instruction is interrupted.

Destination

Addressing Mode

Using Wn

Source

Addressing Mode

Using Wn Status Examples

(Wn = W2)

Direct Direct Allowed ADD.w W0, W1, W2

MOV.w W2, W3

Direct Indirect Stall ADD.w W0, W1, W2

MOV.w [W2], W3

Direct Indirect with

modification Stall ADD.w W0, W1, W2

MOV.w [W2++], W3

Indirect Direct Allowed ADD.w W0, W1, [W2]

MOV.w W2, W3

Indirect Indirect Allowed ADD.w W0, W1, [W2]

MOV.w [W2], W3

Indirect Indirect with

modification Allowed ADD.w W0, W1, [W2]

MOV.w [W2++], W3

Indirect with

modification Direct Allowed ADD.w W0, W1, [W2++]

MOV.w W2, W3

Indirect Indirect Stall ADD.w W0, W1, [W2]

MOV.w [W2], W3

; W2=0x0004 (mapped W2)

Indirect Indirect with

modification Stall ADD.w W0, W1, [W2]

MOV.w [W2++], W3

; W2=0x0004 (mapped W2)

Indirect with

modification Indirect Stall ADD.w W0, W1, [W2++]

MOV.w [W2], W3

Indirect with

modification Indirect with

modification Stall ADD.w W0, W1, [W2++]

MOV.w [W2++], W3

Section 2. CPU

CPU

2.10.2.2 INSTRUCTION STALL CYCLES AND FLOW CHANGE INSTRUCTIONS

The CALL and RCALL instructions write to the stack using working register W15 and can,

therefore, force an instruction stall prior to the next instruction, if the source read of the next

instruction uses W15.

The RETFIE and RETURN instructions can never force an instruction stall prior to the next

instruction because they only perform read operations. However, the RETLW instruction can force

a stall, because it writes to a W register during the last cycle.

The GOTO and branch instructions can never force an instruction stall because they do not

perform write operations.

2.10.2.3 INSTRUCTION STALLS AND DO AND REPEAT LOOPS

Other than the addition of instruction stall cycles, RAW data dependencies do not affect the

operation of either DO or REPEAT loops.

The prefetched instruction within a REPEAT loop does not change until the loop is complete or

an exception occurs. Although register dependency checks occur across instruction boundaries,

the dsPIC33F/PIC24H CPU effectively compares the source and destination of the same

instruction during a REPEAT loop.

The last instruction of a DO loop will either prefetch the instruction at the loop start address or the

next instruction (outside the loop). The instruction stall decision is based on the last instruction

in the loop and the contents of the prefetched instruction.

2.10.2.4 INSTRUCTION STALLS AND PROGRAM SPACE VISIBILITY (PSV)

When Program Space (PS) is mapped to data space by enabling the PSV bit (CORCON<2>),

and the X space EA falls within the visible program space window, the read cycle redirects to the

address in program space. Accessing data from program space takes up to three instruction

cycles.

Instructions operating in PSV address space are subject to RAW data dependencies and

consequent instruction stalls, just like any other instruction. Consider the code segment, as

shown in Example 2-10.

Example 2-10: Code Example to Operate on the PSV

This sequence of instructions would take five instruction cycles to execute. Two instruction cycles

are added to perform the PSV access via W1. An instruction stall cycle is inserted to resolve the

RAW data dependency caused by W2.

ADD W0, [W1], [W2++] ; PSV = 1, W1=0x8000, PSVPAG=0xAA

MOV [W2], [W3]

dsPIC33F/PIC24H Family Reference Manual

2.11 REGISTER MAPS

A summary of the registers associated with the dsPIC33F/PIC24H CPU is provided in Table 2-3.

Table 2-10: CPU Register Map

Name Bit 15 Bit 14 Bit 13 Bit 12 Bit 11 Bit 10 Bit 9 Bit 8 Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0 All Resets

W0 W0 (WREG) 0000 0000 0000 0000

W1 W1 0000 0000 0000 0000

W2 W2 0000 0000 0000 0000

W3 W3 0000 0000 0000 0000

W4 W4 0000 0000 0000 0000

W5 W5 0000 0000 0000 0000

W6 W6 0000 0000 0000 0000

W7 W7 0000 0000 0000 0000

W8 W8 0000 0000 0000 0000

W9 W9 0000 0000 0000 0000

W10 W10 0000 0000 0000 0000

W11 W11 0000 0000 0000 0000

W12 W12 0000 0000 0000 0000

W13 W13 0000 0000 0000 0000

W14 W14 0000 0000 0000 0000

W15 W15 0000 0000 0000 0000

SPLIM SPLIM 0000 0000 0000 0000

ACCAL ACCAL 0000 0000 0000 0000

ACCAH ACCAH 0000 0000 0000 0000

ACCAU Sign-extension of ACCA<39> ACCAU 0000 0000 0000 0000

ACCBL ACCBL 0000 0000 0000 0000

ACCBH ACCBH 0000 0000 0000 0000

ACCBU Sign-extension of ACCB<39> ACCBU 0000 0000 0000 0000

PCL PCL 0000 0000 0000 0000

PCH — — — — — — — — — PCH 00000 0000 0000 0000

TBLPAG — — — — — — — — TBLPAG 0000 0000 0000 0000

PSVPAG — — — — — — — — PSVPAG 0000 0000 0000 0000

RCOUNT RCOUNT xxxx xxxx xxxx xxxx

DCOUNT DCOUNT xxxx xxxx xxxx xxxx

DOSTARTL DOSTARTL 0xxxx xxxx xxxx xxx0

DOSTARTH — — — — — — — — — — DOSTARTH 0000 0000 00xx xxxx

Legend: x = unknown value on Reset, — = unimplemented, read as ‘0’. Reset values are shown in hexadecimal.

Note: Refer to the device data sheet for more details on the specific core register map.

Section 2. CPU

CPU

DOENDL DOENDL 0xxxx xxxx xxxx xxx0

DOENDH — — — — — — — — — — DOENDH 0000 0000 00xx xxxx

SR OA OB SA SB OAB SAB DA DC IPL2 IPL1 IPL0 RA NOV Z C 0000 0000 0000 0000

CORCON — — — US EDT DL<2:0> SATA SATB SATDW ACCSAT IPL3 PSV RND IF 0000 0000 0010 0000

MODCON XMODEN YMODEN — — BWM<3:0> YWM<3:0> XWM<3:0> 0000 0000 0000 0000

XMODSRT XMODSRT<15:0> 0xxxx xxxx xxxx xxx0

XMODEND XMODEND<15:0> 1xxxx xxxx xxxx xxx1

YMODSRT YMODSRT<15:0> 0xxxx xxxx xxxx xxx0

YMODEND YMODEND<15:0> 1xxxx xxxx xxxx xxx1

XBREV BREN XBREV<14:0> xxxx xxxx xxxx xxxx

DISICNT — — DISICNT<13:0> 0000 0000 0000 0000

Table 2-10: CPU Register Map (Continued)

Name Bit 15 Bit 14 Bit 13 Bit 12 Bit 11 Bit 10 Bit 9 Bit 8 Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0 All Resets

Legend: x = unknown value on Reset, — = unimplemented, read as ‘0’. Reset values are shown in hexadecimal.

Note: Refer to the device data sheet for more details on the specific core register map.

dsPIC33F/PIC24H Family Reference Manual

2.12 RELATED APPLICATION NOTES

This section lists application notes that are related to this section of the manual. These

application notes may not be written specifically for the dsPIC33F/PIC24H Product Family, but

the concepts are pertinent and could be used with modification and possible limitations. The

current application notes related to the CPU module include the following:

Title Application Note #

No related application notes at this time. N/A

Note: Please visit the Microchip web site (www.microchip.com) for additional Application

Notes and code examples for the dsPIC33F/PIC24H family of devices.

Section 2. CPU

CPU

2.13 REVISION HISTORY

Revision A (April 2007)

This is the initial released version of this document.

Revision B (September 2009)

This revision incorporates the following updates:

• Note:

- Added a note on program memory error in 2.8.7 “Address Register Dependencies”

•Registers:

- Removed the Address column in Table 2-10 in 2.11 “Register Maps”

• Sections:

- Updated 2.1 “Introduction” with the following data: All instructions execute in a single

cycle, except the instructions that change the program flow, double-word move

(MOV.D) instruction, table instructions and also the instructions accessing Program

Space Visibility (PSV) take more than one cycle

- Updated the REPEAT count value as “DO iteration count value” in 2.9.2 “DO Loop

Construct”

- Updated the exception process in 2.10.2.1 “Instruction Stall Cycles and

Interrupts”as follows: The exception process proceeds as if an ordinary one-cycle

instruction or two-cycle instruction is interrupted

- Updated the read or write cycle as “read cycle” in 2.10.2.4 “Instruction Stalls and

Program Space Visibility (PSV)”

• Tables:

- Added a table (see Table 2-5) on Multiplication Options in 2.6.2.2 “MCU Multiply

Instructions”

• Updated the incorrect references to external documents specified in this document.

• Additional minor corrections such as language and formatting updates were incorporated

throughout the document

Revision C (September 2011)

This revision includes the following updates:

• All occurrences of dsPIC33F were changed to dsPIC33F/PIC24H

• The Preliminary document status was removed

• The STATUS register bit, shown as ‘SZ’, was corrected to ‘Z’ in the Programmer’s Model

(see Figure 2-2)

• Updated Accessing W Register in File Register Instruction (see Example 2-3)

• Added a sentence with a cross-reference to the CORCON register in 2.4.2 “Core Control

• Minor corrections to text and formatting were incorporated throughout the document

dsPIC33F/PIC24H Family Reference Manual

NOTES:

Information contained in this publication regarding device

applications and the like is provided only for your convenience

and may be superseded by updates. It is your responsibility to

ensure that your application meets with your specifications.

MICROCHIP MAKES NO REPRESENTATIONS OR

WARRANTIES OF ANY KIND WHETHER EXPRESS OR

IMPLIED, WRITTEN OR ORAL, STATUTORY OR

OTHERWISE, RELATED TO THE INFORMATION,

INCLUDING BUT NOT LIMITED TO ITS CONDITION,

QUALITY, PERFORMANCE, MERCHANTABILITY OR

FITNESS FOR PURPOSE. Microchip disclaims all liability

arising from this information and its use. Use of Microchip

devices in life support and/or safety applications is entirely at

the buyer’s risk, and the buyer agrees to defend, indemnify and

hold harmless Microchip from any and all damages, claims,

suits, or expenses resulting from such use. No licenses are

conveyed, implicitly or otherwise, under any Microchip

intellectual property rights.

Trademarks

The Microchip name and logo, the Microchip logo, dsPIC,

KEELOQ, KEELOQ logo, MPLAB, PIC, PICmicro, PICSTART,

PIC32 logo, rfPIC and UNI/O are registered trademarks of

Microchip Technology Incorporated in the U.S.A. and other

countries.

FilterLab, Hampshire, HI-TECH C, Linear Active Thermistor,

MXDEV, MXLAB, SEEVAL and The Embedded Control

Solutions Company are registered trademarks of Microchip

Technology Incorporated in the U.S.A.

Analog-for-the-Digital Age, Application Maestro, chipKIT,

chipKIT logo, CodeGuard, dsPICDEM, dsPICDEM.net,

dsPICworks, dsSPEAK, ECAN, ECONOMONITOR,

FanSense, HI-TIDE, In-Circuit Serial Programming, ICSP,

Mindi, MiWi, MPASM, MPLAB Certified logo, MPLIB,

MPLINK, mTouch, Omniscient Code Generation, PICC,

PICC-18, PICDEM, PICDEM.net, PICkit, PICtail, REAL ICE,

rfLAB, Select Mode, Total Endurance, TSHARC,

UniWinDriver, WiperLock and ZENA are trademarks of

Microchip Technology Incorporated in the U.S.A. and other

countries.

SQTP is a service mark of Microchip Technology Incorporated

in the U.S.A.

All other trademarks mentioned herein are property of their

respective companies.

Printed on recycled paper.

ISBN: 978-1-61341-659-4

Note the following details of the code protection feature on Microchip devices:

• Microchip products meet the specification contained in their particular Microchip Data Sheet.

• Microchip believes that its family of products is one of the most secure families of its kind on the market today, when used in the

intended manner and under normal conditions.

• There are dishonest and possibly illegal methods used to breach the code protection feature. All of these methods, to our

knowledge, require using the Microchip products in a manner outside the operating specifications contained in Microchip’s Data

Sheets. Most likely, the person doing so is engaged in theft of intellectual property.

• Microchip is willing to work with the customer who is concerned about the integrity of their code.

• Neither Microchip nor any other semiconductor manufacturer can guarantee the security of their code. Code protection does not

mean that we are guaranteeing the product as “unbreakable.”

Code protection is constantly evolving. We at Microchip are committed to continuously improving the code protection features of our

products. Attempts to break Microchip’s code protection feature may be a violation of the Digital Millennium Copyright Act. If such acts

allow unauthorized access to your software or other copyrighted work, you may have a right to sue for relief under that Act.

Microchip received ISO/TS-16949:2009 certification for its worldwide

headquarters, design and wafer fabrication facilities in Chandler and

Tempe, Arizona; Gresham, Oregon and design centers in California

and India. The Company’s quality system processes and procedures

are for its PIC® MCUs and dsPIC® DSCs, KEELOQ® code hopping

devices, Serial EEPROMs, microperipherals, nonvolatile memory and

analog products. In addition, Microchip’s quality system for the design

and manufacture of development systems is ISO 9001:2000 certified.

AMERICAS

Corporate Office

2355 West Chandler Blvd.

Chandler, AZ 85224-6199

Tel: 480-792-7200

Fax: 480-792-7277

Technical Support:

http://www.microchip.com/

support

Web Address:

www.microchip.com

Atlanta

Duluth, GA

Tel: 678-957-9614

Fax: 678-957-1455

Boston

Westborough, MA

Tel: 774-760-0087

Fax: 774-760-0088

Chicago

Itasca, IL

Tel: 630-285-0071

Fax: 630-285-0075

Cleveland

Independence, OH

Tel: 216-447-0464

Fax: 216-447-0643

Dallas

Addison, TX

Tel: 972-818-7423

Fax: 972-818-2924

Detroit

Farmington Hills, MI

Tel: 248-538-2250

Fax: 248-538-2260

Indianapolis

Noblesville, IN

Tel: 317-773-8323

Fax: 317-773-5453

Los Angeles

Mission Viejo, CA

Tel: 949-462-9523

Fax: 949-462-9608

Santa Clara

Santa Clara, CA

Tel: 408-961-6444

Fax: 408-961-6445

Toronto

Mississauga, Ontario,

Canada

Tel: 905-673-0699

Fax: 905-673-6509

ASIA/PACIFIC

Asia Pacific Office

Suites 3707-14, 37th Floor

Tower 6, The Gateway

Harbour City, Kowloon

Hong Kong

Tel: 852-2401-1200

Fax: 852-2401-3431

Australia - Sydney

Tel: 61-2-9868-6733

Fax: 61-2-9868-6755

China - Beijing

Tel: 86-10-8569-7000

Fax: 86-10-8528-2104

China - Chengdu

Tel: 86-28-8665-5511

Fax: 86-28-8665-7889

China - Chongqing

Tel: 86-23-8980-9588

Fax: 86-23-8980-9500

China - Hangzhou

Tel: 86-571-2819-3187

Fax: 86-571-2819-3189

China - Hong Kong SAR

Tel: 852-2401-1200

Fax: 852-2401-3431

China - Nanjing

Tel: 86-25-8473-2460

Fax: 86-25-8473-2470

China - Qingdao

Tel: 86-532-8502-7355

Fax: 86-532-8502-7205

China - Shanghai

Tel: 86-21-5407-5533

Fax: 86-21-5407-5066

China - Shenyang

Tel: 86-24-2334-2829

Fax: 86-24-2334-2393

China - Shenzhen

Tel: 86-755-8203-2660

Fax: 86-755-8203-1760

China - Wuhan

Tel: 86-27-5980-5300

Fax: 86-27-5980-5118

China - Xian

Tel: 86-29-8833-7252

Fax: 86-29-8833-7256

China - Xiamen

Tel: 86-592-2388138

Fax: 86-592-2388130

China - Zhuhai

Tel: 86-756-3210040

Fax: 86-756-3210049

ASIA/PACIFIC

India - Bangalore

Tel: 91-80-3090-4444

Fax: 91-80-3090-4123

India - New Delhi

Tel: 91-11-4160-8631

Fax: 91-11-4160-8632

India - Pune

Tel: 91-20-2566-1512

Fax: 91-20-2566-1513

Japan - Yokohama

Tel: 81-45-471- 6166

Fax: 81-45-471-6122

Korea - Daegu

Tel: 82-53-744-4301

Fax: 82-53-744-4302

Korea - Seoul

Tel: 82-2-554-7200

Fax: 82-2-558-5932 or

82-2-558-5934

Malaysia - Kuala Lumpur

Tel: 60-3-6201-9857

Fax: 60-3-6201-9859

Malaysia - Penang

Tel: 60-4-227-8870

Fax: 60-4-227-4068

Philippines - Manila

Tel: 63-2-634-9065

Fax: 63-2-634-9069

Singapore

Tel: 65-6334-8870

Fax: 65-6334-8850

Taiwan - Hsin Chu

Tel: 886-3-5778-366

Fax: 886-3-5770-955

Taiwan - Kaohsiung

Tel: 886-7-536-4818

Fax: 886-7-330-9305

Taiwan - Taipei

Tel: 886-2-2500-6610

Fax: 886-2-2508-0102

Thailand - Bangkok

Tel: 66-2-694-1351

Fax: 66-2-694-1350

EUROPE

Austria - Wels

Tel: 43-7242-2244-39

Fax: 43-7242-2244-393

Denmark - Copenhagen

Tel: 45-4450-2828

Fax: 45-4485-2829

France - Paris

Tel: 33-1-69-53-63-20

Fax: 33-1-69-30-90-79

Germany - Munich

Tel: 49-89-627-144-0

Fax: 49-89-627-144-44

Italy - Milan

Tel: 39-0331-742611

Fax: 39-0331-466781

Netherlands - Drunen

Tel: 31-416-690399

Fax: 31-416-690340

Spain - Madrid

Tel: 34-91-708-08-90

Fax: 34-91-708-08-91

UK - Wokingham

Tel: 44-118-921-5869

Fax: 44-118-921-5820

Worldwide Sales and Service

08/02/11

Section 2. CPU 70204 70204C

Navigation menu

Versions of this User Manual:

Views

Navigation