wiki:GcnInstrsVop2

Version 4 (modified by trac, 5 years ago) (diff)

--

Back to Table of content

GCN ISA VOP2/VOP3 instructions

VOP2 instructions can be encoded in the VOP2 encoding and the VOP3a/VOP3b encoding. List of fields for VOP2 encoding:

Bits Name Description
0-8 SRC0 First (scalar or vector) source operand
9-16 VSRC1 Second vector source operand
17-24 VDST Destination vector operand
25-30 OPCODE Operation code
31 ENCODING Encoding type. Must be 0

Syntax: INSTRUCTION VDST, SRC0, VSRC1

List of fields for VOP3A/VOP3B encoding (GCN 1.0/1.1):

Bits Name Description
0-7 VDST Vector destination operand
8-10 ABS Absolute modifiers for source operands (VOP3A)
8-14 SDST Scalar destination operand (VOP3B)
11 CLAMP CLAMP modifier (VOP3A)
15 CLAMP CLAMP modifier (VOP3B)
17-25 OPCODE Operation code
26-31 ENCODING Encoding type. Must be 0b110100
32-40 SRC0 First (scalar or vector) source operand
41-49 SRC1 Second (scalar or vector) source operand
50-58 SRC2 Third (scalar or vector) source operand
59-60 OMOD OMOD modifier. Multiplication modifier
61-63 NEG Negation modifier for source operands

List of fields for VOP3A encoding (GCN 1.2):

Bits Name Description
0-7 VDST Destination vector operand
8-10 ABS Absolute modifiers for source operands (VOP3A)
8-14 SDST Scalar destination operand (VOP3B)
15 CLAMP CLAMP modifier
16-25 OPCODE Operation code
26-31 ENCODING Encoding type. Must be 0b110100
32-40 SRC0 First (scalar or vector) source operand
41-49 SRC1 Second (scalar or vector) source operand
50-58 SRC2 Third (scalar or vector) source operand
59-60 OMOD OMOD modifier. Multiplication modifier
61-63 NEG Negation modifier for source operands

Syntax: INSTRUCTION VDST, SRC0, SRC1 [MODIFIERS]

Modifiers:

  • CLAMP - clamps destination floating point value in range 0.0-1.0
  • MUL:2, MUL:4, DIV:2 - OMOD modifiers. Multiply destination floating point value by 2.0, 4.0 or 0.5 respectively
  • -SRC - negate floating point value from source operand
  • ABS(SRC) - apply absolute value to source operand

Negation and absolute value can be combined: -ABS(V0). Modifiers CLAMP and OMOD (MUL:2, MUL:4 and DIV:2) can be given in random order.

Limitations for operands:

  • only one SGPR can be read by instruction. Multiple occurrences of this same SGPR is allowed
  • only one literal constant can be used, and only when a SGPR or M0 is not used in source operands
  • only SRC0 can holds LDS_DIRECT

VOP2 opcodes (0-63) are reflected in VOP3 in range: 256-319. List of the instructions by opcode:

Opcode Mnemonic (GCN1.0/1.1) Mnemonic (GCN 1.2)
0 (0x0) V_CNDMASK_B32 V_CNDMASK_B32
1 (0x1) V_READLANE_B32 V_ADD_F32
2 (0x2) V_WRITELANE_B32 V_SUB_F32
3 (0x3) V_ADD_F32 V_SUBREV_F32
4 (0x4) V_SUB_F32 V_MUL_LEGACY_F32
5 (0x5) V_SUBREV_F32 V_MUL_F32
6 (0x6) V_MAC_LEGACY_F32 V_MUL_I32_I24
7 (0x7) V_MUL_LEGACY_F32 V_MUL_HI_I32_I24
8 (0x8) V_MUL_F32 V_MUL_U32_U24
9 (0x9) V_MUL_I32_I24 V_MUL_HI_U32_U24
10 (0xa) V_MUL_HI_I32_I24 V_MIN_F32
11 (0xb) V_MUL_U32_U24 V_MAX_F32
12 (0xc) V_MUL_HI_U32_U24 V_MIN_I32
13 (0xd) V_MIN_LEGACY_F32 V_MAX_I32
14 (0xe) V_MAX_LEGACY_F32 V_MIN_U32
15 (0xf) V_MIN_F32 V_MAX_U32

Instruction set

Alphabetically sorted instruction list:

V_ADD_F32

Opcode VOP2: 3 (0x3) for GCN 1.0/1.1; 1 (0x1) for GCN 1.2
Opcode VOP3a: 259 (0x103) for GCN 1.0/1.1; 257 (0x101) for GCN 1.2
Syntax: V_ADD_F32 VDST, SRC0, SRC1
Description: Add two FP value from SRC0 and SRC1 and store result to VDST.
Operation:
VDST = (FLOAT)SRC0 + (FLOAT)SRC1

V_CNDMASK_B32

Opcode VOP2: 0 (0x0) for GCN 1.0/1.1; 1 (0x0) for GCN 1.2
Opcode VOP3a: 259 (0x100) for GCN 1.0/1.1; 256 (0x100) for GCN 1.2
Syntax VOP2: V_CNDMASK_B32 VDST, SRC0, SRC1, VCC
Syntax VOP3a: V_CNDMASK_B32 VDST, SRC0, SRC1, SSRC2(2)
Description: If bit for current thread of VCC or SDST is set then store SRC1 to VDST, otherwise store SRC0 to VDST. CLAMP and OMOD modifier doesn't affect on result.
Operation:
VDST = SSRC2&(1ULL<<THREADID) ? SRC1 : SRC0

V_MAC_LEGACY_F32

Opcode VOP2: 6 (0x6) for GCN 1.0/1.1 Opcode VOP3a: 262 (0x106) for GCN 1.0/1.1 Syntax: V_MUL_LEGACY_F32 VDST, SRC0, SRC1
Description: Multiply FP value from SRC0 by FP value from SRC1 and add result to VDST. If one of value is 0.0 then always do not change VDST (do not apply IEEE rules for 0.0*x).
Operation:
if ((FLOAT)SRC0!=0.0 && (FLOAT)SRC1!=0.0) VDST = (FLOAT)SRC0 * (FLOAT)SRC1 + (FLOAT)VDST

V_MUL_LEGACY_F32

Opcode VOP2: 7 (0x7) for GCN 1.0/1.1; 5 (0x4) for GCN 1.2
Opcode VOP3a: 263 (0x107) for GCN 1.0/1.1; 260 (0x104) for GCN 1.2
Syntax: V_MUL_LEGACY_F32 VDST, SRC0, SRC1
Description: Multiply FP value from SRC0 by FP value from SRC1 and store result to VDST. If one of value is 0.0 then always store 0.0 to VDST (do not apply IEEE rules for 0.0*x).
Operation:
if ((FLOAT)SRC0!=0.0 && (FLOAT)SRC1!=0.0) VDST = (FLOAT)SRC0 * (FLOAT)SRC1 else VDST = 0.0

V_MUL_F32

Opcode VOP2: 8 (0x8) for GCN 1.0/1.1; 5 (0x5) for GCN 1.2
Opcode VOP3a: 264 (0x108) for GCN 1.0/1.1; 261 (0x105) for GCN 1.2
Syntax: V_MUL_F32 VDST, SRC0, SRC1
Description: Multiply FP value from SRC0 by FP value from SRC1 and store result to VDST.
Operation:
VDST = (FLOAT)SRC0 * (FLOAT)SRC1

V_MUL_HI_I32_24

Opcode VOP2: 10 (0xa) for GCN 1.0/1.1; 7 (0x7) for GCN 1.2
Opcode VOP3a: 266 (0x10a) for GCN 1.0/1.1; 263 (0x107) for GCN 1.2
Syntax: V_MUL_HI_I32_24 VDST, SRC0, SRC1
Description: Multiply 24-bit signed integer value from SRC0 by 24-bit signed value from SRC1 and store higher 16-bit of the result to VDST with sign extension. Any modifier doesn't affect on result.
Operation:
INT32 V0 = (INT32)((SRC0&0x7fffff) | (SSRC0&0x800000 ? 0xff800000 : 0)) INT32 V1 = (INT32)((SRC1&0x7fffff) | (SSRC1&0x800000 ? 0xff800000 : 0)) VDST = ((INT64)V0 * V1)>>32

V_MUL_HI_U32_U24

Opcode VOP2: 12 (0xc) for GCN 1.0/1.1; 9 (0x9) for GCN 1.2
Opcode VOP3a: 268 (0x10c) for GCN 1.0/1.1; 265 (0x109) for GCN 1.2
Syntax: V_MUL_HI_U32_U24 VDST, SRC0, SRC1
Description: Multiply 24-bit unsigned integer value from SRC0 by 24-bit unsigned value from SRC1 and store higher 16-bit of the result to VDST. Any modifier doesn't affect to result.
Operation:
VDST = ((UINT64)(SRC0&0xffffff) * (UINT32)(SRC1&0xffffff)) >> 32

V_MUL_I32_I24

Opcode VOP2: 9 (0x9) for GCN 1.0/1.1; 6 (0x6) for GCN 1.2
Opcode VOP3a: 265 (0x109) for GCN 1.0/1.1; 262 (0x106) for GCN 1.2
Syntax: V_MUL_I32_I24 VDST, SRC0, SRC1
Description: Multiply 24-bit signed integer value from SRC0 by 24-bit signed value from SRC1 and store result to VDST. Any modifier doesn't affect to result.
Operation:
INT32 V0 = (INT32)((SRC0&0x7fffff) | (SSRC0&0x800000 ? 0xff800000 : 0)) INT32 V1 = (INT32)((SRC1&0x7fffff) | (SSRC1&0x800000 ? 0xff800000 : 0)) VDST = V0 * V1

V_MUL_U32_U24

Opcode VOP2: 11 (0xb) for GCN 1.0/1.1; 8 (0x8) for GCN 1.2
Opcode VOP3a: 267 (0x10b) for GCN 1.0/1.1; 264 (0x108) for GCN 1.2
Syntax: V_MUL_U32_U24 VDST, SRC0, SRC1
Description: Multiply 24-bit unsigned integer value from SRC0 by 24-bit unsigned value from SRC1 and store result to VDST. Any modifier doesn't affect to result.
Operation:
VDST = (UINT32)(SRC0&0xffffff) * (UINT32)(SRC1&0xffffff)

V_READLANE_B32

Opcode VOP2: 1 (0x1) for GCN 1.0/1.1
Opcode VOP3a: 257 (0x101) for GCN 1.0/1.1
Syntax: V_READLANE_B32 SDST, VSRC0, SSRC1
Description: Copy one VSRC0 lane value to one SDST. Lane (thread id) choosen from SSRC1&63. SSRC1 can be SGPR or M0.
Operation:
SDST = VSRC0[SSRC1 & 63]

V_WRITELANE_B32

Opcode VOP2: 2 (0x2) for GCN 1.0/1.1
Opcode VOP3a: 258 (0x102) for GCN 1.0/1.1
Syntax: V_WRITELANE_B32 VDST, VSRC0, SSRC1
Description: Copy SGPR to one lane of VDST. Lane choosen (thread id) from SSRC1&63. SSRC1 can be SGPR or M0.
Operation:
VDST[SSRC1 & 63] = SSRC0

V_SUB_F32

Opcode VOP2: 4 (0x4) for GCN 1.0/1.1; 2 (0x2) for GCN 1.2
Opcode VOP3a: 260 (0x104) for GCN 1.0/1.1; 258 (0x102) for GCN 1.2
Syntax: V_SUB_F32 VDST, SRC0, SRC1
Description: Subtract FP value from SRC0 and FP value from SRC1 and store result to VDST.
Operation:
VDST = (FLOAT)SRC0 - (FLOAT)SRC1

V_SUBREV_F32

Opcode VOP2: 5 (0x5) for GCN 1.0/1.1; 2 (0x3) for GCN 1.2
Opcode VOP3a: 261 (0x105) for GCN 1.0/1.1; 259 (0x103) for GCN 1.2
Syntax: V_SUBREV_F32 VDST, SRC0, SRC1
Description: Subtract FP value from SRC1 and FP value from SRC0 and store result to VDST.
Operation:
VDST = (FLOAT)SRC1 - (FLOAT)SRC0