v_pk_fmac_f16 GPU Native ISA AMD Vector

Vector Arithmetic

v_pk_fmac_f16

Multiply two packed half-precision float inputs component-wise and accumulate the result into the destination register using fused multiply add.

Encoding

Binary Layout (AMD CDNA 5)
0
31
111100
30:25
VDST
24:17
VSRC1
16:9
SRC0
8:0
 

The ENC_VOP2 layout from AMD's machine-readable ISA specification (AMD CDNA 5). Opcode 60 in OP rebuilds 0x78000000, an identifier AMD lists for this encoding. The same in the other 9 generations that define it.

Format VOP2
Width 32 bits
Opcode 60
Identifier 0x78000000

Operands

  • VDST
    Written. Operand must be a vector GPR. Uses an 8-bit operand field. Data format: 16-bit half-precision floating point value. Packed 2-component value. (OPR_VGPR, FMT_NUM_PK2_F16)
  • SRC0
    Read. All operands. Covers all operands that are allowed as scalar or vector sources. Data format: 16-bit half-precision floating point value. Packed 2-component value. (OPR_SRC, FMT_NUM_PK2_F16)
  • VSRC1
    Read. Operand must be a vector GPR. Uses an 8-bit operand field. Data format: 16-bit half-precision floating point value. Packed 2-component value. (OPR_VGPR, FMT_NUM_PK2_F16)

In AMD's order, as its machine-readable ISA specification lists them for the ENC_VOP2 encoding (AMD CDNA 5).

GFX Target Compatibility

TargetSupport
gfx1100✅ Supported

Related

More in Vector Arithmetic

Reference

AMDGPU / GFX ISA

Semantics

D0[15 : 0].f16 = fma(S0[15 : 0].f16, S1[15 : 0].f16, D0[15 : 0].f16); D0[31 : 16].f16 = fma(S0[31 : 16].f16, S1[31 : 16].f16, D0[31 : 16].f16)

Example

v_pk_fmac_f16 v5, -1, v2

A real instruction accepted by the LLVM assembler, taken verbatim from LLVM's own AMDGPU MC test suite. Not from an AMD document, and not authored here.

Sources