This is an archive of the discontinued LLVM Phabricator instance.

Differential D21129

[AMDGPU] Disassembler: Support for sdwa instructions
ClosedPublic

Authored by SamWot on Jun 8 2016, 6:22 AM.

Details

Reviewers
• tstellarAMD
vpykhtin
Commits
rGc9bdcb75c461: [AMDGPU] Disassembler: Support for sdwa instructions
rL272255: [AMDGPU] Disassembler: Support for sdwa instructions

Diff Detail

Repository
rL LLVM

Event Timeline

SamWot updated this revision to Diff 60021.Jun 8 2016, 6:22 AM
SamWot retitled this revision from to [AMDGPU] Disassembler: Support for sdwa instructions.
SamWot updated this object.
SamWot added a reviewer: vpykhtin.
Herald added a reviewer: • tstellarAMD. · View Herald TranscriptJun 8 2016, 6:22 AM
Herald added subscribers: kzhuravl, arsenm. · View Herald Transcript
vpykhtin accepted this revision.Jun 9 2016, 3:27 AM
vpykhtin edited edge metadata.

LGTM.

This revision is now accepted and ready to land.Jun 9 2016, 3:27 AM
Closed by commit rL272255: [AMDGPU] Disassembler: Support for sdwa instructions (authored by skolton). · Explain WhyJun 9 2016, 4:11 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
llvm/
trunk/
lib/
Target/
AMDGPU/
Disassembler/
6 lines
test/
MC/
Disassembler/
AMDGPU/
347 lines

Diff 60155

llvm/trunk/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp

Show First 20 Lines • Show All 119 Lines • ▼ Show 20 LinesDecodeStatus AMDGPUDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
const unsigned MaxInstBytesNum = (std::min)((size_t)8, Bytes_.size()); const unsigned MaxInstBytesNum = (std::min)((size_t)8, Bytes_.size());
Bytes = Bytes_.slice(0, MaxInstBytesNum); Bytes = Bytes_.slice(0, MaxInstBytesNum);
DecodeStatus Res = MCDisassembler::Fail; DecodeStatus Res = MCDisassembler::Fail;
do { do {
// ToDo: better to switch encoding length using some bit predicate // ToDo: better to switch encoding length using some bit predicate
// but it is unknown yet, so try all we can // but it is unknown yet, so try all we can
// Try to decode DPP first to solve conflict with VOP1 and VOP2 encodings // Try to decode DPP and SDWA first to solve conflict with VOP1 and VOP2
// encodings
if (Bytes.size() >= 8) { if (Bytes.size() >= 8) {
const uint64_t QW = eatBytes<uint64_t>(Bytes); const uint64_t QW = eatBytes<uint64_t>(Bytes);
Res = tryDecodeInst(DecoderTableDPP64, MI, QW, Address); Res = tryDecodeInst(DecoderTableDPP64, MI, QW, Address);
if (Res) break; if (Res) break;
Res = tryDecodeInst(DecoderTableSDWA64, MI, QW, Address);
if (Res) break;
} }
// Reinitialize Bytes as DPP64 could have eaten too much // Reinitialize Bytes as DPP64 could have eaten too much
Bytes = Bytes_.slice(0, MaxInstBytesNum); Bytes = Bytes_.slice(0, MaxInstBytesNum);
// Try decode 32-bit instruction // Try decode 32-bit instruction
if (Bytes.size() < 4) break; if (Bytes.size() < 4) break;
const uint32_t DW = eatBytes<uint32_t>(Bytes); const uint32_t DW = eatBytes<uint32_t>(Bytes);
▲ Show 20 Lines • Show All 293 Lines • Show Last 20 Lines

llvm/trunk/test/MC/Disassembler/AMDGPU/sdwa_vi.txt

# RUN: llvm-mc -arch=amdgcn -mcpu=tonga -disassemble -show-encoding < %s | FileCheck %s -check-prefix=VI
# VI: v_mov_b32_sdwa v1, v2 dst_sel:BYTE_0 dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x02,0x10,0x06,0x06]
0xf9 0x02 0x02 0x7e 0x02 0x10 0x06 0x06
# VI: v_mov_b32_sdwa v3, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 ; encoding: [0xf9,0x02,0x06,0x7e,0x04,0x11,0x05,0x06]
0xf9 0x02 0x06 0x7e 0x04 0x11 0x05 0x06
# VI: v_mov_b32_sdwa v15, v99 dst_sel:BYTE_2 dst_unused:UNUSED_SEXT src0_sel:WORD_0 ; encoding: [0xf9,0x02,0x1e,0x7e,0x63,0x0a,0x04,0x06]
0xf9 0x02 0x1e 0x7e 0x63 0x0a 0x04 0x06
# VI: v_min_u32_sdwa v194, v13, v1 dst_sel:BYTE_3 dst_unused:UNUSED_SEXT src0_sel:BYTE_3 src1_sel:BYTE_2 ; encoding: [0xf9,0x02,0x84,0x1d,0x0d,0x0b,0x03,0x02]
0xf9 0x02 0x84 0x1d 0x0d 0x0b 0x03 0x02
# VI: v_min_u32_sdwa v255, v4, v1 dst_sel:WORD_0 dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:WORD_1 ; encoding: [0xf9,0x02,0xfe,0x1d,0x04,0x04,0x02,0x05]
0xf9 0x02 0xfe 0x1d 0x04 0x04 0x02 0x05
# VI: v_min_u32_sdwa v200, v200, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD ; encoding: [0xf9,0x02,0x90,0x1d,0xc8,0x05,0x01,0x06]
0xf9 0x02 0x90 0x1d 0xc8 0x05 0x01 0x06
# VI: v_min_u32_sdwa v1, v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x1c,0x01,0x06,0x00,0x06]
0xf9 0x02 0x02 0x1c 0x01 0x06 0x00 0x06
# VI: v_cvt_u32_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x0e,0x00,0x7e,0x00,0x16,0x06,0x06]
0xf9 0x0e 0x00 0x7e 0x00 0x16 0x06 0x06
# VI: v_fract_f32_sdwa v0, v0 clamp dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD ; encoding: [0xf9,0x36,0x00,0x7e,0x00,0x26,0x06,0x06]
0xf9 0x36 0x00 0x7e 0x00 0x26 0x06 0x06
# VI: v_sin_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x52,0x00,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x52 0x00 0x7e 0x00 0x06 0x05 0x06
# VI: v_mov_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 ; encoding: [0xf9,0x02,0x02,0x7e,0x00,0x16,0x05,0x06]
0xf9 0x02 0x02 0x7e 0x00 0x16 0x05 0x06
# VI: v_trunc_f32_sdwa v1, v0 clamp dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:WORD_1 ; encoding: [0xf9,0x38,0x02,0x7e,0x00,0x36,0x05,0x06]
0xf9 0x38 0x02 0x7e 0x00 0x36 0x05 0x06
# VI: v_mov_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD ; encoding: [0xf9,0x02,0x02,0x7e,0x00,0x16,0x06,0x06]
0xf9 0x02 0x02 0x7e 0x00 0x16 0x06 0x06
# VI: v_add_f32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x05,0x06]
0xf9 0x00 0x00 0x02 0x00 0x06 0x05 0x06
# VI: v_min_f32_sdwa v0, v0, v0 clamp dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x14,0x00,0x36,0x06,0x02]
0xf9 0x00 0x00 0x14 0x00 0x36 0x06 0x02
# VI: v_and_b32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x26,0x00,0x06,0x06,0x02]
0xf9 0x00 0x00 0x26 0x00 0x06 0x06 0x02
# VI: v_mul_i32_i24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PRESERVE src0_sel:DWORD src1_sel:DWORD ; encoding: [0xf9,0x06,0x02,0x0c,0x02,0x16,0x06,0x06]
0xf9 0x06 0x02 0x0c 0x02 0x16 0x06 0x06
# VI: v_fract_f32_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x36,0x00,0x7e,0x00,0x06,0x25,0x06]
0xf9 0x36 0x00 0x7e 0x00 0x06 0x25 0x06
# VI: v_sin_f32_sdwa v0, -|v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x52,0x00,0x7e,0x00,0x06,0x35,0x06]
0xf9 0x52 0x00 0x7e 0x00 0x06 0x35 0x06
# VI: v_add_f32_sdwa v0, -|v0|, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x35,0x12]
0xf9 0x00 0x00 0x02 0x00 0x06 0x35 0x12
# VI: v_min_f32_sdwa v0, |v0|, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x14,0x00,0x06,0x25,0x12]
0xf9 0x00 0x00 0x14 0x00 0x06 0x25 0x12
# VI: v_nop ; encoding: [0xf9,0x00,0x00,0x7e,0x00,0x16,0x06,0x06]
0xf9 0x00 0x00 0x7e 0x00 0x16 0x06 0x06
# VI: v_cvt_u32_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x0e,0x00,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x0e 0x00 0x7e 0x00 0x06 0x05 0x06
# VI: v_fract_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x36,0x00,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x36 0x00 0x7e 0x00 0x06 0x05 0x06
# VI: v_sin_f32_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x52,0x00,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x52 0x00 0x7e 0x00 0x06 0x05 0x06
# VI: v_mov_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x02,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x02 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f32_i32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x0a,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x0a 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f32_u32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x0c,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x0c 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x10,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x10 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f16_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x14,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x14 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f32_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x16,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x16 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_rpi_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x18,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x18 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_flr_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x1a,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x1a 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_off_f32_i4_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x1c,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x1c 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f32_ubyte0_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x22,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x22 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f32_ubyte1_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x24,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x24 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f32_ubyte2_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x26,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x26 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f32_ubyte3_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x28,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x28 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_trunc_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x38,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x38 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_ceil_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x3a,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x3a 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_rndne_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x3c,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x3c 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_floor_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x3e,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x3e 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_exp_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x40,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x40 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_log_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x42,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x42 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_rcp_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x44,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x44 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_rcp_iflag_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x46,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x46 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_rsq_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x48,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x48 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_sqrt_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x4e,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x4e 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cos_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x54,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x54 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_not_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x56,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x56 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_bfrev_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x58,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x58 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_ffbh_u32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x5a,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x5a 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_ffbl_b32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x5c,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x5c 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_ffbh_i32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x5e,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x5e 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_frexp_exp_i32_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x66,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x66 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_frexp_mant_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x68,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x68 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_log_legacy_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x98,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x98 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_exp_legacy_f32_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x96,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x96 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f16_u16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x72,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x72 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_f16_i16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x74,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x74 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_u16_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x76,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x76 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cvt_i16_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x78,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x78 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_rcp_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x7a,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x7a 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_sqrt_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x7c,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x7c 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_rsq_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x7e,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x7e 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_log_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x80,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x80 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_exp_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x82,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x82 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_frexp_mant_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x84,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x84 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_frexp_exp_i16_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x86,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x86 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_floor_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x88,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x88 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_ceil_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x8a,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x8a 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_trunc_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x8c,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x8c 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_rndne_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x8e,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x8e 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_fract_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x90,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x90 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_sin_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x92,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x92 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_cos_f16_sdwa v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 ; encoding: [0xf9,0x94,0x02,0x7e,0x00,0x06,0x05,0x06]
0xf9 0x94 0x02 0x7e 0x00 0x06 0x05 0x06
# VI: v_add_f32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x02,0x00,0x06,0x05,0x02]
0xf9 0x00 0x00 0x02 0x00 0x06 0x05 0x02
# VI: v_min_f32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x14,0x00,0x06,0x05,0x02]
0xf9 0x00 0x00 0x14 0x00 0x06 0x05 0x02
# VI: v_and_b32_sdwa v0, v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x00,0x00,0x26,0x00,0x06,0x05,0x02]
0xf9 0x00 0x00 0x26 0x00 0x06 0x05 0x02
# VI: v_mul_i32_i24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x0c,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x0c 0x02 0x06 0x05 0x02
# VI: v_sub_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x04,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x04 0x02 0x06 0x05 0x02
# VI: v_subrev_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x06,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x06 0x02 0x06 0x05 0x02
# VI: v_mul_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x0a,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x0a 0x02 0x06 0x05 0x02
# VI: v_mul_hi_i32_i24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x0e,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x0e 0x02 0x06 0x05 0x02
# VI: v_mul_u32_u24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x10,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x10 0x02 0x06 0x05 0x02
# VI: v_mul_hi_u32_u24_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x12,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x12 0x02 0x06 0x05 0x02
# VI: v_max_f32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x16,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x16 0x02 0x06 0x05 0x02
# VI: v_min_i32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x18,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x18 0x02 0x06 0x05 0x02
# VI: v_max_i32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x1a,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x1a 0x02 0x06 0x05 0x02
# VI: v_min_u32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x1c,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x1c 0x02 0x06 0x05 0x02
# VI: v_max_u32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x1e,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x1e 0x02 0x06 0x05 0x02
# VI: v_lshrrev_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x20,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x20 0x02 0x06 0x05 0x02
# VI: v_ashrrev_i32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x22,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x22 0x02 0x06 0x05 0x02
# VI: v_lshlrev_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x24,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x24 0x02 0x06 0x05 0x02
# VI: v_or_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x28,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x28 0x02 0x06 0x05 0x02
# VI: v_xor_b32_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x2a,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x2a 0x02 0x06 0x05 0x02
# VI: v_add_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x3e,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x3e 0x02 0x06 0x05 0x02
# VI: v_sub_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x40,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x40 0x02 0x06 0x05 0x02
# VI: v_subrev_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x42,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x42 0x02 0x06 0x05 0x02
# VI: v_mul_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x44,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x44 0x02 0x06 0x05 0x02
# VI: v_mac_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x46,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x46 0x02 0x06 0x05 0x02
# VI: v_add_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x4c,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x4c 0x02 0x06 0x05 0x02
# VI: v_sub_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x4e,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x4e 0x02 0x06 0x05 0x02
# VI: v_subrev_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x50,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x50 0x02 0x06 0x05 0x02
# VI: v_mul_lo_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x52,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x52 0x02 0x06 0x05 0x02
# VI: v_lshlrev_b16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x54,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x54 0x02 0x06 0x05 0x02
# VI: v_lshrrev_b16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x56,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x56 0x02 0x06 0x05 0x02
# VI: v_ashrrev_b16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x58,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x58 0x02 0x06 0x05 0x02
# VI: v_max_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x5a,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x5a 0x02 0x06 0x05 0x02
# VI: v_min_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x5c,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x5c 0x02 0x06 0x05 0x02
# VI: v_max_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x5e,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x5e 0x02 0x06 0x05 0x02
# VI: v_max_i16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x60,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x60 0x02 0x06 0x05 0x02
# VI: v_min_u16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x62,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x62 0x02 0x06 0x05 0x02
# VI: v_min_i16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x64,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x64 0x02 0x06 0x05 0x02
# VI: v_ldexp_f16_sdwa v1, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:BYTE_2 ; encoding: [0xf9,0x06,0x02,0x66,0x02,0x06,0x05,0x02]
0xf9 0x06 0x02 0x66 0x02 0x06 0x05 0x02