mawei/StarPilot
1
0
Fork 0
mirror of https://github.com/firestar5683/StarPilot.git synced 2026-11-08 13:33:42 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
63243d7e2d768514b61b5cd5d357200ecab7e208
StarPilot/tinygrad_repo/tinygrad/runtime/autogen/amd/cdna/str_pcode.py
T
firestar5683 22a8575199 Vendor dependencies and resources for standalone builds 
Keep official openpilot ancestry and dependency licenses. Store resources directly in Git, remove LFS setup requirements, and leave the downloadable big model outside the source tree.
2026-10-03 21:01:18 +00:00

1447 lines
225 KiB
Python
Raw Blame History

# autogenerated from AMD ISA PDF - do not edit
# ruff: noqa: E501
from tinygrad.runtime.autogen.amd.cdna.enum import DSOp, FLATOp, GLOBALOp, MTBUFOp, MUBUFOp, SCRATCHOp, SMEMOp, SOP1Op, SOP2Op, SOPCOp, SOPKOp, SOPPOp, VOP1Op, VOP2Op, VOP3Op, VOP3POp, VOP3SDOp, VOPCOp
PCODE = {
DSOp.DS_ADD_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 += DATA.u32;\nRETURN_DATA.u32 = tmp',
DSOp.DS_SUB_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 -= DATA.u32;\nRETURN_DATA.u32 = tmp',
DSOp.DS_RSUB_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 = DATA.u32 - MEM[addr].u32;\nRETURN_DATA.u32 = tmp',
DSOp.DS_INC_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = tmp >= src ? 0U : tmp + 1U;\nRETURN_DATA.u32 = tmp',
DSOp.DS_DEC_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = ((tmp == 0U) || (tmp > src)) ? src : tmp - 1U;\nRETURN_DATA.u32 = tmp',
DSOp.DS_MIN_I32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src < tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
DSOp.DS_MAX_I32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src >= tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
DSOp.DS_MIN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src < tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
DSOp.DS_MAX_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src >= tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
DSOp.DS_AND_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp & DATA.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_OR_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp | DATA.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_XOR_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp ^ DATA.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_MSKOR_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = ((tmp & ~DATA.b32) | DATA2.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_WRITE_B32: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET.u32].b32 = DATA[31 : 0]',
DSOp.DS_WRITE2_B32: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET0.u32 * 4U].b32 = DATA[31 : 0];\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET1.u32 * 4U].b32 = DATA2[31 : 0]',
DSOp.DS_WRITE2ST64_B32: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET0.u32 * 256U].b32 = DATA[31 : 0];\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET1.u32 * 256U].b32 = DATA2[31 : 0]',
DSOp.DS_CMPST_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nsrc = DATA2.b32;\ncmp = DATA.b32;\nMEM[addr].b32 = tmp == cmp ? src : tmp;\nRETURN_DATA.b32 = tmp',
DSOp.DS_CMPST_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nsrc = DATA2.f32;\ncmp = DATA.f32;\nMEM[addr].f32 = tmp == cmp ? src : tmp;\nRETURN_DATA.f32 = tmp',
DSOp.DS_MIN_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nsrc = DATA.f32;\nMEM[addr].f32 = src < tmp ? src : tmp;\nRETURN_DATA.f32 = tmp',
DSOp.DS_MAX_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nsrc = DATA.f32;\nMEM[addr].f32 = src > tmp ? src : tmp;\nRETURN_DATA.f32 = tmp',
DSOp.DS_ADD_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nMEM[addr].f32 += DATA.f32;\nRETURN_DATA.f32 = tmp',
DSOp.DS_PK_ADD_F16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].f16 = tmp[31 : 16].f16 + src[31 : 16].f16;\ndst[15 : 0].f16 = tmp[15 : 0].f16 + src[15 : 0].f16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
DSOp.DS_PK_ADD_BF16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].bf16 = tmp[31 : 16].bf16 + src[31 : 16].bf16;\ndst[15 : 0].bf16 = tmp[15 : 0].bf16 + src[15 : 0].bf16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
DSOp.DS_WRITE_ADDTID_B32: "declare OFFSET0 : 8'U;\ndeclare OFFSET1 : 8'U;\nMEM[32'I({ OFFSET1, OFFSET0 } + M0[15 : 0]) + laneID.i32 * 4].u32 = DATA0.u32",
DSOp.DS_WRITE_B8: 'MEM[ADDR].b8 = DATA[7 : 0]',
DSOp.DS_WRITE_B16: 'MEM[ADDR].b16 = DATA[15 : 0]',
DSOp.DS_ADD_RTN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 += DATA.u32;\nRETURN_DATA.u32 = tmp',
DSOp.DS_SUB_RTN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 -= DATA.u32;\nRETURN_DATA.u32 = tmp',
DSOp.DS_RSUB_RTN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 = DATA.u32 - MEM[addr].u32;\nRETURN_DATA.u32 = tmp',
DSOp.DS_INC_RTN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = tmp >= src ? 0U : tmp + 1U;\nRETURN_DATA.u32 = tmp',
DSOp.DS_DEC_RTN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = ((tmp == 0U) || (tmp > src)) ? src : tmp - 1U;\nRETURN_DATA.u32 = tmp',
DSOp.DS_MIN_RTN_I32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src < tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
DSOp.DS_MAX_RTN_I32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src >= tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
DSOp.DS_MIN_RTN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src < tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
DSOp.DS_MAX_RTN_U32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src >= tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
DSOp.DS_AND_RTN_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp & DATA.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_OR_RTN_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp | DATA.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_XOR_RTN_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp ^ DATA.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_MSKOR_RTN_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = ((tmp & ~DATA.b32) | DATA2.b32);\nRETURN_DATA.b32 = tmp',
DSOp.DS_WRXCHG_RTN_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = DATA.b32;\nRETURN_DATA.b32 = tmp',
DSOp.DS_WRXCHG2_RTN_B32: 'addr1 = ADDR_BASE.u32 + OFFSET0.u32 * 4U;\naddr2 = ADDR_BASE.u32 + OFFSET1.u32 * 4U;\ntmp1 = MEM[addr1].b32;\ntmp2 = MEM[addr2].b32;\nMEM[addr1].b32 = DATA.b32;\nMEM[addr2].b32 = DATA2.b32;\n// Note DATA2 can be any other register\nRETURN_DATA[31 : 0] = tmp1;\nRETURN_DATA[63 : 32] = tmp2',
DSOp.DS_WRXCHG2ST64_RTN_B32: 'addr1 = ADDR_BASE.u32 + OFFSET0.u32 * 256U;\naddr2 = ADDR_BASE.u32 + OFFSET1.u32 * 256U;\ntmp1 = MEM[addr1].b32;\ntmp2 = MEM[addr2].b32;\nMEM[addr1].b32 = DATA.b32;\nMEM[addr2].b32 = DATA2.b32;\n// Note DATA2 can be any other register\nRETURN_DATA[31 : 0] = tmp1;\nRETURN_DATA[63 : 32] = tmp2',
DSOp.DS_CMPST_RTN_B32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b32;\nsrc = DATA2.b32;\ncmp = DATA.b32;\nMEM[addr].b32 = tmp == cmp ? src : tmp;\nRETURN_DATA.b32 = tmp',
DSOp.DS_CMPST_RTN_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nsrc = DATA2.f32;\ncmp = DATA.f32;\nMEM[addr].f32 = tmp == cmp ? src : tmp;\nRETURN_DATA.f32 = tmp',
DSOp.DS_MIN_RTN_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nsrc = DATA.f32;\nMEM[addr].f32 = src < tmp ? src : tmp;\nRETURN_DATA.f32 = tmp',
DSOp.DS_MAX_RTN_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nsrc = DATA.f32;\nMEM[addr].f32 = src > tmp ? src : tmp;\nRETURN_DATA.f32 = tmp',
DSOp.DS_WRAP_RTN_B32: 'tmp = MEM[ADDR].u32;\nMEM[ADDR].u32 = tmp >= DATA.u32 ? tmp - DATA.u32 : tmp + DATA2.u32;\nRETURN_DATA = tmp',
DSOp.DS_ADD_RTN_F32: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f32;\nMEM[addr].f32 += DATA.f32;\nRETURN_DATA.f32 = tmp',
DSOp.DS_READ_B32: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET.u32].b32',
DSOp.DS_READ2_B32: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET0.u32 * 4U].b32;\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[63 : 32] = MEM[addr + OFFSET1.u32 * 4U].b32',
DSOp.DS_READ2ST64_B32: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET0.u32 * 256U].b32;\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[63 : 32] = MEM[addr + OFFSET1.u32 * 256U].b32',
DSOp.DS_READ_I8: "RETURN_DATA.i32 = 32'I(signext(MEM[ADDR].i8))",
DSOp.DS_READ_U8: "RETURN_DATA.u32 = 32'U({ 24'0U, MEM[ADDR].u8 })",
DSOp.DS_READ_I16: "RETURN_DATA.i32 = 32'I(signext(MEM[ADDR].i16))",
DSOp.DS_READ_U16: "RETURN_DATA.u32 = 32'U({ 16'0U, MEM[ADDR].u16 })",
DSOp.DS_SWIZZLE_B32: 'Offset[4:0]: Swizzle\n0x00: {1,11,9,19,5,15,d,1d,3,13,b,1b,7,17,f,1f,2,12,a,1a,6,16,e,1e,4,14,c,1c,8,18,10,20}\n0x10: {1,9,5,d,3,b,7,f,2,a,6,e,4,c,8,10,11,19,15,1d,13,1b,17,1f,12,1a,16,1e,14,1c,18,20}\n0x1f: No swizzle',
DSOp.DS_PERMUTE_B32: "// VGPR[laneId][index] is the VGPR RAM\n// VDST, ADDR and DATA0 are from the microcode DS encoding\ndeclare tmp : 32'B[64];\ndeclare OFFSET : 16'U;\ndeclare DATA0 : 32'U;\ndeclare VDST : 32'U;\nfor i in 0 : 63 do\ntmp[i] = 0x0\nendfor;\nfor i in 0 : 63 do\n// If a source thread is disabled, it does not propagate data.\nif EXEC[i].u1 then\n// ADDR needs to be divided by 4.\n// High-order bits are ignored.\ndst_lane = (VGPR[i][ADDR].u32 + OFFSET.u32) / 4U % 64U;\ntmp[dst_lane] = VGPR[i][DATA0]\nendif\nendfor;\n// Copy data into destination VGPRs. If multiple sources\n// select the same destination thread, the highest-numbered\n// source thread wins.\nfor i in 0 : 63 do\nif EXEC[i].u1 then\nVGPR[i][VDST] = tmp[i]\nendif\nendfor",
DSOp.DS_BPERMUTE_B32: "// VGPR[laneId][index] is the VGPR RAM\n// VDST, ADDR and DATA0 are from the microcode DS encoding\ndeclare tmp : 32'B[64];\ndeclare OFFSET : 16'U;\ndeclare DATA0 : 32'U;\ndeclare VDST : 32'U;\nfor i in 0 : 63 do\ntmp[i] = 0x0\nendfor;\nfor i in 0 : 63 do\n// ADDR needs to be divided by 4.\n// High-order bits are ignored.\nsrc_lane = (VGPR[i][ADDR].u32 + OFFSET.u32) / 4U % 64U;\n// EXEC is applied to the source VGPR reads.\nif EXEC[src_lane].u1 then\ntmp[i] = VGPR[src_lane][DATA0]\nendif\nendfor;\n// Copy data into destination VGPRs. Some source\n// data may be broadcast to multiple lanes.\nfor i in 0 : 63 do\nif EXEC[i].u1 then\nVGPR[i][VDST] = tmp[i]\nendif\nendfor",
DSOp.DS_ADD_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 += DATA.u64;\nRETURN_DATA.u64 = tmp',
DSOp.DS_SUB_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 -= DATA.u64;\nRETURN_DATA.u64 = tmp',
DSOp.DS_RSUB_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 = DATA.u64 - MEM[addr].u64;\nRETURN_DATA.u64 = tmp',
DSOp.DS_INC_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = tmp >= src ? 0ULL : tmp + 1ULL;\nRETURN_DATA.u64 = tmp',
DSOp.DS_DEC_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = ((tmp == 0ULL) || (tmp > src)) ? src : tmp - 1ULL;\nRETURN_DATA.u64 = tmp',
DSOp.DS_MIN_I64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src < tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
DSOp.DS_MAX_I64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src >= tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
DSOp.DS_MIN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src < tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
DSOp.DS_MAX_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src >= tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
DSOp.DS_AND_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp & DATA.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_OR_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp | DATA.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_XOR_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp ^ DATA.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_MSKOR_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = ((tmp & ~DATA.b64) | DATA2.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_WRITE_B64: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET.u32].b32 = DATA[31 : 0];\nMEM[addr + OFFSET.u32 + 4U].b32 = DATA[63 : 32]',
DSOp.DS_WRITE2_B64: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET0.u32 * 8U].b32 = DATA[31 : 0];\nMEM[addr + OFFSET0.u32 * 8U + 4U].b32 = DATA[63 : 32];\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET1.u32 * 8U].b32 = DATA2[31 : 0];\nMEM[addr + OFFSET1.u32 * 8U + 4U].b32 = DATA2[63 : 32]',
DSOp.DS_WRITE2ST64_B64: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET0.u32 * 512U].b32 = DATA[31 : 0];\nMEM[addr + OFFSET0.u32 * 512U + 4U].b32 = DATA[63 : 32];\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET1.u32 * 512U].b32 = DATA2[31 : 0];\nMEM[addr + OFFSET1.u32 * 512U + 4U].b32 = DATA2[63 : 32]',
DSOp.DS_CMPST_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nsrc = DATA2.b64;\ncmp = DATA.b64;\nMEM[addr].b64 = tmp == cmp ? src : tmp;\nRETURN_DATA.b64 = tmp',
DSOp.DS_CMPST_F64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f64;\nsrc = DATA2.f64;\ncmp = DATA.f64;\nMEM[addr].f64 = tmp == cmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
DSOp.DS_MIN_F64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src < tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
DSOp.DS_MAX_F64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src > tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
DSOp.DS_WRITE_B8_D16_HI: 'MEM[ADDR].b8 = DATA[23 : 16]',
DSOp.DS_WRITE_B16_D16_HI: 'MEM[ADDR].b16 = DATA[31 : 16]',
DSOp.DS_READ_U8_D16: "RETURN_DATA[15 : 0].u16 = 16'U({ 8'0U, MEM[ADDR].u8 });\n// RETURN_DATA[31:16] is preserved.",
DSOp.DS_READ_U8_D16_HI: "RETURN_DATA[31 : 16].u16 = 16'U({ 8'0U, MEM[ADDR].u8 });\n// RETURN_DATA[15:0] is preserved.",
DSOp.DS_READ_I8_D16: "RETURN_DATA[15 : 0].i16 = 16'I(signext(MEM[ADDR].i8));\n// RETURN_DATA[31:16] is preserved.",
DSOp.DS_READ_I8_D16_HI: "RETURN_DATA[31 : 16].i16 = 16'I(signext(MEM[ADDR].i8));\n// RETURN_DATA[15:0] is preserved.",
DSOp.DS_READ_U16_D16: 'RETURN_DATA[15 : 0].u16 = MEM[ADDR].u16;\n// RETURN_DATA[31:16] is preserved.',
DSOp.DS_READ_U16_D16_HI: 'RETURN_DATA[31 : 16].u16 = MEM[ADDR].u16;\n// RETURN_DATA[15:0] is preserved.',
DSOp.DS_ADD_F64: 'tmp = MEM[ADDR].f64;\nMEM[ADDR].f64 += DATA.f64;\nRETURN_DATA = tmp',
DSOp.DS_ADD_RTN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 += DATA.u64;\nRETURN_DATA.u64 = tmp',
DSOp.DS_SUB_RTN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 -= DATA.u64;\nRETURN_DATA.u64 = tmp',
DSOp.DS_RSUB_RTN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 = DATA.u64 - MEM[addr].u64;\nRETURN_DATA.u64 = tmp',
DSOp.DS_INC_RTN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = tmp >= src ? 0ULL : tmp + 1ULL;\nRETURN_DATA.u64 = tmp',
DSOp.DS_DEC_RTN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = ((tmp == 0ULL) || (tmp > src)) ? src : tmp - 1ULL;\nRETURN_DATA.u64 = tmp',
DSOp.DS_MIN_RTN_I64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src < tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
DSOp.DS_MAX_RTN_I64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src >= tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
DSOp.DS_MIN_RTN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src < tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
DSOp.DS_MAX_RTN_U64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src >= tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
DSOp.DS_AND_RTN_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp & DATA.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_OR_RTN_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp | DATA.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_XOR_RTN_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp ^ DATA.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_MSKOR_RTN_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = ((tmp & ~DATA.b64) | DATA2.b64);\nRETURN_DATA.b64 = tmp',
DSOp.DS_WRXCHG_RTN_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = DATA.b64;\nRETURN_DATA.b64 = tmp',
DSOp.DS_WRXCHG2_RTN_B64: 'addr1 = ADDR_BASE.u32 + OFFSET0.u32 * 8U;\naddr2 = ADDR_BASE.u32 + OFFSET1.u32 * 8U;\ntmp1 = MEM[addr1].b64;\ntmp2 = MEM[addr2].b64;\nMEM[addr1].b64 = DATA.b64;\nMEM[addr2].b64 = DATA2.b64;\n// Note DATA2 can be any other register\nRETURN_DATA[63 : 0] = tmp1;\nRETURN_DATA[127 : 64] = tmp2',
DSOp.DS_WRXCHG2ST64_RTN_B64: 'addr1 = ADDR_BASE.u32 + OFFSET0.u32 * 512U;\naddr2 = ADDR_BASE.u32 + OFFSET1.u32 * 512U;\ntmp1 = MEM[addr1].b64;\ntmp2 = MEM[addr2].b64;\nMEM[addr1].b64 = DATA.b64;\nMEM[addr2].b64 = DATA2.b64;\n// Note DATA2 can be any other register\nRETURN_DATA[63 : 0] = tmp1;\nRETURN_DATA[127 : 64] = tmp2',
DSOp.DS_CMPST_RTN_B64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].b64;\nsrc = DATA2.b64;\ncmp = DATA.b64;\nMEM[addr].b64 = tmp == cmp ? src : tmp;\nRETURN_DATA.b64 = tmp',
DSOp.DS_CMPST_RTN_F64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f64;\nsrc = DATA2.f64;\ncmp = DATA.f64;\nMEM[addr].f64 = tmp == cmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
DSOp.DS_MIN_RTN_F64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src < tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
DSOp.DS_MAX_RTN_F64: 'addr = CalcDsAddr(ADDR.b32, OFFSET0.b32, OFFSET1.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src > tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
DSOp.DS_READ_B64: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET.u32].b32;\nRETURN_DATA[63 : 32] = MEM[addr + OFFSET.u32 + 4U].b32',
DSOp.DS_READ2_B64: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET0.u32 * 8U].b32;\nRETURN_DATA[63 : 32] = MEM[addr + OFFSET0.u32 * 8U + 4U].b32;\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[95 : 64] = MEM[addr + OFFSET1.u32 * 8U].b32;\nRETURN_DATA[127 : 96] = MEM[addr + OFFSET1.u32 * 8U + 4U].b32',
DSOp.DS_READ2ST64_B64: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET0.u32 * 512U].b32;\nRETURN_DATA[63 : 32] = MEM[addr + OFFSET0.u32 * 512U + 4U].b32;\naddr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[95 : 64] = MEM[addr + OFFSET1.u32 * 512U].b32;\nRETURN_DATA[127 : 96] = MEM[addr + OFFSET1.u32 * 512U + 4U].b32',
DSOp.DS_ADD_RTN_F64: 'tmp = MEM[ADDR].f64;\nMEM[ADDR].f64 += DATA.f64;\nRETURN_DATA = tmp',
DSOp.DS_CONDXCHG32_RTN_B64: "declare OFFSET0 : 8'U;\ndeclare OFFSET1 : 8'U;\ndeclare RETURN_DATA : 32'U[2];\nADDR = S0.u32;\nDATA = S1.u64;\noffset = { OFFSET1, OFFSET0 };\nADDR0 = ((ADDR + offset.u32) & 0xfff8U);\nADDR1 = ADDR0 + 4U;\nRETURN_DATA[0] = LDS[ADDR0].u32;\nif DATA[31] then\nLDS[ADDR0] = { 1'0, DATA[30 : 0] }\nendif;\nRETURN_DATA[1] = LDS[ADDR1].u32;\nif DATA[63] then\nLDS[ADDR1] = { 1'0, DATA[62 : 32] }\nendif",
DSOp.DS_READ_ADDTID_B32: "declare OFFSET0 : 8'U;\ndeclare OFFSET1 : 8'U;\nRETURN_DATA.u32 = MEM[32'I({ OFFSET1, OFFSET0 } + M0[15 : 0]) + laneID.i32 * 4].u32",
DSOp.DS_PK_ADD_RTN_F16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].f16 = tmp[31 : 16].f16 + src[31 : 16].f16;\ndst[15 : 0].f16 = tmp[15 : 0].f16 + src[15 : 0].f16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
DSOp.DS_PK_ADD_RTN_BF16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].bf16 = tmp[31 : 16].bf16 + src[31 : 16].bf16;\ndst[15 : 0].bf16 = tmp[15 : 0].bf16 + src[15 : 0].bf16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
DSOp.DS_WRITE_B96: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET.u32].b32 = DATA[31 : 0];\nMEM[addr + OFFSET.u32 + 4U].b32 = DATA[63 : 32];\nMEM[addr + OFFSET.u32 + 8U].b32 = DATA[95 : 64]',
DSOp.DS_WRITE_B128: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nMEM[addr + OFFSET.u32].b32 = DATA[31 : 0];\nMEM[addr + OFFSET.u32 + 4U].b32 = DATA[63 : 32];\nMEM[addr + OFFSET.u32 + 8U].b32 = DATA[95 : 64];\nMEM[addr + OFFSET.u32 + 12U].b32 = DATA[127 : 96]',
DSOp.DS_READ_B96: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET.u32].b32;\nRETURN_DATA[63 : 32] = MEM[addr + OFFSET.u32 + 4U].b32;\nRETURN_DATA[95 : 64] = MEM[addr + OFFSET.u32 + 8U].b32',
DSOp.DS_READ_B128: 'addr = CalcDsAddr(ADDR.b32, 0x0, 0x0);\nRETURN_DATA[31 : 0] = MEM[addr + OFFSET.u32].b32;\nRETURN_DATA[63 : 32] = MEM[addr + OFFSET.u32 + 4U].b32;\nRETURN_DATA[95 : 64] = MEM[addr + OFFSET.u32 + 8U].b32;\nRETURN_DATA[127 : 96] = MEM[addr + OFFSET.u32 + 12U].b32',
FLATOp.FLAT_LOAD_UBYTE: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 24'0U, MEM[addr].u8 })",
FLATOp.FLAT_LOAD_SBYTE: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i8))",
FLATOp.FLAT_LOAD_USHORT: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 16'0U, MEM[addr].u16 })",
FLATOp.FLAT_LOAD_SSHORT: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i16))",
FLATOp.FLAT_LOAD_DWORD: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32',
FLATOp.FLAT_LOAD_DWORDX2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32',
FLATOp.FLAT_LOAD_DWORDX3: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32',
FLATOp.FLAT_LOAD_DWORDX4: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32;\nVDATA[127 : 96] = MEM[addr + 12U].b32',
FLATOp.FLAT_STORE_BYTE: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[7 : 0]',
FLATOp.FLAT_STORE_BYTE_D16_HI: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[23 : 16]',
FLATOp.FLAT_STORE_SHORT: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[15 : 0]',
FLATOp.FLAT_STORE_SHORT_D16_HI: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[31 : 16]',
FLATOp.FLAT_STORE_DWORD: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0]',
FLATOp.FLAT_STORE_DWORDX2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32]',
FLATOp.FLAT_STORE_DWORDX3: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64]',
FLATOp.FLAT_STORE_DWORDX4: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64];\nMEM[addr + 12U].b32 = VDATA[127 : 96]',
FLATOp.FLAT_LOAD_UBYTE_D16: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[15 : 0].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[31:16] is preserved.",
FLATOp.FLAT_LOAD_UBYTE_D16_HI: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[31 : 16].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[15:0] is preserved.",
FLATOp.FLAT_LOAD_SBYTE_D16: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[15 : 0].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[31:16] is preserved.",
FLATOp.FLAT_LOAD_SBYTE_D16_HI: "addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[31 : 16].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[15:0] is preserved.",
FLATOp.FLAT_LOAD_SHORT_D16: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = MEM[addr].b16;\n// VDATA[31:16] is preserved.',
FLATOp.FLAT_LOAD_SHORT_D16_HI: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\nVDATA[31 : 16].b16 = MEM[addr].b16;\n// VDATA[15:0] is preserved.',
FLATOp.FLAT_ATOMIC_SWAP: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = DATA.b32;\nRETURN_DATA.b32 = tmp',
FLATOp.FLAT_ATOMIC_CMPSWAP: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA[31 : 0].u32;\ncmp = DATA[63 : 32].u32;\nMEM[addr].u32 = tmp == cmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
FLATOp.FLAT_ATOMIC_ADD: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 += DATA.u32;\nRETURN_DATA.u32 = tmp',
FLATOp.FLAT_ATOMIC_SUB: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 -= DATA.u32;\nRETURN_DATA.u32 = tmp',
FLATOp.FLAT_ATOMIC_SMIN: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src < tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
FLATOp.FLAT_ATOMIC_UMIN: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src < tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
FLATOp.FLAT_ATOMIC_SMAX: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src >= tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
FLATOp.FLAT_ATOMIC_UMAX: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src >= tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
FLATOp.FLAT_ATOMIC_AND: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp & DATA.b32);\nRETURN_DATA.b32 = tmp',
FLATOp.FLAT_ATOMIC_OR: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp | DATA.b32);\nRETURN_DATA.b32 = tmp',
FLATOp.FLAT_ATOMIC_XOR: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp ^ DATA.b32);\nRETURN_DATA.b32 = tmp',
FLATOp.FLAT_ATOMIC_INC: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = tmp >= src ? 0U : tmp + 1U;\nRETURN_DATA.u32 = tmp',
FLATOp.FLAT_ATOMIC_DEC: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = ((tmp == 0U) || (tmp > src)) ? src : tmp - 1U;\nRETURN_DATA.u32 = tmp',
FLATOp.FLAT_ATOMIC_ADD_F32: 'tmp = MEM[ADDR].f32;\nMEM[ADDR].f32 += DATA.f32;\nRETURN_DATA = tmp',
FLATOp.FLAT_ATOMIC_PK_ADD_F16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].f16 = tmp[31 : 16].f16 + src[31 : 16].f16;\ndst[15 : 0].f16 = tmp[15 : 0].f16 + src[15 : 0].f16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
FLATOp.FLAT_ATOMIC_ADD_F64: 'tmp = MEM[ADDR].f64;\nMEM[ADDR].f64 += DATA.f64;\nRETURN_DATA = tmp',
FLATOp.FLAT_ATOMIC_MIN_F64: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src < tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
FLATOp.FLAT_ATOMIC_MAX_F64: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src > tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
FLATOp.FLAT_ATOMIC_PK_ADD_BF16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].bf16 = tmp[31 : 16].bf16 + src[31 : 16].bf16;\ndst[15 : 0].bf16 = tmp[15 : 0].bf16 + src[15 : 0].bf16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
FLATOp.FLAT_ATOMIC_SWAP_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = DATA.b64;\nRETURN_DATA.b64 = tmp',
FLATOp.FLAT_ATOMIC_CMPSWAP_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA[63 : 0].u64;\ncmp = DATA[127 : 64].u64;\nMEM[addr].u64 = tmp == cmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
FLATOp.FLAT_ATOMIC_ADD_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 += DATA.u64;\nRETURN_DATA.u64 = tmp',
FLATOp.FLAT_ATOMIC_SUB_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 -= DATA.u64;\nRETURN_DATA.u64 = tmp',
FLATOp.FLAT_ATOMIC_SMIN_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src < tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
FLATOp.FLAT_ATOMIC_UMIN_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src < tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
FLATOp.FLAT_ATOMIC_SMAX_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src >= tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
FLATOp.FLAT_ATOMIC_UMAX_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src >= tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
FLATOp.FLAT_ATOMIC_AND_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp & DATA.b64);\nRETURN_DATA.b64 = tmp',
FLATOp.FLAT_ATOMIC_OR_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp | DATA.b64);\nRETURN_DATA.b64 = tmp',
FLATOp.FLAT_ATOMIC_XOR_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp ^ DATA.b64);\nRETURN_DATA.b64 = tmp',
FLATOp.FLAT_ATOMIC_INC_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = tmp >= src ? 0ULL : tmp + 1ULL;\nRETURN_DATA.u64 = tmp',
FLATOp.FLAT_ATOMIC_DEC_X2: 'addr = CalcFlatAddr(ADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = ((tmp == 0ULL) || (tmp > src)) ? src : tmp - 1ULL;\nRETURN_DATA.u64 = tmp',
GLOBALOp.GLOBAL_LOAD_UBYTE: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 24'0U, MEM[addr].u8 })",
GLOBALOp.GLOBAL_LOAD_SBYTE: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i8))",
GLOBALOp.GLOBAL_LOAD_USHORT: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 16'0U, MEM[addr].u16 })",
GLOBALOp.GLOBAL_LOAD_SSHORT: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i16))",
GLOBALOp.GLOBAL_LOAD_DWORD: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32',
GLOBALOp.GLOBAL_LOAD_DWORDX2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32',
GLOBALOp.GLOBAL_LOAD_DWORDX3: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32',
GLOBALOp.GLOBAL_LOAD_DWORDX4: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32;\nVDATA[127 : 96] = MEM[addr + 12U].b32',
GLOBALOp.GLOBAL_STORE_BYTE: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[7 : 0]',
GLOBALOp.GLOBAL_STORE_BYTE_D16_HI: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[23 : 16]',
GLOBALOp.GLOBAL_STORE_SHORT: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[15 : 0]',
GLOBALOp.GLOBAL_STORE_SHORT_D16_HI: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[31 : 16]',
GLOBALOp.GLOBAL_STORE_DWORD: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0]',
GLOBALOp.GLOBAL_STORE_DWORDX2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32]',
GLOBALOp.GLOBAL_STORE_DWORDX3: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64]',
GLOBALOp.GLOBAL_STORE_DWORDX4: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64];\nMEM[addr + 12U].b32 = VDATA[127 : 96]',
GLOBALOp.GLOBAL_LOAD_UBYTE_D16: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[15 : 0].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[31:16] is preserved.",
GLOBALOp.GLOBAL_LOAD_UBYTE_D16_HI: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 16].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[15:0] is preserved.",
GLOBALOp.GLOBAL_LOAD_SBYTE_D16: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[15 : 0].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[31:16] is preserved.",
GLOBALOp.GLOBAL_LOAD_SBYTE_D16_HI: "addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 16].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[15:0] is preserved.",
GLOBALOp.GLOBAL_LOAD_SHORT_D16: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = MEM[addr].b16;\n// VDATA[31:16] is preserved.',
GLOBALOp.GLOBAL_LOAD_SHORT_D16_HI: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 16].b16 = MEM[addr].b16;\n// VDATA[15:0] is preserved.',
GLOBALOp.GLOBAL_ATOMIC_SWAP: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = DATA.b32;\nRETURN_DATA.b32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_CMPSWAP: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA[31 : 0].u32;\ncmp = DATA[63 : 32].u32;\nMEM[addr].u32 = tmp == cmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_ADD: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 += DATA.u32;\nRETURN_DATA.u32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_SUB: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 -= DATA.u32;\nRETURN_DATA.u32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_SMIN: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src < tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_UMIN: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src < tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_SMAX: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src >= tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_UMAX: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src >= tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_AND: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp & DATA.b32);\nRETURN_DATA.b32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_OR: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp | DATA.b32);\nRETURN_DATA.b32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_XOR: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp ^ DATA.b32);\nRETURN_DATA.b32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_INC: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = tmp >= src ? 0U : tmp + 1U;\nRETURN_DATA.u32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_DEC: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = ((tmp == 0U) || (tmp > src)) ? src : tmp - 1U;\nRETURN_DATA.u32 = tmp',
GLOBALOp.GLOBAL_ATOMIC_ADD_F32: 'tmp = MEM[ADDR].f32;\nMEM[ADDR].f32 += DATA.f32;\nRETURN_DATA = tmp',
GLOBALOp.GLOBAL_ATOMIC_PK_ADD_F16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].f16 = tmp[31 : 16].f16 + src[31 : 16].f16;\ndst[15 : 0].f16 = tmp[15 : 0].f16 + src[15 : 0].f16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
GLOBALOp.GLOBAL_ATOMIC_ADD_F64: 'tmp = MEM[ADDR].f64;\nMEM[ADDR].f64 += DATA.f64;\nRETURN_DATA = tmp',
GLOBALOp.GLOBAL_ATOMIC_MIN_F64: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src < tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_MAX_F64: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src > tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_PK_ADD_BF16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].bf16 = tmp[31 : 16].bf16 + src[31 : 16].bf16;\ndst[15 : 0].bf16 = tmp[15 : 0].bf16 + src[15 : 0].bf16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
GLOBALOp.GLOBAL_ATOMIC_SWAP_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = DATA.b64;\nRETURN_DATA.b64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_CMPSWAP_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA[63 : 0].u64;\ncmp = DATA[127 : 64].u64;\nMEM[addr].u64 = tmp == cmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_ADD_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 += DATA.u64;\nRETURN_DATA.u64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_SUB_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 -= DATA.u64;\nRETURN_DATA.u64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_SMIN_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src < tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_UMIN_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src < tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_SMAX_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src >= tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_UMAX_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src >= tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_AND_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp & DATA.b64);\nRETURN_DATA.b64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_OR_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp | DATA.b64);\nRETURN_DATA.b64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_XOR_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp ^ DATA.b64);\nRETURN_DATA.b64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_INC_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = tmp >= src ? 0ULL : tmp + 1ULL;\nRETURN_DATA.u64 = tmp',
GLOBALOp.GLOBAL_ATOMIC_DEC_X2: 'addr = CalcGlobalAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = ((tmp == 0ULL) || (tmp > src)) ? src : tmp - 1ULL;\nRETURN_DATA.u64 = tmp',
MTBUFOp.TBUFFER_LOAD_FORMAT_X: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format',
MTBUFOp.TBUFFER_LOAD_FORMAT_XY: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format\nVDATA[63 : 32].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetY()])',
MTBUFOp.TBUFFER_LOAD_FORMAT_XYZ: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format\nVDATA[63 : 32].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetY()]);\nVDATA[95 : 64].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetZ()])',
MTBUFOp.TBUFFER_LOAD_FORMAT_XYZW: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format\nVDATA[63 : 32].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetY()]);\nVDATA[95 : 64].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetZ()]);\nVDATA[127 : 96].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetW()])',
MTBUFOp.TBUFFER_STORE_FORMAT_X: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format',
MTBUFOp.TBUFFER_STORE_FORMAT_XY: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(VDATA[63 : 32].b32)',
MTBUFOp.TBUFFER_STORE_FORMAT_XYZ: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(VDATA[63 : 32].b32);\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(VDATA[95 : 64].b32)',
MTBUFOp.TBUFFER_STORE_FORMAT_XYZW: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(VDATA[63 : 32].b32);\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(VDATA[95 : 64].b32);\nMEM[addr + ChannelOffsetW()] = ConvertToFormat(VDATA[127 : 96].b32)',
MTBUFOp.TBUFFER_LOAD_FORMAT_D16_X: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\n// VDATA[31:16].b16 is preserved.",
MTBUFOp.TBUFFER_LOAD_FORMAT_D16_XY: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\nVDATA[31 : 16].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetY()]))",
MTBUFOp.TBUFFER_LOAD_FORMAT_D16_XYZ: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\nVDATA[31 : 16].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetY()]));\nVDATA[47 : 32].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetZ()]));\n// VDATA[63:48].b16 is preserved.",
MTBUFOp.TBUFFER_LOAD_FORMAT_D16_XYZW: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\nVDATA[31 : 16].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetY()]));\nVDATA[47 : 32].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetZ()]));\nVDATA[63 : 48].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetW()]))",
MTBUFOp.TBUFFER_STORE_FORMAT_D16_X: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format",
MTBUFOp.TBUFFER_STORE_FORMAT_D16_XY: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(32'B(VDATA[31 : 16].b16))",
MTBUFOp.TBUFFER_STORE_FORMAT_D16_XYZ: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(32'B(VDATA[31 : 16].b16));\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(32'B(VDATA[47 : 32].b16))",
MTBUFOp.TBUFFER_STORE_FORMAT_D16_XYZW: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(32'B(VDATA[31 : 16].b16));\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(32'B(VDATA[47 : 32].b16));\nMEM[addr + ChannelOffsetW()] = ConvertToFormat(32'B(VDATA[63 : 48].b16))",
MUBUFOp.BUFFER_LOAD_FORMAT_X: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format',
MUBUFOp.BUFFER_LOAD_FORMAT_XY: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format\nVDATA[63 : 32].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetY()])',
MUBUFOp.BUFFER_LOAD_FORMAT_XYZ: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format\nVDATA[63 : 32].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetY()]);\nVDATA[95 : 64].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetZ()])',
MUBUFOp.BUFFER_LOAD_FORMAT_XYZW: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetX()]);\n// Mem access size depends on format\nVDATA[63 : 32].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetY()]);\nVDATA[95 : 64].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetZ()]);\nVDATA[127 : 96].b32 = ConvertFromFormat(MEM[addr + ChannelOffsetW()])',
MUBUFOp.BUFFER_STORE_FORMAT_X: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format',
MUBUFOp.BUFFER_STORE_FORMAT_XY: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(VDATA[63 : 32].b32)',
MUBUFOp.BUFFER_STORE_FORMAT_XYZ: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(VDATA[63 : 32].b32);\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(VDATA[95 : 64].b32)',
MUBUFOp.BUFFER_STORE_FORMAT_XYZW: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(VDATA[31 : 0].b32);\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(VDATA[63 : 32].b32);\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(VDATA[95 : 64].b32);\nMEM[addr + ChannelOffsetW()] = ConvertToFormat(VDATA[127 : 96].b32)',
MUBUFOp.BUFFER_LOAD_FORMAT_D16_X: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\n// VDATA[31:16].b16 is preserved.",
MUBUFOp.BUFFER_LOAD_FORMAT_D16_XY: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\nVDATA[31 : 16].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetY()]))",
MUBUFOp.BUFFER_LOAD_FORMAT_D16_XYZ: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\nVDATA[31 : 16].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetY()]));\nVDATA[47 : 32].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetZ()]));\n// VDATA[63:48].b16 is preserved.",
MUBUFOp.BUFFER_LOAD_FORMAT_D16_XYZW: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\nVDATA[31 : 16].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetY()]));\nVDATA[47 : 32].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetZ()]));\nVDATA[63 : 48].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetW()]))",
MUBUFOp.BUFFER_STORE_FORMAT_D16_X: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format",
MUBUFOp.BUFFER_STORE_FORMAT_D16_XY: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(32'B(VDATA[31 : 16].b16))",
MUBUFOp.BUFFER_STORE_FORMAT_D16_XYZ: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(32'B(VDATA[31 : 16].b16));\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(32'B(VDATA[47 : 32].b16))",
MUBUFOp.BUFFER_STORE_FORMAT_D16_XYZW: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[15 : 0].b16));\n// Mem access size depends on format\nMEM[addr + ChannelOffsetY()] = ConvertToFormat(32'B(VDATA[31 : 16].b16));\nMEM[addr + ChannelOffsetZ()] = ConvertToFormat(32'B(VDATA[47 : 32].b16));\nMEM[addr + ChannelOffsetW()] = ConvertToFormat(32'B(VDATA[63 : 48].b16))",
MUBUFOp.BUFFER_LOAD_UBYTE: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 24'0U, MEM[addr].u8 })",
MUBUFOp.BUFFER_LOAD_SBYTE: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i8))",
MUBUFOp.BUFFER_LOAD_USHORT: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 16'0U, MEM[addr].u16 })",
MUBUFOp.BUFFER_LOAD_SSHORT: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i16))",
MUBUFOp.BUFFER_LOAD_DWORD: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32',
MUBUFOp.BUFFER_LOAD_DWORDX2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32',
MUBUFOp.BUFFER_LOAD_DWORDX3: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32',
MUBUFOp.BUFFER_LOAD_DWORDX4: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32;\nVDATA[127 : 96] = MEM[addr + 12U].b32',
MUBUFOp.BUFFER_STORE_BYTE: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[7 : 0]',
MUBUFOp.BUFFER_STORE_BYTE_D16_HI: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[23 : 16]',
MUBUFOp.BUFFER_STORE_SHORT: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[15 : 0]',
MUBUFOp.BUFFER_STORE_SHORT_D16_HI: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[31 : 16]',
MUBUFOp.BUFFER_STORE_DWORD: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0]',
MUBUFOp.BUFFER_STORE_DWORDX2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32]',
MUBUFOp.BUFFER_STORE_DWORDX3: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64]',
MUBUFOp.BUFFER_STORE_DWORDX4: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64];\nMEM[addr + 12U].b32 = VDATA[127 : 96]',
MUBUFOp.BUFFER_LOAD_UBYTE_D16: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[31:16] is preserved.",
MUBUFOp.BUFFER_LOAD_UBYTE_D16_HI: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 16].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[15:0] is preserved.",
MUBUFOp.BUFFER_LOAD_SBYTE_D16: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[31:16] is preserved.",
MUBUFOp.BUFFER_LOAD_SBYTE_D16_HI: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 16].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[15:0] is preserved.",
MUBUFOp.BUFFER_LOAD_SHORT_D16: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = MEM[addr].b16;\n// VDATA[31:16] is preserved.',
MUBUFOp.BUFFER_LOAD_SHORT_D16_HI: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 16].b16 = MEM[addr].b16;\n// VDATA[15:0] is preserved.',
MUBUFOp.BUFFER_LOAD_FORMAT_D16_HI_X: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nVDATA[31 : 16].b16 = 16'B(ConvertFromFormat(MEM[addr + ChannelOffsetX()]));\n// Mem access size depends on format\n// VDATA[15:0].b16 is preserved.",
MUBUFOp.BUFFER_STORE_FORMAT_D16_HI_X: "addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\nMEM[addr + ChannelOffsetX()] = ConvertToFormat(32'B(VDATA[31 : 16].b16));\n// Mem access size depends on format",
MUBUFOp.BUFFER_ATOMIC_SWAP: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = DATA.b32;\nRETURN_DATA.b32 = tmp',
MUBUFOp.BUFFER_ATOMIC_CMPSWAP: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA[31 : 0].u32;\ncmp = DATA[63 : 32].u32;\nMEM[addr].u32 = tmp == cmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
MUBUFOp.BUFFER_ATOMIC_ADD: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 += DATA.u32;\nRETURN_DATA.u32 = tmp',
MUBUFOp.BUFFER_ATOMIC_SUB: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 -= DATA.u32;\nRETURN_DATA.u32 = tmp',
MUBUFOp.BUFFER_ATOMIC_SMIN: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src < tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
MUBUFOp.BUFFER_ATOMIC_UMIN: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src < tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
MUBUFOp.BUFFER_ATOMIC_SMAX: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src >= tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
MUBUFOp.BUFFER_ATOMIC_UMAX: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src >= tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
MUBUFOp.BUFFER_ATOMIC_AND: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp & DATA.b32);\nRETURN_DATA.b32 = tmp',
MUBUFOp.BUFFER_ATOMIC_OR: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp | DATA.b32);\nRETURN_DATA.b32 = tmp',
MUBUFOp.BUFFER_ATOMIC_XOR: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp ^ DATA.b32);\nRETURN_DATA.b32 = tmp',
MUBUFOp.BUFFER_ATOMIC_INC: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = tmp >= src ? 0U : tmp + 1U;\nRETURN_DATA.u32 = tmp',
MUBUFOp.BUFFER_ATOMIC_DEC: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = ((tmp == 0U) || (tmp > src)) ? src : tmp - 1U;\nRETURN_DATA.u32 = tmp',
MUBUFOp.BUFFER_ATOMIC_ADD_F32: 'tmp = MEM[ADDR].f32;\nMEM[ADDR].f32 += DATA.f32;\nRETURN_DATA = tmp',
MUBUFOp.BUFFER_ATOMIC_PK_ADD_F16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].f16 = tmp[31 : 16].f16 + src[31 : 16].f16;\ndst[15 : 0].f16 = tmp[15 : 0].f16 + src[15 : 0].f16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
MUBUFOp.BUFFER_ATOMIC_ADD_F64: 'tmp = MEM[ADDR].f64;\nMEM[ADDR].f64 += DATA.f64;\nRETURN_DATA = tmp',
MUBUFOp.BUFFER_ATOMIC_MIN_F64: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src < tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
MUBUFOp.BUFFER_ATOMIC_MAX_F64: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].f64;\nsrc = DATA.f64;\nMEM[addr].f64 = src > tmp ? src : tmp;\nRETURN_DATA.f64 = tmp',
MUBUFOp.BUFFER_ATOMIC_PK_ADD_BF16: 'tmp = MEM[ADDR];\nsrc = DATA;\ndst[31 : 16].bf16 = tmp[31 : 16].bf16 + src[31 : 16].bf16;\ndst[15 : 0].bf16 = tmp[15 : 0].bf16 + src[15 : 0].bf16;\nMEM[ADDR] = dst.b32;\nRETURN_DATA = tmp',
MUBUFOp.BUFFER_ATOMIC_SWAP_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = DATA.b64;\nRETURN_DATA.b64 = tmp',
MUBUFOp.BUFFER_ATOMIC_CMPSWAP_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA[63 : 0].u64;\ncmp = DATA[127 : 64].u64;\nMEM[addr].u64 = tmp == cmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
MUBUFOp.BUFFER_ATOMIC_ADD_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 += DATA.u64;\nRETURN_DATA.u64 = tmp',
MUBUFOp.BUFFER_ATOMIC_SUB_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 -= DATA.u64;\nRETURN_DATA.u64 = tmp',
MUBUFOp.BUFFER_ATOMIC_SMIN_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src < tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
MUBUFOp.BUFFER_ATOMIC_UMIN_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src < tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
MUBUFOp.BUFFER_ATOMIC_SMAX_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src >= tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
MUBUFOp.BUFFER_ATOMIC_UMAX_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src >= tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
MUBUFOp.BUFFER_ATOMIC_AND_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp & DATA.b64);\nRETURN_DATA.b64 = tmp',
MUBUFOp.BUFFER_ATOMIC_OR_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp | DATA.b64);\nRETURN_DATA.b64 = tmp',
MUBUFOp.BUFFER_ATOMIC_XOR_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp ^ DATA.b64);\nRETURN_DATA.b64 = tmp',
MUBUFOp.BUFFER_ATOMIC_INC_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = tmp >= src ? 0ULL : tmp + 1ULL;\nRETURN_DATA.u64 = tmp',
MUBUFOp.BUFFER_ATOMIC_DEC_X2: 'addr = CalcBufferAddr(VADDR.b32, SRSRC.b32, SOFFSET.b32, OFFSET.b32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = ((tmp == 0ULL) || (tmp > src)) ? src : tmp - 1ULL;\nRETURN_DATA.u64 = tmp',
SCRATCHOp.SCRATCH_LOAD_UBYTE: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 24'0U, MEM[addr].u8 })",
SCRATCHOp.SCRATCH_LOAD_SBYTE: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i8))",
SCRATCHOp.SCRATCH_LOAD_USHORT: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.u32 = 32'U({ 16'0U, MEM[addr].u16 })",
SCRATCHOp.SCRATCH_LOAD_SSHORT: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA.i32 = 32'I(signext(MEM[addr].i16))",
SCRATCHOp.SCRATCH_LOAD_DWORD: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32',
SCRATCHOp.SCRATCH_LOAD_DWORDX2: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32',
SCRATCHOp.SCRATCH_LOAD_DWORDX3: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32',
SCRATCHOp.SCRATCH_LOAD_DWORDX4: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 0] = MEM[addr].b32;\nVDATA[63 : 32] = MEM[addr + 4U].b32;\nVDATA[95 : 64] = MEM[addr + 8U].b32;\nVDATA[127 : 96] = MEM[addr + 12U].b32',
SCRATCHOp.SCRATCH_STORE_BYTE: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[7 : 0]',
SCRATCHOp.SCRATCH_STORE_BYTE_D16_HI: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b8 = VDATA[23 : 16]',
SCRATCHOp.SCRATCH_STORE_SHORT: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[15 : 0]',
SCRATCHOp.SCRATCH_STORE_SHORT_D16_HI: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b16 = VDATA[31 : 16]',
SCRATCHOp.SCRATCH_STORE_DWORD: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0]',
SCRATCHOp.SCRATCH_STORE_DWORDX2: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32]',
SCRATCHOp.SCRATCH_STORE_DWORDX3: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64]',
SCRATCHOp.SCRATCH_STORE_DWORDX4: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nMEM[addr].b32 = VDATA[31 : 0];\nMEM[addr + 4U].b32 = VDATA[63 : 32];\nMEM[addr + 8U].b32 = VDATA[95 : 64];\nMEM[addr + 12U].b32 = VDATA[127 : 96]',
SCRATCHOp.SCRATCH_LOAD_UBYTE_D16: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[15 : 0].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[31:16] is preserved.",
SCRATCHOp.SCRATCH_LOAD_UBYTE_D16_HI: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 16].u16 = 16'U({ 8'0U, MEM[addr].u8 });\n// VDATA[15:0] is preserved.",
SCRATCHOp.SCRATCH_LOAD_SBYTE_D16: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[15 : 0].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[31:16] is preserved.",
SCRATCHOp.SCRATCH_LOAD_SBYTE_D16_HI: "addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 16].i16 = 16'I(signext(MEM[addr].i8));\n// VDATA[15:0] is preserved.",
SCRATCHOp.SCRATCH_LOAD_SHORT_D16: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[15 : 0].b16 = MEM[addr].b16;\n// VDATA[31:16] is preserved.',
SCRATCHOp.SCRATCH_LOAD_SHORT_D16_HI: 'addr = CalcScratchAddr(ADDR.b32, SADDR.b32, OFFSET.b32);\nVDATA[31 : 16].b16 = MEM[addr].b16;\n// VDATA[15:0] is preserved.',
SMEMOp.S_LOAD_DWORD: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32',
SMEMOp.S_LOAD_DWORDX2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32',
SMEMOp.S_LOAD_DWORDX4: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32;\nSDATA[95 : 64] = MEM[addr + 8U].b32;\nSDATA[127 : 96] = MEM[addr + 12U].b32',
SMEMOp.S_LOAD_DWORDX8: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32;\nSDATA[95 : 64] = MEM[addr + 8U].b32;\nSDATA[127 : 96] = MEM[addr + 12U].b32;\nSDATA[159 : 128] = MEM[addr + 16U].b32;\nSDATA[191 : 160] = MEM[addr + 20U].b32;\nSDATA[223 : 192] = MEM[addr + 24U].b32;\nSDATA[255 : 224] = MEM[addr + 28U].b32',
SMEMOp.S_LOAD_DWORDX16: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32;\nSDATA[95 : 64] = MEM[addr + 8U].b32;\nSDATA[127 : 96] = MEM[addr + 12U].b32;\nSDATA[159 : 128] = MEM[addr + 16U].b32;\nSDATA[191 : 160] = MEM[addr + 20U].b32;\nSDATA[223 : 192] = MEM[addr + 24U].b32;\nSDATA[255 : 224] = MEM[addr + 28U].b32;\nSDATA[287 : 256] = MEM[addr + 32U].b32;\nSDATA[319 : 288] = MEM[addr + 36U].b32;\nSDATA[351 : 320] = MEM[addr + 40U].b32;\nSDATA[383 : 352] = MEM[addr + 44U].b32;\nSDATA[415 : 384] = MEM[addr + 48U].b32;\nSDATA[447 : 416] = MEM[addr + 52U].b32;\nSDATA[479 : 448] = MEM[addr + 56U].b32;\nSDATA[511 : 480] = MEM[addr + 60U].b32',
SMEMOp.S_SCRATCH_LOAD_DWORD: 'addr = CalcScalarScratchAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32',
SMEMOp.S_SCRATCH_LOAD_DWORDX2: 'addr = CalcScalarScratchAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32',
SMEMOp.S_SCRATCH_LOAD_DWORDX4: 'addr = CalcScalarScratchAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32;\nSDATA[95 : 64] = MEM[addr + 8U].b32;\nSDATA[127 : 96] = MEM[addr + 12U].b32',
SMEMOp.S_BUFFER_LOAD_DWORD: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32',
SMEMOp.S_BUFFER_LOAD_DWORDX2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32',
SMEMOp.S_BUFFER_LOAD_DWORDX4: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32;\nSDATA[95 : 64] = MEM[addr + 8U].b32;\nSDATA[127 : 96] = MEM[addr + 12U].b32',
SMEMOp.S_BUFFER_LOAD_DWORDX8: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32;\nSDATA[95 : 64] = MEM[addr + 8U].b32;\nSDATA[127 : 96] = MEM[addr + 12U].b32;\nSDATA[159 : 128] = MEM[addr + 16U].b32;\nSDATA[191 : 160] = MEM[addr + 20U].b32;\nSDATA[223 : 192] = MEM[addr + 24U].b32;\nSDATA[255 : 224] = MEM[addr + 28U].b32',
SMEMOp.S_BUFFER_LOAD_DWORDX16: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nSDATA[31 : 0] = MEM[addr].b32;\nSDATA[63 : 32] = MEM[addr + 4U].b32;\nSDATA[95 : 64] = MEM[addr + 8U].b32;\nSDATA[127 : 96] = MEM[addr + 12U].b32;\nSDATA[159 : 128] = MEM[addr + 16U].b32;\nSDATA[191 : 160] = MEM[addr + 20U].b32;\nSDATA[223 : 192] = MEM[addr + 24U].b32;\nSDATA[255 : 224] = MEM[addr + 28U].b32;\nSDATA[287 : 256] = MEM[addr + 32U].b32;\nSDATA[319 : 288] = MEM[addr + 36U].b32;\nSDATA[351 : 320] = MEM[addr + 40U].b32;\nSDATA[383 : 352] = MEM[addr + 44U].b32;\nSDATA[415 : 384] = MEM[addr + 48U].b32;\nSDATA[447 : 416] = MEM[addr + 52U].b32;\nSDATA[479 : 448] = MEM[addr + 56U].b32;\nSDATA[511 : 480] = MEM[addr + 60U].b32',
SMEMOp.S_STORE_DWORD: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0]',
SMEMOp.S_STORE_DWORDX2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0];\nMEM[addr + 4U].b32 = SDATA[63 : 32]',
SMEMOp.S_STORE_DWORDX4: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0];\nMEM[addr + 4U].b32 = SDATA[63 : 32];\nMEM[addr + 8U].b32 = SDATA[95 : 64];\nMEM[addr + 12U].b32 = SDATA[127 : 96]',
SMEMOp.S_SCRATCH_STORE_DWORD: 'addr = CalcScalarScratchAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0]',
SMEMOp.S_SCRATCH_STORE_DWORDX2: 'addr = CalcScalarScratchAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0];\nMEM[addr + 4U].b32 = SDATA[63 : 32]',
SMEMOp.S_SCRATCH_STORE_DWORDX4: 'addr = CalcScalarScratchAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0];\nMEM[addr + 4U].b32 = SDATA[63 : 32];\nMEM[addr + 8U].b32 = SDATA[95 : 64];\nMEM[addr + 12U].b32 = SDATA[127 : 96]',
SMEMOp.S_BUFFER_STORE_DWORD: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0]',
SMEMOp.S_BUFFER_STORE_DWORDX2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0];\nMEM[addr + 4U].b32 = SDATA[63 : 32]',
SMEMOp.S_BUFFER_STORE_DWORDX4: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\nMEM[addr].b32 = SDATA[31 : 0];\nMEM[addr + 4U].b32 = SDATA[63 : 32];\nMEM[addr + 8U].b32 = SDATA[95 : 64];\nMEM[addr + 12U].b32 = SDATA[127 : 96]',
SMEMOp.S_BUFFER_ATOMIC_SWAP: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = DATA.b32;\nRETURN_DATA.b32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_CMPSWAP: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA[31 : 0].u32;\ncmp = DATA[63 : 32].u32;\nMEM[addr].u32 = tmp == cmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_ADD: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 += DATA.u32;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_SUB: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 -= DATA.u32;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_SMIN: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src < tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_UMIN: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src < tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_SMAX: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src >= tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_UMAX: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src >= tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_AND: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp & DATA.b32);\nRETURN_DATA.b32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_OR: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp | DATA.b32);\nRETURN_DATA.b32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_XOR: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp ^ DATA.b32);\nRETURN_DATA.b32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_INC: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = tmp >= src ? 0U : tmp + 1U;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_DEC: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = ((tmp == 0U) || (tmp > src)) ? src : tmp - 1U;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_BUFFER_ATOMIC_SWAP_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = DATA.b64;\nRETURN_DATA.b64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_CMPSWAP_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA[63 : 0].u64;\ncmp = DATA[127 : 64].u64;\nMEM[addr].u64 = tmp == cmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_ADD_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 += DATA.u64;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_SUB_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 -= DATA.u64;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_SMIN_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src < tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_UMIN_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src < tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_SMAX_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src >= tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_UMAX_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src >= tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_AND_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp & DATA.b64);\nRETURN_DATA.b64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_OR_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp | DATA.b64);\nRETURN_DATA.b64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_XOR_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp ^ DATA.b64);\nRETURN_DATA.b64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_INC_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = tmp >= src ? 0ULL : tmp + 1ULL;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_BUFFER_ATOMIC_DEC_X2: 'addr = CalcScalarBufferAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = ((tmp == 0ULL) || (tmp > src)) ? src : tmp - 1ULL;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_ATOMIC_SWAP: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = DATA.b32;\nRETURN_DATA.b32 = tmp',
SMEMOp.S_ATOMIC_CMPSWAP: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA[31 : 0].u32;\ncmp = DATA[63 : 32].u32;\nMEM[addr].u32 = tmp == cmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_ATOMIC_ADD: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 += DATA.u32;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_ATOMIC_SUB: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nMEM[addr].u32 -= DATA.u32;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_ATOMIC_SMIN: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src < tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
SMEMOp.S_ATOMIC_UMIN: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src < tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_ATOMIC_SMAX: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i32;\nsrc = DATA.i32;\nMEM[addr].i32 = src >= tmp ? src : tmp;\nRETURN_DATA.i32 = tmp',
SMEMOp.S_ATOMIC_UMAX: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = src >= tmp ? src : tmp;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_ATOMIC_AND: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp & DATA.b32);\nRETURN_DATA.b32 = tmp',
SMEMOp.S_ATOMIC_OR: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp | DATA.b32);\nRETURN_DATA.b32 = tmp',
SMEMOp.S_ATOMIC_XOR: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b32;\nMEM[addr].b32 = (tmp ^ DATA.b32);\nRETURN_DATA.b32 = tmp',
SMEMOp.S_ATOMIC_INC: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = tmp >= src ? 0U : tmp + 1U;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_ATOMIC_DEC: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u32;\nsrc = DATA.u32;\nMEM[addr].u32 = ((tmp == 0U) || (tmp > src)) ? src : tmp - 1U;\nRETURN_DATA.u32 = tmp',
SMEMOp.S_ATOMIC_SWAP_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = DATA.b64;\nRETURN_DATA.b64 = tmp',
SMEMOp.S_ATOMIC_CMPSWAP_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA[63 : 0].u64;\ncmp = DATA[127 : 64].u64;\nMEM[addr].u64 = tmp == cmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_ATOMIC_ADD_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 += DATA.u64;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_ATOMIC_SUB_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nMEM[addr].u64 -= DATA.u64;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_ATOMIC_SMIN_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src < tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
SMEMOp.S_ATOMIC_UMIN_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src < tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_ATOMIC_SMAX_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].i64;\nsrc = DATA.i64;\nMEM[addr].i64 = src >= tmp ? src : tmp;\nRETURN_DATA.i64 = tmp',
SMEMOp.S_ATOMIC_UMAX_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = src >= tmp ? src : tmp;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_ATOMIC_AND_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp & DATA.b64);\nRETURN_DATA.b64 = tmp',
SMEMOp.S_ATOMIC_OR_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp | DATA.b64);\nRETURN_DATA.b64 = tmp',
SMEMOp.S_ATOMIC_XOR_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].b64;\nMEM[addr].b64 = (tmp ^ DATA.b64);\nRETURN_DATA.b64 = tmp',
SMEMOp.S_ATOMIC_INC_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = tmp >= src ? 0ULL : tmp + 1ULL;\nRETURN_DATA.u64 = tmp',
SMEMOp.S_ATOMIC_DEC_X2: 'addr = CalcScalarGlobalAddr(SBASE.b32, SOFFSET.b32, OFFSET.i32);\ntmp = MEM[addr].u64;\nsrc = DATA.u64;\nMEM[addr].u64 = ((tmp == 0ULL) || (tmp > src)) ? src : tmp - 1ULL;\nRETURN_DATA.u64 = tmp',
SOP1Op.S_MOV_B32: 'D0.b32 = S0.b32',
SOP1Op.S_MOV_B64: 'D0.b64 = S0.b64',
SOP1Op.S_CMOV_B32: 'if SCC then\nD0.b32 = S0.b32\nendif',
SOP1Op.S_CMOV_B64: 'if SCC then\nD0.b64 = S0.b64\nendif',
SOP1Op.S_NOT_B32: 'D0.u32 = ~S0.u32;\nSCC = D0.u32 != 0U',
SOP1Op.S_NOT_B64: 'D0.u64 = ~S0.u64;\nSCC = D0.u64 != 0ULL',
SOP1Op.S_WQM_B32: "tmp = 0U;\ndeclare i : 6'U;\nfor i in 6'0U : 6'31U do\ntmp[i] = S0.u32[i & 6'60U +: 6'4U] != 0U\nendfor;\nD0.u32 = tmp;\nSCC = D0.u32 != 0U",
SOP1Op.S_WQM_B64: "tmp = 0ULL;\ndeclare i : 6'U;\nfor i in 6'0U : 6'63U do\ntmp[i] = S0.u64[i & 6'60U +: 6'4U] != 0ULL\nendfor;\nD0.u64 = tmp;\nSCC = D0.u64 != 0ULL",
SOP1Op.S_BREV_B32: 'D0.u32[31 : 0] = S0.u32[0 : 31]',
SOP1Op.S_BREV_B64: 'D0.u64[63 : 0] = S0.u64[0 : 63]',
SOP1Op.S_BCNT0_I32_B32: "tmp = 0;\nfor i in 0 : 31 do\ntmp += S0.u32[i] == 1'0U ? 1 : 0\nendfor;\nD0.i32 = tmp;\nSCC = D0.u32 != 0U",
SOP1Op.S_BCNT0_I32_B64: "tmp = 0;\nfor i in 0 : 63 do\ntmp += S0.u64[i] == 1'0U ? 1 : 0\nendfor;\nD0.i32 = tmp;\nSCC = D0.u64 != 0ULL",
SOP1Op.S_BCNT1_I32_B32: "tmp = 0;\nfor i in 0 : 31 do\ntmp += S0.u32[i] == 1'1U ? 1 : 0\nendfor;\nD0.i32 = tmp;\nSCC = D0.u32 != 0U",
SOP1Op.S_BCNT1_I32_B64: "tmp = 0;\nfor i in 0 : 63 do\ntmp += S0.u64[i] == 1'1U ? 1 : 0\nendfor;\nD0.i32 = tmp;\nSCC = D0.u64 != 0ULL",
SOP1Op.S_FF0_I32_B32: "tmp = -1;\n// Set if no zeros are found\nfor i in 0 : 31 do\n// Search from LSB\nif S0.u32[i] == 1'0U then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp",
SOP1Op.S_FF0_I32_B64: "tmp = -1;\n// Set if no zeros are found\nfor i in 0 : 63 do\n// Search from LSB\nif S0.u64[i] == 1'0U then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp",
SOP1Op.S_FF1_I32_B32: "tmp = -1;\n// Set if no ones are found\nfor i in 0 : 31 do\n// Search from LSB\nif S0.u32[i] == 1'1U then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp",
SOP1Op.S_FF1_I32_B64: "tmp = -1;\n// Set if no ones are found\nfor i in 0 : 63 do\n// Search from LSB\nif S0.u64[i] == 1'1U then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp",
SOP1Op.S_FLBIT_I32_B32: "tmp = -1;\n// Set if no ones are found\nfor i in 0 : 31 do\n// Search from MSB\nif S0.u32[31 - i] == 1'1U then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp",
SOP1Op.S_FLBIT_I32_B64: "tmp = -1;\n// Set if no ones are found\nfor i in 0 : 63 do\n// Search from MSB\nif S0.u64[63 - i] == 1'1U then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp",
SOP1Op.S_FLBIT_I32: 'tmp = -1;\n// Set if all bits are the same\nfor i in 1 : 31 do\n// Search from MSB\nif S0.u32[31 - i] != S0.u32[31] then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp',
SOP1Op.S_FLBIT_I32_I64: 'tmp = -1;\n// Set if all bits are the same\nfor i in 1 : 63 do\n// Search from MSB\nif S0.u64[63 - i] != S0.u64[63] then\ntmp = i;\nbreak\nendif\nendfor;\nD0.i32 = tmp',
SOP1Op.S_SEXT_I32_I8: "D0.i32 = 32'I(signext(S0.i8))",
SOP1Op.S_SEXT_I32_I16: "D0.i32 = 32'I(signext(S0.i16))",
SOP1Op.S_BITSET0_B32: "D0.u32[S0.u32[4 : 0]] = 1'0U",
SOP1Op.S_BITSET0_B64: "D0.u64[S0.u32[5 : 0]] = 1'0U",
SOP1Op.S_BITSET1_B32: "D0.u32[S0.u32[4 : 0]] = 1'1U",
SOP1Op.S_BITSET1_B64: "D0.u64[S0.u32[5 : 0]] = 1'1U",
SOP1Op.S_GETPC_B64: 'D0.i64 = PC + 4LL',
SOP1Op.S_SETPC_B64: 'PC = S0.i64',
SOP1Op.S_SWAPPC_B64: 'jump_addr = S0.i64;\nD0.i64 = PC + 4LL;\nPC = jump_addr.i64',
SOP1Op.S_RFE_B64: "WAVE_STATUS.PRIV = 1'0U;\nPC = S0.i64",
SOP1Op.S_AND_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = (S0.u64 & EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_OR_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = (S0.u64 | EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_XOR_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = (S0.u64 ^ EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_ANDN2_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = (S0.u64 & ~EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_ORN2_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = (S0.u64 | ~EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_NAND_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = ~(S0.u64 & EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_NOR_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = ~(S0.u64 | EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_XNOR_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = ~(S0.u64 ^ EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_QUADMASK_B32: 'tmp = 0U;\nfor i in 0 : 7 do\ntmp[i] = S0.u32[i * 4 +: 4] != 0U\nendfor;\nD0.u32 = tmp;\nSCC = D0.u32 != 0U',
SOP1Op.S_QUADMASK_B64: 'tmp = 0ULL;\nfor i in 0 : 15 do\ntmp[i] = S0.u64[i * 4 +: 4] != 0ULL\nendfor;\nD0.u64 = tmp;\nSCC = D0.u64 != 0ULL',
SOP1Op.S_MOVRELS_B32: 'addr = SRC0.u32;\n// Raw value from instruction\naddr += M0.u32[31 : 0];\nD0.b32 = SGPR[addr].b32',
SOP1Op.S_MOVRELS_B64: 'addr = SRC0.u32;\n// Raw value from instruction\naddr += M0.u32[31 : 0];\nD0.b64 = SGPR[addr].b64',
SOP1Op.S_MOVRELD_B32: 'addr = DST.u32;\n// Raw value from instruction\naddr += M0.u32[31 : 0];\nSGPR[addr].b32 = S0.b32',
SOP1Op.S_MOVRELD_B64: 'addr = DST.u32;\n// Raw value from instruction\naddr += M0.u32[31 : 0];\nSGPR[addr].b64 = S0.b64',
SOP1Op.S_CBRANCH_JOIN: "saved_csp = S0.u32;\nif WAVE_MODE.CSP.u32 == saved_csp then\nPC += 4LL;\n// Second time to JOIN: continue with program.\nelse\nWAVE_MODE.CSP -= 3'1U;\n// First time to JOIN; jump to other FORK path.\n{ PC, EXEC } = SGPR[WAVE_MODE.CSP.u32 * 4U].b128;\n// Read 128 bits from 4 consecutive SGPRs.\nendif",
SOP1Op.S_ABS_I32: 'D0.i32 = S0.i32 < 0 ? -S0.i32 : S0.i32;\nSCC = D0.i32 != 0',
SOP1Op.S_SET_GPR_IDX_IDX: 'M0[7 : 0] = S0.u32[7 : 0].b8',
SOP1Op.S_ANDN1_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = (~S0.u64 & EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_ORN1_SAVEEXEC_B64: 'saveexec = EXEC.u64;\nEXEC.u64 = (~S0.u64 | EXEC.u64);\nD0.u64 = saveexec.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_ANDN1_WREXEC_B64: 'EXEC.u64 = (~S0.u64 & EXEC.u64);\nD0.u64 = EXEC.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_ANDN2_WREXEC_B64: 'EXEC.u64 = (S0.u64 & ~EXEC.u64);\nD0.u64 = EXEC.u64;\nSCC = EXEC.u64 != 0ULL',
SOP1Op.S_BITREPLICATE_B64_B32: 'tmp = S0.u32;\nfor i in 0 : 31 do\nD0.u64[i * 2] = tmp[i];\nD0.u64[i * 2 + 1] = tmp[i]\nendfor',
SOP2Op.S_ADD_U32: "tmp = 64'U(S0.u32) + 64'U(S1.u32);\nSCC = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// unsigned overflow or carry-out for S_ADDC_U32.\nD0.u32 = tmp.u32",
SOP2Op.S_SUB_U32: "tmp = S0.u32 - S1.u32;\nSCC = S1.u32 > S0.u32 ? 1'1U : 1'0U;\n// unsigned overflow or carry-out for S_SUBB_U32.\nD0.u32 = tmp.u32",
SOP2Op.S_ADD_I32: 'tmp = S0.i32 + S1.i32;\nSCC = ((S0.u32[31] == S1.u32[31]) && (S0.u32[31] != tmp.u32[31]));\n// signed overflow.\nD0.i32 = tmp.i32',
SOP2Op.S_SUB_I32: 'tmp = S0.i32 - S1.i32;\nSCC = ((S0.u32[31] != S1.u32[31]) && (S0.u32[31] != tmp.u32[31]));\n// signed overflow.\nD0.i32 = tmp.i32',
SOP2Op.S_ADDC_U32: "tmp = 64'U(S0.u32) + 64'U(S1.u32) + SCC.u64;\nSCC = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// unsigned overflow or carry-out for S_ADDC_U32.\nD0.u32 = tmp.u32",
SOP2Op.S_SUBB_U32: "tmp = S0.u32 - S1.u32 - SCC.u32;\nSCC = 64'U(S1.u32) + SCC.u64 > 64'U(S0.u32) ? 1'1U : 1'0U;\n// unsigned overflow or carry-out for S_SUBB_U32.\nD0.u32 = tmp.u32",
SOP2Op.S_MIN_I32: 'SCC = S0.i32 < S1.i32;\nD0.i32 = SCC ? S0.i32 : S1.i32',
SOP2Op.S_MIN_U32: 'SCC = S0.u32 < S1.u32;\nD0.u32 = SCC ? S0.u32 : S1.u32',
SOP2Op.S_MAX_I32: 'SCC = S0.i32 >= S1.i32;\nD0.i32 = SCC ? S0.i32 : S1.i32',
SOP2Op.S_MAX_U32: 'SCC = S0.u32 >= S1.u32;\nD0.u32 = SCC ? S0.u32 : S1.u32',
SOP2Op.S_CSELECT_B32: 'D0.u32 = SCC ? S0.u32 : S1.u32',
SOP2Op.S_CSELECT_B64: 'D0.u64 = SCC ? S0.u64 : S1.u64',
SOP2Op.S_AND_B32: 'D0.u32 = (S0.u32 & S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_AND_B64: 'D0.u64 = (S0.u64 & S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_OR_B32: 'D0.u32 = (S0.u32 | S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_OR_B64: 'D0.u64 = (S0.u64 | S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_XOR_B32: 'D0.u32 = (S0.u32 ^ S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_XOR_B64: 'D0.u64 = (S0.u64 ^ S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_ANDN2_B32: 'D0.u32 = (S0.u32 & ~S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_ANDN2_B64: 'D0.u64 = (S0.u64 & ~S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_ORN2_B32: 'D0.u32 = (S0.u32 | ~S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_ORN2_B64: 'D0.u64 = (S0.u64 | ~S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_NAND_B32: 'D0.u32 = ~(S0.u32 & S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_NAND_B64: 'D0.u64 = ~(S0.u64 & S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_NOR_B32: 'D0.u32 = ~(S0.u32 | S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_NOR_B64: 'D0.u64 = ~(S0.u64 | S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_XNOR_B32: 'D0.u32 = ~(S0.u32 ^ S1.u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_XNOR_B64: 'D0.u64 = ~(S0.u64 ^ S1.u64);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_LSHL_B32: 'D0.u32 = (S0.u32 << S1[4 : 0].u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_LSHL_B64: 'D0.u64 = (S0.u64 << S1[5 : 0].u32);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_LSHR_B32: 'D0.u32 = (S0.u32 >> S1[4 : 0].u32);\nSCC = D0.u32 != 0U',
SOP2Op.S_LSHR_B64: 'D0.u64 = (S0.u64 >> S1[5 : 0].u32);\nSCC = D0.u64 != 0ULL',
SOP2Op.S_ASHR_I32: "D0.i32 = 32'I(signext(S0.i32) >> S1[4 : 0].u32);\nSCC = D0.i32 != 0",
SOP2Op.S_ASHR_I64: 'D0.i64 = (signext(S0.i64) >> S1[5 : 0].u32);\nSCC = D0.i64 != 0LL',
SOP2Op.S_BFM_B32: 'D0.u32 = (((1U << S0[4 : 0].u32) - 1U) << S1[4 : 0].u32)',
SOP2Op.S_BFM_B64: 'D0.u64 = (((1ULL << S0[5 : 0].u32) - 1ULL) << S1[5 : 0].u32)',
SOP2Op.S_MUL_I32: 'D0.i32 = S0.i32 * S1.i32',
SOP2Op.S_BFE_U32: 'D0.u32 = ((S0.u32 >> S1[4 : 0].u32) & ((1U << S1[22 : 16].u32) - 1U));\nSCC = D0.u32 != 0U',
SOP2Op.S_BFE_I32: 'tmp.i32 = ((S0.i32 >> S1[4 : 0].u32) & ((1 << S1[22 : 16].u32) - 1));\nD0.i32 = signext_from_bit(tmp.i32, S1[22 : 16].u32);\nSCC = D0.i32 != 0',
SOP2Op.S_BFE_U64: 'D0.u64 = ((S0.u64 >> S1[5 : 0].u32) & ((1ULL << S1[22 : 16].u32) - 1ULL));\nSCC = D0.u64 != 0ULL',
SOP2Op.S_BFE_I64: 'tmp.i64 = ((S0.i64 >> S1[5 : 0].u32) & ((1LL << S1[22 : 16].u32) - 1LL));\nD0.i64 = signext_from_bit(tmp.i64, S1[22 : 16].u32);\nSCC = D0.i64 != 0LL',
SOP2Op.S_CBRANCH_G_FORK: "mask_pass = (S0.u64 & EXEC.u64);\nmask_fail = (~S0.u64 & EXEC.u64);\nif mask_pass == EXEC.u64 then\nPC = 64'I(S1.u64)\nelsif mask_fail == EXEC.u64 then\nPC += 4LL\nelsif bitCount(mask_fail.b64) < bitCount(mask_pass.b64) then\nEXEC = mask_fail.b64;\nSGPR[WAVE_MODE.CSP.u32 * 4U].b128 = { S1.u64, mask_pass };\nWAVE_MODE.CSP += 3'1U;\nPC += 4LL\nelse\nEXEC = mask_pass.b64;\nSGPR[WAVE_MODE.CSP.u32 * 4U].b128 = { (PC + 4LL), mask_fail };\nWAVE_MODE.CSP += 3'1U;\nPC = 64'I(S1.u64)\nendif",
SOP2Op.S_ABSDIFF_I32: 'D0.i32 = S0.i32 - S1.i32;\nif D0.i32 < 0 then\nD0.i32 = -D0.i32\nendif;\nSCC = D0.i32 != 0',
SOP2Op.S_MUL_HI_U32: "D0.u32 = 32'U((64'U(S0.u32) * 64'U(S1.u32)) >> 32U)",
SOP2Op.S_MUL_HI_I32: "D0.i32 = 32'I((64'I(S0.i32) * 64'I(S1.i32)) >> 32U)",
SOP2Op.S_LSHL1_ADD_U32: "tmp = (64'U(S0.u32) << 1U) + 64'U(S1.u32);\nSCC = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// unsigned overflow.\nD0.u32 = tmp.u32",
SOP2Op.S_LSHL2_ADD_U32: "tmp = (64'U(S0.u32) << 2U) + 64'U(S1.u32);\nSCC = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// unsigned overflow.\nD0.u32 = tmp.u32",
SOP2Op.S_LSHL3_ADD_U32: "tmp = (64'U(S0.u32) << 3U) + 64'U(S1.u32);\nSCC = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// unsigned overflow.\nD0.u32 = tmp.u32",
SOP2Op.S_LSHL4_ADD_U32: "tmp = (64'U(S0.u32) << 4U) + 64'U(S1.u32);\nSCC = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// unsigned overflow.\nD0.u32 = tmp.u32",
SOP2Op.S_PACK_LL_B32_B16: 'D0 = { S1[15 : 0].u16, S0[15 : 0].u16 }',
SOP2Op.S_PACK_LH_B32_B16: 'D0 = { S1[31 : 16].u16, S0[15 : 0].u16 }',
SOP2Op.S_PACK_HH_B32_B16: 'D0 = { S1[31 : 16].u16, S0[31 : 16].u16 }',
SOPCOp.S_CMP_EQ_I32: 'SCC = S0.i32 == S1.i32',
SOPCOp.S_CMP_LG_I32: 'SCC = S0.i32 <> S1.i32',
SOPCOp.S_CMP_GT_I32: 'SCC = S0.i32 > S1.i32',
SOPCOp.S_CMP_GE_I32: 'SCC = S0.i32 >= S1.i32',
SOPCOp.S_CMP_LT_I32: 'SCC = S0.i32 < S1.i32',
SOPCOp.S_CMP_LE_I32: 'SCC = S0.i32 <= S1.i32',
SOPCOp.S_CMP_EQ_U32: 'SCC = S0.u32 == S1.u32',
SOPCOp.S_CMP_LG_U32: 'SCC = S0.u32 <> S1.u32',
SOPCOp.S_CMP_GT_U32: 'SCC = S0.u32 > S1.u32',
SOPCOp.S_CMP_GE_U32: 'SCC = S0.u32 >= S1.u32',
SOPCOp.S_CMP_LT_U32: 'SCC = S0.u32 < S1.u32',
SOPCOp.S_CMP_LE_U32: 'SCC = S0.u32 <= S1.u32',
SOPCOp.S_BITCMP0_B32: "SCC = S0.u32[S1.u32[4 : 0]] == 1'0U",
SOPCOp.S_BITCMP1_B32: "SCC = S0.u32[S1.u32[4 : 0]] == 1'1U",
SOPCOp.S_BITCMP0_B64: "SCC = S0.u64[S1.u32[5 : 0]] == 1'0U",
SOPCOp.S_BITCMP1_B64: "SCC = S0.u64[S1.u32[5 : 0]] == 1'1U",
SOPCOp.S_SETVSKIP: 'VSKIP = S0.u32[S1.u32[4 : 0]]',
SOPCOp.S_SET_GPR_IDX_ON: "WAVE_MODE.GPR_IDX_EN = 1'1U;\nM0[7 : 0] = S0.u32[7 : 0].b8;\nM0[15 : 12] = SRC1.u32[3 : 0].b4;\n// this is the direct content of raw S1 field\n// Remaining bits of M0 are unmodified.",
SOPCOp.S_CMP_EQ_U64: 'SCC = S0.u64 == S1.u64',
SOPCOp.S_CMP_LG_U64: 'SCC = S0.u64 <> S1.u64',
SOPKOp.S_MOVK_I32: "D0.i32 = 32'I(signext(S0.i16))",
SOPKOp.S_CMOVK_I32: "if SCC then\nD0.i32 = 32'I(signext(S0.i16))\nendif",
SOPKOp.S_CMPK_EQ_I32: "SCC = S0.i32 == 32'I(signext(S1.i16))",
SOPKOp.S_CMPK_LG_I32: "SCC = S0.i32 != 32'I(signext(S1.i16))",
SOPKOp.S_CMPK_GT_I32: "SCC = S0.i32 > 32'I(signext(S1.i16))",
SOPKOp.S_CMPK_GE_I32: "SCC = S0.i32 >= 32'I(signext(S1.i16))",
SOPKOp.S_CMPK_LT_I32: "SCC = S0.i32 < 32'I(signext(S1.i16))",
SOPKOp.S_CMPK_LE_I32: "SCC = S0.i32 <= 32'I(signext(S1.i16))",
SOPKOp.S_CMPK_EQ_U32: "SCC = S0.u32 == 32'U(S1.u16)",
SOPKOp.S_CMPK_LG_U32: "SCC = S0.u32 != 32'U(S1.u16)",
SOPKOp.S_CMPK_GT_U32: "SCC = S0.u32 > 32'U(S1.u16)",
SOPKOp.S_CMPK_GE_U32: "SCC = S0.u32 >= 32'U(S1.u16)",
SOPKOp.S_CMPK_LT_U32: "SCC = S0.u32 < 32'U(S1.u16)",
SOPKOp.S_CMPK_LE_U32: "SCC = S0.u32 <= 32'U(S1.u16)",
SOPKOp.S_ADDK_I32: "tmp = D0.i32;\n// Save value to check sign bits for overflow later.\nD0.i32 = D0.i32 + 32'I(signext(S0.i16));\nSCC = ((tmp[31] == S0.i16[15]) && (tmp[31] != D0.i32[31]));\n// signed overflow.",
SOPKOp.S_MULK_I32: "D0.i32 = D0.i32 * 32'I(signext(S0.i16))",
SOPKOp.S_CBRANCH_I_FORK: "// Initial setup.\nmask_pass = (S0.u64 & EXEC.u64);\nmask_fail = (~S0.u64 & EXEC.u64);\ntarget_addr = PC + signext(SIMM16.i32 * 4) + 4LL;\n// Decide where to jump to.\nif mask_pass == EXEC.u64 then\nPC = target_addr\nelsif mask_fail == EXEC.u64 then\nPC += 4LL\nelsif bitCount(mask_fail.b64) < bitCount(mask_pass.b64) then\nEXEC = mask_fail.b64;\nSGPR[WAVE_MODE.CSP.u32 * 4U].b128 = { target_addr, mask_pass };\nWAVE_MODE.CSP += 3'1U;\nPC += 4LL\nelse\nEXEC = mask_pass.b64;\nSGPR[WAVE_MODE.CSP.u32 * 4U].b128 = { (PC + 4LL), mask_fail };\nWAVE_MODE.CSP += 3'1U;\nPC = target_addr\nendif",
SOPKOp.S_GETREG_B32: "hwRegId = SIMM16.u16[5 : 0];\noffset = SIMM16.u16[10 : 6];\nsize = SIMM16.u16[15 : 11].u32 + 1U;\n// logical size is in range 1:32\nvalue = HW_REGISTERS[hwRegId];\nD0.u32 = 32'U(32'I(value >> offset.u32) & ((1 << size) - 1))",
SOPKOp.S_SETREG_B32: "hwRegId = SIMM16.u16[5 : 0];\noffset = SIMM16.u16[10 : 6];\nsize = SIMM16.u16[15 : 11].u32 + 1U;\n// logical size is in range 1:32\nmask = (1 << size) - 1;\nmask = (mask << offset.u32);\nmask = (mask & HwRegWriteMask(hwRegId, WAVE_STATUS.PRIV));\n// Mask of bits that can be modified\nvalue = ((S0.u32 << offset.u32) & mask.u32);\nvalue = (value | 32'U(HW_REGISTERS[hwRegId].i32 & ~mask));\nHW_REGISTERS[hwRegId] = value.b32;\n// Side-effects may trigger here if certain bits are modified",
SOPKOp.S_SETREG_IMM32_B32: "hwRegId = SIMM16.u16[5 : 0];\noffset = SIMM16.u16[10 : 6];\nsize = SIMM16.u16[15 : 11].u32 + 1U;\n// logical size is in range 1:32\nmask = (1 << size) - 1;\nmask = (mask << offset.u32);\nmask = (mask & HwRegWriteMask(hwRegId, WAVE_STATUS.PRIV));\n// Mask of bits that can be modified\nvalue = ((SIMM32.u32 << offset.u32) & mask.u32);\nvalue = (value | 32'U(HW_REGISTERS[hwRegId].i32 & ~mask));\nHW_REGISTERS[hwRegId] = value.b32;\n// Side-effects may trigger here if certain bits are modified",
SOPKOp.S_CALL_B64: "D0.i64 = PC + 4LL;\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL",
SOPPOp.S_NOP: 'for i in 0U : SIMM16.u16[3 : 0].u32 do\nnop()\nendfor',
SOPPOp.S_BRANCH: "PC = PC + signext(SIMM16.i16 * 16'4) + 4LL;\n// short jump.",
SOPPOp.S_CBRANCH_SCC0: "if SCC == 1'0U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_SCC1: "if SCC == 1'1U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_VCCZ: "if VCCZ.u1 == 1'1U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_VCCNZ: "if VCCZ.u1 == 1'0U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_EXECZ: "if EXECZ.u1 == 1'1U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_EXECNZ: "if EXECZ.u1 == 1'0U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_WAITCNT: 'SIMM16[3:0] = vmcount (vector memory operations) lower bits [3:0],',
SOPPOp.S_SLEEP: 's_sleep 0 // Wait for 0 clocks.\ns_sleep 1 // Wait for 1-64 clocks.\ns_sleep 2 // Wait for 65-128 clocks.',
SOPPOp.S_TRAP: 'TrapID = SIMM16.u16[7 : 0];\n"Wait for all instructions to complete";\n// PC passed into trap handler points to S_TRAP itself,\n// *not* to the next instruction.\n{ TTMP[1], TTMP[0] } = { 3\'0, PCRewind[3 : 0], HT[0], TrapID[7 : 0], PC[47 : 0] };\nPC = TBA.i64;\n// trap base address\nWAVE_STATUS.PRIV = 1\'1U',
SOPPOp.S_CBRANCH_CDBGSYS: "if WAVE_STATUS.COND_DBG_SYS.u32 != 0U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_CDBGUSER: "if WAVE_STATUS.COND_DBG_USER.u32 != 0U then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_CDBGSYS_OR_USER: "if (WAVE_STATUS.COND_DBG_SYS || WAVE_STATUS.COND_DBG_USER) then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_CBRANCH_CDBGSYS_AND_USER: "if (WAVE_STATUS.COND_DBG_SYS && WAVE_STATUS.COND_DBG_USER) then\nPC = PC + signext(SIMM16.i16 * 16'4) + 4LL\nelse\nPC = PC + 4LL\nendif",
SOPPOp.S_SET_GPR_IDX_OFF: "WAVE_MODE.GPR_IDX_EN = 1'0U",
SOPPOp.S_SET_GPR_IDX_MODE: 'M0[15 : 12] = SIMM16.u16[3 : 0].b4',
VOP1Op.V_MOV_B32_E32: 'D0.b32 = S0.b32',
VOP1Op.V_READFIRSTLANE_B32_E32: "declare lane : 32'I;\nif EXEC == 0x0LL then\nlane = 0;\n// Force lane 0 if all lanes are disabled\nelse\nlane = s_ff1_i32_b64(EXEC);\n// Lowest active lane\nendif;\nD0.b32 = VGPR[lane][SRC0.u32]",
VOP1Op.V_CVT_I32_F64_E32: 'D0.i32 = f64_to_i32(S0.f64)',
VOP1Op.V_CVT_F64_I32_E32: 'D0.f64 = i32_to_f64(S0.i32)',
VOP1Op.V_CVT_F32_I32_E32: 'D0.f32 = i32_to_f32(S0.i32)',
VOP1Op.V_CVT_F32_U32_E32: 'D0.f32 = u32_to_f32(S0.u32)',
VOP1Op.V_CVT_U32_F32_E32: 'D0.u32 = f32_to_u32(S0.f32)',
VOP1Op.V_CVT_I32_F32_E32: 'D0.i32 = f32_to_i32(S0.f32)',
VOP1Op.V_CVT_F16_F32_E32: 'D0.f16 = f32_to_f16(S0.f32)',
VOP1Op.V_CVT_F32_F16_E32: 'D0.f32 = f16_to_f32(S0.f16)',
VOP1Op.V_CVT_RPI_I32_F32_E32: 'D0.i32 = f32_to_i32(floor(S0.f32 + 0.5F))',
VOP1Op.V_CVT_FLR_I32_F32_E32: 'D0.i32 = f32_to_i32(floor(S0.f32))',
VOP1Op.V_CVT_OFF_F32_I4_E32: "declare CVT_OFF_TABLE : 32'F[16];\nD0.f32 = CVT_OFF_TABLE[S0.u32[3 : 0]]",
VOP1Op.V_CVT_F32_F64_E32: 'D0.f32 = f64_to_f32(S0.f64)',
VOP1Op.V_CVT_F64_F32_E32: 'D0.f64 = f32_to_f64(S0.f32)',
VOP1Op.V_CVT_F32_UBYTE0_E32: 'D0.f32 = u32_to_f32(S0[7 : 0].u32)',
VOP1Op.V_CVT_F32_UBYTE1_E32: 'D0.f32 = u32_to_f32(S0[15 : 8].u32)',
VOP1Op.V_CVT_F32_UBYTE2_E32: 'D0.f32 = u32_to_f32(S0[23 : 16].u32)',
VOP1Op.V_CVT_F32_UBYTE3_E32: 'D0.f32 = u32_to_f32(S0[31 : 24].u32)',
VOP1Op.V_CVT_U32_F64_E32: 'D0.u32 = f64_to_u32(S0.f64)',
VOP1Op.V_CVT_F64_U32_E32: 'D0.f64 = u32_to_f64(S0.u32)',
VOP1Op.V_TRUNC_F64_E32: 'D0.f64 = trunc(S0.f64)',
VOP1Op.V_CEIL_F64_E32: 'D0.f64 = trunc(S0.f64);\nif ((S0.f64 > 0.0) && (S0.f64 != D0.f64)) then\nD0.f64 += 1.0\nendif',
VOP1Op.V_RNDNE_F64_E32: 'D0.f64 = floor(S0.f64 + 0.5);\nif (isEven(floor(S0.f64)) && (fract(S0.f64) == 0.5)) then\nD0.f64 -= 1.0\nendif',
VOP1Op.V_FLOOR_F64_E32: 'D0.f64 = trunc(S0.f64);\nif ((S0.f64 < 0.0) && (S0.f64 != D0.f64)) then\nD0.f64 += -1.0\nendif',
VOP1Op.V_FRACT_F32_E32: 'D0.f32 = S0.f32 + -floor(S0.f32)',
VOP1Op.V_TRUNC_F32_E32: 'D0.f32 = trunc(S0.f32)',
VOP1Op.V_CEIL_F32_E32: 'D0.f32 = trunc(S0.f32);\nif ((S0.f32 > 0.0F) && (S0.f32 != D0.f32)) then\nD0.f32 += 1.0F\nendif',
VOP1Op.V_RNDNE_F32_E32: "D0.f32 = floor(S0.f32 + 0.5F);\nif (isEven(64'F(floor(S0.f32))) && (fract(S0.f32) == 0.5F)) then\nD0.f32 -= 1.0F\nendif",
VOP1Op.V_FLOOR_F32_E32: 'D0.f32 = trunc(S0.f32);\nif ((S0.f32 < 0.0F) && (S0.f32 != D0.f32)) then\nD0.f32 += -1.0F\nendif',
VOP1Op.V_EXP_F32_E32: 'D0.f32 = pow(2.0F, S0.f32)',
VOP1Op.V_LOG_F32_E32: 'D0.f32 = log2(S0.f32)',
VOP1Op.V_RCP_F32_E32: 'D0.f32 = 1.0F / S0.f32',
VOP1Op.V_RCP_IFLAG_F32_E32: 'D0.f32 = 1.0F / S0.f32;\n// Can only raise integer DIV_BY_ZERO exception',
VOP1Op.V_RSQ_F32_E32: 'D0.f32 = 1.0F / sqrt(S0.f32)',
VOP1Op.V_RCP_F64_E32: 'D0.f64 = 1.0 / S0.f64',
VOP1Op.V_RSQ_F64_E32: 'D0.f64 = 1.0 / sqrt(S0.f64)',
VOP1Op.V_SQRT_F32_E32: 'D0.f32 = sqrt(S0.f32)',
VOP1Op.V_SQRT_F64_E32: 'D0.f64 = sqrt(S0.f64)',
VOP1Op.V_SIN_F32_E32: "D0.f32 = sin(S0.f32 * 32'F(PI * 2.0))",
VOP1Op.V_COS_F32_E32: "D0.f32 = cos(S0.f32 * 32'F(PI * 2.0))",
VOP1Op.V_NOT_B32_E32: 'D0.u32 = ~S0.u32',
VOP1Op.V_BFREV_B32_E32: 'D0.u32[31 : 0] = S0.u32[0 : 31]',
VOP1Op.V_FFBH_U32_E32: "D0.i32 = -1;\n// Set if no ones are found\nfor i in 0 : 31 do\n// Search from MSB\nif S0.u32[31 - i] == 1'1U then\nD0.i32 = i;\nbreak\nendif\nendfor",
VOP1Op.V_FFBL_B32_E32: "D0.i32 = -1;\n// Set if no ones are found\nfor i in 0 : 31 do\n// Search from LSB\nif S0.u32[i] == 1'1U then\nD0.i32 = i;\nbreak\nendif\nendfor",
VOP1Op.V_FFBH_I32_E32: 'D0.i32 = -1;\n// Set if all bits are the same\nfor i in 1 : 31 do\n// Search from MSB\nif S0.i32[31 - i] != S0.i32[31] then\nD0.i32 = i;\nbreak\nendif\nendfor',
VOP1Op.V_FREXP_EXP_I32_F64_E32: 'if ((S0.f64 == +INF) || (S0.f64 == -INF) || isNAN(S0.f64)) then\nD0.i32 = 0\nelse\nD0.i32 = exponent(S0.f64) - 1023 + 1\nendif',
VOP1Op.V_FREXP_MANT_F64_E32: 'if ((S0.f64 == +INF) || (S0.f64 == -INF) || isNAN(S0.f64)) then\nD0.f64 = S0.f64\nelse\nD0.f64 = mantissa(S0.f64)\nendif',
VOP1Op.V_FRACT_F64_E32: 'D0.f64 = S0.f64 + -floor(S0.f64)',
VOP1Op.V_FREXP_EXP_I32_F32_E32: "if ((64'F(S0.f32) == +INF) || (64'F(S0.f32) == -INF) || isNAN(64'F(S0.f32))) then\nD0.i32 = 0\nelse\nD0.i32 = exponent(S0.f32) - 127 + 1\nendif",
VOP1Op.V_FREXP_MANT_F32_E32: "if ((64'F(S0.f32) == +INF) || (64'F(S0.f32) == -INF) || isNAN(64'F(S0.f32))) then\nD0.f32 = S0.f32\nelse\nD0.f32 = mantissa(S0.f32)\nendif",
VOP1Op.V_MOV_B64_E32: 'D0.b64 = S0.b64',
VOP1Op.V_CVT_F16_U16_E32: 'D0.f16 = u16_to_f16(S0.u16)',
VOP1Op.V_CVT_F16_I16_E32: 'D0.f16 = i16_to_f16(S0.i16)',
VOP1Op.V_CVT_U16_F16_E32: 'D0.u16 = f16_to_u16(S0.f16)',
VOP1Op.V_CVT_I16_F16_E32: 'D0.i16 = f16_to_i16(S0.f16)',
VOP1Op.V_RCP_F16_E32: "D0.f16 = 16'1.0 / S0.f16",
VOP1Op.V_SQRT_F16_E32: 'D0.f16 = sqrt(S0.f16)',
VOP1Op.V_RSQ_F16_E32: "D0.f16 = 16'1.0 / sqrt(S0.f16)",
VOP1Op.V_LOG_F16_E32: 'D0.f16 = log2(S0.f16)',
VOP1Op.V_EXP_F16_E32: "D0.f16 = pow(16'2.0, S0.f16)",
VOP1Op.V_FREXP_MANT_F16_E32: "if ((64'F(S0.f16) == +INF) || (64'F(S0.f16) == -INF) || isNAN(64'F(S0.f16))) then\nD0.f16 = S0.f16\nelse\nD0.f16 = mantissa(S0.f16)\nendif",
VOP1Op.V_FREXP_EXP_I16_F16_E32: "if ((64'F(S0.f16) == +INF) || (64'F(S0.f16) == -INF) || isNAN(64'F(S0.f16))) then\nD0.i16 = 16'0\nelse\nD0.i16 = 16'I(exponent(S0.f16) - 15 + 1)\nendif",
VOP1Op.V_FLOOR_F16_E32: "D0.f16 = trunc(S0.f16);\nif ((S0.f16 < 16'0.0) && (S0.f16 != D0.f16)) then\nD0.f16 += -16'1.0\nendif",
VOP1Op.V_CEIL_F16_E32: "D0.f16 = trunc(S0.f16);\nif ((S0.f16 > 16'0.0) && (S0.f16 != D0.f16)) then\nD0.f16 += 16'1.0\nendif",
VOP1Op.V_TRUNC_F16_E32: 'D0.f16 = trunc(S0.f16)',
VOP1Op.V_RNDNE_F16_E32: "D0.f16 = floor(S0.f16 + 16'0.5);\nif (isEven(64'F(floor(S0.f16))) && (fract(S0.f16) == 16'0.5)) then\nD0.f16 -= 16'1.0\nendif",
VOP1Op.V_FRACT_F16_E32: 'D0.f16 = S0.f16 + -floor(S0.f16)',
VOP1Op.V_SIN_F16_E32: "D0.f16 = sin(S0.f16 * 16'F(PI * 2.0))",
VOP1Op.V_COS_F16_E32: "D0.f16 = cos(S0.f16 * 16'F(PI * 2.0))",
VOP1Op.V_CVT_NORM_I16_F16_E32: 'D0.i16 = f16_to_snorm(S0.f16)',
VOP1Op.V_CVT_NORM_U16_F16_E32: 'D0.u16 = f16_to_unorm(S0.f16)',
VOP1Op.V_SAT_PK_U8_I16_E32: "SAT8 = lambda(n) (\nif n <= 16'0 then\nreturn 8'0U\nelsif n >= 16'255 then\nreturn 8'255U\nelse\nreturn n[7 : 0].u8\nendif);\ntmp = 16'0;\ntmp[7 : 0].u8 = SAT8(S0[15 : 0].i16);\ntmp[15 : 8].u8 = SAT8(S0[31 : 16].i16);\nD0.b16 = tmp.b16",
VOP1Op.V_SWAP_B32_E32: 'tmp = D0.b32;\nD0.b32 = S0.b32;\nS0.b32 = tmp',
VOP1Op.V_CVT_F32_FP8_E32: 'if SDWA_SRC0_SEL == BYTE1.b3 then\nD0.f32 = fp8_to_f32(S0[15 : 8].fp8)\nelsif SDWA_SRC0_SEL == BYTE2.b3 then\nD0.f32 = fp8_to_f32(S0[23 : 16].fp8)\nelsif SDWA_SRC0_SEL == BYTE3.b3 then\nD0.f32 = fp8_to_f32(S0[31 : 24].fp8)\nelse\n// BYTE0 implied\nD0.f32 = fp8_to_f32(S0[7 : 0].fp8)\nendif',
VOP1Op.V_CVT_F32_BF8_E32: 'if SDWA_SRC0_SEL == BYTE1.b3 then\nD0.f32 = bf8_to_f32(S0[15 : 8].bf8)\nelsif SDWA_SRC0_SEL == BYTE2.b3 then\nD0.f32 = bf8_to_f32(S0[23 : 16].bf8)\nelsif SDWA_SRC0_SEL == BYTE3.b3 then\nD0.f32 = bf8_to_f32(S0[31 : 24].bf8)\nelse\n// BYTE0 implied\nD0.f32 = bf8_to_f32(S0[7 : 0].bf8)\nendif',
VOP1Op.V_CVT_PK_F32_FP8_E32: 'tmp = SDWA_SRC0_SEL[1 : 0] == WORD1.b2 ? S0[31 : 16] : S0[15 : 0];\nD0[31 : 0].f32 = fp8_to_f32(tmp[7 : 0].fp8);\nD0[63 : 32].f32 = fp8_to_f32(tmp[15 : 8].fp8)',
VOP1Op.V_CVT_PK_F32_BF8_E32: 'tmp = SDWA_SRC0_SEL[1 : 0] == WORD1.b2 ? S0[31 : 16] : S0[15 : 0];\nD0[31 : 0].f32 = bf8_to_f32(tmp[7 : 0].bf8);\nD0[63 : 32].f32 = bf8_to_f32(tmp[15 : 8].bf8)',
VOP1Op.V_PRNG_B32_E32: 'in = S0.u32;\nD0.u32 = ((in << 1U) ^ (in[31] ? 197U : 0U))',
VOP1Op.V_PERMLANE16_SWAP_B32_E32: 'for pass in 0 : 1 do\nfor lane in 0 : 15 do\ntmp = VGPR[pass * 32 + lane][SRC0.u32];\nVGPR[pass * 32 + lane][SRC0.u32] = VGPR[pass * 32 + lane + 16][VDST.u32];\nVGPR[pass * 32 + lane + 16][VDST.u32] = tmp\nendfor\nendfor',
VOP1Op.V_PERMLANE32_SWAP_B32_E32: 'for lane in 0 : 31 do\ntmp = VGPR[lane][SRC0.u32];\nVGPR[lane][SRC0.u32] = VGPR[lane + 32][VDST.u32];\nVGPR[lane + 32][VDST.u32] = tmp\nendfor',
VOP1Op.V_CVT_F32_BF16_E32: "D0.f32 = 32'F({ S0.b16, 16'0U })",
VOP2Op.V_CNDMASK_B32_E32: 'D0.u32 = VCC.u64[laneId] ? S1.u32 : S0.u32',
VOP2Op.V_ADD_F32_E32: 'D0.f32 = S0.f32 + S1.f32',
VOP2Op.V_SUB_F32_E32: 'D0.f32 = S0.f32 - S1.f32',
VOP2Op.V_SUBREV_F32_E32: 'D0.f32 = S1.f32 - S0.f32',
VOP2Op.V_FMAC_F64_E32: 'D0.f64 = fma(S0.f64, S1.f64, D0.f64)',
VOP2Op.V_MUL_F32_E32: 'D0.f32 = S0.f32 * S1.f32',
VOP2Op.V_MUL_I32_I24_E32: "D0.i32 = 32'I(S0.i24) * 32'I(S1.i24)",
VOP2Op.V_MUL_HI_I32_I24_E32: "D0.i32 = 32'I((64'I(S0.i24) * 64'I(S1.i24)) >> 32U)",
VOP2Op.V_MUL_U32_U24_E32: "D0.u32 = 32'U(S0.u24) * 32'U(S1.u24)",
VOP2Op.V_MUL_HI_U32_U24_E32: "D0.u32 = 32'U((64'U(S0.u24) * 64'U(S1.u24)) >> 32U)",
VOP2Op.V_MIN_F32_E32: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S0.f32)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S1.f32)))\nelsif isNAN(64'F(S0.f32)) then\nD0.f32 = S1.f32\nelsif isNAN(64'F(S1.f32)) then\nD0.f32 = S0.f32\nelsif ((64'F(S0.f32) == +0.0) && (64'F(S1.f32) == -0.0)) then\nD0.f32 = S1.f32\nelsif ((64'F(S0.f32) == -0.0) && (64'F(S1.f32) == +0.0)) then\nD0.f32 = S0.f32\nelse\n// Note: there's no IEEE case here like there is for V_MAX_F32.\nD0.f32 = S0.f32 < S1.f32 ? S0.f32 : S1.f32\nendif",
VOP2Op.V_MAX_F32_E32: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S0.f32)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S1.f32)))\nelsif isNAN(64'F(S0.f32)) then\nD0.f32 = S1.f32\nelsif isNAN(64'F(S1.f32)) then\nD0.f32 = S0.f32\nelsif ((64'F(S0.f32) == +0.0) && (64'F(S1.f32) == -0.0)) then\nD0.f32 = S0.f32\nelsif ((64'F(S0.f32) == -0.0) && (64'F(S1.f32) == +0.0)) then\nD0.f32 = S1.f32\nelsif WAVE_MODE.IEEE then\nD0.f32 = S0.f32 >= S1.f32 ? S0.f32 : S1.f32\nelse\nD0.f32 = S0.f32 > S1.f32 ? S0.f32 : S1.f32\nendif",
VOP2Op.V_MIN_I32_E32: 'D0.i32 = S0.i32 < S1.i32 ? S0.i32 : S1.i32',
VOP2Op.V_MAX_I32_E32: 'D0.i32 = S0.i32 >= S1.i32 ? S0.i32 : S1.i32',
VOP2Op.V_MIN_U32_E32: 'D0.u32 = S0.u32 < S1.u32 ? S0.u32 : S1.u32',
VOP2Op.V_MAX_U32_E32: 'D0.u32 = S0.u32 >= S1.u32 ? S0.u32 : S1.u32',
VOP2Op.V_LSHRREV_B32_E32: 'D0.u32 = (S1.u32 >> S0[4 : 0].u32)',
VOP2Op.V_ASHRREV_I32_E32: 'D0.i32 = (S1.i32 >> S0[4 : 0].u32)',
VOP2Op.V_LSHLREV_B32_E32: 'D0.u32 = (S1.u32 << S0[4 : 0].u32)',
VOP2Op.V_AND_B32_E32: 'D0.u32 = (S0.u32 & S1.u32)',
VOP2Op.V_OR_B32_E32: 'D0.u32 = (S0.u32 | S1.u32)',
VOP2Op.V_XOR_B32_E32: 'D0.u32 = (S0.u32 ^ S1.u32)',
VOP2Op.V_DOT2C_F32_BF16_E32: 'tmp = D0.f32;\ntmp += bf16_to_f32(S0[15 : 0].bf16) * bf16_to_f32(S1[15 : 0].bf16);\ntmp += bf16_to_f32(S0[31 : 16].bf16) * bf16_to_f32(S1[31 : 16].bf16);\nD0.f32 = tmp',
VOP2Op.V_FMAMK_F32_E32: 'D0.f32 = fma(S0.f32, SIMM32.f32, S1.f32)',
VOP2Op.V_FMAAK_F32_E32: 'D0.f32 = fma(S0.f32, S1.f32, SIMM32.f32)',
VOP2Op.V_ADD_CO_U32_E32: "tmp = 64'U(S0.u32) + 64'U(S1.u32);\nVCC.u64[laneId] = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_ADDC_CO_U32.\nD0.u32 = tmp.u32",
VOP2Op.V_SUB_CO_U32_E32: "tmp = S0.u32 - S1.u32;\nVCC.u64[laneId] = S1.u32 > S0.u32 ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP2Op.V_SUBREV_CO_U32_E32: "tmp = S1.u32 - S0.u32;\nVCC.u64[laneId] = S0.u32 > S1.u32 ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP2Op.V_ADDC_CO_U32_E32: "tmp = 64'U(S0.u32) + 64'U(S1.u32) + VCC.u64[laneId].u64;\nVCC.u64[laneId] = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_ADDC_CO_U32.\nD0.u32 = tmp.u32",
VOP2Op.V_SUBB_CO_U32_E32: "tmp = S0.u32 - S1.u32 - VCC.u64[laneId].u32;\nVCC.u64[laneId] = 64'U(S1.u32) + VCC.u64[laneId].u64 > 64'U(S0.u32) ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP2Op.V_SUBBREV_CO_U32_E32: "tmp = S1.u32 - S0.u32 - VCC.u64[laneId].u32;\nVCC.u64[laneId] = 64'U(S0.u32) + VCC.u64[laneId].u64 > 64'U(S1.u32) ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP2Op.V_ADD_F16_E32: 'D0.f16 = S0.f16 + S1.f16',
VOP2Op.V_SUB_F16_E32: 'D0.f16 = S0.f16 - S1.f16',
VOP2Op.V_SUBREV_F16_E32: 'D0.f16 = S1.f16 - S0.f16',
VOP2Op.V_MUL_F16_E32: 'D0.f16 = S0.f16 * S1.f16',
VOP2Op.V_MAC_F16_E32: "tmp = S0.f16 * S1.f16 + D0.f16;\nif OPSEL.u4[3] then\nD0 = { tmp.f16, D0[15 : 0] }\nelse\nD0 = { 16'0, tmp.f16 }\nendif",
VOP2Op.V_MADMK_F16_E32: "tmp = S0.f16 * SIMM16.f16 + S1.f16;\nD0 = { 16'0, tmp.f16 }",
VOP2Op.V_MADAK_F16_E32: "tmp = S0.f16 * S1.f16 + SIMM16.f16;\nD0 = { 16'0, tmp.f16 }",
VOP2Op.V_ADD_U16_E32: 'D0.u16 = S0.u16 + S1.u16',
VOP2Op.V_SUB_U16_E32: 'D0.u16 = S0.u16 - S1.u16',
VOP2Op.V_SUBREV_U16_E32: 'D0.u16 = S1.u16 - S0.u16',
VOP2Op.V_MUL_LO_U16_E32: 'D0.u16 = S0.u16 * S1.u16',
VOP2Op.V_LSHLREV_B16_E32: 'D0.u16 = (S1.u16 << S0[3 : 0].u32)',
VOP2Op.V_LSHRREV_B16_E32: 'D0.u16 = (S1.u16 >> S0[3 : 0].u32)',
VOP2Op.V_ASHRREV_I16_E32: 'D0.i16 = (S1.i16 >> S0[3 : 0].u32)',
VOP2Op.V_MAX_F16_E32: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S0.f16)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S1.f16)))\nelsif isNAN(64'F(S0.f16)) then\nD0.f16 = S1.f16\nelsif isNAN(64'F(S1.f16)) then\nD0.f16 = S0.f16\nelsif ((64'F(S0.f16) == +0.0) && (64'F(S1.f16) == -0.0)) then\nD0.f16 = S0.f16\nelsif ((64'F(S0.f16) == -0.0) && (64'F(S1.f16) == +0.0)) then\nD0.f16 = S1.f16\nelsif WAVE_MODE.IEEE then\nD0.f16 = S0.f16 >= S1.f16 ? S0.f16 : S1.f16\nelse\nD0.f16 = S0.f16 > S1.f16 ? S0.f16 : S1.f16\nendif",
VOP2Op.V_MIN_F16_E32: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S0.f16)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S1.f16)))\nelsif isNAN(64'F(S0.f16)) then\nD0.f16 = S1.f16\nelsif isNAN(64'F(S1.f16)) then\nD0.f16 = S0.f16\nelsif ((64'F(S0.f16) == +0.0) && (64'F(S1.f16) == -0.0)) then\nD0.f16 = S1.f16\nelsif ((64'F(S0.f16) == -0.0) && (64'F(S1.f16) == +0.0)) then\nD0.f16 = S0.f16\nelse\n// Note: there's no IEEE case here like there is for V_MAX_F16.\nD0.f16 = S0.f16 < S1.f16 ? S0.f16 : S1.f16\nendif",
VOP2Op.V_MAX_U16_E32: 'D0.u16 = S0.u16 >= S1.u16 ? S0.u16 : S1.u16',
VOP2Op.V_MAX_I16_E32: 'D0.i16 = S0.i16 >= S1.i16 ? S0.i16 : S1.i16',
VOP2Op.V_MIN_U16_E32: 'D0.u16 = S0.u16 < S1.u16 ? S0.u16 : S1.u16',
VOP2Op.V_MIN_I16_E32: 'D0.i16 = S0.i16 < S1.i16 ? S0.i16 : S1.i16',
VOP2Op.V_LDEXP_F16_E32: "D0.f16 = S0.f16 * 16'F(2.0F ** 32'I(S1.i16))",
VOP2Op.V_ADD_U32_E32: 'D0.u32 = S0.u32 + S1.u32',
VOP2Op.V_SUB_U32_E32: 'D0.u32 = S0.u32 - S1.u32',
VOP2Op.V_SUBREV_U32_E32: 'D0.u32 = S1.u32 - S0.u32',
VOP2Op.V_DOT2C_F32_F16_E32: 'tmp = D0.f32;\ntmp += f16_to_f32(S0[15 : 0].f16) * f16_to_f32(S1[15 : 0].f16);\ntmp += f16_to_f32(S0[31 : 16].f16) * f16_to_f32(S1[31 : 16].f16);\nD0.f32 = tmp',
VOP2Op.V_DOT2C_I32_I16_E32: 'tmp = D0.i32;\ntmp += i16_to_i32(S0[15 : 0].i16) * i16_to_i32(S1[15 : 0].i16);\ntmp += i16_to_i32(S0[31 : 16].i16) * i16_to_i32(S1[31 : 16].i16);\nD0.i32 = tmp',
VOP2Op.V_DOT4C_I32_I8_E32: 'tmp = D0.i32;\ntmp += i8_to_i32(S0[7 : 0].i8) * i8_to_i32(S1[7 : 0].i8);\ntmp += i8_to_i32(S0[15 : 8].i8) * i8_to_i32(S1[15 : 8].i8);\ntmp += i8_to_i32(S0[23 : 16].i8) * i8_to_i32(S1[23 : 16].i8);\ntmp += i8_to_i32(S0[31 : 24].i8) * i8_to_i32(S1[31 : 24].i8);\nD0.i32 = tmp',
VOP2Op.V_DOT8C_I32_I4_E32: 'tmp = D0.i32;\ntmp += i4_to_i32(S0[3 : 0].i4) * i4_to_i32(S1[3 : 0].i4);\ntmp += i4_to_i32(S0[7 : 4].i4) * i4_to_i32(S1[7 : 4].i4);\ntmp += i4_to_i32(S0[11 : 8].i4) * i4_to_i32(S1[11 : 8].i4);\ntmp += i4_to_i32(S0[15 : 12].i4) * i4_to_i32(S1[15 : 12].i4);\ntmp += i4_to_i32(S0[19 : 16].i4) * i4_to_i32(S1[19 : 16].i4);\ntmp += i4_to_i32(S0[23 : 20].i4) * i4_to_i32(S1[23 : 20].i4);\ntmp += i4_to_i32(S0[27 : 24].i4) * i4_to_i32(S1[27 : 24].i4);\ntmp += i4_to_i32(S0[31 : 28].i4) * i4_to_i32(S1[31 : 28].i4);\nD0.i32 = tmp',
VOP2Op.V_FMAC_F32_E32: 'D0.f32 = fma(S0.f32, S1.f32, D0.f32)',
VOP2Op.V_PK_FMAC_F16_E32: 'D0[15 : 0].f16 = fma(S0[15 : 0].f16, S1[15 : 0].f16, D0[15 : 0].f16);\nD0[31 : 16].f16 = fma(S0[31 : 16].f16, S1[31 : 16].f16, D0[31 : 16].f16)',
VOP2Op.V_XNOR_B32_E32: 'D0.u32 = ~(S0.u32 ^ S1.u32)',
VOP3Op.V_CMP_CLASS_F32_E64: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f32)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f32)) then\nresult = S1.u32[1]\nelsif exponent(S0.f32) == 255 then\n// +-INF\nresult = S1.u32[sign(S0.f32) ? 2 : 9]\nelsif exponent(S0.f32) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f32) ? 3 : 8]\nelsif 64'F(abs(S0.f32)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f32) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f32) ? 5 : 6]\nendif;\nD0.u64[laneId] = result;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_CLASS_F32_E64: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f32)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f32)) then\nresult = S1.u32[1]\nelsif exponent(S0.f32) == 255 then\n// +-INF\nresult = S1.u32[sign(S0.f32) ? 2 : 9]\nelsif exponent(S0.f32) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f32) ? 3 : 8]\nelsif 64'F(abs(S0.f32)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f32) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f32) ? 5 : 6]\nendif;\nEXEC.u64[laneId] = D0.u64[laneId] = result",
VOP3Op.V_CMP_CLASS_F64_E64: "declare result : 1'U;\nif isSignalNAN(S0.f64) then\nresult = S1.u32[0]\nelsif isQuietNAN(S0.f64) then\nresult = S1.u32[1]\nelsif exponent(S0.f64) == 2047 then\n// +-INF\nresult = S1.u32[sign(S0.f64) ? 2 : 9]\nelsif exponent(S0.f64) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f64) ? 3 : 8]\nelsif abs(S0.f64) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f64) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f64) ? 5 : 6]\nendif;\nD0.u64[laneId] = result;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_CLASS_F64_E64: "declare result : 1'U;\nif isSignalNAN(S0.f64) then\nresult = S1.u32[0]\nelsif isQuietNAN(S0.f64) then\nresult = S1.u32[1]\nelsif exponent(S0.f64) == 2047 then\n// +-INF\nresult = S1.u32[sign(S0.f64) ? 2 : 9]\nelsif exponent(S0.f64) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f64) ? 3 : 8]\nelsif abs(S0.f64) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f64) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f64) ? 5 : 6]\nendif;\nEXEC.u64[laneId] = D0.u64[laneId] = result",
VOP3Op.V_CMP_CLASS_F16_E64: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f16)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f16)) then\nresult = S1.u32[1]\nelsif exponent(S0.f16) == 31 then\n// +-INF\nresult = S1.u32[sign(S0.f16) ? 2 : 9]\nelsif exponent(S0.f16) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f16) ? 3 : 8]\nelsif 64'F(abs(S0.f16)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f16) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f16) ? 5 : 6]\nendif;\nD0.u64[laneId] = result;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_CLASS_F16_E64: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f16)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f16)) then\nresult = S1.u32[1]\nelsif exponent(S0.f16) == 31 then\n// +-INF\nresult = S1.u32[sign(S0.f16) ? 2 : 9]\nelsif exponent(S0.f16) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f16) ? 3 : 8]\nelsif 64'F(abs(S0.f16)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f16) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f16) ? 5 : 6]\nendif;\nEXEC.u64[laneId] = D0.u64[laneId] = result",
VOP3Op.V_CMP_F_F16_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_F16_E64: 'D0.u64[laneId] = S0.f16 < S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_F16_E64: 'D0.u64[laneId] = S0.f16 == S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_F16_E64: 'D0.u64[laneId] = S0.f16 <= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_F16_E64: 'D0.u64[laneId] = S0.f16 > S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LG_F16_E64: 'D0.u64[laneId] = S0.f16 <> S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_F16_E64: 'D0.u64[laneId] = S0.f16 >= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_O_F16_E64: "D0.u64[laneId] = (!isNAN(64'F(S0.f16)) && !isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_U_F16_E64: "D0.u64[laneId] = (isNAN(64'F(S0.f16)) || isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_NGE_F16_E64: 'D0.u64[laneId] = !(S0.f16 >= S1.f16);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLG_F16_E64: 'D0.u64[laneId] = !(S0.f16 <> S1.f16);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NGT_F16_E64: 'D0.u64[laneId] = !(S0.f16 > S1.f16);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLE_F16_E64: 'D0.u64[laneId] = !(S0.f16 <= S1.f16);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NEQ_F16_E64: 'D0.u64[laneId] = !(S0.f16 == S1.f16);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLT_F16_E64: 'D0.u64[laneId] = !(S0.f16 < S1.f16);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_TRU_F16_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_F16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 < S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 == S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 <= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 > S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LG_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 <> S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 >= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_O_F16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = (!isNAN(64'F(S0.f16)) && !isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_U_F16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = (isNAN(64'F(S0.f16)) || isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_NGE_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 >= S1.f16);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLG_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 <> S1.f16);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NGT_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 > S1.f16);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLE_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 <= S1.f16);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NEQ_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 == S1.f16);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLT_F16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 < S1.f16);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_TRU_F16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_F32_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_F32_E64: 'D0.u64[laneId] = S0.f32 < S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_F32_E64: 'D0.u64[laneId] = S0.f32 == S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_F32_E64: 'D0.u64[laneId] = S0.f32 <= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_F32_E64: 'D0.u64[laneId] = S0.f32 > S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LG_F32_E64: 'D0.u64[laneId] = S0.f32 <> S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_F32_E64: 'D0.u64[laneId] = S0.f32 >= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_O_F32_E64: "D0.u64[laneId] = (!isNAN(64'F(S0.f32)) && !isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_U_F32_E64: "D0.u64[laneId] = (isNAN(64'F(S0.f32)) || isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_NGE_F32_E64: 'D0.u64[laneId] = !(S0.f32 >= S1.f32);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLG_F32_E64: 'D0.u64[laneId] = !(S0.f32 <> S1.f32);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NGT_F32_E64: 'D0.u64[laneId] = !(S0.f32 > S1.f32);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLE_F32_E64: 'D0.u64[laneId] = !(S0.f32 <= S1.f32);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NEQ_F32_E64: 'D0.u64[laneId] = !(S0.f32 == S1.f32);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLT_F32_E64: 'D0.u64[laneId] = !(S0.f32 < S1.f32);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_TRU_F32_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_F32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 < S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 == S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 <= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 > S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LG_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 <> S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 >= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_O_F32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = (!isNAN(64'F(S0.f32)) && !isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_U_F32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = (isNAN(64'F(S0.f32)) || isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_NGE_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 >= S1.f32);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLG_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 <> S1.f32);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NGT_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 > S1.f32);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLE_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 <= S1.f32);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NEQ_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 == S1.f32);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLT_F32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 < S1.f32);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_TRU_F32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_F64_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_F64_E64: 'D0.u64[laneId] = S0.f64 < S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_F64_E64: 'D0.u64[laneId] = S0.f64 == S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_F64_E64: 'D0.u64[laneId] = S0.f64 <= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_F64_E64: 'D0.u64[laneId] = S0.f64 > S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LG_F64_E64: 'D0.u64[laneId] = S0.f64 <> S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_F64_E64: 'D0.u64[laneId] = S0.f64 >= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_O_F64_E64: 'D0.u64[laneId] = (!isNAN(S0.f64) && !isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_U_F64_E64: 'D0.u64[laneId] = (isNAN(S0.f64) || isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NGE_F64_E64: 'D0.u64[laneId] = !(S0.f64 >= S1.f64);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLG_F64_E64: 'D0.u64[laneId] = !(S0.f64 <> S1.f64);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NGT_F64_E64: 'D0.u64[laneId] = !(S0.f64 > S1.f64);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLE_F64_E64: 'D0.u64[laneId] = !(S0.f64 <= S1.f64);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NEQ_F64_E64: 'D0.u64[laneId] = !(S0.f64 == S1.f64);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NLT_F64_E64: 'D0.u64[laneId] = !(S0.f64 < S1.f64);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_TRU_F64_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_F64_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 < S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 == S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 <= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 > S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LG_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 <> S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 >= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_O_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = (!isNAN(S0.f64) && !isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_U_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = (isNAN(S0.f64) || isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NGE_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 >= S1.f64);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLG_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 <> S1.f64);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NGT_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 > S1.f64);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLE_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 <= S1.f64);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NEQ_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 == S1.f64);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NLT_F64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 < S1.f64);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_TRU_F64_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_I16_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_I16_E64: 'D0.u64[laneId] = S0.i16 < S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_I16_E64: 'D0.u64[laneId] = S0.i16 == S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_I16_E64: 'D0.u64[laneId] = S0.i16 <= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_I16_E64: 'D0.u64[laneId] = S0.i16 > S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NE_I16_E64: 'D0.u64[laneId] = S0.i16 <> S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_I16_E64: 'D0.u64[laneId] = S0.i16 >= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_T_I16_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_U16_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_U16_E64: 'D0.u64[laneId] = S0.u16 < S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_U16_E64: 'D0.u64[laneId] = S0.u16 == S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_U16_E64: 'D0.u64[laneId] = S0.u16 <= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_U16_E64: 'D0.u64[laneId] = S0.u16 > S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NE_U16_E64: 'D0.u64[laneId] = S0.u16 <> S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_U16_E64: 'D0.u64[laneId] = S0.u16 >= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_T_U16_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_I16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_I16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 < S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_I16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 == S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_I16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 <= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_I16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 > S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NE_I16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 <> S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_I16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 >= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_T_I16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_U16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_U16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 < S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_U16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 == S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_U16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 <= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_U16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 > S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NE_U16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 <> S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_U16_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 >= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_T_U16_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_I32_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_I32_E64: 'D0.u64[laneId] = S0.i32 < S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_I32_E64: 'D0.u64[laneId] = S0.i32 == S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_I32_E64: 'D0.u64[laneId] = S0.i32 <= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_I32_E64: 'D0.u64[laneId] = S0.i32 > S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NE_I32_E64: 'D0.u64[laneId] = S0.i32 <> S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_I32_E64: 'D0.u64[laneId] = S0.i32 >= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_T_I32_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_U32_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_U32_E64: 'D0.u64[laneId] = S0.u32 < S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_U32_E64: 'D0.u64[laneId] = S0.u32 == S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_U32_E64: 'D0.u64[laneId] = S0.u32 <= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_U32_E64: 'D0.u64[laneId] = S0.u32 > S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NE_U32_E64: 'D0.u64[laneId] = S0.u32 <> S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_U32_E64: 'D0.u64[laneId] = S0.u32 >= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_T_U32_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_I32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_I32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 < S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_I32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 == S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_I32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 <= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_I32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 > S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NE_I32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 <> S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_I32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 >= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_T_I32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_U32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_U32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 < S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_U32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 == S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_U32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 <= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_U32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 > S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NE_U32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 <> S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_U32_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 >= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_T_U32_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_I64_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_I64_E64: 'D0.u64[laneId] = S0.i64 < S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_I64_E64: 'D0.u64[laneId] = S0.i64 == S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_I64_E64: 'D0.u64[laneId] = S0.i64 <= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_I64_E64: 'D0.u64[laneId] = S0.i64 > S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NE_I64_E64: 'D0.u64[laneId] = S0.i64 <> S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_I64_E64: 'D0.u64[laneId] = S0.i64 >= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_T_I64_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_F_U64_E64: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMP_LT_U64_E64: 'D0.u64[laneId] = S0.u64 < S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_EQ_U64_E64: 'D0.u64[laneId] = S0.u64 == S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_LE_U64_E64: 'D0.u64[laneId] = S0.u64 <= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GT_U64_E64: 'D0.u64[laneId] = S0.u64 > S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_NE_U64_E64: 'D0.u64[laneId] = S0.u64 <> S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_GE_U64_E64: 'D0.u64[laneId] = S0.u64 >= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMP_T_U64_E64: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_I64_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_I64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 < S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_I64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 == S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_I64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 <= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_I64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 > S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NE_I64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 <> S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_I64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 >= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_T_I64_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_F_U64_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CMPX_LT_U64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 < S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_EQ_U64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 == S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_LE_U64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 <= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GT_U64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 > S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_NE_U64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 <> S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_GE_U64_E64: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 >= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOP3Op.V_CMPX_T_U64_E64: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOP3Op.V_CNDMASK_B32_E64: 'D0.u32 = VCC.u64[laneId] ? S1.u32 : S0.u32',
VOP3Op.V_ADD_F32_E64: 'D0.f32 = S0.f32 + S1.f32',
VOP3Op.V_SUB_F32_E64: 'D0.f32 = S0.f32 - S1.f32',
VOP3Op.V_SUBREV_F32_E64: 'D0.f32 = S1.f32 - S0.f32',
VOP3Op.V_FMAC_F64_E64: 'D0.f64 = fma(S0.f64, S1.f64, D0.f64)',
VOP3Op.V_MUL_F32_E64: 'D0.f32 = S0.f32 * S1.f32',
VOP3Op.V_MUL_I32_I24_E64: "D0.i32 = 32'I(S0.i24) * 32'I(S1.i24)",
VOP3Op.V_MUL_HI_I32_I24_E64: "D0.i32 = 32'I((64'I(S0.i24) * 64'I(S1.i24)) >> 32U)",
VOP3Op.V_MUL_U32_U24_E64: "D0.u32 = 32'U(S0.u24) * 32'U(S1.u24)",
VOP3Op.V_MUL_HI_U32_U24_E64: "D0.u32 = 32'U((64'U(S0.u24) * 64'U(S1.u24)) >> 32U)",
VOP3Op.V_MIN_F32_E64: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S0.f32)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S1.f32)))\nelsif isNAN(64'F(S0.f32)) then\nD0.f32 = S1.f32\nelsif isNAN(64'F(S1.f32)) then\nD0.f32 = S0.f32\nelsif ((64'F(S0.f32) == +0.0) && (64'F(S1.f32) == -0.0)) then\nD0.f32 = S1.f32\nelsif ((64'F(S0.f32) == -0.0) && (64'F(S1.f32) == +0.0)) then\nD0.f32 = S0.f32\nelse\n// Note: there's no IEEE case here like there is for V_MAX_F32.\nD0.f32 = S0.f32 < S1.f32 ? S0.f32 : S1.f32\nendif",
VOP3Op.V_MAX_F32_E64: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S0.f32)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f32))) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S1.f32)))\nelsif isNAN(64'F(S0.f32)) then\nD0.f32 = S1.f32\nelsif isNAN(64'F(S1.f32)) then\nD0.f32 = S0.f32\nelsif ((64'F(S0.f32) == +0.0) && (64'F(S1.f32) == -0.0)) then\nD0.f32 = S0.f32\nelsif ((64'F(S0.f32) == -0.0) && (64'F(S1.f32) == +0.0)) then\nD0.f32 = S1.f32\nelsif WAVE_MODE.IEEE then\nD0.f32 = S0.f32 >= S1.f32 ? S0.f32 : S1.f32\nelse\nD0.f32 = S0.f32 > S1.f32 ? S0.f32 : S1.f32\nendif",
VOP3Op.V_MIN_I32_E64: 'D0.i32 = S0.i32 < S1.i32 ? S0.i32 : S1.i32',
VOP3Op.V_MAX_I32_E64: 'D0.i32 = S0.i32 >= S1.i32 ? S0.i32 : S1.i32',
VOP3Op.V_MIN_U32_E64: 'D0.u32 = S0.u32 < S1.u32 ? S0.u32 : S1.u32',
VOP3Op.V_MAX_U32_E64: 'D0.u32 = S0.u32 >= S1.u32 ? S0.u32 : S1.u32',
VOP3Op.V_LSHRREV_B32_E64: 'D0.u32 = (S1.u32 >> S0[4 : 0].u32)',
VOP3Op.V_ASHRREV_I32_E64: 'D0.i32 = (S1.i32 >> S0[4 : 0].u32)',
VOP3Op.V_LSHLREV_B32_E64: 'D0.u32 = (S1.u32 << S0[4 : 0].u32)',
VOP3Op.V_AND_B32_E64: 'D0.u32 = (S0.u32 & S1.u32)',
VOP3Op.V_OR_B32_E64: 'D0.u32 = (S0.u32 | S1.u32)',
VOP3Op.V_XOR_B32_E64: 'D0.u32 = (S0.u32 ^ S1.u32)',
VOP3Op.V_DOT2C_F32_BF16_E64: 'tmp = D0.f32;\ntmp += bf16_to_f32(S0[15 : 0].bf16) * bf16_to_f32(S1[15 : 0].bf16);\ntmp += bf16_to_f32(S0[31 : 16].bf16) * bf16_to_f32(S1[31 : 16].bf16);\nD0.f32 = tmp',
VOP3Op.V_ADD_F16_E64: 'D0.f16 = S0.f16 + S1.f16',
VOP3Op.V_SUB_F16_E64: 'D0.f16 = S0.f16 - S1.f16',
VOP3Op.V_SUBREV_F16_E64: 'D0.f16 = S1.f16 - S0.f16',
VOP3Op.V_MUL_F16_E64: 'D0.f16 = S0.f16 * S1.f16',
VOP3Op.V_MAC_F16_E64: "tmp = S0.f16 * S1.f16 + D0.f16;\nif OPSEL.u4[3] then\nD0 = { tmp.f16, D0[15 : 0] }\nelse\nD0 = { 16'0, tmp.f16 }\nendif",
VOP3Op.V_ADD_U16_E64: 'D0.u16 = S0.u16 + S1.u16',
VOP3Op.V_SUB_U16_E64: 'D0.u16 = S0.u16 - S1.u16',
VOP3Op.V_SUBREV_U16_E64: 'D0.u16 = S1.u16 - S0.u16',
VOP3Op.V_MUL_LO_U16_E64: 'D0.u16 = S0.u16 * S1.u16',
VOP3Op.V_LSHLREV_B16_E64: 'D0.u16 = (S1.u16 << S0[3 : 0].u32)',
VOP3Op.V_LSHRREV_B16_E64: 'D0.u16 = (S1.u16 >> S0[3 : 0].u32)',
VOP3Op.V_ASHRREV_I16_E64: 'D0.i16 = (S1.i16 >> S0[3 : 0].u32)',
VOP3Op.V_MAX_F16_E64: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S0.f16)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S1.f16)))\nelsif isNAN(64'F(S0.f16)) then\nD0.f16 = S1.f16\nelsif isNAN(64'F(S1.f16)) then\nD0.f16 = S0.f16\nelsif ((64'F(S0.f16) == +0.0) && (64'F(S1.f16) == -0.0)) then\nD0.f16 = S0.f16\nelsif ((64'F(S0.f16) == -0.0) && (64'F(S1.f16) == +0.0)) then\nD0.f16 = S1.f16\nelsif WAVE_MODE.IEEE then\nD0.f16 = S0.f16 >= S1.f16 ? S0.f16 : S1.f16\nelse\nD0.f16 = S0.f16 > S1.f16 ? S0.f16 : S1.f16\nendif",
VOP3Op.V_MIN_F16_E64: "if (WAVE_MODE.IEEE && isSignalNAN(64'F(S0.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S0.f16)))\nelsif (WAVE_MODE.IEEE && isSignalNAN(64'F(S1.f16))) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S1.f16)))\nelsif isNAN(64'F(S0.f16)) then\nD0.f16 = S1.f16\nelsif isNAN(64'F(S1.f16)) then\nD0.f16 = S0.f16\nelsif ((64'F(S0.f16) == +0.0) && (64'F(S1.f16) == -0.0)) then\nD0.f16 = S1.f16\nelsif ((64'F(S0.f16) == -0.0) && (64'F(S1.f16) == +0.0)) then\nD0.f16 = S0.f16\nelse\n// Note: there's no IEEE case here like there is for V_MAX_F16.\nD0.f16 = S0.f16 < S1.f16 ? S0.f16 : S1.f16\nendif",
VOP3Op.V_MAX_U16_E64: 'D0.u16 = S0.u16 >= S1.u16 ? S0.u16 : S1.u16',
VOP3Op.V_MAX_I16_E64: 'D0.i16 = S0.i16 >= S1.i16 ? S0.i16 : S1.i16',
VOP3Op.V_MIN_U16_E64: 'D0.u16 = S0.u16 < S1.u16 ? S0.u16 : S1.u16',
VOP3Op.V_MIN_I16_E64: 'D0.i16 = S0.i16 < S1.i16 ? S0.i16 : S1.i16',
VOP3Op.V_LDEXP_F16_E64: "D0.f16 = S0.f16 * 16'F(2.0F ** 32'I(S1.i16))",
VOP3Op.V_ADD_U32_E64: 'D0.u32 = S0.u32 + S1.u32',
VOP3Op.V_SUB_U32_E64: 'D0.u32 = S0.u32 - S1.u32',
VOP3Op.V_SUBREV_U32_E64: 'D0.u32 = S1.u32 - S0.u32',
VOP3Op.V_DOT2C_F32_F16_E64: 'tmp = D0.f32;\ntmp += f16_to_f32(S0[15 : 0].f16) * f16_to_f32(S1[15 : 0].f16);\ntmp += f16_to_f32(S0[31 : 16].f16) * f16_to_f32(S1[31 : 16].f16);\nD0.f32 = tmp',
VOP3Op.V_DOT2C_I32_I16_E64: 'tmp = D0.i32;\ntmp += i16_to_i32(S0[15 : 0].i16) * i16_to_i32(S1[15 : 0].i16);\ntmp += i16_to_i32(S0[31 : 16].i16) * i16_to_i32(S1[31 : 16].i16);\nD0.i32 = tmp',
VOP3Op.V_DOT4C_I32_I8_E64: 'tmp = D0.i32;\ntmp += i8_to_i32(S0[7 : 0].i8) * i8_to_i32(S1[7 : 0].i8);\ntmp += i8_to_i32(S0[15 : 8].i8) * i8_to_i32(S1[15 : 8].i8);\ntmp += i8_to_i32(S0[23 : 16].i8) * i8_to_i32(S1[23 : 16].i8);\ntmp += i8_to_i32(S0[31 : 24].i8) * i8_to_i32(S1[31 : 24].i8);\nD0.i32 = tmp',
VOP3Op.V_DOT8C_I32_I4_E64: 'tmp = D0.i32;\ntmp += i4_to_i32(S0[3 : 0].i4) * i4_to_i32(S1[3 : 0].i4);\ntmp += i4_to_i32(S0[7 : 4].i4) * i4_to_i32(S1[7 : 4].i4);\ntmp += i4_to_i32(S0[11 : 8].i4) * i4_to_i32(S1[11 : 8].i4);\ntmp += i4_to_i32(S0[15 : 12].i4) * i4_to_i32(S1[15 : 12].i4);\ntmp += i4_to_i32(S0[19 : 16].i4) * i4_to_i32(S1[19 : 16].i4);\ntmp += i4_to_i32(S0[23 : 20].i4) * i4_to_i32(S1[23 : 20].i4);\ntmp += i4_to_i32(S0[27 : 24].i4) * i4_to_i32(S1[27 : 24].i4);\ntmp += i4_to_i32(S0[31 : 28].i4) * i4_to_i32(S1[31 : 28].i4);\nD0.i32 = tmp',
VOP3Op.V_FMAC_F32_E64: 'D0.f32 = fma(S0.f32, S1.f32, D0.f32)',
VOP3Op.V_PK_FMAC_F16_E64: 'D0[15 : 0].f16 = fma(S0[15 : 0].f16, S1[15 : 0].f16, D0[15 : 0].f16);\nD0[31 : 16].f16 = fma(S0[31 : 16].f16, S1[31 : 16].f16, D0[31 : 16].f16)',
VOP3Op.V_XNOR_B32_E64: 'D0.u32 = ~(S0.u32 ^ S1.u32)',
VOP3Op.V_MAD_I32_I24_E64: "D0.i32 = 32'I(S0.i24) * 32'I(S1.i24) + S2.i32",
VOP3Op.V_MAD_U32_U24_E64: "D0.u32 = 32'U(S0.u24) * 32'U(S1.u24) + S2.u32",
VOP3Op.V_CUBEID_F32_E64: '// Set D0.f = cubemap face ID ({0.0, 1.0, ..., 5.0}).\n// XYZ coordinate is given in (S0.f, S1.f, S2.f).\n// S0.f = x\n// S1.f = y\n// S2.f = z\nif ((abs(S2.f32) >= abs(S0.f32)) && (abs(S2.f32) >= abs(S1.f32))) then\nif S2.f32 < 0.0F then\nD0.f32 = 5.0F\nelse\nD0.f32 = 4.0F\nendif\nelsif abs(S1.f32) >= abs(S0.f32) then\nif S1.f32 < 0.0F then\nD0.f32 = 3.0F\nelse\nD0.f32 = 2.0F\nendif\nelse\nif S0.f32 < 0.0F then\nD0.f32 = 1.0F\nelse\nD0.f32 = 0.0F\nendif\nendif',
VOP3Op.V_CUBESC_F32_E64: '// D0.f = cubemap S coordinate.\n// XYZ coordinate is given in (S0.f, S1.f, S2.f).\n// S0.f = x\n// S1.f = y\n// S2.f = z\nif ((abs(S2.f32) >= abs(S0.f32)) && (abs(S2.f32) >= abs(S1.f32))) then\nif S2.f32 < 0.0F then\nD0.f32 = -S0.f32\nelse\nD0.f32 = S0.f32\nendif\nelsif abs(S1.f32) >= abs(S0.f32) then\nD0.f32 = S0.f32\nelse\nif S0.f32 < 0.0F then\nD0.f32 = S2.f32\nelse\nD0.f32 = -S2.f32\nendif\nendif',
VOP3Op.V_CUBETC_F32_E64: '// D0.f = cubemap T coordinate.\n// XYZ coordinate is given in (S0.f, S1.f, S2.f).\n// S0.f = x\n// S1.f = y\n// S2.f = z\nif ((abs(S2.f32) >= abs(S0.f32)) && (abs(S2.f32) >= abs(S1.f32))) then\nD0.f32 = -S1.f32\nelsif abs(S1.f32) >= abs(S0.f32) then\nif S1.f32 < 0.0F then\nD0.f32 = -S2.f32\nelse\nD0.f32 = S2.f32\nendif\nelse\nD0.f32 = -S1.f32\nendif',
VOP3Op.V_CUBEMA_F32_E64: '// D0.f = 2.0 * cubemap major axis.\n// XYZ coordinate is given in (S0.f, S1.f, S2.f).\n// S0.f = x\n// S1.f = y\n// S2.f = z\nif ((abs(S2.f32) >= abs(S0.f32)) && (abs(S2.f32) >= abs(S1.f32))) then\nD0.f32 = S2.f32 * 2.0F\nelsif abs(S1.f32) >= abs(S0.f32) then\nD0.f32 = S1.f32 * 2.0F\nelse\nD0.f32 = S0.f32 * 2.0F\nendif',
VOP3Op.V_BFE_U32_E64: 'D0.u32 = ((S0.u32 >> S1[4 : 0].u32) & ((1U << S2[4 : 0].u32) - 1U))',
VOP3Op.V_BFE_I32_E64: 'tmp.i32 = ((S0.i32 >> S1[4 : 0].u32) & ((1 << S2[4 : 0].u32) - 1));\nD0.i32 = signext_from_bit(tmp.i32, S2[4 : 0].u32)',
VOP3Op.V_BFI_B32_E64: 'D0.u32 = ((S0.u32 & S1.u32) | (~S0.u32 & S2.u32))',
VOP3Op.V_FMA_F32_E64: 'D0.f32 = fma(S0.f32, S1.f32, S2.f32)',
VOP3Op.V_FMA_F64_E64: 'D0.f64 = fma(S0.f64, S1.f64, S2.f64)',
VOP3Op.V_LERP_U8_E64: 'tmp = ((S0.u32[31 : 24] + S1.u32[31 : 24] + S2.u32[24].u8) >> 1U << 24U);\ntmp += ((S0.u32[23 : 16] + S1.u32[23 : 16] + S2.u32[16].u8) >> 1U << 16U);\ntmp += ((S0.u32[15 : 8] + S1.u32[15 : 8] + S2.u32[8].u8) >> 1U << 8U);\ntmp += ((S0.u32[7 : 0] + S1.u32[7 : 0] + S2.u32[0].u8) >> 1U);\nD0.u32 = tmp.u32',
VOP3Op.V_ALIGNBIT_B32_E64: "D0.u32 = 32'U(({ S0.u32, S1.u32 } >> S2.u32[4 : 0]) & 0xffffffffLL)",
VOP3Op.V_ALIGNBYTE_B32_E64: "D0.u32 = 32'U(({ S0.u32, S1.u32 } >> (S2.u32[1 : 0] * 8U)) & 0xffffffffLL)",
VOP3Op.V_MIN3_F32_E64: 'D0.f32 = v_min_f32(v_min_f32(S0.f32, S1.f32), S2.f32)',
VOP3Op.V_MIN3_I32_E64: 'D0.i32 = v_min_i32(v_min_i32(S0.i32, S1.i32), S2.i32)',
VOP3Op.V_MIN3_U32_E64: 'D0.u32 = v_min_u32(v_min_u32(S0.u32, S1.u32), S2.u32)',
VOP3Op.V_MAX3_F32_E64: 'D0.f32 = v_max_f32(v_max_f32(S0.f32, S1.f32), S2.f32)',
VOP3Op.V_MAX3_I32_E64: 'D0.i32 = v_max_i32(v_max_i32(S0.i32, S1.i32), S2.i32)',
VOP3Op.V_MAX3_U32_E64: 'D0.u32 = v_max_u32(v_max_u32(S0.u32, S1.u32), S2.u32)',
VOP3Op.V_MED3_F32_E64: "if (isNAN(64'F(S0.f32)) || isNAN(64'F(S1.f32)) || isNAN(64'F(S2.f32))) then\nD0.f32 = v_min3_f32(S0.f32, S1.f32, S2.f32)\nelsif v_max3_f32(S0.f32, S1.f32, S2.f32) == S0.f32 then\nD0.f32 = v_max_f32(S1.f32, S2.f32)\nelsif v_max3_f32(S0.f32, S1.f32, S2.f32) == S1.f32 then\nD0.f32 = v_max_f32(S0.f32, S2.f32)\nelse\nD0.f32 = v_max_f32(S0.f32, S1.f32)\nendif",
VOP3Op.V_MED3_I32_E64: 'if v_max3_i32(S0.i32, S1.i32, S2.i32) == S0.i32 then\nD0.i32 = v_max_i32(S1.i32, S2.i32)\nelsif v_max3_i32(S0.i32, S1.i32, S2.i32) == S1.i32 then\nD0.i32 = v_max_i32(S0.i32, S2.i32)\nelse\nD0.i32 = v_max_i32(S0.i32, S1.i32)\nendif',
VOP3Op.V_MED3_U32_E64: 'if v_max3_u32(S0.u32, S1.u32, S2.u32) == S0.u32 then\nD0.u32 = v_max_u32(S1.u32, S2.u32)\nelsif v_max3_u32(S0.u32, S1.u32, S2.u32) == S1.u32 then\nD0.u32 = v_max_u32(S0.u32, S2.u32)\nelse\nD0.u32 = v_max_u32(S0.u32, S1.u32)\nendif',
VOP3Op.V_SAD_U8_E64: "ABSDIFF = lambda(x, y) (\nx > y ? x - y : y - x);\n// UNSIGNED comparison\ntmp = S2.u32;\ntmp += 32'U(ABSDIFF(S0.u32[7 : 0], S1.u32[7 : 0]));\ntmp += 32'U(ABSDIFF(S0.u32[15 : 8], S1.u32[15 : 8]));\ntmp += 32'U(ABSDIFF(S0.u32[23 : 16], S1.u32[23 : 16]));\ntmp += 32'U(ABSDIFF(S0.u32[31 : 24], S1.u32[31 : 24]));\nD0.u32 = tmp",
VOP3Op.V_SAD_HI_U8_E64: "D0.u32 = (32'U(v_sad_u8(S0, S1, 0U)) << 16U) + S2.u32",
VOP3Op.V_SAD_U16_E64: 'ABSDIFF = lambda(x, y) (\nx > y ? x - y : y - x);\n// UNSIGNED comparison\ntmp = S2.u32;\ntmp += ABSDIFF(S0[15 : 0].u16, S1[15 : 0].u16);\ntmp += ABSDIFF(S0[31 : 16].u16, S1[31 : 16].u16);\nD0.u32 = tmp',
VOP3Op.V_SAD_U32_E64: 'ABSDIFF = lambda(x, y) (\nx > y ? x - y : y - x);\n// UNSIGNED comparison\nD0.u32 = ABSDIFF(S0.u32, S1.u32) + S2.u32',
VOP3Op.V_CVT_PK_U8_F32_E64: "tmp = (S2.u32 & 32'U(~(0xff << (S1.u32[1 : 0].u32 * 8U))));\ntmp = (tmp | ((32'U(f32_to_u8(S0.f32)) & 255U) << (S1.u32[1 : 0].u32 * 8U)));\nD0.u32 = tmp",
VOP3Op.V_DIV_FIXUP_F32_E64: "sign_out = (sign(S1.f32) ^ sign(S2.f32));\nif isNAN(64'F(S2.f32)) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S2.f32)))\nelsif isNAN(64'F(S1.f32)) then\nD0.f32 = 32'F(cvtToQuietNAN(64'F(S1.f32)))\nelsif ((64'F(S1.f32) == 0.0) && (64'F(S2.f32) == 0.0)) then\n// 0/0\nD0.f32 = 32'F(0xffc00000)\nelsif ((64'F(abs(S1.f32)) == +INF) && (64'F(abs(S2.f32)) == +INF)) then\n// inf/inf\nD0.f32 = 32'F(0xffc00000)\nelsif ((64'F(S1.f32) == 0.0) || (64'F(abs(S2.f32)) == +INF)) then\n// x/0, or inf/y\nD0.f32 = sign_out ? -INF.f32 : +INF.f32\nelsif ((64'F(abs(S1.f32)) == +INF) || (64'F(S2.f32) == 0.0)) then\n// x/inf, 0/y\nD0.f32 = sign_out ? -0.0F : 0.0F\nelsif exponent(S2.f32) - exponent(S1.f32) < -150 then\nD0.f32 = sign_out ? -UNDERFLOW_F32 : UNDERFLOW_F32\nelsif exponent(S1.f32) == 255 then\nD0.f32 = sign_out ? -OVERFLOW_F32 : OVERFLOW_F32\nelse\nD0.f32 = sign_out ? -abs(S0.f32) : abs(S0.f32)\nendif",
VOP3Op.V_DIV_FIXUP_F64_E64: "sign_out = (sign(S1.f64) ^ sign(S2.f64));\nif isNAN(S2.f64) then\nD0.f64 = cvtToQuietNAN(S2.f64)\nelsif isNAN(S1.f64) then\nD0.f64 = cvtToQuietNAN(S1.f64)\nelsif ((S1.f64 == 0.0) && (S2.f64 == 0.0)) then\n// 0/0\nD0.f64 = 64'F(0xfff8000000000000LL)\nelsif ((abs(S1.f64) == +INF) && (abs(S2.f64) == +INF)) then\n// inf/inf\nD0.f64 = 64'F(0xfff8000000000000LL)\nelsif ((S1.f64 == 0.0) || (abs(S2.f64) == +INF)) then\n// x/0, or inf/y\nD0.f64 = sign_out ? -INF : +INF\nelsif ((abs(S1.f64) == +INF) || (S2.f64 == 0.0)) then\n// x/inf, 0/y\nD0.f64 = sign_out ? -0.0 : 0.0\nelsif exponent(S2.f64) - exponent(S1.f64) < -1075 then\nD0.f64 = sign_out ? -UNDERFLOW_F64 : UNDERFLOW_F64\nelsif exponent(S1.f64) == 2047 then\nD0.f64 = sign_out ? -OVERFLOW_F64 : OVERFLOW_F64\nelse\nD0.f64 = sign_out ? -abs(S0.f64) : abs(S0.f64)\nendif",
VOP3Op.V_DIV_FMAS_F32_E64: 'if VCC.u64[laneId] then\nD0.f32 = 2.0F ** 32 * fma(S0.f32, S1.f32, S2.f32)\nelse\nD0.f32 = fma(S0.f32, S1.f32, S2.f32)\nendif',
VOP3Op.V_DIV_FMAS_F64_E64: 'if VCC.u64[laneId] then\nD0.f64 = 2.0 ** 64 * fma(S0.f64, S1.f64, S2.f64)\nelse\nD0.f64 = fma(S0.f64, S1.f64, S2.f64)\nendif',
VOP3Op.V_MSAD_U8_E64: "ABSDIFF = lambda(x, y) (\nx > y ? x - y : y - x);\n// UNSIGNED comparison\ntmp = S2.u32;\ntmp += S1.u32[7 : 0] == 8'0U ? 0U : 32'U(ABSDIFF(S0.u32[7 : 0], S1.u32[7 : 0]));\ntmp += S1.u32[15 : 8] == 8'0U ? 0U : 32'U(ABSDIFF(S0.u32[15 : 8], S1.u32[15 : 8]));\ntmp += S1.u32[23 : 16] == 8'0U ? 0U : 32'U(ABSDIFF(S0.u32[23 : 16], S1.u32[23 : 16]));\ntmp += S1.u32[31 : 24] == 8'0U ? 0U : 32'U(ABSDIFF(S0.u32[31 : 24], S1.u32[31 : 24]));\nD0.u32 = tmp",
VOP3Op.V_QSAD_PK_U16_U8_E64: "tmp[63 : 48] = 16'B(v_sad_u8(S0[55 : 24], S1[31 : 0], S2[63 : 48].u32));\ntmp[47 : 32] = 16'B(v_sad_u8(S0[47 : 16], S1[31 : 0], S2[47 : 32].u32));\ntmp[31 : 16] = 16'B(v_sad_u8(S0[39 : 8], S1[31 : 0], S2[31 : 16].u32));\ntmp[15 : 0] = 16'B(v_sad_u8(S0[31 : 0], S1[31 : 0], S2[15 : 0].u32));\nD0.b64 = tmp.b64",
VOP3Op.V_MQSAD_PK_U16_U8_E64: "tmp[63 : 48] = 16'B(v_msad_u8(S0[55 : 24], S1[31 : 0], S2[63 : 48].u32));\ntmp[47 : 32] = 16'B(v_msad_u8(S0[47 : 16], S1[31 : 0], S2[47 : 32].u32));\ntmp[31 : 16] = 16'B(v_msad_u8(S0[39 : 8], S1[31 : 0], S2[31 : 16].u32));\ntmp[15 : 0] = 16'B(v_msad_u8(S0[31 : 0], S1[31 : 0], S2[15 : 0].u32));\nD0.b64 = tmp.b64",
VOP3Op.V_MQSAD_U32_U8_E64: "tmp[127 : 96] = 32'B(v_msad_u8(S0[55 : 24], S1[31 : 0], S2[127 : 96].u32));\ntmp[95 : 64] = 32'B(v_msad_u8(S0[47 : 16], S1[31 : 0], S2[95 : 64].u32));\ntmp[63 : 32] = 32'B(v_msad_u8(S0[39 : 8], S1[31 : 0], S2[63 : 32].u32));\ntmp[31 : 0] = 32'B(v_msad_u8(S0[31 : 0], S1[31 : 0], S2[31 : 0].u32));\nD0.b128 = tmp.b128",
VOP3Op.V_MAD_LEGACY_F16_E64: "tmp = S0.f16 * S1.f16 + S2.f16;\nif OPSEL.u4[3] then\nD0 = { tmp.f16, D0[15 : 0] }\nelse\nD0 = { 16'0, tmp.f16 }\nendif",
VOP3Op.V_MAD_LEGACY_U16_E64: "tmp = S0.u16 * S1.u16 + S2.u16;\nif OPSEL.u4[3] then\nD0 = { tmp.u16, D0[15 : 0] }\nelse\nD0 = { 16'0, tmp.u16 }\nendif",
VOP3Op.V_MAD_LEGACY_I16_E64: "tmp = S0.i16 * S1.i16 + S2.i16;\nif OPSEL.u4[3] then\nD0 = { tmp.i16, D0[15 : 0] }\nelse\nD0 = { 16'0, tmp.i16 }\nendif",
VOP3Op.V_PERM_B32_E64: "BYTE_PERMUTE = lambda(data, sel) (\ndeclare in : 8'B[8];\nfor i in 0 : 7 do\nin[i] = data[i * 8 + 7 : i * 8].b8\nendfor;\nif sel.u32 >= 13U then\nreturn 8'0xff\nelsif sel.u32 == 12U then\nreturn 8'0x0\nelsif sel.u32 == 11U then\nreturn in[7][7].b8 * 8'0xff\nelsif sel.u32 == 10U then\nreturn in[5][7].b8 * 8'0xff\nelsif sel.u32 == 9U then\nreturn in[3][7].b8 * 8'0xff\nelsif sel.u32 == 8U then\nreturn in[1][7].b8 * 8'0xff\nelse\nreturn in[sel]\nendif);\nD0[31 : 24] = BYTE_PERMUTE({ S0.u32, S1.u32 }, S2.u32[31 : 24]);\nD0[23 : 16] = BYTE_PERMUTE({ S0.u32, S1.u32 }, S2.u32[23 : 16]);\nD0[15 : 8] = BYTE_PERMUTE({ S0.u32, S1.u32 }, S2.u32[15 : 8]);\nD0[7 : 0] = BYTE_PERMUTE({ S0.u32, S1.u32 }, S2.u32[7 : 0])",
VOP3Op.V_FMA_LEGACY_F16_E64: "tmp = fma(S0.f16, S1.f16, S2.f16);\nif OPSEL.u4[3] then\nD0 = { tmp.f16, D0[15 : 0] }\nelse\nD0 = { 16'0, tmp.f16 }\nendif",
VOP3Op.V_DIV_FIXUP_LEGACY_F16_E64: "sign_out = (sign(S1.f16) ^ sign(S2.f16));\nif isNAN(64'F(S2.f16)) then\ntmp = cvtToQuietNAN(64'F(S2.f16))\nelsif isNAN(64'F(S1.f16)) then\ntmp = cvtToQuietNAN(64'F(S1.f16))\nelsif ((64'F(S1.f16) == 0.0) && (64'F(S2.f16) == 0.0)) then\n// 0/0\ntmp = 16'F(0xfe00)\nelsif ((64'F(abs(S1.f16)) == +INF) && (64'F(abs(S2.f16)) == +INF)) then\n// inf/inf\ntmp = 16'F(0xfe00)\nelsif ((64'F(S1.f16) == 0.0) || (64'F(abs(S2.f16)) == +INF)) then\n// x/0, or inf/y\ntmp = sign_out ? -INF : +INF\nelsif ((64'F(abs(S1.f16)) == +INF) || (64'F(S2.f16) == 0.0)) then\n// x/inf, 0/y\ntmp = sign_out ? -0.0 : 0.0\nelse\ntmp = sign_out ? -abs(S0.f16) : abs(S0.f16)\nendif;\nif OPSEL.u4[3] then\nD0 = { tmp.f16, D0[15 : 0] }\nelse\nD0 = { 16'0, tmp.f16 }\nendif",
VOP3Op.V_CVT_PKACCUM_U8_F32_E64: "byte = S1.u32[1 : 0];\nbit = byte.u32 * 8U;\nD0.u32[bit + 7U : bit] = 32'U(f32_to_u8(S0.f32))",
VOP3Op.V_MAD_U32_U16_E64: "D0.u32 = 32'U(S0.u16) * 32'U(S1.u16) + S2.u32",
VOP3Op.V_MAD_I32_I16_E64: "D0.i32 = 32'I(S0.i16) * 32'I(S1.i16) + S2.i32",
VOP3Op.V_XAD_U32_E64: 'D0.u32 = (S0.u32 ^ S1.u32) + S2.u32',
VOP3Op.V_MIN3_F16_E64: 'D0.f16 = v_min_f16(v_min_f16(S0.f16, S1.f16), S2.f16)',
VOP3Op.V_MIN3_I16_E64: 'D0.i16 = v_min_i16(v_min_i16(S0.i16, S1.i16), S2.i16)',
VOP3Op.V_MIN3_U16_E64: 'D0.u16 = v_min_u16(v_min_u16(S0.u16, S1.u16), S2.u16)',
VOP3Op.V_MAX3_F16_E64: 'D0.f16 = v_max_f16(v_max_f16(S0.f16, S1.f16), S2.f16)',
VOP3Op.V_MAX3_I16_E64: 'D0.i16 = v_max_i16(v_max_i16(S0.i16, S1.i16), S2.i16)',
VOP3Op.V_MAX3_U16_E64: 'D0.u16 = v_max_u16(v_max_u16(S0.u16, S1.u16), S2.u16)',
VOP3Op.V_MED3_F16_E64: "if (isNAN(64'F(S0.f16)) || isNAN(64'F(S1.f16)) || isNAN(64'F(S2.f16))) then\nD0.f16 = v_min3_f16(S0.f16, S1.f16, S2.f16)\nelsif v_max3_f16(S0.f16, S1.f16, S2.f16) == S0.f16 then\nD0.f16 = v_max_f16(S1.f16, S2.f16)\nelsif v_max3_f16(S0.f16, S1.f16, S2.f16) == S1.f16 then\nD0.f16 = v_max_f16(S0.f16, S2.f16)\nelse\nD0.f16 = v_max_f16(S0.f16, S1.f16)\nendif",
VOP3Op.V_MED3_I16_E64: 'if v_max3_i16(S0.i16, S1.i16, S2.i16) == S0.i16 then\nD0.i16 = v_max_i16(S1.i16, S2.i16)\nelsif v_max3_i16(S0.i16, S1.i16, S2.i16) == S1.i16 then\nD0.i16 = v_max_i16(S0.i16, S2.i16)\nelse\nD0.i16 = v_max_i16(S0.i16, S1.i16)\nendif',
VOP3Op.V_MED3_U16_E64: 'if v_max3_u16(S0.u16, S1.u16, S2.u16) == S0.u16 then\nD0.u16 = v_max_u16(S1.u16, S2.u16)\nelsif v_max3_u16(S0.u16, S1.u16, S2.u16) == S1.u16 then\nD0.u16 = v_max_u16(S0.u16, S2.u16)\nelse\nD0.u16 = v_max_u16(S0.u16, S1.u16)\nendif',
VOP3Op.V_LSHL_ADD_U32_E64: 'D0.u32 = (S0.u32 << S1.u32[4 : 0].u32) + S2.u32',
VOP3Op.V_ADD_LSHL_U32_E64: 'D0.u32 = ((S0.u32 + S1.u32) << S2.u32[4 : 0].u32)',
VOP3Op.V_ADD3_U32_E64: 'D0.u32 = S0.u32 + S1.u32 + S2.u32',
VOP3Op.V_LSHL_OR_B32: 'D0.u32 = ((S0.u32 << S1.u32[4 : 0].u32) | S2.u32)',
VOP3Op.V_AND_OR_B32: 'D0.u32 = ((S0.u32 & S1.u32) | S2.u32)',
VOP3Op.V_OR3_B32: 'D0.u32 = (S0.u32 | S1.u32 | S2.u32)',
VOP3Op.V_MAD_F16: 'D0.f16 = S0.f16 * S1.f16 + S2.f16',
VOP3Op.V_MAD_U16: 'D0.u16 = S0.u16 * S1.u16 + S2.u16',
VOP3Op.V_MAD_I16: 'D0.i16 = S0.i16 * S1.i16 + S2.i16',
VOP3Op.V_FMA_F16: 'D0.f16 = fma(S0.f16, S1.f16, S2.f16)',
VOP3Op.V_DIV_FIXUP_F16: "sign_out = (sign(S1.f16) ^ sign(S2.f16));\nif isNAN(64'F(S2.f16)) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S2.f16)))\nelsif isNAN(64'F(S1.f16)) then\nD0.f16 = 16'F(cvtToQuietNAN(64'F(S1.f16)))\nelsif ((64'F(S1.f16) == 0.0) && (64'F(S2.f16) == 0.0)) then\n// 0/0\nD0.f16 = 16'F(0xfe00)\nelsif ((64'F(abs(S1.f16)) == +INF) && (64'F(abs(S2.f16)) == +INF)) then\n// inf/inf\nD0.f16 = 16'F(0xfe00)\nelsif ((64'F(S1.f16) == 0.0) || (64'F(abs(S2.f16)) == +INF)) then\n// x/0, or inf/y\nD0.f16 = sign_out ? -INF.f16 : +INF.f16\nelsif ((64'F(abs(S1.f16)) == +INF) || (64'F(S2.f16) == 0.0)) then\n// x/inf, 0/y\nD0.f16 = sign_out ? -16'0.0 : 16'0.0\nelse\nD0.f16 = sign_out ? -abs(S0.f16) : abs(S0.f16)\nendif",
VOP3Op.V_LSHL_ADD_U64: 'D0.u64 = (S0.u64 << S1.u32[2 : 0].u32) + S2.u64',
VOP3Op.V_BITOP3_B16: "TTBL = { INST.OMOD[1 : 0], INST.ABS[2 : 0], INST.NEG[2 : 0] };\ntmp = 16'0U;\ntmp = (tmp | (32'I(TTBL.b32 & 0x1) != 0 ? 16'U(~S0.b16 & ~S1.b16 & ~S2.b16) : 16'0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x2) != 0 ? 16'U(~S0.b16 & ~S1.b16 & S2.b16) : 16'0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x4) != 0 ? 16'U(~S0.b16 & S1.b16 & ~S2.b16) : 16'0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x8) != 0 ? 16'U(~S0.b16 & S1.b16 & S2.b16) : 16'0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x10) != 0 ? 16'U(S0.b16 & ~S1.b16 & ~S2.b16) : 16'0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x20) != 0 ? 16'U(S0.b16 & ~S1.b16 & S2.b16) : 16'0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x40) != 0 ? 16'U(S0.b16 & S1.b16 & ~S2.b16) : 16'0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x80) != 0 ? 16'U(S0.b16 & S1.b16 & S2.b16) : 16'0U));\nD.b16 = tmp.b16",
VOP3Op.V_BITOP3_B32: "TTBL = { INST.OMOD[1 : 0], INST.ABS[2 : 0], INST.NEG[2 : 0] };\ntmp = 0U;\ntmp = (tmp | (32'I(TTBL.b32 & 0x1) != 0 ? 32'U(~S0.b32 & ~S1.b32 & ~S2.b32) : 0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x2) != 0 ? 32'U(~S0.b32 & ~S1.b32 & S2.b32) : 0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x4) != 0 ? 32'U(~S0.b32 & S1.b32 & ~S2.b32) : 0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x8) != 0 ? 32'U(~S0.b32 & S1.b32 & S2.b32) : 0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x10) != 0 ? 32'U(S0.b32 & ~S1.b32 & ~S2.b32) : 0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x20) != 0 ? 32'U(S0.b32 & ~S1.b32 & S2.b32) : 0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x40) != 0 ? 32'U(S0.b32 & S1.b32 & ~S2.b32) : 0U));\ntmp = (tmp | (32'I(TTBL.b32 & 0x80) != 0 ? 32'U(S0.b32 & S1.b32 & S2.b32) : 0U));\nD.b32 = tmp.b32",
VOP3Op.V_CVT_SCALEF32_PK_FP8_F32: "scale = 32'U(exponent(S2.f32));\ntmp0 = f32_to_fp8_scale(S0.f32, scale.u8);\ntmp1 = f32_to_fp8_scale(S1.f32, scale.u8);\ndstword = OPSEL[3].i32 * 16;\nVGPR[laneId][VDST.u32][dstword + 15 : dstword].b16 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_BF8_F32: "scale = 32'U(exponent(S2.f32));\ntmp0 = f32_to_bf8_scale(S0.f32, scale.u8);\ntmp1 = f32_to_bf8_scale(S1.f32, scale.u8);\ndstword = OPSEL[3].i32 * 16;\nVGPR[laneId][VDST.u32][dstword + 15 : dstword].b16 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_FP8_F32: "scale = 32'U(exponent(S2.f32));\ntmp = f32_to_fp8_sr_scale(S0.f32, S1.u32, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].fp8 = tmp;\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_BF8_F32: "scale = 32'U(exponent(S2.f32));\ntmp = f32_to_bf8_sr_scale(S0.f32, S1.u32, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].bf8 = tmp;\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_F32_FP8: "scale = 32'U(exponent(S1.f32));\nsrcword = OPSEL[0].i32 * 16;\nsrc = VGPR[laneId][SRC0.u32][srcword + 15 : srcword].b16;\ntmp0 = fp8_to_f32_scale(src[7 : 0].fp8, scale.u8);\ntmp1 = fp8_to_f32_scale(src[15 : 8].fp8, scale.u8);\nD0[31 : 0].f32 = tmp0;\nD0[63 : 32].f32 = tmp1",
VOP3Op.V_CVT_SCALEF32_PK_F32_BF8: "scale = 32'U(exponent(S1.f32));\nsrcword = OPSEL[0].i32 * 16;\nsrc = VGPR[laneId][SRC0.u32][srcword + 15 : srcword].b16;\ntmp0 = bf8_to_f32_scale(src[7 : 0].bf8, scale.u8);\ntmp1 = bf8_to_f32_scale(src[15 : 8].bf8, scale.u8);\nD0[31 : 0].f32 = tmp0;\nD0[63 : 32].f32 = tmp1",
VOP3Op.V_CVT_SCALEF32_F32_FP8: "scale = 32'U(exponent(S1.f32));\nsrcbyte = OPSEL[1 : 0].i32 * 8;\nsrc = VGPR[laneId][SRC0.u32][srcbyte + 7 : srcbyte].fp8;\ntmp = fp8_to_f32_scale(src, scale.u8);\nD0 = tmp.b32",
VOP3Op.V_CVT_SCALEF32_F32_BF8: "scale = 32'U(exponent(S1.f32));\nsrcbyte = OPSEL[1 : 0].i32 * 8;\nsrc = VGPR[laneId][SRC0.u32][srcbyte + 7 : srcbyte].bf8;\ntmp = bf8_to_f32_scale(src, scale.u8);\nD0 = tmp.b32",
VOP3Op.V_CVT_SCALEF32_PK_FP4_F32: "scale = 32'U(exponent(S2.f32));\ntmp0 = f32_to_fp4_scale(S0.f32, scale.u8);\ntmp1 = f32_to_fp4_scale(S1.f32, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].b8 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_PK_FP4_F32: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ntmp0 = f32_to_fp4_sr_scale(S0[31 : 0].f32, randomVal, scale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32));\ntmp1 = f32_to_fp4_sr_scale(S0[63 : 32].f32, randomVal, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].b8 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_F32_FP4: "scale = 32'U(exponent(S1.f32));\nsrcbyte = OPSEL[1 : 0].i32 * 8;\nsrc = VGPR[laneId][SRC0.u32][srcbyte + 7 : srcbyte].b8;\ntmp0 = fp4_to_f32_scale(src[3 : 0].fp4, scale.u8);\ntmp1 = fp4_to_f32_scale(src[7 : 4].fp4, scale.u8);\nD0[31 : 0].f32 = tmp0;\nD0[63 : 32].f32 = tmp1",
VOP3Op.V_CVT_SCALEF32_PK_FP8_F16: "scale = 32'U(exponent(S1.f32));\ntmp0 = f16_to_fp8_scale(S0[15 : 0].f16, scale.u8);\ntmp1 = f16_to_fp8_scale(S0[31 : 16].f16, scale.u8);\ndstword = OPSEL[3].i32 * 16;\nVGPR[laneId][VDST.u32][dstword + 15 : dstword].b16 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_BF8_F16: "scale = 32'U(exponent(S1.f32));\ntmp0 = f16_to_bf8_scale(S0[15 : 0].f16, scale.u8);\ntmp1 = f16_to_bf8_scale(S0[31 : 16].f16, scale.u8);\ndstword = OPSEL[3].i32 * 16;\nVGPR[laneId][VDST.u32][dstword + 15 : dstword].b16 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_FP8_F16: "scale = 32'U(exponent(S2.f32));\ntmp = f16_to_fp8_sr_scale(S0.f16, S1.u32, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].fp8 = tmp;\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_BF8_F16: "scale = 32'U(exponent(S2.f32));\ntmp = f16_to_bf8_sr_scale(S0.f16, S1.u32, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].bf8 = tmp;\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_FP8_BF16: "scale = 32'U(exponent(S1.f32));\ntmp0 = bf16_to_fp8_scale(S0[15 : 0].bf16, scale.u8);\ntmp1 = bf16_to_fp8_scale(S0[31 : 16].bf16, scale.u8);\ndstword = OPSEL[3].i32 * 16;\nVGPR[laneId][VDST.u32][dstword + 15 : dstword].b16 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_BF8_BF16: "scale = 32'U(exponent(S1.f32));\ntmp0 = bf16_to_bf8_scale(S0[15 : 0].bf16, scale.u8);\ntmp1 = bf16_to_bf8_scale(S0[31 : 16].bf16, scale.u8);\ndstword = OPSEL[3].i32 * 16;\nVGPR[laneId][VDST.u32][dstword + 15 : dstword].b16 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_FP8_BF16: "scale = 32'U(exponent(S2.f32));\ntmp = bf16_to_fp8_sr_scale(S0.bf16, S1.u32, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].fp8 = tmp;\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_BF8_BF16: "scale = 32'U(exponent(S2.f32));\ntmp = bf16_to_bf8_sr_scale(S0.bf16, S1.u32, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].bf8 = tmp;\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_F16_FP8: "scale = 32'U(exponent(S1.f32));\nsrcword = OPSEL[0].i32 * 16;\nsrc = VGPR[laneId][SRC0.u32][srcword + 15 : srcword].b16;\ntmp0 = fp8_to_f16_scale(src[7 : 0].fp8, scale.u8);\ntmp1 = fp8_to_f16_scale(src[15 : 8].fp8, scale.u8);\nD0[15 : 0].f16 = tmp0;\nD0[31 : 16].f16 = tmp1",
VOP3Op.V_CVT_SCALEF32_PK_F16_BF8: "scale = 32'U(exponent(S1.f32));\nsrcword = OPSEL[0].i32 * 16;\nsrc = VGPR[laneId][SRC0.u32][srcword + 15 : srcword].b16;\ntmp0 = bf8_to_f16_scale(src[7 : 0].bf8, scale.u8);\ntmp1 = bf8_to_f16_scale(src[15 : 8].bf8, scale.u8);\nD0[15 : 0].f16 = tmp0;\nD0[31 : 16].f16 = tmp1",
VOP3Op.V_CVT_SCALEF32_F16_FP8: "scale = 32'U(exponent(S1.f32));\nsrcbyte = OPSEL[1 : 0].i32 * 8;\nsrc = VGPR[laneId][SRC0.u32][srcbyte + 7 : srcbyte].fp8;\ntmp = fp8_to_f16_scale(src, scale.u8);\n// OPSEL[3] controls destination hi/lo\nD0 = tmp.b32",
VOP3Op.V_CVT_SCALEF32_F16_BF8: "scale = 32'U(exponent(S1.f32));\nsrcbyte = OPSEL[1 : 0].i32 * 8;\nsrc = VGPR[laneId][SRC0.u32][srcbyte + 7 : srcbyte].bf8;\ntmp = bf8_to_f16_scale(src, scale.u8);\n// OPSEL[3] controls destination hi/lo\nD0 = tmp.b32",
VOP3Op.V_CVT_SCALEF32_PK_FP4_F16: "scale = 32'U(exponent(S1.f32));\ntmp0 = f16_to_fp4_scale(S0[15 : 0].f16, scale.u8);\ntmp1 = f16_to_fp4_scale(S0[31 : 16].f16, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].b8 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_FP4_BF16: "scale = 32'U(exponent(S1.f32));\ntmp0 = bf16_to_fp4_scale(S0[15 : 0].bf16, scale.u8);\ntmp1 = bf16_to_fp4_scale(S0[31 : 16].bf16, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].b8 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_PK_FP4_F16: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ntmp0 = f16_to_fp4_sr_scale(S0[15 : 0].f16, randomVal, scale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32));\ntmp1 = f16_to_fp4_sr_scale(S0[31 : 16].f16, randomVal, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].b8 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_SR_PK_FP4_BF16: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ntmp0 = bf16_to_fp4_sr_scale(S0[15 : 0].bf16, randomVal, scale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32));\ntmp1 = bf16_to_fp4_sr_scale(S0[31 : 16].bf16, randomVal, scale.u8);\ndstbyte = OPSEL[3 : 2].i32 * 8;\nVGPR[laneId][VDST.u32][dstbyte + 7 : dstbyte].b8 = { tmp1, tmp0 };\n// Other destination bits are preserved",
VOP3Op.V_CVT_SCALEF32_PK_F16_FP4: "scale = 32'U(exponent(S1.f32));\nsrcbyte = OPSEL[1 : 0].i32 * 8;\nsrc = VGPR[laneId][SRC0.u32][srcbyte + 7 : srcbyte].b8;\ntmp0 = fp4_to_f16_scale(src[3 : 0].fp4, scale.u8);\ntmp1 = fp4_to_f16_scale(src[7 : 4].fp4, scale.u8);\nD0[15 : 0].f16 = tmp0;\nD0[31 : 16].f16 = tmp1",
VOP3Op.V_CVT_SCALEF32_PK_BF16_FP4: "scale = 32'U(exponent(S1.f32));\nsrcbyte = OPSEL[1 : 0].i32 * 8;\nsrc = VGPR[laneId][SRC0.u32][srcbyte + 7 : srcbyte].b8;\ntmp0 = fp4_to_bf16_scale(src[3 : 0].fp4, scale.u8);\ntmp1 = fp4_to_bf16_scale(src[7 : 4].fp4, scale.u8);\nD0[15 : 0].bf16 = tmp0;\nD0[31 : 16].bf16 = tmp1",
VOP3Op.V_CVT_SCALEF32_2XPK16_FP6_F32: "scale = 32'U(exponent(S2.f32));\ndeclare tmp : 192'B;\nfor pass in 0 : 15 do\ndOffset = pass * 12;\nsOffset = pass * 32;\n// Note that S0 and S1 inputs are interleaved in the packed result.\ntmp[dOffset + 5 : dOffset].fp6 = f32_to_fp6_scale(S0[sOffset + 31 : sOffset].f32, scale.u8);\ntmp[dOffset + 11 : dOffset + 6].fp6 = f32_to_fp6_scale(S1[sOffset + 31 : sOffset].f32, scale.u8)\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_2XPK16_BF6_F32: "scale = 32'U(exponent(S2.f32));\ndeclare tmp : 192'B;\nfor pass in 0 : 15 do\ndOffset = pass * 12;\nsOffset = pass * 32;\n// Note that S0 and S1 inputs are interleaved in the packed result.\ntmp[dOffset + 5 : dOffset].bf6 = f32_to_bf6_scale(S0[sOffset + 31 : sOffset].f32, scale.u8);\ntmp[dOffset + 11 : dOffset + 6].bf6 = f32_to_bf6_scale(S1[sOffset + 31 : sOffset].f32, scale.u8)\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_SR_PK32_FP6_F32: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 32;\ntmp[dOffset + 5 : dOffset].fp6 = f32_to_fp6_sr_scale(S0[sOffset + 31 : sOffset].f32, randomVal,\nscale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32))\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_SR_PK32_BF6_F32: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 32;\ntmp[dOffset + 5 : dOffset].bf6 = f32_to_bf6_sr_scale(S0[sOffset + 31 : sOffset].f32, randomVal,\nscale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32))\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_PK32_F32_FP6: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 1024'B;\nfor pass in 0 : 31 do\ndOffset = pass * 32;\nsOffset = pass * 6;\ntmp[dOffset + 31 : dOffset].f32 = fp6_to_f32_scale(S0[sOffset + 5 : sOffset].fp6, scale.u8)\nendfor;\nD0[1023 : 0] = tmp.b1024",
VOP3Op.V_CVT_SCALEF32_PK32_F32_BF6: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 1024'B;\nfor pass in 0 : 31 do\ndOffset = pass * 32;\nsOffset = pass * 6;\ntmp[dOffset + 31 : dOffset].f32 = bf6_to_f32_scale(S0[sOffset + 5 : sOffset].bf6, scale.u8)\nendfor;\nD0[1023 : 0] = tmp.b1024",
VOP3Op.V_CVT_SCALEF32_PK32_FP6_F16: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].fp6 = f16_to_fp6_scale(S0[sOffset + 15 : sOffset].f16, scale.u8)\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_PK32_FP6_BF16: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].fp6 = bf16_to_fp6_scale(S0[sOffset + 15 : sOffset].bf16, scale.u8)\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_PK32_BF6_F16: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].bf6 = f16_to_bf6_scale(S0[sOffset + 15 : sOffset].f16, scale.u8)\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_PK32_BF6_BF16: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].bf6 = bf16_to_bf6_scale(S0[sOffset + 15 : sOffset].bf16, scale.u8)\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_SR_PK32_FP6_F16: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].fp6 = f16_to_fp6_sr_scale(S0[sOffset + 15 : sOffset].f16, randomVal,\nscale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32))\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_SR_PK32_FP6_BF16: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].fp6 = bf16_to_fp6_sr_scale(S0[sOffset + 15 : sOffset].bf16, randomVal,\nscale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32))\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_SR_PK32_BF6_F16: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].bf6 = f16_to_bf6_sr_scale(S0[sOffset + 15 : sOffset].f16, randomVal,\nscale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32))\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_SR_PK32_BF6_BF16: "scale = 32'U(exponent(S2.f32));\nrandomVal = S1.u32;\ndeclare tmp : 192'B;\nfor pass in 0 : 31 do\ndOffset = pass * 6;\nsOffset = pass * 16;\ntmp[dOffset + 5 : dOffset].bf6 = bf16_to_bf6_sr_scale(S0[sOffset + 15 : sOffset].bf16, randomVal,\nscale.u8);\nrandomVal = 32'U(v_prng_b32(randomVal.b32))\nendfor;\nD0[191 : 0] = tmp.b192",
VOP3Op.V_CVT_SCALEF32_PK32_F16_FP6: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 512'B;\nfor pass in 0 : 31 do\ndOffset = pass * 16;\nsOffset = pass * 6;\ntmp[dOffset + 15 : dOffset].f16 = fp6_to_f16_scale(S0[sOffset + 5 : sOffset].fp6, scale.u8)\nendfor;\nD0[511 : 0] = tmp.b512",
VOP3Op.V_CVT_SCALEF32_PK32_BF16_FP6: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 512'B;\nfor pass in 0 : 31 do\ndOffset = pass * 16;\nsOffset = pass * 6;\ntmp[dOffset + 15 : dOffset].bf16 = fp6_to_bf16_scale(S0[sOffset + 5 : sOffset].fp6, scale.u8)\nendfor;\nD0[511 : 0] = tmp.b512",
VOP3Op.V_CVT_SCALEF32_PK32_F16_BF6: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 512'B;\nfor pass in 0 : 31 do\ndOffset = pass * 16;\nsOffset = pass * 6;\ntmp[dOffset + 15 : dOffset].f16 = bf6_to_f16_scale(S0[sOffset + 5 : sOffset].bf6, scale.u8)\nendfor;\nD0[511 : 0] = tmp.b512",
VOP3Op.V_CVT_SCALEF32_PK32_BF16_BF6: "scale = 32'U(exponent(S1.f32));\ndeclare tmp : 512'B;\nfor pass in 0 : 31 do\ndOffset = pass * 16;\nsOffset = pass * 6;\ntmp[dOffset + 15 : dOffset].bf16 = bf6_to_bf16_scale(S0[sOffset + 5 : sOffset].bf6, scale.u8)\nendfor;\nD0[511 : 0] = tmp.b512",
VOP3Op.V_ASHR_PK_I8_I32: "SAT8 = lambda(n) (\nif n <= -128 then\nreturn 8'0x80\nelsif n >= 127 then\nreturn 8'0x7f\nelse\nreturn n[7 : 0].b8\nendif);\ndeclare tmp : 16'B;\ntmp[7 : 0] = SAT8(S0.i32 >> S2[4 : 0].u32);\ntmp[15 : 8] = SAT8(S1.i32 >> S2[4 : 0].u32);\nD0[15 : 0] = tmp",
VOP3Op.V_ASHR_PK_U8_I32: "SAT8 = lambda(n) (\nif n <= 0 then\nreturn 8'0x0\nelsif n >= 255 then\nreturn 8'0xff\nelse\nreturn n[7 : 0].b8\nendif);\ndeclare tmp : 16'B;\ntmp[7 : 0] = SAT8(S0.i32 >> S2[4 : 0].u32);\ntmp[15 : 8] = SAT8(S1.i32 >> S2[4 : 0].u32);\nD0[15 : 0] = tmp",
VOP3Op.V_CVT_PK_F16_F32: 'prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\ntmp[15 : 0].f16 = f32_to_f16(S0.f32);\ntmp[31 : 16].f16 = f32_to_f16(S1.f32);\nD0 = tmp.b32;\nROUND_MODE = prev_mode',
VOP3Op.V_CVT_PK_BF16_F32: 'prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\ntmp[15 : 0].bf16 = f32_to_bf16(S0.f32);\ntmp[31 : 16].bf16 = f32_to_bf16(S1.f32);\nD0 = tmp.b32;\nROUND_MODE = prev_mode',
VOP3Op.V_CVT_SCALEF32_PK_BF16_FP8: "scale = 32'U(exponent(S1.f32));\nsrcword = OPSEL[0].i32 * 16;\nsrc = VGPR[laneId][SRC0.u32][srcword + 15 : srcword].b16;\ntmp0 = fp8_to_bf16_scale(src[7 : 0].fp8, scale);\ntmp1 = fp8_to_bf16_scale(src[15 : 8].fp8, scale);\nD0[15 : 0].bf16 = tmp0.bf16;\nD0[31 : 16].bf16 = tmp1.bf16",
VOP3Op.V_CVT_SCALEF32_PK_BF16_BF8: "scale = 32'U(exponent(S1.f32));\nsrcword = OPSEL[0].i32 * 16;\nsrc = VGPR[laneId][SRC0.u32][srcword + 15 : srcword].b16;\ntmp0 = bf8_to_bf16_scale(src[7 : 0].bf8, scale);\ntmp1 = bf8_to_bf16_scale(src[15 : 8].bf8, scale);\nD0[15 : 0].bf16 = tmp0.bf16;\nD0[31 : 16].bf16 = tmp1.bf16",
VOP3Op.V_ADD_F64: 'D0.f64 = S0.f64 + S1.f64',
VOP3Op.V_MUL_F64: 'D0.f64 = S0.f64 * S1.f64',
VOP3Op.V_MIN_F64: "if (WAVE_MODE.IEEE && isSignalNAN(S0.f64)) then\nD0.f64 = cvtToQuietNAN(S0.f64)\nelsif (WAVE_MODE.IEEE && isSignalNAN(S1.f64)) then\nD0.f64 = cvtToQuietNAN(S1.f64)\nelsif isNAN(S0.f64) then\nD0.f64 = S1.f64\nelsif isNAN(S1.f64) then\nD0.f64 = S0.f64\nelsif ((S0.f64 == +0.0) && (S1.f64 == -0.0)) then\nD0.f64 = S1.f64\nelsif ((S0.f64 == -0.0) && (S1.f64 == +0.0)) then\nD0.f64 = S0.f64\nelse\n// Note: there's no IEEE case here like there is for V_MAX_F64.\nD0.f64 = S0.f64 < S1.f64 ? S0.f64 : S1.f64\nendif",
VOP3Op.V_MAX_F64: 'if (WAVE_MODE.IEEE && isSignalNAN(S0.f64)) then\nD0.f64 = cvtToQuietNAN(S0.f64)\nelsif (WAVE_MODE.IEEE && isSignalNAN(S1.f64)) then\nD0.f64 = cvtToQuietNAN(S1.f64)\nelsif isNAN(S0.f64) then\nD0.f64 = S1.f64\nelsif isNAN(S1.f64) then\nD0.f64 = S0.f64\nelsif ((S0.f64 == +0.0) && (S1.f64 == -0.0)) then\nD0.f64 = S0.f64\nelsif ((S0.f64 == -0.0) && (S1.f64 == +0.0)) then\nD0.f64 = S1.f64\nelsif WAVE_MODE.IEEE then\nD0.f64 = S0.f64 >= S1.f64 ? S0.f64 : S1.f64\nelse\nD0.f64 = S0.f64 > S1.f64 ? S0.f64 : S1.f64\nendif',
VOP3Op.V_LDEXP_F64: 'D0.f64 = S0.f64 * 2.0 ** S1.i32',
VOP3Op.V_MUL_LO_U32: 'D0.u32 = S0.u32 * S1.u32',
VOP3Op.V_MUL_HI_U32: "D0.u32 = 32'U((64'U(S0.u32) * 64'U(S1.u32)) >> 32U)",
VOP3Op.V_MUL_HI_I32: "D0.i32 = 32'I((64'I(S0.i32) * 64'I(S1.i32)) >> 32U)",
VOP3Op.V_LDEXP_F32: 'D0.f32 = S0.f32 * 2.0F ** S1.i32',
VOP3Op.V_READLANE_B32: 'lane = S1.u32[5 : 0];\n// Lane select\nD0.b32 = VGPR[lane][SRC0.u32]',
VOP3Op.V_WRITELANE_B32: 'lane = S1.u32[5 : 0];\n// Lane select\nVGPR[lane][VDST.u32] = S0.b32',
VOP3Op.V_BCNT_U32_B32: "tmp = S1.u32;\nfor i in 0 : 31 do\ntmp += S0[i].u32;\n// count i'th bit\nendfor;\nD0.u32 = tmp",
VOP3Op.V_MBCNT_LO_U32_B32: "ThreadMask = (1LL << laneId.u32) - 1LL;\nMaskedValue = (S0.u32 & ThreadMask[31 : 0].u32);\ntmp = S1.u32;\nfor i in 0 : 31 do\ntmp += MaskedValue[i] == 1'1U ? 1U : 0U\nendfor;\nD0.u32 = tmp",
VOP3Op.V_MBCNT_HI_U32_B32: "ThreadMask = (1LL << laneId.u32) - 1LL;\nMaskedValue = (S0.u32 & ThreadMask[63 : 32].u32);\ntmp = S1.u32;\nfor i in 0 : 31 do\ntmp += MaskedValue[i] == 1'1U ? 1U : 0U\nendfor;\nD0.u32 = tmp",
VOP3Op.V_LSHLREV_B64: 'D0.u64 = (S1.u64 << S0[5 : 0].u32)',
VOP3Op.V_LSHRREV_B64: 'D0.u64 = (S1.u64 >> S0[5 : 0].u32)',
VOP3Op.V_ASHRREV_I64: 'D0.i64 = (S1.i64 >> S0[5 : 0].u32)',
VOP3Op.V_TRIG_PREOP_F64: "shift = 32'I(S1[4 : 0].u32) * 53;\nif exponent(S0.f64) > 1077 then\nshift += exponent(S0.f64) - 1077\nendif;\n// (2.0/PI) == 0.{b_1200, b_1199, b_1198, ..., b_1, b_0}\n// b_1200 is the MSB of the fractional part of 2.0/PI\n// Left shift operation indicates which bits are brought\n// into the whole part of the number.\n// Only whole part of result is kept.\nresult = 64'F((1201'B(2.0 / PI)[1200 : 0] << shift.u32) & 1201'0x1fffffffffffff);\nscale = -53 - shift;\nif exponent(S0.f64) >= 1968 then\nscale += 128\nendif;\nD0.f64 = ldexp(result, scale)",
VOP3Op.V_BFM_B32: 'D0.u32 = (((1U << S0[4 : 0].u32) - 1U) << S1[4 : 0].u32)',
VOP3Op.V_CVT_PKNORM_I16_F32: "declare tmp : 32'B;\ntmp[15 : 0].i16 = f32_to_snorm(S0.f32);\ntmp[31 : 16].i16 = f32_to_snorm(S1.f32);\nD0 = tmp.b32",
VOP3Op.V_CVT_PKNORM_U16_F32: "declare tmp : 32'B;\ntmp[15 : 0].u16 = f32_to_unorm(S0.f32);\ntmp[31 : 16].u16 = f32_to_unorm(S1.f32);\nD0 = tmp.b32",
VOP3Op.V_CVT_PKRTZ_F16_F32: 'prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_TOWARD_ZERO;\ntmp[15 : 0].f16 = f32_to_f16(S0.f32);\ntmp[31 : 16].f16 = f32_to_f16(S1.f32);\nD0 = tmp.b32;\nROUND_MODE = prev_mode;\n// Round-toward-zero regardless of current round mode setting in hardware.',
VOP3Op.V_CVT_PK_U16_U32: "declare tmp : 32'B;\ntmp[15 : 0].u16 = u32_to_u16(S0.u32);\ntmp[31 : 16].u16 = u32_to_u16(S1.u32);\nD0 = tmp.b32",
VOP3Op.V_CVT_PK_I16_I32: "declare tmp : 32'B;\ntmp[15 : 0].i16 = i32_to_i16(S0.i32);\ntmp[31 : 16].i16 = i32_to_i16(S1.i32);\nD0 = tmp.b32",
VOP3Op.V_CVT_PKNORM_I16_F16: "declare tmp : 32'B;\ntmp[15 : 0].i16 = f16_to_snorm(S0.f16);\ntmp[31 : 16].i16 = f16_to_snorm(S1.f16);\nD0 = tmp.b32",
VOP3Op.V_CVT_PKNORM_U16_F16: "declare tmp : 32'B;\ntmp[15 : 0].u16 = f16_to_unorm(S0.f16);\ntmp[31 : 16].u16 = f16_to_unorm(S1.f16);\nD0 = tmp.b32",
VOP3Op.V_ADD_I32: 'D0.i32 = S0.i32 + S1.i32',
VOP3Op.V_SUB_I32: 'D0.i32 = S0.i32 - S1.i32',
VOP3Op.V_ADD_I16: 'D0.i16 = S0.i16 + S1.i16',
VOP3Op.V_SUB_I16: 'D0.i16 = S0.i16 - S1.i16',
VOP3Op.V_PACK_B32_F16: 'D0[31 : 16].f16 = S1.f16;\nD0[15 : 0].f16 = S0.f16',
VOP3Op.V_MUL_LEGACY_F32: "if ((64'F(S0.f32) == 0.0) || (64'F(S1.f32) == 0.0)) then\n// DX9 rules, 0.0 * x = 0.0\nD0.f32 = 0.0F\nelse\nD0.f32 = S0.f32 * S1.f32\nendif",
VOP3Op.V_CVT_PK_FP8_F32: 'prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\nif OPSEL[3].u32 == 0U then\nVGPR[laneId][VDST.u32][15 : 0].b16 = { f32_to_fp8(S1.f32), f32_to_fp8(S0.f32) };\n// D0[31:16] are preserved\nelse\nVGPR[laneId][VDST.u32][31 : 16].b16 = { f32_to_fp8(S1.f32), f32_to_fp8(S0.f32) };\n// D0[15:0] are preserved\nendif;\nROUND_MODE = prev_mode',
VOP3Op.V_CVT_PK_BF8_F32: 'prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\nif OPSEL[3].u32 == 0U then\nVGPR[laneId][VDST.u32][15 : 0].b16 = { f32_to_bf8(S1.f32), f32_to_bf8(S0.f32) };\n// D0[31:16] are preserved\nelse\nVGPR[laneId][VDST.u32][31 : 16].b16 = { f32_to_bf8(S1.f32), f32_to_bf8(S0.f32) };\n// D0[15:0] are preserved\nendif;\nROUND_MODE = prev_mode',
VOP3Op.V_CVT_SR_FP8_F32: "prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\ns = sign(S0.f32);\ne = exponent(S0.f32);\nm = 23'U(32'U(23'B(mantissa(S0.f32))) + S1[31 : 12].u32);\ntmp = float32(s, e, m);\n// Add stochastic value to mantissa, wrap around on overflow\nif OPSEL[3 : 2].u2 == 2'0U then\nVGPR[laneId][VDST.u32][7 : 0].fp8 = f32_to_fp8(tmp.f32)\nelsif OPSEL[3 : 2].u2 == 2'1U then\nVGPR[laneId][VDST.u32][15 : 8].fp8 = f32_to_fp8(tmp.f32)\nelsif OPSEL[3 : 2].u2 == 2'2U then\nVGPR[laneId][VDST.u32][23 : 16].fp8 = f32_to_fp8(tmp.f32)\nelse\nVGPR[laneId][VDST.u32][31 : 24].fp8 = f32_to_fp8(tmp.f32)\nendif;\n// Unwritten bytes of D are preserved.\nROUND_MODE = prev_mode",
VOP3Op.V_CVT_SR_BF8_F32: "prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\ns = sign(S0.f32);\ne = exponent(S0.f32);\nm = 23'U(32'U(23'B(mantissa(S0.f32))) + S1[31 : 11].u32);\ntmp = float32(s, e, m);\n// Add stochastic value to mantissa, wrap around on overflow\nif OPSEL[3 : 2].u2 == 2'0U then\nVGPR[laneId][VDST.u32][7 : 0].bf8 = f32_to_bf8(tmp.f32)\nelsif OPSEL[3 : 2].u2 == 2'1U then\nVGPR[laneId][VDST.u32][15 : 8].bf8 = f32_to_bf8(tmp.f32)\nelsif OPSEL[3 : 2].u2 == 2'2U then\nVGPR[laneId][VDST.u32][23 : 16].bf8 = f32_to_bf8(tmp.f32)\nelse\nVGPR[laneId][VDST.u32][31 : 24].bf8 = f32_to_bf8(tmp.f32)\nendif;\n// Unwritten bytes of D are preserved.\nROUND_MODE = prev_mode",
VOP3Op.V_CVT_SR_F16_F32: "prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\nif OPSEL[3].u2 == 2'0U then\nVGPR[laneId][VDST.u32][15 : 0].f16 = 16'F(f32_to_f16_SR(S0.f32, S1.u32))\nelse\nVGPR[laneId][VDST.u32][31 : 16].f16 = 16'F(f32_to_f16_sr(S0.f32, S1.u32))\nendif;\n// Unwritten bytes of D are preserved.\nROUND_MODE = prev_mode",
VOP3Op.V_CVT_SR_BF16_F32: "prev_mode = ROUND_MODE;\nROUND_MODE = ROUND_NEAREST_EVEN;\nif OPSEL[3].u2 == 2'0U then\nVGPR[laneId][VDST.u32][15 : 0].bf16 = 16'BF(f32_to_bf16_SR(S0.f32, S1.u32))\nelse\nVGPR[laneId][VDST.u32][31 : 16].bf16 = 16'BF(f32_to_bf16_sr(S0.f32, S1.u32))\nendif;\n// Unwritten bytes of D are preserved.\nROUND_MODE = prev_mode",
VOP3Op.V_MINIMUM3_F32: "D0.f32 = 32'F(v_minimum_f32(v_minimum_f32(S0.f32, S1.f32), S2.f32))",
VOP3Op.V_MAXIMUM3_F32: "D0.f32 = 32'F(v_maximum_f32(v_maximum_f32(S0.f32, S1.f32), S2.f32))",
VOP3POp.V_PK_MAD_I16: "declare tmp : 32'B;\ntmp[15 : 0].i16 = S0[15 : 0].i16 * S1[15 : 0].i16 + S2[15 : 0].i16;\ntmp[31 : 16].i16 = S0[31 : 16].i16 * S1[31 : 16].i16 + S2[31 : 16].i16;\nD0.b32 = tmp",
VOP3POp.V_PK_MUL_LO_U16: 'tmp[31 : 16].u16 = S0[31 : 16].u16 * S1[31 : 16].u16;\ntmp[15 : 0].u16 = S0[15 : 0].u16 * S1[15 : 0].u16;\nD0.b32 = tmp.b32',
VOP3POp.V_PK_ADD_I16: "declare tmp : 32'B;\ntmp[15 : 0].i16 = S0[15 : 0].i16 + S1[15 : 0].i16;\ntmp[31 : 16].i16 = S0[31 : 16].i16 + S1[31 : 16].i16;\nD0.b32 = tmp",
VOP3POp.V_PK_SUB_I16: "declare tmp : 32'B;\ntmp[15 : 0].i16 = S0[15 : 0].i16 - S1[15 : 0].i16;\ntmp[31 : 16].i16 = S0[31 : 16].i16 - S1[31 : 16].i16;\nD0.b32 = tmp",
VOP3POp.V_PK_LSHLREV_B16: 'tmp[31 : 16].u16 = (S1[31 : 16].u16 << S0.u32[19 : 16].u32);\ntmp[15 : 0].u16 = (S1[15 : 0].u16 << S0.u32[3 : 0].u32);\nD0.b32 = tmp.b32',
VOP3POp.V_PK_LSHRREV_B16: 'tmp[31 : 16].u16 = (S1[31 : 16].u16 >> S0.u32[19 : 16].u32);\ntmp[15 : 0].u16 = (S1[15 : 0].u16 >> S0.u32[3 : 0].u32);\nD0.b32 = tmp.b32',
VOP3POp.V_PK_ASHRREV_I16: 'tmp[31 : 16].i16 = (S1[31 : 16].i16 >> S0.u32[19 : 16].u32);\ntmp[15 : 0].i16 = (S1[15 : 0].i16 >> S0.u32[3 : 0].u32);\nD0.b32 = tmp.b32',
VOP3POp.V_PK_MAX_I16: "declare tmp : 32'B;\ntmp[15 : 0].i16 = S0[15 : 0].i16 >= S1[15 : 0].i16 ? S0[15 : 0].i16 : S1[15 : 0].i16;\ntmp[31 : 16].i16 = S0[31 : 16].i16 >= S1[31 : 16].i16 ? S0[31 : 16].i16 : S1[31 : 16].i16;\nD0.b32 = tmp",
VOP3POp.V_PK_MIN_I16: "declare tmp : 32'B;\ntmp[15 : 0].i16 = S0[15 : 0].i16 < S1[15 : 0].i16 ? S0[15 : 0].i16 : S1[15 : 0].i16;\ntmp[31 : 16].i16 = S0[31 : 16].i16 < S1[31 : 16].i16 ? S0[31 : 16].i16 : S1[31 : 16].i16;\nD0.b32 = tmp",
VOP3POp.V_PK_MAD_U16: "declare tmp : 32'B;\ntmp[15 : 0].u16 = S0[15 : 0].u16 * S1[15 : 0].u16 + S2[15 : 0].u16;\ntmp[31 : 16].u16 = S0[31 : 16].u16 * S1[31 : 16].u16 + S2[31 : 16].u16;\nD0.b32 = tmp",
VOP3POp.V_PK_ADD_U16: "declare tmp : 32'B;\ntmp[15 : 0].u16 = S0[15 : 0].u16 + S1[15 : 0].u16;\ntmp[31 : 16].u16 = S0[31 : 16].u16 + S1[31 : 16].u16;\nD0.b32 = tmp",
VOP3POp.V_PK_SUB_U16: "declare tmp : 32'B;\ntmp[15 : 0].u16 = S0[15 : 0].u16 - S1[15 : 0].u16;\ntmp[31 : 16].u16 = S0[31 : 16].u16 - S1[31 : 16].u16;\nD0.b32 = tmp",
VOP3POp.V_PK_MAX_U16: "declare tmp : 32'B;\ntmp[15 : 0].u16 = S0[15 : 0].u16 >= S1[15 : 0].u16 ? S0[15 : 0].u16 : S1[15 : 0].u16;\ntmp[31 : 16].u16 = S0[31 : 16].u16 >= S1[31 : 16].u16 ? S0[31 : 16].u16 : S1[31 : 16].u16;\nD0.b32 = tmp",
VOP3POp.V_PK_MIN_U16: "declare tmp : 32'B;\ntmp[15 : 0].u16 = S0[15 : 0].u16 < S1[15 : 0].u16 ? S0[15 : 0].u16 : S1[15 : 0].u16;\ntmp[31 : 16].u16 = S0[31 : 16].u16 < S1[31 : 16].u16 ? S0[31 : 16].u16 : S1[31 : 16].u16;\nD0.b32 = tmp",
VOP3POp.V_PK_FMA_F16: "declare tmp : 32'B;\ntmp[15 : 0].f16 = fma(S0[15 : 0].f16, S1[15 : 0].f16, S2[15 : 0].f16);\ntmp[31 : 16].f16 = fma(S0[31 : 16].f16, S1[31 : 16].f16, S2[31 : 16].f16);\nD0.b32 = tmp",
VOP3POp.V_PK_ADD_F16: "declare tmp : 32'B;\ntmp[15 : 0].f16 = S0[15 : 0].f16 + S1[15 : 0].f16;\ntmp[31 : 16].f16 = S0[31 : 16].f16 + S1[31 : 16].f16;\nD0.b32 = tmp",
VOP3POp.V_PK_MUL_F16: "declare tmp : 32'B;\ntmp[15 : 0].f16 = S0[15 : 0].f16 * S1[15 : 0].f16;\ntmp[31 : 16].f16 = S0[31 : 16].f16 * S1[31 : 16].f16;\nD0.b32 = tmp",
VOP3POp.V_PK_MIN_F16: "declare tmp : 32'B;\ntmp[15 : 0].f16 = v_min_f16(S0[15 : 0].f16, S1[15 : 0].f16);\ntmp[31 : 16].f16 = v_min_f16(S0[31 : 16].f16, S1[31 : 16].f16);\nD0.b32 = tmp",
VOP3POp.V_PK_MAX_F16: "declare tmp : 32'B;\ntmp[15 : 0].f16 = v_max_f16(S0[15 : 0].f16, S1[15 : 0].f16);\ntmp[31 : 16].f16 = v_max_f16(S0[31 : 16].f16, S1[31 : 16].f16);\nD0.b32 = tmp",
VOP3POp.V_DOT2_F32_BF16: "tmp = 32'F(S0[15 : 0].bf16) * 32'F(S1[15 : 0].bf16);\ntmp += 32'F(S0[31 : 16].bf16) * 32'F(S1[31 : 16].bf16);\ntmp += S2.f32;\nD0.f32 = tmp",
VOP3POp.V_PK_MINIMUM3_F16: "tmp[31 : 16].f16 = 16'F(v_minimum3_f16(S0[31 : 16].f16, S1[31 : 16].f16, S2[31 : 16].f16));\ntmp[15 : 0].f16 = 16'F(v_minimum3_f16(S0[15 : 0].f16, S1[15 : 0].f16, S2[15 : 0].f16));\nD0.b32 = tmp.b32",
VOP3POp.V_PK_MAXIMUM3_F16: "tmp[31 : 16].f16 = 16'F(v_maximum3_f16(S0[31 : 16].f16, S1[31 : 16].f16, S2[31 : 16].f16));\ntmp[15 : 0].f16 = 16'F(v_maximum3_f16(S0[15 : 0].f16, S1[15 : 0].f16, S2[15 : 0].f16));\nD0.b32 = tmp.b32",
VOP3POp.V_MAD_MIX_F32: "declare in : 32'F[3];\ndeclare S : 32'B[3];\nfor i in 0 : 2 do\nif !OPSEL_HI.u3[i] then\nin[i] = S[i].f32\nelsif OPSEL.u3[i] then\nin[i] = f16_to_f32(S[i][31 : 16].f16)\nelse\nin[i] = f16_to_f32(S[i][15 : 0].f16)\nendif\nendfor;\nD0[31 : 0].f32 = in[0] * in[1] + in[2]",
VOP3POp.V_MAD_MIXLO_F16: "declare in : 32'F[3];\ndeclare S : 32'B[3];\nfor i in 0 : 2 do\nif !OPSEL_HI.u3[i] then\nin[i] = S[i].f32\nelsif OPSEL.u3[i] then\nin[i] = f16_to_f32(S[i][31 : 16].f16)\nelse\nin[i] = f16_to_f32(S[i][15 : 0].f16)\nendif\nendfor;\nD0[15 : 0].f16 = f32_to_f16(in[0] * in[1] + in[2])",
VOP3POp.V_MAD_MIXHI_F16: "declare in : 32'F[3];\ndeclare S : 32'B[3];\nfor i in 0 : 2 do\nif !OPSEL_HI.u3[i] then\nin[i] = S[i].f32\nelsif OPSEL.u3[i] then\nin[i] = f16_to_f32(S[i][31 : 16].f16)\nelse\nin[i] = f16_to_f32(S[i][15 : 0].f16)\nendif\nendfor;\nD0[31 : 16].f16 = f32_to_f16(in[0] * in[1] + in[2])",
VOP3POp.V_DOT2_F32_F16: 'tmp = S2.f32;\ntmp += f16_to_f32(S0[15 : 0].f16) * f16_to_f32(S1[15 : 0].f16);\ntmp += f16_to_f32(S0[31 : 16].f16) * f16_to_f32(S1[31 : 16].f16);\nD0.f32 = tmp',
VOP3POp.V_DOT2_I32_I16: 'tmp = S2.i32;\ntmp += i16_to_i32(S0[15 : 0].i16) * i16_to_i32(S1[15 : 0].i16);\ntmp += i16_to_i32(S0[31 : 16].i16) * i16_to_i32(S1[31 : 16].i16);\nD0.i32 = tmp',
VOP3POp.V_DOT2_U32_U16: 'tmp = S2.u32;\ntmp += u16_to_u32(S0[15 : 0].u16) * u16_to_u32(S1[15 : 0].u16);\ntmp += u16_to_u32(S0[31 : 16].u16) * u16_to_u32(S1[31 : 16].u16);\nD0.u32 = tmp',
VOP3POp.V_DOT4_I32_I8: 'tmp = S2.i32;\ntmp += i8_to_i32(S0[7 : 0].i8) * i8_to_i32(S1[7 : 0].i8);\ntmp += i8_to_i32(S0[15 : 8].i8) * i8_to_i32(S1[15 : 8].i8);\ntmp += i8_to_i32(S0[23 : 16].i8) * i8_to_i32(S1[23 : 16].i8);\ntmp += i8_to_i32(S0[31 : 24].i8) * i8_to_i32(S1[31 : 24].i8);\nD0.i32 = tmp',
VOP3POp.V_DOT4_U32_U8: 'tmp = S2.u32;\ntmp += u8_to_u32(S0[7 : 0].u8) * u8_to_u32(S1[7 : 0].u8);\ntmp += u8_to_u32(S0[15 : 8].u8) * u8_to_u32(S1[15 : 8].u8);\ntmp += u8_to_u32(S0[23 : 16].u8) * u8_to_u32(S1[23 : 16].u8);\ntmp += u8_to_u32(S0[31 : 24].u8) * u8_to_u32(S1[31 : 24].u8);\nD0.u32 = tmp',
VOP3POp.V_DOT8_I32_I4: 'tmp = S2.i32;\ntmp += i4_to_i32(S0[3 : 0].i4) * i4_to_i32(S1[3 : 0].i4);\ntmp += i4_to_i32(S0[7 : 4].i4) * i4_to_i32(S1[7 : 4].i4);\ntmp += i4_to_i32(S0[11 : 8].i4) * i4_to_i32(S1[11 : 8].i4);\ntmp += i4_to_i32(S0[15 : 12].i4) * i4_to_i32(S1[15 : 12].i4);\ntmp += i4_to_i32(S0[19 : 16].i4) * i4_to_i32(S1[19 : 16].i4);\ntmp += i4_to_i32(S0[23 : 20].i4) * i4_to_i32(S1[23 : 20].i4);\ntmp += i4_to_i32(S0[27 : 24].i4) * i4_to_i32(S1[27 : 24].i4);\ntmp += i4_to_i32(S0[31 : 28].i4) * i4_to_i32(S1[31 : 28].i4);\nD0.i32 = tmp',
VOP3POp.V_DOT8_U32_U4: 'tmp = S2.u32;\ntmp += u4_to_u32(S0[3 : 0].u4) * u4_to_u32(S1[3 : 0].u4);\ntmp += u4_to_u32(S0[7 : 4].u4) * u4_to_u32(S1[7 : 4].u4);\ntmp += u4_to_u32(S0[11 : 8].u4) * u4_to_u32(S1[11 : 8].u4);\ntmp += u4_to_u32(S0[15 : 12].u4) * u4_to_u32(S1[15 : 12].u4);\ntmp += u4_to_u32(S0[19 : 16].u4) * u4_to_u32(S1[19 : 16].u4);\ntmp += u4_to_u32(S0[23 : 20].u4) * u4_to_u32(S1[23 : 20].u4);\ntmp += u4_to_u32(S0[27 : 24].u4) * u4_to_u32(S1[27 : 24].u4);\ntmp += u4_to_u32(S0[31 : 28].u4) * u4_to_u32(S1[31 : 28].u4);\nD0.u32 = tmp',
VOP3POp.V_MFMA_F32_16X16X128_F8F6F4: 'D = A (16x128) * B (128x16) + C (16x16)',
VOP3POp.V_MFMA_F32_32X32X64_F8F6F4: 'D = A (32x64) * B (64x32) + C (32x32)',
VOP3POp.V_PK_FMA_F32: "declare tmp : 64'B;\ntmp[31 : 0].f32 = fma(S0[31 : 0].f32, S1[31 : 0].f32, S2[31 : 0].f32);\ntmp[63 : 32].f32 = fma(S0[63 : 32].f32, S1[63 : 32].f32, S2[63 : 32].f32);\nD0.b64 = tmp",
VOP3POp.V_PK_MUL_F32: "declare tmp : 64'B;\ntmp[31 : 0].f32 = S0[31 : 0].f32 * S1[31 : 0].f32;\ntmp[63 : 32].f32 = S0[63 : 32].f32 * S1[63 : 32].f32;\nD0.b64 = tmp",
VOP3POp.V_PK_ADD_F32: "declare tmp : 64'B;\ntmp[31 : 0].f32 = S0[31 : 0].f32 + S1[31 : 0].f32;\ntmp[63 : 32].f32 = S0[63 : 32].f32 + S1[63 : 32].f32;\nD0.b64 = tmp",
VOP3POp.V_PK_MOV_B32: 'tmp0.u32 = S0.u32[OPSEL[0].i32 * 32 + 31 : OPSEL[0].i32 * 32];\ntmp1.u32 = S1.u32[OPSEL[1].i32 * 32 + 31 : OPSEL[1].i32 * 32];\nD0.u32[31 : 0] = tmp0.u32;\nD0.u32[63 : 32] = tmp1.u32',
VOP3POp.V_MFMA_F32_16X16X32_BF16: 'D = A (16x32) * B (32x16) + C (16x16)',
VOP3POp.V_MFMA_I32_16X16X64_I8: 'D = A (16x64) * B (64x16) + C (16x16)',
VOP3POp.V_MFMA_F32_32X32X16_BF16: 'D = A (32x16) * B (16x32) + C (32x32)',
VOP3POp.V_MFMA_I32_32X32X32_I8: 'D = A (32x32) * B (32x32) + C (32x32)',
VOP3POp.V_SMFMAC_F32_16X16X64_BF16: 'D = A (sparse 16x64) * B (64x16) + D (16x16)',
VOP3POp.V_SMFMAC_I32_16X16X128_I8: 'D = A (sparse 16x128) * B (128x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X128_BF8_BF8: 'D = A (sparse 16x128) * B (128x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X128_BF8_FP8: 'D = A (sparse 16x128) * B (128x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X128_FP8_BF8: 'D = A (sparse 16x128) * B (128x16) + D (16x16)',
VOP3POp.V_MFMA_F32_32X32X1_2B_F32: 'D = A (32x1) * B (1x32) + C (32x32)',
VOP3POp.V_MFMA_F32_16X16X1_4B_F32: 'D = A (16x1) * B (1x16) + C (16x16)',
VOP3POp.V_MFMA_F32_4X4X1_16B_F32: 'D = A (4x1) * B (1x4) + C (4x4)',
VOP3POp.V_SMFMAC_F32_16X16X128_FP8_FP8: 'D = A (sparse 16x128) * B (128x16) + D (16x16)',
VOP3POp.V_MFMA_F32_32X32X2_F32: 'D = A (32x2) * B (2x32) + C (32x32)',
VOP3POp.V_MFMA_F32_16X16X4_F32: 'D = A (16x4) * B (4x16) + C (16x16)',
VOP3POp.V_SMFMAC_F32_32X32X32_BF16: 'D = A (sparse 32x32) * B (32x32) + D (32x32)',
VOP3POp.V_SMFMAC_I32_32X32X64_I8: 'D = A (sparse 32x64) * B (64x32) + D (32x32)',
VOP3POp.V_MFMA_F32_32X32X4_2B_F16: 'D = A (32x4) * B (4x32) + C (32x32)',
VOP3POp.V_MFMA_F32_16X16X4_4B_F16: 'D = A (16x4) * B (4x16) + C (16x16)',
VOP3POp.V_MFMA_F32_4X4X4_16B_F16: 'D = A (4x4) * B (4x4) + C (4x4)',
VOP3POp.V_SMFMAC_F32_32X32X64_BF8_BF8: 'D = A (sparse 32x64) * B (64x32) + D (32x32)',
VOP3POp.V_MFMA_F32_32X32X8_F16: 'D = A (32x8) * B (8x32) + C (32x32)',
VOP3POp.V_MFMA_F32_16X16X16_F16: 'D = A (16x16) * B (16x16) + C (16x16)',
VOP3POp.V_SMFMAC_F32_32X32X64_BF8_FP8: 'D = A (sparse 32x64) * B (64x32) + D (32x32)',
VOP3POp.V_SMFMAC_F32_32X32X64_FP8_BF8: 'D = A (sparse 32x64) * B (64x32) + D (32x32)',
VOP3POp.V_MFMA_I32_32X32X4_2B_I8: 'D = A (32x4) * B (4x32) + C (32x32)',
VOP3POp.V_MFMA_I32_16X16X4_4B_I8: 'D = A (16x4) * B (4x16) + C (16x16)',
VOP3POp.V_MFMA_I32_4X4X4_16B_I8: 'D = A (4x4) * B (4x4) + C (4x4)',
VOP3POp.V_SMFMAC_F32_32X32X64_FP8_FP8: 'D = A (sparse 32x64) * B (64x32) + D (32x32)',
VOP3POp.V_MFMA_F32_16X16X32_F16: 'D = A (16x32) * B (32x16) + C (16x16)',
VOP3POp.V_MFMA_F32_32X32X16_F16: 'D = A (32x16) * B (16x32) + C (32x32)',
VOP3POp.V_MFMA_I32_32X32X16_I8: 'D = A (32x16) * B (16x32) + C (32x32)',
VOP3POp.V_MFMA_I32_16X16X32_I8: 'D = A (16x32) * B (32x16) + C (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X64_F16: 'D = A (sparse 16x64) * B (64x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_32X32X32_F16: 'D = A (sparse 32x32) * B (32x32) + D (32x32)',
VOP3POp.V_MFMA_F32_32X32X4_2B_BF16: 'D = A (32x4) * B (4x32) + C (32x32)',
VOP3POp.V_MFMA_F32_16X16X4_4B_BF16: 'D = A (16x4) * B (4x16) + C (16x16)',
VOP3POp.V_MFMA_F32_4X4X4_16B_BF16: 'D = A (4x4) * B (4x4) + C (4x4)',
VOP3POp.V_MFMA_F32_32X32X8_BF16: 'D = A (32x8) * B (8x32) + C (32x32)',
VOP3POp.V_MFMA_F32_16X16X16_BF16: 'D = A (16x16) * B (16x16) + C (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X32_F16: 'D = A (sparse 16x32) * B (32x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_32X32X16_F16: 'D = A (sparse 32x16) * B (16x32) + D (32x32)',
VOP3POp.V_SMFMAC_F32_16X16X32_BF16: 'D = A (sparse 16x32) * B (32x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_32X32X16_BF16: 'D = A (sparse 32x16) * B (16x32) + D (32x32)',
VOP3POp.V_SMFMAC_I32_16X16X64_I8: 'D = A (sparse 16x64) * B (64x16) + D (16x16)',
VOP3POp.V_SMFMAC_I32_32X32X32_I8: 'D = A (sparse 32x32) * B (32x32) + D (32x32)',
VOP3POp.V_MFMA_F64_16X16X4_F64: 'D = A (16x4) * B (4x16) + C (16x16)',
VOP3POp.V_MFMA_F64_4X4X4_4B_F64: 'D = A (4x4) * B (4x4) + C (4x4)',
VOP3POp.V_MFMA_F32_16X16X32_BF8_BF8: 'D = A (16x32) * B (32x16) + C (16x16)',
VOP3POp.V_MFMA_F32_16X16X32_BF8_FP8: 'D = A (16x32) * B (32x16) + C (16x16)',
VOP3POp.V_MFMA_F32_16X16X32_FP8_BF8: 'D = A (16x32) * B (32x16) + C (16x16)',
VOP3POp.V_MFMA_F32_16X16X32_FP8_FP8: 'D = A (16x32) * B (32x16) + C (16x16)',
VOP3POp.V_MFMA_F32_32X32X16_BF8_BF8: 'D = A (32x16) * B (16x32) + C (32x32)',
VOP3POp.V_MFMA_F32_32X32X16_BF8_FP8: 'D = A (32x16) * B (16x32) + C (32x32)',
VOP3POp.V_MFMA_F32_32X32X16_FP8_BF8: 'D = A (32x16) * B (16x32) + C (32x32)',
VOP3POp.V_MFMA_F32_32X32X16_FP8_FP8: 'D = A (32x16) * B (16x32) + C (32x32)',
VOP3POp.V_SMFMAC_F32_16X16X64_BF8_BF8: 'D = A (sparse 16x64) * B (64x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X64_BF8_FP8: 'D = A (sparse 16x64) * B (64x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X64_FP8_BF8: 'D = A (sparse 16x64) * B (64x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_16X16X64_FP8_FP8: 'D = A (sparse 16x64) * B (64x16) + D (16x16)',
VOP3POp.V_SMFMAC_F32_32X32X32_BF8_BF8: 'D = A (sparse 32x32) * B (32x32) + D (32x32)',
VOP3POp.V_SMFMAC_F32_32X32X32_BF8_FP8: 'D = A (sparse 32x32) * B (32x32) + D (32x32)',
VOP3POp.V_SMFMAC_F32_32X32X32_FP8_BF8: 'D = A (sparse 32x32) * B (32x32) + D (32x32)',
VOP3POp.V_SMFMAC_F32_32X32X32_FP8_FP8: 'D = A (sparse 32x32) * B (32x32) + D (32x32)',
VOP3SDOp.V_ADD_CO_U32: "tmp = 64'U(S0.u32) + 64'U(S1.u32);\nVCC.u64[laneId] = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_ADDC_CO_U32.\nD0.u32 = tmp.u32",
VOP3SDOp.V_SUB_CO_U32: "tmp = S0.u32 - S1.u32;\nVCC.u64[laneId] = S1.u32 > S0.u32 ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP3SDOp.V_SUBREV_CO_U32: "tmp = S1.u32 - S0.u32;\nVCC.u64[laneId] = S0.u32 > S1.u32 ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP3SDOp.V_ADDC_CO_U32: "tmp = 64'U(S0.u32) + 64'U(S1.u32) + VCC.u64[laneId].u64;\nVCC.u64[laneId] = tmp >= 0x100000000ULL ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_ADDC_CO_U32.\nD0.u32 = tmp.u32",
VOP3SDOp.V_SUBB_CO_U32: "tmp = S0.u32 - S1.u32 - VCC.u64[laneId].u32;\nVCC.u64[laneId] = 64'U(S1.u32) + VCC.u64[laneId].u64 > 64'U(S0.u32) ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP3SDOp.V_SUBBREV_CO_U32: "tmp = S1.u32 - S0.u32 - VCC.u64[laneId].u32;\nVCC.u64[laneId] = 64'U(S0.u32) + VCC.u64[laneId].u64 > 64'U(S1.u32) ? 1'1U : 1'0U;\n// VCC is an UNSIGNED overflow/carry-out for V_SUBB_CO_U32.\nD0.u32 = tmp.u32",
VOP3SDOp.V_DIV_SCALE_F32: "VCC = 0x0LL;\nif ((64'F(S2.f32) == 0.0) || (64'F(S1.f32) == 0.0)) then\nD0.f32 = NAN.f32\nelsif exponent(S2.f32) - exponent(S1.f32) >= 96 then\n// N/D near MAX_FLOAT_F32\nVCC = 0x1LL;\nif S0.f32 == S1.f32 then\n// Only scale the denominator\nD0.f32 = ldexp(S0.f32, 64)\nendif\nelsif S1.f32 == DENORM.f32 then\nD0.f32 = ldexp(S0.f32, 64)\nelsif ((1.0 / 64'F(S1.f32) == DENORM.f64) && (S2.f32 / S1.f32 == DENORM.f32)) then\nVCC = 0x1LL;\nif S0.f32 == S1.f32 then\n// Only scale the denominator\nD0.f32 = ldexp(S0.f32, 64)\nendif\nelsif 1.0 / 64'F(S1.f32) == DENORM.f64 then\nD0.f32 = ldexp(S0.f32, -64)\nelsif S2.f32 / S1.f32 == DENORM.f32 then\nVCC = 0x1LL;\nif S0.f32 == S2.f32 then\n// Only scale the numerator\nD0.f32 = ldexp(S0.f32, 64)\nendif\nelsif exponent(S2.f32) <= 23 then\n// Numerator is tiny\nD0.f32 = ldexp(S0.f32, 64)\nendif",
VOP3SDOp.V_DIV_SCALE_F64: 'VCC = 0x0LL;\nif ((S2.f64 == 0.0) || (S1.f64 == 0.0)) then\nD0.f64 = NAN.f64\nelsif exponent(S2.f64) - exponent(S1.f64) >= 768 then\n// N/D near MAX_FLOAT_F64\nVCC = 0x1LL;\nif S0.f64 == S1.f64 then\n// Only scale the denominator\nD0.f64 = ldexp(S0.f64, 128)\nendif\nelsif S1.f64 == DENORM.f64 then\nD0.f64 = ldexp(S0.f64, 128)\nelsif ((1.0 / S1.f64 == DENORM.f64) && (S2.f64 / S1.f64 == DENORM.f64)) then\nVCC = 0x1LL;\nif S0.f64 == S1.f64 then\n// Only scale the denominator\nD0.f64 = ldexp(S0.f64, 128)\nendif\nelsif 1.0 / S1.f64 == DENORM.f64 then\nD0.f64 = ldexp(S0.f64, -128)\nelsif S2.f64 / S1.f64 == DENORM.f64 then\nVCC = 0x1LL;\nif S0.f64 == S2.f64 then\n// Only scale the numerator\nD0.f64 = ldexp(S0.f64, 128)\nendif\nelsif exponent(S2.f64) <= 53 then\n// Numerator is tiny\nD0.f64 = ldexp(S0.f64, 128)\nendif',
VOP3SDOp.V_MAD_U64_U32: "{ D1.u1, D0.u64 } = 65'B(65'U(S0.u32) * 65'U(S1.u32) + 65'U(S2.u64))",
VOP3SDOp.V_MAD_I64_I32: "{ D1.i1, D0.i64 } = 65'B(65'I(S0.i32) * 65'I(S1.i32) + 65'I(S2.i64))",
VOPCOp.V_CMP_CLASS_F32_E32: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f32)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f32)) then\nresult = S1.u32[1]\nelsif exponent(S0.f32) == 255 then\n// +-INF\nresult = S1.u32[sign(S0.f32) ? 2 : 9]\nelsif exponent(S0.f32) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f32) ? 3 : 8]\nelsif 64'F(abs(S0.f32)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f32) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f32) ? 5 : 6]\nendif;\nD0.u64[laneId] = result;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_CLASS_F32_E32: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f32)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f32)) then\nresult = S1.u32[1]\nelsif exponent(S0.f32) == 255 then\n// +-INF\nresult = S1.u32[sign(S0.f32) ? 2 : 9]\nelsif exponent(S0.f32) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f32) ? 3 : 8]\nelsif 64'F(abs(S0.f32)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f32) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f32) ? 5 : 6]\nendif;\nEXEC.u64[laneId] = D0.u64[laneId] = result",
VOPCOp.V_CMP_CLASS_F64_E32: "declare result : 1'U;\nif isSignalNAN(S0.f64) then\nresult = S1.u32[0]\nelsif isQuietNAN(S0.f64) then\nresult = S1.u32[1]\nelsif exponent(S0.f64) == 2047 then\n// +-INF\nresult = S1.u32[sign(S0.f64) ? 2 : 9]\nelsif exponent(S0.f64) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f64) ? 3 : 8]\nelsif abs(S0.f64) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f64) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f64) ? 5 : 6]\nendif;\nD0.u64[laneId] = result;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_CLASS_F64_E32: "declare result : 1'U;\nif isSignalNAN(S0.f64) then\nresult = S1.u32[0]\nelsif isQuietNAN(S0.f64) then\nresult = S1.u32[1]\nelsif exponent(S0.f64) == 2047 then\n// +-INF\nresult = S1.u32[sign(S0.f64) ? 2 : 9]\nelsif exponent(S0.f64) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f64) ? 3 : 8]\nelsif abs(S0.f64) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f64) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f64) ? 5 : 6]\nendif;\nEXEC.u64[laneId] = D0.u64[laneId] = result",
VOPCOp.V_CMP_CLASS_F16_E32: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f16)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f16)) then\nresult = S1.u32[1]\nelsif exponent(S0.f16) == 31 then\n// +-INF\nresult = S1.u32[sign(S0.f16) ? 2 : 9]\nelsif exponent(S0.f16) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f16) ? 3 : 8]\nelsif 64'F(abs(S0.f16)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f16) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f16) ? 5 : 6]\nendif;\nD0.u64[laneId] = result;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_CLASS_F16_E32: "declare result : 1'U;\nif isSignalNAN(64'F(S0.f16)) then\nresult = S1.u32[0]\nelsif isQuietNAN(64'F(S0.f16)) then\nresult = S1.u32[1]\nelsif exponent(S0.f16) == 31 then\n// +-INF\nresult = S1.u32[sign(S0.f16) ? 2 : 9]\nelsif exponent(S0.f16) > 0 then\n// +-normal value\nresult = S1.u32[sign(S0.f16) ? 3 : 8]\nelsif 64'F(abs(S0.f16)) > 0.0 then\n// +-denormal value\nresult = S1.u32[sign(S0.f16) ? 4 : 7]\nelse\n// +-0.0\nresult = S1.u32[sign(S0.f16) ? 5 : 6]\nendif;\nEXEC.u64[laneId] = D0.u64[laneId] = result",
VOPCOp.V_CMP_F_F16_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_F16_E32: 'D0.u64[laneId] = S0.f16 < S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_F16_E32: 'D0.u64[laneId] = S0.f16 == S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_F16_E32: 'D0.u64[laneId] = S0.f16 <= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_F16_E32: 'D0.u64[laneId] = S0.f16 > S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LG_F16_E32: 'D0.u64[laneId] = S0.f16 <> S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_F16_E32: 'D0.u64[laneId] = S0.f16 >= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_O_F16_E32: "D0.u64[laneId] = (!isNAN(64'F(S0.f16)) && !isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_U_F16_E32: "D0.u64[laneId] = (isNAN(64'F(S0.f16)) || isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_NGE_F16_E32: 'D0.u64[laneId] = !(S0.f16 >= S1.f16);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLG_F16_E32: 'D0.u64[laneId] = !(S0.f16 <> S1.f16);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NGT_F16_E32: 'D0.u64[laneId] = !(S0.f16 > S1.f16);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLE_F16_E32: 'D0.u64[laneId] = !(S0.f16 <= S1.f16);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NEQ_F16_E32: 'D0.u64[laneId] = !(S0.f16 == S1.f16);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLT_F16_E32: 'D0.u64[laneId] = !(S0.f16 < S1.f16);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_TRU_F16_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_F16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 < S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 == S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 <= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 > S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LG_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 <> S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f16 >= S1.f16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_O_F16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = (!isNAN(64'F(S0.f16)) && !isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_U_F16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = (isNAN(64'F(S0.f16)) || isNAN(64'F(S1.f16)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_NGE_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 >= S1.f16);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLG_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 <> S1.f16);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NGT_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 > S1.f16);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLE_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 <= S1.f16);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NEQ_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 == S1.f16);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLT_F16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f16 < S1.f16);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_TRU_F16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_F32_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_F32_E32: 'D0.u64[laneId] = S0.f32 < S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_F32_E32: 'D0.u64[laneId] = S0.f32 == S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_F32_E32: 'D0.u64[laneId] = S0.f32 <= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_F32_E32: 'D0.u64[laneId] = S0.f32 > S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LG_F32_E32: 'D0.u64[laneId] = S0.f32 <> S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_F32_E32: 'D0.u64[laneId] = S0.f32 >= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_O_F32_E32: "D0.u64[laneId] = (!isNAN(64'F(S0.f32)) && !isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_U_F32_E32: "D0.u64[laneId] = (isNAN(64'F(S0.f32)) || isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_NGE_F32_E32: 'D0.u64[laneId] = !(S0.f32 >= S1.f32);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLG_F32_E32: 'D0.u64[laneId] = !(S0.f32 <> S1.f32);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NGT_F32_E32: 'D0.u64[laneId] = !(S0.f32 > S1.f32);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLE_F32_E32: 'D0.u64[laneId] = !(S0.f32 <= S1.f32);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NEQ_F32_E32: 'D0.u64[laneId] = !(S0.f32 == S1.f32);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLT_F32_E32: 'D0.u64[laneId] = !(S0.f32 < S1.f32);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_TRU_F32_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_F32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 < S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 == S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 <= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 > S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LG_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 <> S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f32 >= S1.f32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_O_F32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = (!isNAN(64'F(S0.f32)) && !isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_U_F32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = (isNAN(64'F(S0.f32)) || isNAN(64'F(S1.f32)));\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_NGE_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 >= S1.f32);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLG_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 <> S1.f32);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NGT_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 > S1.f32);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLE_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 <= S1.f32);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NEQ_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 == S1.f32);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLT_F32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f32 < S1.f32);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_TRU_F32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_F64_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_F64_E32: 'D0.u64[laneId] = S0.f64 < S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_F64_E32: 'D0.u64[laneId] = S0.f64 == S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_F64_E32: 'D0.u64[laneId] = S0.f64 <= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_F64_E32: 'D0.u64[laneId] = S0.f64 > S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LG_F64_E32: 'D0.u64[laneId] = S0.f64 <> S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_F64_E32: 'D0.u64[laneId] = S0.f64 >= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_O_F64_E32: 'D0.u64[laneId] = (!isNAN(S0.f64) && !isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_U_F64_E32: 'D0.u64[laneId] = (isNAN(S0.f64) || isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NGE_F64_E32: 'D0.u64[laneId] = !(S0.f64 >= S1.f64);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLG_F64_E32: 'D0.u64[laneId] = !(S0.f64 <> S1.f64);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NGT_F64_E32: 'D0.u64[laneId] = !(S0.f64 > S1.f64);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLE_F64_E32: 'D0.u64[laneId] = !(S0.f64 <= S1.f64);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NEQ_F64_E32: 'D0.u64[laneId] = !(S0.f64 == S1.f64);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NLT_F64_E32: 'D0.u64[laneId] = !(S0.f64 < S1.f64);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_TRU_F64_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_F64_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 < S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 == S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 <= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 > S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LG_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 <> S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.f64 >= S1.f64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_O_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = (!isNAN(S0.f64) && !isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_U_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = (isNAN(S0.f64) || isNAN(S1.f64));\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NGE_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 >= S1.f64);\n// With NAN inputs this is not the same operation as <\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLG_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 <> S1.f64);\n// With NAN inputs this is not the same operation as ==\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NGT_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 > S1.f64);\n// With NAN inputs this is not the same operation as <=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLE_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 <= S1.f64);\n// With NAN inputs this is not the same operation as >\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NEQ_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 == S1.f64);\n// With NAN inputs this is not the same operation as !=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NLT_F64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = !(S0.f64 < S1.f64);\n// With NAN inputs this is not the same operation as >=\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_TRU_F64_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_I16_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_I16_E32: 'D0.u64[laneId] = S0.i16 < S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_I16_E32: 'D0.u64[laneId] = S0.i16 == S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_I16_E32: 'D0.u64[laneId] = S0.i16 <= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_I16_E32: 'D0.u64[laneId] = S0.i16 > S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NE_I16_E32: 'D0.u64[laneId] = S0.i16 <> S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_I16_E32: 'D0.u64[laneId] = S0.i16 >= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_T_I16_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_U16_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_U16_E32: 'D0.u64[laneId] = S0.u16 < S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_U16_E32: 'D0.u64[laneId] = S0.u16 == S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_U16_E32: 'D0.u64[laneId] = S0.u16 <= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_U16_E32: 'D0.u64[laneId] = S0.u16 > S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NE_U16_E32: 'D0.u64[laneId] = S0.u16 <> S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_U16_E32: 'D0.u64[laneId] = S0.u16 >= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_T_U16_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_I16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_I16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 < S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_I16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 == S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_I16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 <= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_I16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 > S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NE_I16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 <> S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_I16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i16 >= S1.i16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_T_I16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_U16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_U16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 < S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_U16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 == S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_U16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 <= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_U16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 > S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NE_U16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 <> S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_U16_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u16 >= S1.u16;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_T_U16_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_I32_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_I32_E32: 'D0.u64[laneId] = S0.i32 < S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_I32_E32: 'D0.u64[laneId] = S0.i32 == S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_I32_E32: 'D0.u64[laneId] = S0.i32 <= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_I32_E32: 'D0.u64[laneId] = S0.i32 > S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NE_I32_E32: 'D0.u64[laneId] = S0.i32 <> S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_I32_E32: 'D0.u64[laneId] = S0.i32 >= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_T_I32_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_U32_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_U32_E32: 'D0.u64[laneId] = S0.u32 < S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_U32_E32: 'D0.u64[laneId] = S0.u32 == S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_U32_E32: 'D0.u64[laneId] = S0.u32 <= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_U32_E32: 'D0.u64[laneId] = S0.u32 > S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NE_U32_E32: 'D0.u64[laneId] = S0.u32 <> S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_U32_E32: 'D0.u64[laneId] = S0.u32 >= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_T_U32_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_I32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_I32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 < S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_I32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 == S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_I32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 <= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_I32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 > S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NE_I32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 <> S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_I32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i32 >= S1.i32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_T_I32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_U32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_U32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 < S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_U32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 == S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_U32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 <= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_U32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 > S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NE_U32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 <> S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_U32_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u32 >= S1.u32;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_T_U32_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_I64_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_I64_E32: 'D0.u64[laneId] = S0.i64 < S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_I64_E32: 'D0.u64[laneId] = S0.i64 == S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_I64_E32: 'D0.u64[laneId] = S0.i64 <= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_I64_E32: 'D0.u64[laneId] = S0.i64 > S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NE_I64_E32: 'D0.u64[laneId] = S0.i64 <> S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_I64_E32: 'D0.u64[laneId] = S0.i64 >= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_T_I64_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_F_U64_E32: "D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMP_LT_U64_E32: 'D0.u64[laneId] = S0.u64 < S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_EQ_U64_E32: 'D0.u64[laneId] = S0.u64 == S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_LE_U64_E32: 'D0.u64[laneId] = S0.u64 <= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GT_U64_E32: 'D0.u64[laneId] = S0.u64 > S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_NE_U64_E32: 'D0.u64[laneId] = S0.u64 <> S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_GE_U64_E32: 'D0.u64[laneId] = S0.u64 >= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMP_T_U64_E32: "D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_I64_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_I64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 < S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_I64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 == S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_I64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 <= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_I64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 > S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NE_I64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 <> S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_I64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.i64 >= S1.i64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_T_I64_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_F_U64_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'0U;\n// D0 = VCC in VOPC encoding.",
VOPCOp.V_CMPX_LT_U64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 < S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_EQ_U64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 == S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_LE_U64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 <= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GT_U64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 > S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_NE_U64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 <> S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_GE_U64_E32: 'EXEC.u64[laneId] = D0.u64[laneId] = S0.u64 >= S1.u64;\n// D0 = VCC in VOPC encoding.',
VOPCOp.V_CMPX_T_U64_E32: "EXEC.u64[laneId] = D0.u64[laneId] = 1'1U;\n// D0 = VCC in VOPC encoding.",
}
Reference in New Issue View Git Blame Copy Permalink