diff --git a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp index 7c293c1a5e512..06df08feda8fa 100644 --- a/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp +++ b/llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp @@ -345,6 +345,25 @@ static DecodeStatus decodeOperand_VSrcT16_Lo128(MCInst &Inst, unsigned Imm, (AMDGPU::OperandSemantics)OperandSemantics)); } +template +static DecodeStatus +decodeOperand_VSrcT16_Lo128_Deferred(MCInst &Inst, unsigned Imm, + uint64_t /*Addr*/, + const MCDisassembler *Decoder) { + const auto *DAsm = static_cast(Decoder); + assert(isUInt<9>(Imm) && "9-bit encoding expected"); + + if (Imm & AMDGPU::EncValues::IS_VGPR) { + bool IsHi = Imm & (1 << 7); + unsigned RegIdx = Imm & 0x7f; + return addOperand(Inst, DAsm->createVGPR16Operand(RegIdx, IsHi)); + } + return addOperand(Inst, DAsm->decodeNonVGPRSrcOp( + OpWidth, Imm & 0xFF, true, ImmWidth, + (AMDGPU::OperandSemantics)OperandSemantics)); +} + template static DecodeStatus decodeOperand_VSrcT16(MCInst &Inst, unsigned Imm, diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.td b/llvm/lib/Target/AMDGPU/SIRegisterInfo.td index d47ff9fe96c94..46b2b4a389200 100644 --- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.td +++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.td @@ -1161,7 +1161,8 @@ def SSrcOrLds_b32 : SrcRegOrImm9 : RegOrImmOperand { - let DecoderMethod = "decodeSrcRegOrImmDeferred9"; } @@ -1222,6 +1223,13 @@ def VSrc_bf16_Deferred : SrcRegOrImmDeferred9; def VSrc_f32_Deferred : SrcRegOrImmDeferred9; +// True 16 Operands +def VSrcT_f16_Lo128_Deferred : SrcRegOrImmDeferred9 { + let DecoderMethodName = "decodeOperand_VSrcT16_Lo128_Deferred"; + let EncoderMethod = "getMachineOpValueT16Lo128"; +} + def VSrcFake16_bf16_Lo128_Deferred : SrcRegOrImmDeferred9; def VSrcFake16_f16_Lo128_Deferred diff --git a/llvm/lib/Target/AMDGPU/VOP2Instructions.td b/llvm/lib/Target/AMDGPU/VOP2Instructions.td index c0d38fa52b344..103575dc351f2 100644 --- a/llvm/lib/Target/AMDGPU/VOP2Instructions.td +++ b/llvm/lib/Target/AMDGPU/VOP2Instructions.td @@ -374,6 +374,12 @@ class VOP_MADAK : VOP_MADK_Base { } def VOP_MADAK_F16 : VOP_MADAK ; +def VOP_MADAK_F16_t16 : VOP_MADAK { + let IsTrue16 = 1; + let IsRealTrue16 = 1; + let DstRC = getVALUDstForVT.ret; + let Ins32 = (ins VSrcT_f16_Lo128_Deferred:$src0, VGPRSrc_16_Lo128:$src1, ImmOpType:$imm); +} def VOP_MADAK_F16_fake16 : VOP_MADAK { let IsTrue16 = 1; let DstRC = getVALUDstForVT_fake16.ret; @@ -399,6 +405,12 @@ class VOP_MADMK : VOP_MADK_Base { } def VOP_MADMK_F16 : VOP_MADMK ; +def VOP_MADMK_F16_t16 : VOP_MADMK { + let IsTrue16 = 1; + let IsRealTrue16 = 1; + let DstRC = getVALUDstForVT.ret; + let Ins32 = (ins VSrcT_f16_Lo128_Deferred:$src0, ImmOpType:$imm, VGPRSrc_16_Lo128:$src1); +} def VOP_MADMK_F16_fake16 : VOP_MADMK { let IsTrue16 = 1; let DstRC = getVALUDstForVT_fake16.ret; @@ -467,6 +479,42 @@ class VOP_MAC : VOPProfile <[vt0, vt1, vt1, v } def VOP_MAC_F16 : VOP_MAC ; +def VOP_MAC_F16_t16 : VOP_MAC { + let IsTrue16 = 1; + let IsRealTrue16 = 1; + let HasOpSel = 1; + let DstRC = VOPDstOperand_t16Lo128; + let Src0RC32 = getVOPSrc0ForVT.ret; + let Src1RC32 = getVregSrcForVT.ret; + let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1, getVregSrcForVT.ret:$src2); + let Src0DPP = getVregSrcForVT.ret; + let Src1DPP = getVregSrcForVT.ret; + let Src2DPP = getVregSrcForVT.ret; + let Src0ModDPP = getSrcModDPP_t16.ret; + let Src1ModDPP = getSrcModDPP_t16.ret; + let Src2ModDPP = getSrcModDPP_t16.ret; + let InsDPP = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0, + Src1ModDPP:$src1_modifiers, Src1DPP:$src1, + getVregSrcForVT.ret:$src2, // stub argument + dpp_ctrl:$dpp_ctrl, DppRowMask:$row_mask, + DppBankMask:$bank_mask, DppBoundCtrl:$bound_ctrl); + let InsDPP8 = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0, + Src1ModDPP:$src1_modifiers, Src1DPP:$src1, + getVregSrcForVT.ret:$src2, // stub argument + dpp8:$dpp8, Dpp8FI:$fi); + let DstRC64 = getVALUDstForVT.ret; + let Src0RC64 = getVOP3SrcForVT.ret; + let Src1RC64 = getVOP3SrcForVT.ret; + let Src0VOP3DPP = VGPRSrc_16; + let Src1VOP3DPP = getVOP3DPPSrcForVT.ret; + let Src2VOP3DPP = getVOP3DPPSrcForVT.ret; + let Src0ModVOP3DPP = getSrc0ModVOP3DPP.ret; + let Src1ModVOP3DPP = getSrcModVOP3DPP.ret; + let Src2ModVOP3DPP = getSrcModVOP3DPP.ret; + let Src0Mod = getSrc0Mod.ret; + let Src1Mod = getSrcMod.ret; + let Src2Mod = getSrcMod.ret; +} def VOP_MAC_F16_fake16 : VOP_MAC { let IsTrue16 = 1; let DstRC = getVALUDstForVT_fake16.ret; @@ -998,6 +1046,9 @@ let FPDPRounding = 1, isReMaterializable = 1, FixedSize = 1 in { let SubtargetPredicate = isGFX10Plus, True16Predicate = NotHasTrue16BitInsts in { def V_FMAMK_F16 : VOP2_Pseudo <"v_fmamk_f16", VOP_MADMK_F16, [], "">; } +let True16Predicate = UseRealTrue16Insts in { + def V_FMAMK_F16_t16 : VOP2_Pseudo <"v_fmamk_f16_t16", VOP_MADMK_F16_t16, [], "">; +} let True16Predicate = UseFakeTrue16Insts in { def V_FMAMK_F16_fake16 : VOP2_Pseudo <"v_fmamk_f16_fake16", VOP_MADMK_F16_fake16, [], "">; } @@ -1006,6 +1057,9 @@ let isCommutable = 1 in { let SubtargetPredicate = isGFX10Plus, True16Predicate = NotHasTrue16BitInsts in { def V_FMAAK_F16 : VOP2_Pseudo <"v_fmaak_f16", VOP_MADAK_F16, [], "">; } +let True16Predicate = UseRealTrue16Insts in { + def V_FMAAK_F16_t16 : VOP2_Pseudo <"v_fmaak_f16_t16", VOP_MADAK_F16_t16, [], "">; +} let True16Predicate = UseFakeTrue16Insts in { def V_FMAAK_F16_fake16 : VOP2_Pseudo <"v_fmaak_f16_fake16", VOP_MADAK_F16_fake16, [], "">; } @@ -1020,6 +1074,9 @@ let SubtargetPredicate = isGFX10Plus in { let True16Predicate = NotHasTrue16BitInsts in { defm V_FMAC_F16 : VOP2Inst <"v_fmac_f16", VOP_MAC_F16>; } +let True16Predicate = UseRealTrue16Insts in { + defm V_FMAC_F16_t16 : VOP2Inst <"v_fmac_f16_t16", VOP_MAC_F16_t16>; +} let True16Predicate = UseFakeTrue16Insts in { defm V_FMAC_F16_fake16 : VOP2Inst <"v_fmac_f16_fake16", VOP_MAC_F16_fake16>; } @@ -1692,8 +1749,8 @@ multiclass VOP3Only_Realtriple_t16_gfx11_gfx12 op, string asmName, stri VOP3_Realtriple_t16_gfx12; multiclass VOP3Only_Realtriple_t16_and_fake16_gfx11_gfx12 op, string asmName, string OpName = NAME> { - defm OpName#"_t16": VOP3Only_Realtriple_t16_gfx11_gfx12; - defm OpName#"_fake16": VOP3Only_Realtriple_t16_gfx11_gfx12; + defm _t16: VOP3Only_Realtriple_t16_gfx11_gfx12; + defm _fake16: VOP3Only_Realtriple_t16_gfx11_gfx12; } multiclass VOP3beOnly_Realtriple_gfx11_gfx12 op> : @@ -1712,7 +1769,14 @@ multiclass VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12 op, string asmN multiclass VOP2_Real_FULL_t16_gfx11_gfx12 op, string asmName, string opName = NAME> : - VOP2_Real_FULL_with_name_gfx11_gfx12; + VOP2_Real_FULL_with_name, + VOP2_Real_FULL_with_name; + +multiclass VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12 op, string asmName, + string opName = NAME> { + defm _t16: VOP2_Real_FULL_t16_gfx11_gfx12; + defm _fake16: VOP2_Real_FULL_t16_gfx11_gfx12; +} multiclass VOP2_Real_FULL_gfx11 op> : VOP2_Real_FULL; @@ -1747,15 +1811,15 @@ defm V_SUBREV_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16 defm V_SUBREV_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x034, "v_subrev_f16">; defm V_MUL_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">; defm V_MUL_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x035, "v_mul_f16">; -defm V_FMAC_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x036, "v_fmac_f16">; +defm V_FMAC_F16 : VOP2_Real_FULL_t16_and_fake16_gfx11_gfx12<0x036, "v_fmac_f16">; defm V_LDEXP_F16_t16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">; defm V_LDEXP_F16_fake16 : VOP2_Real_FULL_t16_gfx11_gfx12<0x03b, "v_ldexp_f16">; defm V_MAX_F16_t16 : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">; defm V_MAX_F16_fake16 : VOP2_Real_FULL_t16_gfx11<0x039, "v_max_f16">; defm V_MIN_F16_t16 : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">; defm V_MIN_F16_fake16 : VOP2_Real_FULL_t16_gfx11<0x03a, "v_min_f16">; -defm V_FMAMK_F16_fake16 : VOP2Only_Real_MADK_t16_gfx11_gfx12<0x037, "v_fmamk_f16">; -defm V_FMAAK_F16_fake16 : VOP2Only_Real_MADK_t16_gfx11_gfx12<0x038, "v_fmaak_f16">; +defm V_FMAMK_F16 : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x037, "v_fmamk_f16">; +defm V_FMAAK_F16 : VOP2Only_Real_MADK_t16_and_fake16_gfx11_gfx12<0x038, "v_fmaak_f16">; // VOP3 only. defm V_CNDMASK_B16 : VOP3Only_Realtriple_gfx11_gfx12<0x25d>; diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s index 2a4b3ea201701..0f2852fc531ed 100644 --- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s +++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2.s @@ -648,50 +648,56 @@ v_dot2c_f32_f16 v5, src_scc, v2 v_dot2c_f32_f16 v255, 0xfe0b, v255 // GFX11: encoding: [0xff,0xfe,0xff,0x05,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, v1, v2, 0xfe0b +v_fmaak_f16 v5.l, v1.l, v2.l, 0xfe0b // GFX11: encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, v127, v2, 0xfe0b +v_fmaak_f16 v5.l, v127.l, v2.l, 0xfe0b // GFX11: encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, s1, v2, 0xfe0b +v_fmaak_f16 v5.l, v1.h, v2.l, 0xfe0b +// GFX11: encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +v_fmaak_f16 v5.l, v127.h, v2.l, 0xfe0b +// GFX11: encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +v_fmaak_f16 v5.l, s1, v2.l, 0xfe0b // GFX11: encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, s105, v2, 0xfe0b +v_fmaak_f16 v5.l, s105, v2.l, 0xfe0b // GFX11: encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b +v_fmaak_f16 v5.l, vcc_lo, v2.l, 0xfe0b // GFX11: encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b +v_fmaak_f16 v5.l, vcc_hi, v2.l, 0xfe0b // GFX11: encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, ttmp15, v2, 0xfe0b +v_fmaak_f16 v5.l, ttmp15, v2.l, 0xfe0b // GFX11: encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, m0, v2, 0xfe0b +v_fmaak_f16 v5.l, m0, v2.l, 0xfe0b // GFX11: encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, exec_lo, v2, 0xfe0b +v_fmaak_f16 v5.l, exec_lo, v2.l, 0xfe0b // GFX11: encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, exec_hi, v2, 0xfe0b +v_fmaak_f16 v5.l, exec_hi, v2.l, 0xfe0b // GFX11: encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, null, v2, 0xfe0b +v_fmaak_f16 v5.l, null, v2.l, 0xfe0b // GFX11: encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, -1, v2, 0xfe0b +v_fmaak_f16 v5.l, -1, v2.l, 0xfe0b // GFX11: encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, 0.5, v2, 0xfe0b -// GFX11: encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +v_fmaak_f16 v127.l, 0.5, v127.l, 0xfe0b +// GFX11: encoding: [0xf0,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, src_scc, v2, 0xfe0b -// GFX11: encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +v_fmaak_f16 v5.h, src_scc, v2.h, 0xfe0b +// GFX11: encoding: [0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v127, 0xfe0b, v127, 0xfe0b -// GFX11: encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] +v_fmaak_f16 v127.h, 0xfe0b, v127.h, 0xfe0b +// GFX11: encoding: [0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00] v_fmaak_f32 v5, v1, v2, 0xaf123456 // GFX11: encoding: [0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf] @@ -783,50 +789,56 @@ v_fmac_dx9_zero_f32 v5, src_scc, v2 v_fmac_dx9_zero_f32 v255, 0xaf123456, v255 // GFX11: encoding: [0xff,0xfe,0xff,0x0d,0x56,0x34,0x12,0xaf] -v_fmac_f16 v5, v1, v2 +v_fmac_f16 v5.l, v1.l, v2.l // GFX11: encoding: [0x01,0x05,0x0a,0x6c] -v_fmac_f16 v5, v127, v2 +v_fmac_f16 v5.l, v127.l, v2.l // GFX11: encoding: [0x7f,0x05,0x0a,0x6c] -v_fmac_f16 v5, s1, v2 +v_fmac_f16 v5.l, v1.h, v2.l +// GFX11: encoding: [0x81,0x05,0x0a,0x6c] + +v_fmac_f16 v5.l, v127.h, v2.l +// GFX11: encoding: [0xff,0x05,0x0a,0x6c] + +v_fmac_f16 v5.l, s1, v2.l // GFX11: encoding: [0x01,0x04,0x0a,0x6c] -v_fmac_f16 v5, s105, v2 +v_fmac_f16 v5.l, s105, v2.l // GFX11: encoding: [0x69,0x04,0x0a,0x6c] -v_fmac_f16 v5, vcc_lo, v2 +v_fmac_f16 v5.l, vcc_lo, v2.l // GFX11: encoding: [0x6a,0x04,0x0a,0x6c] -v_fmac_f16 v5, vcc_hi, v2 +v_fmac_f16 v5.l, vcc_hi, v2.l // GFX11: encoding: [0x6b,0x04,0x0a,0x6c] -v_fmac_f16 v5, ttmp15, v2 +v_fmac_f16 v5.l, ttmp15, v2.l // GFX11: encoding: [0x7b,0x04,0x0a,0x6c] -v_fmac_f16 v5, m0, v2 +v_fmac_f16 v5.l, m0, v2.l // GFX11: encoding: [0x7d,0x04,0x0a,0x6c] -v_fmac_f16 v5, exec_lo, v2 +v_fmac_f16 v5.l, exec_lo, v2.l // GFX11: encoding: [0x7e,0x04,0x0a,0x6c] -v_fmac_f16 v5, exec_hi, v2 +v_fmac_f16 v5.l, exec_hi, v2.l // GFX11: encoding: [0x7f,0x04,0x0a,0x6c] -v_fmac_f16 v5, null, v2 +v_fmac_f16 v5.l, null, v2.l // GFX11: encoding: [0x7c,0x04,0x0a,0x6c] -v_fmac_f16 v5, -1, v2 +v_fmac_f16 v5.l, -1, v2.l // GFX11: encoding: [0xc1,0x04,0x0a,0x6c] -v_fmac_f16 v5, 0.5, v2 -// GFX11: encoding: [0xf0,0x04,0x0a,0x6c] +v_fmac_f16 v127.l, 0.5, v127.l +// GFX11: encoding: [0xf0,0xfe,0xfe,0x6c] -v_fmac_f16 v5, src_scc, v2 -// GFX11: encoding: [0xfd,0x04,0x0a,0x6c] +v_fmac_f16 v5.h, src_scc, v2.h +// GFX11: encoding: [0xfd,0x04,0x0b,0x6d] -v_fmac_f16 v127, 0xfe0b, v127 -// GFX11: encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00] +v_fmac_f16 v127.h, 0xfe0b, v127.h +// GFX11: encoding: [0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00] v_fmac_f32 v5, v1, v2 // GFX11: encoding: [0x01,0x05,0x0a,0x56] @@ -918,50 +930,56 @@ v_fmac_legacy_f32 v5, src_scc, v2 v_fmac_legacy_f32 v255, 0xaf123456, v255 // GFX11: encoding: [0xff,0xfe,0xff,0x0d,0x56,0x34,0x12,0xaf] -v_fmamk_f16 v5, v1, 0xfe0b, v3 +v_fmamk_f16 v5.l, v1.l, 0xfe0b, v3.l // GFX11: encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, v127, 0xfe0b, v3 +v_fmamk_f16 v5.l, v127.l, 0xfe0b, v3.l // GFX11: encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, s1, 0xfe0b, v3 +v_fmamk_f16 v5.l, v1.h, 0xfe0b, v3.l +// GFX11: encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +v_fmamk_f16 v5.l, v127.h, 0xfe0b, v3.l +// GFX11: encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +v_fmamk_f16 v5.l, s1, 0xfe0b, v3.l // GFX11: encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, s105, 0xfe0b, v3 +v_fmamk_f16 v5.l, s105, 0xfe0b, v3.l // GFX11: encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3 +v_fmamk_f16 v5.l, vcc_lo, 0xfe0b, v3.l // GFX11: encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3 +v_fmamk_f16 v5.l, vcc_hi, 0xfe0b, v3.l // GFX11: encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, ttmp15, 0xfe0b, v3 +v_fmamk_f16 v5.l, ttmp15, 0xfe0b, v3.l // GFX11: encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, m0, 0xfe0b, v3 +v_fmamk_f16 v5.l, m0, 0xfe0b, v3.l // GFX11: encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, exec_lo, 0xfe0b, v3 +v_fmamk_f16 v5.l, exec_lo, 0xfe0b, v3.l // GFX11: encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, exec_hi, 0xfe0b, v3 +v_fmamk_f16 v5.l, exec_hi, 0xfe0b, v3.l // GFX11: encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, null, 0xfe0b, v3 +v_fmamk_f16 v5.l, null, 0xfe0b, v3.l // GFX11: encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, -1, 0xfe0b, v3 +v_fmamk_f16 v5.l, -1, 0xfe0b, v3.l // GFX11: encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, 0.5, 0xfe0b, v3 -// GFX11: encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +v_fmamk_f16 v127.l, 0.5, 0xfe0b, v127.l +// GFX11: encoding: [0xf0,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, src_scc, 0xfe0b, v3 -// GFX11: encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +v_fmamk_f16 v5.h, src_scc, 0xfe0b, v3.h +// GFX11: encoding: [0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v127, 0xfe0b, 0xfe0b, v127 -// GFX11: encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] +v_fmamk_f16 v127.h, 0xfe0b, 0xfe0b, v127.h +// GFX11: encoding: [0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00] v_fmamk_f32 v5, v1, 0xaf123456, v3 // GFX11: encoding: [0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf] diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s index 3eff00bb96e47..f40278cb9c42e 100644 --- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s +++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp16.s @@ -629,47 +629,47 @@ v_dot2c_f32_f16 v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 f v_dot2c_f32_f16 v255, -|v255|, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1 // GFX11: encoding: [0xfa,0xfe,0xff,0x05,0xff,0x6f,0xf5,0x30] -v_fmac_f16 v5, v1, v2 quad_perm:[3,2,1,0] +v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[3,2,1,0] // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] -v_fmac_f16 v5, v1, v2 quad_perm:[0,1,2,3] +v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[0,1,2,3] // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] -v_fmac_f16 v5, v1, v2 row_mirror +v_fmac_f16 v5.l, v1.l, v2.l row_mirror // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_half_mirror +v_fmac_f16 v5.l, v1.l, v2.l row_half_mirror // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shl:1 +v_fmac_f16 v5.l, v1.l, v2.l row_shl:1 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shl:15 +v_fmac_f16 v5.l, v1.l, v2.l row_shl:15 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shr:1 +v_fmac_f16 v5.l, v1.l, v2.l row_shr:1 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shr:15 +v_fmac_f16 v5.l, v1.l, v2.l row_shr:15 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_ror:1 +v_fmac_f16 v5.l, v1.l, v2.l row_ror:1 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_ror:15 +v_fmac_f16 v5.l, v1.l, v2.l row_ror:15 // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf +v_fmac_f16 v5.l, v1.l, v2.l row_share:0 row_mask:0xf bank_mask:0xf // GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_share:15 row_mask:0x0 bank_mask:0x1 -// GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01] +v_fmac_f16 v127.l, v127.l, v127.l row_share:15 row_mask:0x0 bank_mask:0x1 +// GFX11: encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x5f,0x01,0x01] -v_fmac_f16 v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0 -// GFX11: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x60,0x09,0x13] +v_fmac_f16 v5.h, v1.h, v2.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0 +// GFX11: encoding: [0xfa,0x04,0x0b,0x6d,0x81,0x60,0x09,0x13] -v_fmac_f16 v127, -|v127|, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1 -// GFX11: encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xf5,0x30] +v_fmac_f16 v127.h, -|v127.h|, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1 +// GFX11: encoding: [0xfa,0xfe,0xff,0x6d,0xff,0x6f,0xf5,0x30] v_fmac_f32 v5, v1, v2 quad_perm:[3,2,1,0] // GFX11: encoding: [0xfa,0x04,0x0a,0x56,0x01,0x1b,0x00,0xff] diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp8.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp8.s index 0f19cf0028525..ffec9f3a7ec09 100644 --- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp8.s +++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_dpp8.s @@ -132,14 +132,17 @@ v_dot2c_f32_f16 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] fi:1 v_dot2c_f32_f16 v255, v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0 // GFX11: encoding: [0xe9,0xfe,0xff,0x05,0xff,0x00,0x00,0x00] -v_fmac_f16 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] +v_fmac_f16 v5.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] // GFX11: encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] -v_fmac_f16 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] fi:1 -// GFX11: encoding: [0xea,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] +v_fmac_f16 v127.l, v127.l, v127.l dpp8:[7,6,5,4,3,2,1,0] +// GFX11: encoding: [0xe9,0xfe,0xfe,0x6c,0x7f,0x77,0x39,0x05] -v_fmac_f16 v127, v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0 -// GFX11: encoding: [0xe9,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00] +v_fmac_f16 v5.h, v1.h, v2.h dpp8:[7,6,5,4,3,2,1,0] fi:1 +// GFX11: encoding: [0xea,0x04,0x0b,0x6d,0x81,0x77,0x39,0x05] + +v_fmac_f16 v127.h, v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0 +// GFX11: encoding: [0xe9,0xfe,0xff,0x6d,0xff,0x00,0x00,0x00] v_fmac_f32 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] // GFX11: encoding: [0xe9,0x04,0x0a,0x56,0x01,0x77,0x39,0x05] diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_t16_err.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_t16_err.s index dd619f3077f70..2027a436fa72f 100644 --- a/llvm/test/MC/AMDGPU/gfx11_asm_vop2_t16_err.s +++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop2_t16_err.s @@ -29,50 +29,95 @@ v_add_f16_e32 v5, v1, v255 v_add_f16_e32 v5, v255, v2 // GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode -v_fmaak_f16_e32 v255, v1, v2, 0xfe0b -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v255.h, v1.h, v2.h, 0xfe0b +// GFX11: :[[@LINE-1]]:17: error: invalid operand for instruction -v_fmaak_f16_e32 v5, v1, v255, 0xfe0b -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v255.l, v1.l, v2.l, 0xfe0b +// GFX11: :[[@LINE-1]]:17: error: invalid operand for instruction -v_fmaak_f16_e32 v5, v255, v2, 0xfe0b -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.h, v1.h, v255.h, 0xfe0b +// GFX11: :[[@LINE-1]]:29: error: invalid operand for instruction -v_fmac_f16_dpp v255, v1, v2 dpp8:[7,6,5,4,3,2,1,0] -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.h, v255.h, v2.h, 0xfe0b +// GFX11: :[[@LINE-1]]:23: error: invalid operand for instruction -v_fmac_f16_dpp v255, v1, v2 quad_perm:[3,2,1,0] -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.l, v1.l, v255.l, 0xfe0b +// GFX11: :[[@LINE-1]]:29: error: invalid operand for instruction -v_fmac_f16_dpp v5, v1, v255 dpp8:[7,6,5,4,3,2,1,0] -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.l, v255.l, v2.l, 0xfe0b +// GFX11: :[[@LINE-1]]:23: error: invalid operand for instruction -v_fmac_f16_dpp v5, v1, v255 quad_perm:[3,2,1,0] -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.h, v1.h, v2.h dpp8:[7,6,5,4,3,2,1,0] +// GFX11: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_dpp v5, v255, v2 dpp8:[7,6,5,4,3,2,1,0] -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.h, v1.h, v2.h quad_perm:[3,2,1,0] +// GFX11: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_dpp v5, v255, v2 quad_perm:[3,2,1,0] -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] +// GFX11: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_e32 v255, v1, v2 -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.l, v1.l, v2.l quad_perm:[3,2,1,0] +// GFX11: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_e32 v5, v1, v255 -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v1.h, v255.h dpp8:[7,6,5,4,3,2,1,0] +// GFX11: :[[@LINE-1]]:28: error: invalid operand for instruction -v_fmac_f16_e32 v5, v255, v2 -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v1.h, v255.h quad_perm:[3,2,1,0] +// GFX11: :[[@LINE-1]]:28: error: invalid operand for instruction -v_fmamk_f16_e32 v255, v1, 0xfe0b, v3 -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v255.h, v2.h dpp8:[7,6,5,4,3,2,1,0] +// GFX11: :[[@LINE-1]]:22: error: invalid operand for instruction -v_fmamk_f16_e32 v5, v1, 0xfe0b, v255 -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v255.h, v2.h quad_perm:[3,2,1,0] +// GFX11: :[[@LINE-1]]:22: error: invalid operand for instruction -v_fmamk_f16_e32 v5, v255, 0xfe0b, v3 -// GFX11: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.l, v1.l, v255.l dpp8:[7,6,5,4,3,2,1,0] +// GFX11: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_dpp v5.l, v1.l, v255.l quad_perm:[3,2,1,0] +// GFX11: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_dpp v5.l, v255.l, v2.l dpp8:[7,6,5,4,3,2,1,0] +// GFX11: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmac_f16_dpp v5.l, v255.l, v2.l quad_perm:[3,2,1,0] +// GFX11: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmac_f16_e32 v255.h, v1.h, v2.h +// GFX11: :[[@LINE-1]]:16: error: invalid operand for instruction + +v_fmac_f16_e32 v255.l, v1.l, v2.l +// GFX11: :[[@LINE-1]]:16: error: invalid operand for instruction + +v_fmac_f16_e32 v5.h, v1.h, v255.h +// GFX11: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_e32 v5.h, v255.h, v2.h +// GFX11: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmac_f16_e32 v5.l, v1.l, v255.l +// GFX11: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_e32 v5.l, v255.l, v2.l +// GFX11: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmamk_f16_e32 v255.h, v1.h, 0xfe0b, v3.h +// GFX11: :[[@LINE-1]]:17: error: invalid operand for instruction + +v_fmamk_f16_e32 v255.l, v1.l, 0xfe0b, v3.l +// GFX11: :[[@LINE-1]]:17: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.h, v1.h, 0xfe0b, v255.h +// GFX11: :[[@LINE-1]]:37: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.h, v255.h, 0xfe0b, v3.h +// GFX11: :[[@LINE-1]]:23: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.l, v1.l, 0xfe0b, v255.l +// GFX11: :[[@LINE-1]]:37: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.l, v255.l, 0xfe0b, v3.l +// GFX11: :[[@LINE-1]]:23: error: invalid operand for instruction v_ldexp_f16_dpp v255.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] // GFX11: :[[@LINE-1]]:17: error: invalid operand for instruction diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop2.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop2.s index 5593ea77d9424..d777f6d29a613 100644 --- a/llvm/test/MC/AMDGPU/gfx12_asm_vop2.s +++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop2.s @@ -594,51 +594,63 @@ v_cvt_pkrtz_f16_f32 v5, src_scc, v2 v_cvt_pkrtz_f16_f32 v255, 0xaf123456, v255 // GFX12: encoding: [0xff,0xfe,0xff,0x5f,0x56,0x34,0x12,0xaf] -v_fmaak_f16 v5, v1, v2, 0xfe0b +v_fmaak_f16 v5.l, v1.l, v2.l, 0xfe0b // GFX12: encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, v127, v2, 0xfe0b +v_fmaak_f16 v5.l, v127.l, v2.l, 0xfe0b // GFX12: encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, s1, v2, 0xfe0b +v_fmaak_f16 v5.l, s1, v2.l, 0xfe0b // GFX12: encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, s105, v2, 0xfe0b +v_fmaak_f16 v5.l, s105, v2.l, 0xfe0b // GFX12: encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b +v_fmaak_f16 v5.l, vcc_lo, v2.l, 0xfe0b // GFX12: encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b +v_fmaak_f16 v5.l, vcc_hi, v2.l, 0xfe0b // GFX12: encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, ttmp15, v2, 0xfe0b +v_fmaak_f16 v5.l, ttmp15, v2.l, 0xfe0b // GFX12: encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, m0, v2, 0xfe0b +v_fmaak_f16 v5.l, m0, v2.l, 0xfe0b // GFX12: encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, exec_lo, v2, 0xfe0b +v_fmaak_f16 v5.l, exec_lo, v2.l, 0xfe0b // GFX12: encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, exec_hi, v2, 0xfe0b +v_fmaak_f16 v5.l, exec_hi, v2.l, 0xfe0b // GFX12: encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, null, v2, 0xfe0b +v_fmaak_f16 v5.l, null, v2.l, 0xfe0b // GFX12: encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, -1, v2, 0xfe0b +v_fmaak_f16 v5.l, -1, v2.l, 0xfe0b // GFX12: encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, 0.5, v2, 0xfe0b +v_fmaak_f16 v5.l, 0.5, v2.l, 0xfe0b // GFX12: encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v5, src_scc, v2, 0xfe0b +v_fmaak_f16 v5.l, src_scc, v2.l, 0xfe0b // GFX12: encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -v_fmaak_f16 v127, 0xfe0b, v127, 0xfe0b +v_fmaak_f16 v127.l, 0xfe0b, v127.l, 0xfe0b // GFX12: encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] +v_fmaak_f16 v5.l, v1.h, v2.l, 0xfe0b +// GFX12: encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +v_fmaak_f16 v5.l, v127.h, v2.l, 0xfe0b +// GFX12: encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +v_fmaak_f16 v5.h, src_scc, v2.h, 0xfe0b +// GFX12: encoding: [0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00] + +v_fmaak_f16 v127.h, 0xfe0b, v127.h, 0xfe0b +// GFX12: encoding: [0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00] + v_fmaak_f32 v5, v1, v2, 0xaf123456 // GFX12: encoding: [0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf] @@ -684,51 +696,63 @@ v_fmaak_f32 v5, src_scc, v2, 0xaf123456 v_fmaak_f32 v255, 0xaf123456, v255, 0xaf123456 // GFX12: encoding: [0xff,0xfe,0xff,0x5b,0x56,0x34,0x12,0xaf] -v_fmac_f16 v5, v1, v2 +v_fmac_f16 v5.l, v1.l, v2.l // GFX12: encoding: [0x01,0x05,0x0a,0x6c] -v_fmac_f16 v5, v127, v2 +v_fmac_f16 v5.l, v127.l, v2.l // GFX12: encoding: [0x7f,0x05,0x0a,0x6c] -v_fmac_f16 v5, s1, v2 +v_fmac_f16 v5.l, s1, v2.l // GFX12: encoding: [0x01,0x04,0x0a,0x6c] -v_fmac_f16 v5, s105, v2 +v_fmac_f16 v5.l, s105, v2.l // GFX12: encoding: [0x69,0x04,0x0a,0x6c] -v_fmac_f16 v5, vcc_lo, v2 +v_fmac_f16 v5.l, vcc_lo, v2.l // GFX12: encoding: [0x6a,0x04,0x0a,0x6c] -v_fmac_f16 v5, vcc_hi, v2 +v_fmac_f16 v5.l, vcc_hi, v2.l // GFX12: encoding: [0x6b,0x04,0x0a,0x6c] -v_fmac_f16 v5, ttmp15, v2 +v_fmac_f16 v5.l, ttmp15, v2.l // GFX12: encoding: [0x7b,0x04,0x0a,0x6c] -v_fmac_f16 v5, m0, v2 +v_fmac_f16 v5.l, m0, v2.l // GFX12: encoding: [0x7d,0x04,0x0a,0x6c] -v_fmac_f16 v5, exec_lo, v2 +v_fmac_f16 v5.l, exec_lo, v2.l // GFX12: encoding: [0x7e,0x04,0x0a,0x6c] -v_fmac_f16 v5, exec_hi, v2 +v_fmac_f16 v5.l, exec_hi, v2.l // GFX12: encoding: [0x7f,0x04,0x0a,0x6c] -v_fmac_f16 v5, null, v2 +v_fmac_f16 v5.l, null, v2.l // GFX12: encoding: [0x7c,0x04,0x0a,0x6c] -v_fmac_f16 v5, -1, v2 +v_fmac_f16 v5.l, -1, v2.l // GFX12: encoding: [0xc1,0x04,0x0a,0x6c] -v_fmac_f16 v5, 0.5, v2 +v_fmac_f16 v5.l, 0.5, v2.l // GFX12: encoding: [0xf0,0x04,0x0a,0x6c] -v_fmac_f16 v5, src_scc, v2 +v_fmac_f16 v5.l, src_scc, v2.l // GFX12: encoding: [0xfd,0x04,0x0a,0x6c] -v_fmac_f16 v127, 0xfe0b, v127 +v_fmac_f16 v127.l, 0xfe0b, v127.l // GFX12: encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00] +v_fmac_f16 v5.l, v1.h, v2.l +// GFX12: encoding: [0x81,0x05,0x0a,0x6c] + +v_fmac_f16 v5.l, v127.h, v2.l +// GFX12: encoding: [0xff,0x05,0x0a,0x6c] + +v_fmac_f16 v5.h, src_scc, v2.h +// GFX12: encoding: [0xfd,0x04,0x0b,0x6d] + +v_fmac_f16 v127.h, 0xfe0b, v127.h +// GFX12: encoding: [0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00] + v_fmac_f32 v5, v1, v2 // GFX12: encoding: [0x01,0x05,0x0a,0x56] @@ -774,51 +798,63 @@ v_fmac_f32 v5, src_scc, v2 v_fmac_f32 v255, 0xaf123456, v255 // GFX12: encoding: [0xff,0xfe,0xff,0x57,0x56,0x34,0x12,0xaf] -v_fmamk_f16 v5, v1, 0xfe0b, v3 +v_fmamk_f16 v5.l, v1.l, 0xfe0b, v3.l // GFX12: encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, v127, 0xfe0b, v3 +v_fmamk_f16 v5.l, v127.l, 0xfe0b, v3.l // GFX12: encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, s1, 0xfe0b, v3 +v_fmamk_f16 v5.l, s1, 0xfe0b, v3.l // GFX12: encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, s105, 0xfe0b, v3 +v_fmamk_f16 v5.l, s105, 0xfe0b, v3.l // GFX12: encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3 +v_fmamk_f16 v5.l, vcc_lo, 0xfe0b, v3.l // GFX12: encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3 +v_fmamk_f16 v5.l, vcc_hi, 0xfe0b, v3.l // GFX12: encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, ttmp15, 0xfe0b, v3 +v_fmamk_f16 v5.l, ttmp15, 0xfe0b, v3.l // GFX12: encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, m0, 0xfe0b, v3 +v_fmamk_f16 v5.l, m0, 0xfe0b, v3.l // GFX12: encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, exec_lo, 0xfe0b, v3 +v_fmamk_f16 v5.l, exec_lo, 0xfe0b, v3.l // GFX12: encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, exec_hi, 0xfe0b, v3 +v_fmamk_f16 v5.l, exec_hi, 0xfe0b, v3.l // GFX12: encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, null, 0xfe0b, v3 +v_fmamk_f16 v5.l, null, 0xfe0b, v3.l // GFX12: encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, -1, 0xfe0b, v3 +v_fmamk_f16 v5.l, -1, 0xfe0b, v3.l // GFX12: encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, 0.5, 0xfe0b, v3 +v_fmamk_f16 v5.l, 0.5, 0xfe0b, v3.l // GFX12: encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v5, src_scc, 0xfe0b, v3 +v_fmamk_f16 v5.l, src_scc, 0xfe0b, v3.l // GFX12: encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -v_fmamk_f16 v127, 0xfe0b, 0xfe0b, v127 +v_fmamk_f16 v127.l, 0xfe0b, 0xfe0b, v127.l // GFX12: encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] +v_fmamk_f16 v5.l, v1.h, 0xfe0b, v3.l +// GFX12: encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +v_fmamk_f16 v5.l, v127.h, 0xfe0b, v3.l +// GFX12: encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +v_fmamk_f16 v5.h, src_scc, 0xfe0b, v3.h +// GFX12: encoding: [0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00] + +v_fmamk_f16 v127.h, 0xfe0b, 0xfe0b, v127.h +// GFX12: encoding: [0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00] + v_fmamk_f32 v5, v1, 0xaf123456, v3 // GFX12: encoding: [0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf] diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp16.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp16.s index a0f93f459f915..4424fced3e3ea 100644 --- a/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp16.s +++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp16.s @@ -521,48 +521,54 @@ v_cvt_pkrtz_f16_f32 v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl v_cvt_pkrtz_f16_f32 v255, -|v255|, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1 // GFX12: encoding: [0xfa,0xfe,0xff,0x5f,0xff,0x6f,0xf5,0x30] -v_fmac_f16 v5, v1, v2 quad_perm:[3,2,1,0] +v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[3,2,1,0] // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] -v_fmac_f16 v5, v1, v2 quad_perm:[0,1,2,3] +v_fmac_f16 v5.l, v1.l, v2.l quad_perm:[0,1,2,3] // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] -v_fmac_f16 v5, v1, v2 row_mirror +v_fmac_f16 v5.l, v1.l, v2.l row_mirror // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_half_mirror +v_fmac_f16 v5.l, v1.l, v2.l row_half_mirror // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shl:1 +v_fmac_f16 v5.l, v1.l, v2.l row_shl:1 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shl:15 +v_fmac_f16 v5.l, v1.l, v2.l row_shl:15 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shr:1 +v_fmac_f16 v5.l, v1.l, v2.l row_shr:1 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_shr:15 +v_fmac_f16 v5.l, v1.l, v2.l row_shr:15 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_ror:1 +v_fmac_f16 v5.l, v1.l, v2.l row_ror:1 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_ror:15 +v_fmac_f16 v5.l, v1.l, v2.l row_ror:15 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf +v_fmac_f16 v5.l, v1.l, v2.l row_share:0 row_mask:0xf bank_mask:0xf // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] -v_fmac_f16 v5, v1, v2 row_share:15 row_mask:0x0 bank_mask:0x1 +v_fmac_f16 v5.l, v1.l, v2.l row_share:15 row_mask:0x0 bank_mask:0x1 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01] -v_fmac_f16 v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0 +v_fmac_f16 v5.l, v1.l, v2.l row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0 // GFX12: encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x60,0x09,0x13] -v_fmac_f16 v127, -|v127|, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1 +v_fmac_f16 v127.l, -|v127.l|, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1 // GFX12: encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xf5,0x30] +v_fmac_f16 v5.h, v1.h, v2.h row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0 +// GFX12: encoding: [0xfa,0x04,0x0b,0x6d,0x81,0x60,0x09,0x13] + +v_fmac_f16 v127.h, -|v127.h|, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1 +// GFX12: encoding: [0xfa,0xfe,0xff,0x6d,0xff,0x6f,0xf5,0x30] + v_fmac_f32 v5, v1, v2 quad_perm:[3,2,1,0] // GFX12: encoding: [0xfa,0x04,0x0a,0x56,0x01,0x1b,0x00,0xff] diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp8.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp8.s index 81fcb323e2711..05971722a7268 100644 --- a/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp8.s +++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop2_dpp8.s @@ -114,15 +114,21 @@ v_cvt_pkrtz_f16_f32 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] fi:1 v_cvt_pkrtz_f16_f32 v255, v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:0 // GFX12: encoding: [0xe9,0xfe,0xff,0x5f,0xff,0x00,0x00,0x00] -v_fmac_f16 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] +v_fmac_f16 v5.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] // GFX12: encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] -v_fmac_f16 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] fi:1 +v_fmac_f16 v5.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] fi:1 // GFX12: encoding: [0xea,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] -v_fmac_f16 v127, v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:0 +v_fmac_f16 v127.l, v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:0 // GFX12: encoding: [0xe9,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00] +v_fmac_f16 v5.h, v1.h, v2.h dpp8:[7,6,5,4,3,2,1,0] fi:1 +// GFX12: encoding: [0xea,0x04,0x0b,0x6d,0x81,0x77,0x39,0x05] + +v_fmac_f16 v127.h, v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:0 +// GFX12: encoding: [0xe9,0xfe,0xff,0x6d,0xff,0x00,0x00,0x00] + v_fmac_f32 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] // GFX12: encoding: [0xe9,0x04,0x0a,0x56,0x01,0x77,0x39,0x05] diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop2_t16_err.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop2_t16_err.s index b339bc1960f3e..92729d4bca3ce 100644 --- a/llvm/test/MC/AMDGPU/gfx12_asm_vop2_t16_err.s +++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop2_t16_err.s @@ -28,50 +28,95 @@ v_add_f16_e32 v5, v1, v255 v_add_f16_e32 v5, v255, v2 // GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode -v_fmaak_f16_e32 v255, v1, v2, 0xfe0b -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v255.h, v1.h, v2.h, 0xfe0b +// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction -v_fmaak_f16_e32 v5, v1, v255, 0xfe0b -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v255.l, v1.l, v2.l, 0xfe0b +// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction -v_fmaak_f16_e32 v5, v255, v2, 0xfe0b -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.h, v1.h, v255.h, 0xfe0b +// GFX12: :[[@LINE-1]]:29: error: invalid operand for instruction -v_fmac_f16_dpp v255, v1, v2 dpp8:[7,6,5,4,3,2,1,0] -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.h, v255.h, v2.h, 0xfe0b +// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction -v_fmac_f16_dpp v255, v1, v2 quad_perm:[3,2,1,0] -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.l, v1.l, v255.l, 0xfe0b +// GFX12: :[[@LINE-1]]:29: error: invalid operand for instruction -v_fmac_f16_dpp v5, v1, v255 dpp8:[7,6,5,4,3,2,1,0] -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmaak_f16_e32 v5.l, v255.l, v2.l, 0xfe0b +// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction -v_fmac_f16_dpp v5, v1, v255 quad_perm:[3,2,1,0] -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.h, v1.h, v2.h dpp8:[7,6,5,4,3,2,1,0] +// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_dpp v5, v255, v2 dpp8:[7,6,5,4,3,2,1,0] -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.h, v1.h, v2.h quad_perm:[3,2,1,0] +// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_dpp v5, v255, v2 quad_perm:[3,2,1,0] -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] +// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_e32 v255, v1, v2 -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v255.l, v1.l, v2.l quad_perm:[3,2,1,0] +// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction -v_fmac_f16_e32 v5, v1, v255 -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v1.h, v255.h dpp8:[7,6,5,4,3,2,1,0] +// GFX12: :[[@LINE-1]]:28: error: invalid operand for instruction -v_fmac_f16_e32 v5, v255, v2 -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v1.h, v255.h quad_perm:[3,2,1,0] +// GFX12: :[[@LINE-1]]:28: error: invalid operand for instruction -v_fmamk_f16_e32 v255, v1, 0xfe0b, v3 -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v255.h, v2.h dpp8:[7,6,5,4,3,2,1,0] +// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction -v_fmamk_f16_e32 v5, v1, 0xfe0b, v255 -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.h, v255.h, v2.h quad_perm:[3,2,1,0] +// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction -v_fmamk_f16_e32 v5, v255, 0xfe0b, v3 -// GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode +v_fmac_f16_dpp v5.l, v1.l, v255.l dpp8:[7,6,5,4,3,2,1,0] +// GFX12: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_dpp v5.l, v1.l, v255.l quad_perm:[3,2,1,0] +// GFX12: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_dpp v5.l, v255.l, v2.l dpp8:[7,6,5,4,3,2,1,0] +// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmac_f16_dpp v5.l, v255.l, v2.l quad_perm:[3,2,1,0] +// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmac_f16_e32 v255.h, v1.h, v2.h +// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction + +v_fmac_f16_e32 v255.l, v1.l, v2.l +// GFX12: :[[@LINE-1]]:16: error: invalid operand for instruction + +v_fmac_f16_e32 v5.h, v1.h, v255.h +// GFX12: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_e32 v5.h, v255.h, v2.h +// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmac_f16_e32 v5.l, v1.l, v255.l +// GFX12: :[[@LINE-1]]:28: error: invalid operand for instruction + +v_fmac_f16_e32 v5.l, v255.l, v2.l +// GFX12: :[[@LINE-1]]:22: error: invalid operand for instruction + +v_fmamk_f16_e32 v255.h, v1.h, 0xfe0b, v3.h +// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction + +v_fmamk_f16_e32 v255.l, v1.l, 0xfe0b, v3.l +// GFX12: :[[@LINE-1]]:17: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.h, v1.h, 0xfe0b, v255.h +// GFX12: :[[@LINE-1]]:37: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.h, v255.h, 0xfe0b, v3.h +// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.l, v1.l, 0xfe0b, v255.l +// GFX12: :[[@LINE-1]]:37: error: invalid operand for instruction + +v_fmamk_f16_e32 v5.l, v255.l, 0xfe0b, v3.l +// GFX12: :[[@LINE-1]]:23: error: invalid operand for instruction v_ldexp_f16_dpp v255, v1, v2 dpp8:[7,6,5,4,3,2,1,0] // GFX12: :[[@LINE-1]]:1: error: operands are not valid for this GPU or mode diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2.txt index 4f638cd8ff54f..372721a17d4d7 100644 --- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2.txt +++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2.txt @@ -469,49 +469,70 @@ # GFX11: v_dot2acc_f32_f16 v255, 0xfe0b, v255 ; encoding: [0xff,0xfe,0xff,0x05,0x0b,0xfe,0x00,0x00] 0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, v1, v2, 0xfe0b ; encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, v1.l, v2.l, 0xfe0b ; encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, v1, v2, 0xfe0b ; encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, v127, v2, 0xfe0b ; encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, v127.l, v2.l, 0xfe0b ; encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, v127, v2, 0xfe0b ; encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmaak_f16 v5.l, v1.h, v2.l, 0xfe0b ; encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, v129/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2, 0xfe0b ; encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmaak_f16 v5.l, v127.h, v2.l, 0xfe0b ; encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, v255/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2, 0xfe0b ; encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, s1, v2, 0xfe0b ; encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, s1, v2.l, 0xfe0b ; encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, s1, v2, 0xfe0b ; encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, s105, v2, 0xfe0b ; encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, s105, v2.l, 0xfe0b ; encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, s105, v2, 0xfe0b ; encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b ; encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, vcc_lo, v2.l, 0xfe0b ; encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b ; encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b ; encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, vcc_hi, v2.l, 0xfe0b ; encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b ; encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, ttmp15, v2, 0xfe0b ; encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, ttmp15, v2.l, 0xfe0b ; encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, ttmp15, v2, 0xfe0b ; encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, m0, v2, 0xfe0b ; encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, m0, v2.l, 0xfe0b ; encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, m0, v2, 0xfe0b ; encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, exec_lo, v2, 0xfe0b ; encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, exec_lo, v2.l, 0xfe0b ; encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, exec_lo, v2, 0xfe0b ; encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, exec_hi, v2, 0xfe0b ; encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, exec_hi, v2.l, 0xfe0b ; encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, exec_hi, v2, 0xfe0b ; encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, null, v2, 0xfe0b ; encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, null, v2.l, 0xfe0b ; encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, null, v2, 0xfe0b ; encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, -1, v2, 0xfe0b ; encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmaak_f16 v5.l, -1, v2.l, 0xfe0b ; encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v5, -1, v2, 0xfe0b ; encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] -0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, 0.5, v2, 0xfe0b ; encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +0xf0,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmaak_f16 v127.l, 0.5, v127.l, 0xfe0b ; encoding: [0xf0,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmaak_f16 v127, 0.5, v127, 0xfe0b ; encoding: [0xf0,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] -0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v5, src_scc, v2, 0xfe0b ; encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmaak_f16 v5.h, src_scc, v2.h, 0xfe0b ; encoding: [0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00] -0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmaak_f16 v127, 0xfe0b, v127, 0xfe0b ; encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] +0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmaak_f16 v127.h, 0xfe0b, v127.h, 0xfe0b ; encoding: [0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00] 0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf # GFX11: v_fmaak_f32 v5, v1, v2, 0xaf123456 ; encoding: [0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf] @@ -604,49 +625,70 @@ # GFX11: v_fmac_dx9_zero_f32_e32 v255, 0xaf123456, v255 ; encoding: [0xff,0xfe,0xff,0x0d,0x56,0x34,0x12,0xaf] 0x01,0x05,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, v1.l, v2.l ; encoding: [0x01,0x05,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x6c] 0x7f,0x05,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, v127.l, v2.l ; encoding: [0x7f,0x05,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x6c] + +0x81,0x05,0x0a,0x6c +# GFX11-REAL16: v_fmac_f16_e32 v5.l, v1.h, v2.l ; encoding: [0x81,0x05,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, v129/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2 ; encoding: [0x81,0x05,0x0a,0x6c] + +0xff,0x05,0x0a,0x6c +# GFX11-REAL16: v_fmac_f16_e32 v5.l, v127.h, v2.l ; encoding: [0xff,0x05,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, v255/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2 ; encoding: [0xff,0x05,0x0a,0x6c] 0x01,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, s1, v2.l ; encoding: [0x01,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x6c] 0x69,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, s105, v2.l ; encoding: [0x69,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x6c] 0x6a,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, vcc_lo, v2.l ; encoding: [0x6a,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x6c] 0x6b,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, vcc_hi, v2.l ; encoding: [0x6b,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x6c] 0x7b,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, ttmp15, v2.l ; encoding: [0x7b,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x6c] 0x7d,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, m0, v2.l ; encoding: [0x7d,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x6c] 0x7e,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, exec_lo, v2.l ; encoding: [0x7e,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x6c] 0x7f,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, exec_hi, v2.l ; encoding: [0x7f,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x6c] 0x7c,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, null, v2.l ; encoding: [0x7c,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x6c] 0xc1,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x6c] +# GFX11-REAL16: v_fmac_f16_e32 v5.l, -1, v2.l ; encoding: [0xc1,0x04,0x0a,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x6c] -0xf0,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, 0.5, v2 ; encoding: [0xf0,0x04,0x0a,0x6c] +0xf0,0xfe,0xfe,0x6c +# GFX11-REAL16: v_fmac_f16_e32 v127.l, 0.5, v127.l ; encoding: [0xf0,0xfe,0xfe,0x6c] +# GFX11-FAKE16: v_fmac_f16_e32 v127, 0.5, v127 ; encoding: [0xf0,0xfe,0xfe,0x6c] -0xfd,0x04,0x0a,0x6c -# GFX11: v_fmac_f16_e32 v5, src_scc, v2 ; encoding: [0xfd,0x04,0x0a,0x6c] +0xfd,0x04,0x0b,0x6d +# GFX11-REAL16: v_fmac_f16_e32 v5.h, src_scc, v2.h ; encoding: [0xfd,0x04,0x0b,0x6d] -0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmac_f16_e32 v127, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00] +0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmac_f16_e32 v127.h, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00] 0x01,0x05,0x0a,0x56 # GFX11: v_fmac_f32_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x56] @@ -694,49 +736,70 @@ # GFX11: v_fmac_f32_e32 v255, 0xaf123456, v255 ; encoding: [0xff,0xfe,0xff,0x57,0x56,0x34,0x12,0xaf] 0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, v1, 0xfe0b, v3 ; encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, v1.l, 0xfe0b, v3.l ; encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, v1, 0xfe0b, v3 ; encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, v127, 0xfe0b, v3 ; encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, v127.l, 0xfe0b, v3.l ; encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, v127, 0xfe0b, v3 ; encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmamk_f16 v5.l, v1.h, 0xfe0b, v3.l ; encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, v129/*Invalid register, operand has 'VS_32_Lo128' register class*/, 0xfe0b, v3 ; encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmamk_f16 v5.l, v127.h, 0xfe0b, v3.l ; encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, v255/*Invalid register, operand has 'VS_32_Lo128' register class*/, 0xfe0b, v3 ; encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, s1, 0xfe0b, v3 ; encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, s1, 0xfe0b, v3.l ; encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, s1, 0xfe0b, v3 ; encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, s105, 0xfe0b, v3 ; encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, s105, 0xfe0b, v3.l ; encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, s105, 0xfe0b, v3 ; encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3 ; encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, vcc_lo, 0xfe0b, v3.l ; encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3 ; encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3 ; encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, vcc_hi, 0xfe0b, v3.l ; encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3 ; encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, ttmp15, 0xfe0b, v3 ; encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, ttmp15, 0xfe0b, v3.l ; encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, ttmp15, 0xfe0b, v3 ; encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, m0, 0xfe0b, v3 ; encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, m0, 0xfe0b, v3.l ; encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, m0, 0xfe0b, v3 ; encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, exec_lo, 0xfe0b, v3 ; encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, exec_lo, 0xfe0b, v3.l ; encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, exec_lo, 0xfe0b, v3 ; encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, exec_hi, 0xfe0b, v3 ; encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, exec_hi, 0xfe0b, v3.l ; encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, exec_hi, 0xfe0b, v3 ; encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, null, 0xfe0b, v3 ; encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, null, 0xfe0b, v3.l ; encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, null, 0xfe0b, v3 ; encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, -1, 0xfe0b, v3 ; encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-REAL16: v_fmamk_f16 v5.l, -1, 0xfe0b, v3.l ; encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v5, -1, 0xfe0b, v3 ; encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] -0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, 0.5, 0xfe0b, v3 ; encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +0xf0,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmamk_f16 v127.l, 0.5, 0xfe0b, v127.l ; encoding: [0xf0,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] +# GFX11-FAKE16: v_fmamk_f16 v127, 0.5, 0xfe0b, v127 ; encoding: [0xf0,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] -0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v5, src_scc, 0xfe0b, v3 ; encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmamk_f16 v5.h, src_scc, 0xfe0b, v3.h ; encoding: [0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00] -0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00 -# GFX11: v_fmamk_f16 v127, 0xfe0b, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] +0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00 +# GFX11-REAL16: v_fmamk_f16 v127.h, 0xfe0b, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00] 0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf # GFX11: v_fmamk_f32 v5, v1, 0xaf123456, v3 ; encoding: [0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf] diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp16.txt index a8a40f883cc48..84465624da6ba 100644 --- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp16.txt +++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp16.txt @@ -437,46 +437,60 @@ # GFX11: v_dot2acc_f32_f16_dpp v255, -|v255|, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xff,0x05,0xff,0x6f,0xfd,0x30] 0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] -0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01 -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01] +0xfa,0xfe,0xfe,0x6c,0x7f,0x5f,0x01,0x01 +# GFX11-REAL16: v_fmac_f16_dpp v127.l, v127.l, v127.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x5f,0x01,0x01] +# GFX11-FAKE16: v_fmac_f16_dpp v127, v127, v127 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x5f,0x01,0x01] -0xfa,0x04,0x0a,0x6c,0x01,0x60,0x01,0x13 -# GFX11: v_fmac_f16_dpp v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x60,0x01,0x13] +0xfa,0x04,0x0b,0x6d,0x81,0x60,0x01,0x13 +# GFX11-REAL16: v_fmac_f16_dpp v5.h, v1.h, v2.h row_xmask:0 row_mask:0x1 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0b,0x6d,0x81,0x60,0x01,0x13] +# GFX11-FAKE16: v_mul_i32_i24_e32 v128, 1, v176 ; encoding: [0x81,0x60,0x01,0x13] -0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xfd,0x30 -# GFX11: v_fmac_f16_dpp v127, -|v127|, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xfd,0x30] +0xfa,0xfe,0xff,0x6d,0xff,0x6f,0xfd,0x30 +# GFX11-REAL16: v_fmac_f16_dpp v127.h, -|v127.h|, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xff,0x6d,0xff,0x6f,0xfd,0x30] +# GFX11-FAKE16: v_lshlrev_b32_e32 v126, v255, v183 ; encoding: [0xff,0x6f,0xfd,0x30] 0xfa,0x04,0x0a,0x56,0x01,0x1b,0x00,0xff # GFX11: v_fmac_f32_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x56,0x01,0x1b,0x00,0xff] diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp8.txt index a1d2c34f09f2b..3d4a16d41880a 100644 --- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp8.txt +++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop2_dpp8.txt @@ -65,10 +65,19 @@ # GFX11: v_dot2acc_f32_f16_dpp v255, v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xff,0x05,0xff,0x00,0x00,0x00] 0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05 -# GFX11: v_fmac_f16_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] +# GFX11-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] +# GFX11-FAKE16: v_fmac_f16_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] -0xea,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00 -# GFX11: v_fmac_f16_dpp v127, v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00] +0xe9,0xfe,0xfe,0x6c,0x7f,0x77,0x39,0x05 +# GFX11-REAL16: v_fmac_f16_dpp v127.l, v127.l, v127.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xfe,0xfe,0x6c,0x7f,0x77,0x39,0x05] +# GFX11-FAKE16: v_fmac_f16_dpp v127, v127, v127 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0xfe,0xfe,0x6c,0x7f,0x77,0x39,0x05] + +0xe9,0x04,0x0b,0x6d,0x81,0x77,0x39,0x05 +# GFX11-REAL16: v_fmac_f16_dpp v5.h, v1.h, v2.h dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0b,0x6d,0x81,0x77,0x39,0x05] +# GFX11-FAKE16: v_dot2acc_f32_f16 v156, v129, v187 ; encoding: [0x81,0x77,0x39,0x05] + +0xea,0xfe,0xff,0x6d,0xff,0x00,0x00,0x00 +# GFX11-REAL16: v_fmac_f16_dpp v127.h, v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xff,0x6d,0xff,0x00,0x00,0x00] 0xe9,0x04,0x0a,0x56,0x01,0x77,0x39,0x05 # GFX11: v_fmac_f32_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x56,0x01,0x77,0x39,0x05] diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2.txt index 1276d898160b3..b120d6a030cc9 100644 --- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2.txt +++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2.txt @@ -446,49 +446,78 @@ # GFX12: v_cvt_pk_rtz_f16_f32_e32 v255, 0xaf123456, v255 ; encoding: [0xff,0xfe,0xff,0x5f,0x56,0x34,0x12,0xaf] 0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, v1, v2, 0xfe0b ; encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, v1.l, v2.l, 0xfe0b ; encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, v1, v2, 0xfe0b ; encoding: [0x01,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, v127, v2, 0xfe0b ; encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, v127.l, v2.l, 0xfe0b ; encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, v127, v2, 0xfe0b ; encoding: [0x7f,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, s1, v2, 0xfe0b ; encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, s1, v2.l, 0xfe0b ; encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, s1, v2, 0xfe0b ; encoding: [0x01,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, s105, v2, 0xfe0b ; encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, s105, v2.l, 0xfe0b ; encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, s105, v2, 0xfe0b ; encoding: [0x69,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b ; encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, vcc_lo, v2.l, 0xfe0b ; encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, vcc_lo, v2, 0xfe0b ; encoding: [0x6a,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b ; encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, vcc_hi, v2.l, 0xfe0b ; encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, vcc_hi, v2, 0xfe0b ; encoding: [0x6b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, ttmp15, v2, 0xfe0b ; encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, ttmp15, v2.l, 0xfe0b ; encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, ttmp15, v2, 0xfe0b ; encoding: [0x7b,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, m0, v2, 0xfe0b ; encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, m0, v2.l, 0xfe0b ; encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, m0, v2, 0xfe0b ; encoding: [0x7d,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, exec_lo, v2, 0xfe0b ; encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, exec_lo, v2.l, 0xfe0b ; encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, exec_lo, v2, 0xfe0b ; encoding: [0x7e,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, exec_hi, v2, 0xfe0b ; encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, exec_hi, v2.l, 0xfe0b ; encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, exec_hi, v2, 0xfe0b ; encoding: [0x7f,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, null, v2, 0xfe0b ; encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, null, v2.l, 0xfe0b ; encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, null, v2, 0xfe0b ; encoding: [0x7c,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, -1, v2, 0xfe0b ; encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, -1, v2.l, 0xfe0b ; encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, -1, v2, 0xfe0b ; encoding: [0xc1,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, 0.5, v2, 0xfe0b ; encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, 0.5, v2.l, 0xfe0b ; encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, 0.5, v2, 0xfe0b ; encoding: [0xf0,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v5, src_scc, v2, 0xfe0b ; encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v5.l, src_scc, v2.l, 0xfe0b ; encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, src_scc, v2, 0xfe0b ; encoding: [0xfd,0x04,0x0a,0x70,0x0b,0xfe,0x00,0x00] 0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmaak_f16 v127, 0xfe0b, v127, 0xfe0b ; encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmaak_f16 v127.l, 0xfe0b, v127.l, 0xfe0b ; encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v127, 0xfe0b, v127, 0xfe0b ; encoding: [0xff,0xfe,0xfe,0x70,0x0b,0xfe,0x00,0x00] + +0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmaak_f16 v5.l, v1.h, v2.l, 0xfe0b ; encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, v129/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2, 0xfe0b ; encoding: [0x81,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmaak_f16 v5.l, v127.h, v2.l, 0xfe0b ; encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmaak_f16 v5, v255/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2, 0xfe0b ; encoding: [0xff,0x05,0x0a,0x70,0x0b,0xfe,0x00,0x00] + +0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmaak_f16 v5.h, src_scc, v2.h, 0xfe0b ; encoding: [0xfd,0x04,0x0b,0x71,0x0b,0xfe,0x00,0x00] + +0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmaak_f16 v127.h, 0xfe0b, v127.h, 0xfe0b ; encoding: [0xff,0xfe,0xff,0x71,0x0b,0xfe,0x00,0x00] 0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf # GFX12: v_fmaak_f32 v5, v1, v2, 0xaf123456 ; encoding: [0x01,0x05,0x0a,0x5a,0x56,0x34,0x12,0xaf] @@ -536,49 +565,78 @@ # GFX12: v_fmaak_f32 v255, 0xaf123456, v255, 0xaf123456 ; encoding: [0xff,0xfe,0xff,0x5b,0x56,0x34,0x12,0xaf] 0x01,0x05,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, v1.l, v2.l ; encoding: [0x01,0x05,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x6c] 0x7f,0x05,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, v127.l, v2.l ; encoding: [0x7f,0x05,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, v127, v2 ; encoding: [0x7f,0x05,0x0a,0x6c] 0x01,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, s1, v2.l ; encoding: [0x01,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, s1, v2 ; encoding: [0x01,0x04,0x0a,0x6c] 0x69,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, s105, v2.l ; encoding: [0x69,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, s105, v2 ; encoding: [0x69,0x04,0x0a,0x6c] 0x6a,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, vcc_lo, v2.l ; encoding: [0x6a,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, vcc_lo, v2 ; encoding: [0x6a,0x04,0x0a,0x6c] 0x6b,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, vcc_hi, v2.l ; encoding: [0x6b,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, vcc_hi, v2 ; encoding: [0x6b,0x04,0x0a,0x6c] 0x7b,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, ttmp15, v2.l ; encoding: [0x7b,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, ttmp15, v2 ; encoding: [0x7b,0x04,0x0a,0x6c] 0x7d,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, m0, v2.l ; encoding: [0x7d,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, m0, v2 ; encoding: [0x7d,0x04,0x0a,0x6c] 0x7e,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, exec_lo, v2.l ; encoding: [0x7e,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, exec_lo, v2 ; encoding: [0x7e,0x04,0x0a,0x6c] 0x7f,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, exec_hi, v2.l ; encoding: [0x7f,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, exec_hi, v2 ; encoding: [0x7f,0x04,0x0a,0x6c] 0x7c,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, null, v2.l ; encoding: [0x7c,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, null, v2 ; encoding: [0x7c,0x04,0x0a,0x6c] 0xc1,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, -1, v2.l ; encoding: [0xc1,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, -1, v2 ; encoding: [0xc1,0x04,0x0a,0x6c] 0xf0,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, 0.5, v2 ; encoding: [0xf0,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, 0.5, v2.l ; encoding: [0xf0,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, 0.5, v2 ; encoding: [0xf0,0x04,0x0a,0x6c] 0xfd,0x04,0x0a,0x6c -# GFX12: v_fmac_f16_e32 v5, src_scc, v2 ; encoding: [0xfd,0x04,0x0a,0x6c] +# GFX12-REAL16: v_fmac_f16_e32 v5.l, src_scc, v2.l ; encoding: [0xfd,0x04,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, src_scc, v2 ; encoding: [0xfd,0x04,0x0a,0x6c] 0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmac_f16_e32 v127, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmac_f16_e32 v127.l, 0xfe0b, v127.l ; encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmac_f16_e32 v127, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x6c,0x0b,0xfe,0x00,0x00] + +0x81,0x05,0x0a,0x6c +# GFX12-REAL16: v_fmac_f16_e32 v5.l, v1.h, v2.l ; encoding: [0x81,0x05,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, v129/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2 ; encoding: [0x81,0x05,0x0a,0x6c] + +0xff,0x05,0x0a,0x6c +# GFX12-REAL16: v_fmac_f16_e32 v5.l, v127.h, v2.l ; encoding: [0xff,0x05,0x0a,0x6c] +# GFX12-FAKE16: v_fmac_f16_e32 v5, v255/*Invalid register, operand has 'VS_32_Lo128' register class*/, v2 ; encoding: [0xff,0x05,0x0a,0x6c] + +0xfd,0x04,0x0b,0x6d +# GFX12-REAL16: v_fmac_f16_e32 v5.h, src_scc, v2.h ; encoding: [0xfd,0x04,0x0b,0x6d] + +0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmac_f16_e32 v127.h, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x6d,0x0b,0xfe,0x00,0x00] 0x01,0x05,0x0a,0x56 # GFX12: v_fmac_f32_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x56] @@ -626,49 +684,78 @@ # GFX12: v_fmac_f32_e32 v255, 0xaf123456, v255 ; encoding: [0xff,0xfe,0xff,0x57,0x56,0x34,0x12,0xaf] 0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, v1, 0xfe0b, v3 ; encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, v1.l, 0xfe0b, v3.l ; encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, v1, 0xfe0b, v3 ; encoding: [0x01,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, v127, 0xfe0b, v3 ; encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, v127.l, 0xfe0b, v3.l ; encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, v127, 0xfe0b, v3 ; encoding: [0x7f,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, s1, 0xfe0b, v3 ; encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, s1, 0xfe0b, v3.l ; encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, s1, 0xfe0b, v3 ; encoding: [0x01,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, s105, 0xfe0b, v3 ; encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, s105, 0xfe0b, v3.l ; encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, s105, 0xfe0b, v3 ; encoding: [0x69,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3 ; encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, vcc_lo, 0xfe0b, v3.l ; encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, vcc_lo, 0xfe0b, v3 ; encoding: [0x6a,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3 ; encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, vcc_hi, 0xfe0b, v3.l ; encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, vcc_hi, 0xfe0b, v3 ; encoding: [0x6b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, ttmp15, 0xfe0b, v3 ; encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, ttmp15, 0xfe0b, v3.l ; encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, ttmp15, 0xfe0b, v3 ; encoding: [0x7b,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, m0, 0xfe0b, v3 ; encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, m0, 0xfe0b, v3.l ; encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, m0, 0xfe0b, v3 ; encoding: [0x7d,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, exec_lo, 0xfe0b, v3 ; encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, exec_lo, 0xfe0b, v3.l ; encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, exec_lo, 0xfe0b, v3 ; encoding: [0x7e,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, exec_hi, 0xfe0b, v3 ; encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, exec_hi, 0xfe0b, v3.l ; encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, exec_hi, 0xfe0b, v3 ; encoding: [0x7f,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, null, 0xfe0b, v3 ; encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, null, 0xfe0b, v3.l ; encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, null, 0xfe0b, v3 ; encoding: [0x7c,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, -1, 0xfe0b, v3 ; encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, -1, 0xfe0b, v3.l ; encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, -1, 0xfe0b, v3 ; encoding: [0xc1,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, 0.5, 0xfe0b, v3 ; encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, 0.5, 0xfe0b, v3.l ; encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, 0.5, 0xfe0b, v3 ; encoding: [0xf0,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v5, src_scc, 0xfe0b, v3 ; encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v5.l, src_scc, 0xfe0b, v3.l ; encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, src_scc, 0xfe0b, v3 ; encoding: [0xfd,0x06,0x0a,0x6e,0x0b,0xfe,0x00,0x00] 0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00 -# GFX12: v_fmamk_f16 v127, 0xfe0b, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-REAL16: v_fmamk_f16 v127.l, 0xfe0b, 0xfe0b, v127.l ; encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v127, 0xfe0b, 0xfe0b, v127 ; encoding: [0xff,0xfe,0xfe,0x6e,0x0b,0xfe,0x00,0x00] + +0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmamk_f16 v5.l, v1.h, 0xfe0b, v3.l ; encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, v129/*Invalid register, operand has 'VS_32_Lo128' register class*/, 0xfe0b, v3 ; encoding: [0x81,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmamk_f16 v5.l, v127.h, 0xfe0b, v3.l ; encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] +# GFX12-FAKE16: v_fmamk_f16 v5, v255/*Invalid register, operand has 'VS_32_Lo128' register class*/, 0xfe0b, v3 ; encoding: [0xff,0x07,0x0a,0x6e,0x0b,0xfe,0x00,0x00] + +0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmamk_f16 v5.h, src_scc, 0xfe0b, v3.h ; encoding: [0xfd,0x06,0x0b,0x6f,0x0b,0xfe,0x00,0x00] + +0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00 +# GFX12-REAL16: v_fmamk_f16 v127.h, 0xfe0b, 0xfe0b, v127.h ; encoding: [0xff,0xfe,0xff,0x6f,0x0b,0xfe,0x00,0x00] 0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf # GFX12: v_fmamk_f32 v5, v1, 0xaf123456, v3 ; encoding: [0x01,0x07,0x0a,0x58,0x56,0x34,0x12,0xaf] diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp16.txt index 551fb0d311188..9a9a1eb2cf959 100644 --- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp16.txt +++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp16.txt @@ -383,46 +383,68 @@ # GFX12: v_cvt_pk_rtz_f16_f32_dpp v255, -|v255|, -|v255| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xff,0x5f,0xff,0x6f,0xfd,0x30] 0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1b,0x00,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0xe4,0x00,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x40,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_half_mirror row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x41,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shl:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x01,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shl:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x0f,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x11,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_shr:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x1f,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_ror:1 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x21,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_ror:15 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x2f,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x50,0x01,0xff] 0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01 -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_share:15 row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x5f,0x01,0x01] 0xfa,0x04,0x0a,0x6c,0x01,0x60,0x01,0x13 -# GFX12: v_fmac_f16_dpp v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x60,0x01,0x13] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l row_xmask:0 row_mask:0x1 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x60,0x01,0x13] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 row_xmask:0 row_mask:0x1 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0a,0x6c,0x01,0x60,0x01,0x13] 0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xfd,0x30 -# GFX12: v_fmac_f16_dpp v127, -|v127|, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xfd,0x30] +# GFX12-REAL16: v_fmac_f16_dpp v127.l, -|v127.l|, -|v127.l| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xfd,0x30] +# GFX12-FAKE16: v_fmac_f16_dpp v127, -|v127|, -|v127| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xfe,0x6c,0x7f,0x6f,0xfd,0x30] + +0xfa,0x04,0x0b,0x6d,0x81,0x60,0x01,0x13 +# GFX12-REAL16: v_fmac_f16_dpp v5.h, v1.h, v2.h row_xmask:0 row_mask:0x1 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0b,0x6d,0x81,0x60,0x01,0x13] +# GFX12-FAKE16: v_mul_i32_i24_e32 v128, 1, v176 ; encoding: [0x81,0x60,0x01,0x13] + +0xfa,0xfe,0xff,0x6d,0xff,0x6f,0xfd,0x30 +# GFX12-REAL16: v_fmac_f16_dpp v127.h, -|v127.h|, -|v127.h| row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:1 fi:1 ; encoding: [0xfa,0xfe,0xff,0x6d,0xff,0x6f,0xfd,0x30] +# GFX12-FAKE16: v_lshlrev_b32_e32 v126, v255, v183 ; encoding: [0xff,0x6f,0xfd,0x30] 0xfa,0x04,0x0a,0x56,0x01,0x1b,0x00,0xff # GFX12: v_fmac_f32_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x56,0x01,0x1b,0x00,0xff] diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp8.txt index bbf494c153fd3..0ef2c082f7dda 100644 --- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp8.txt +++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop2_dpp8.txt @@ -59,10 +59,19 @@ # GFX12: v_cvt_pk_rtz_f16_f32_dpp v255, v255, v255 dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xff,0x5f,0xff,0x00,0x00,0x00] 0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05 -# GFX12: v_fmac_f16_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] +# GFX12-REAL16: v_fmac_f16_dpp v5.l, v1.l, v2.l dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] +# GFX12-FAKE16: v_fmac_f16_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x6c,0x01,0x77,0x39,0x05] 0xea,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00 -# GFX12: v_fmac_f16_dpp v127, v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00] +# GFX12-REAL16: v_fmac_f16_dpp v127.l, v127.l, v127.l dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00] +# GFX12-FAKE16: v_fmac_f16_dpp v127, v127, v127 dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xfe,0x6c,0x7f,0x00,0x00,0x00] + +0xe9,0x04,0x0b,0x6d,0x81,0x77,0x39,0x05 +# GFX12-REAL16: v_fmac_f16_dpp v5.h, v1.h, v2.h dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0b,0x6d,0x81,0x77,0x39,0x05] +# GFX12-FAKE16: v_add_f64_e32 v[156:157], v[129:130], v[187:188] ; encoding: [0x81,0x77,0x39,0x05] + +0xea,0xfe,0xff,0x6d,0xff,0x00,0x00,0x00 +# GFX12-REAL16: v_fmac_f16_dpp v127.h, v127.h, v127.h dpp8:[0,0,0,0,0,0,0,0] fi:1 ; encoding: [0xea,0xfe,0xff,0x6d,0xff,0x00,0x00,0x00] 0xe9,0x04,0x0a,0x56,0x01,0x77,0x39,0x05 # GFX12: v_fmac_f32_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0xe9,0x04,0x0a,0x56,0x01,0x77,0x39,0x05]