diff options
Diffstat (limited to 'contrib/llvm/lib/Target/AMDGPU/VOP2Instructions.td')
| -rw-r--r-- | contrib/llvm/lib/Target/AMDGPU/VOP2Instructions.td | 186 |
1 files changed, 132 insertions, 54 deletions
diff --git a/contrib/llvm/lib/Target/AMDGPU/VOP2Instructions.td b/contrib/llvm/lib/Target/AMDGPU/VOP2Instructions.td index 00e5ab3..d5acb49 100644 --- a/contrib/llvm/lib/Target/AMDGPU/VOP2Instructions.td +++ b/contrib/llvm/lib/Target/AMDGPU/VOP2Instructions.td @@ -40,12 +40,24 @@ class VOP2_MADKe <bits<6> op, VOPProfile P> : Enc64 { class VOP2_SDWAe <bits<6> op, VOPProfile P> : VOP_SDWAe <P> { bits<8> vdst; bits<8> src1; - + + let Inst{8-0} = 0xf9; // sdwa + let Inst{16-9} = !if(P.HasSrc1, src1{7-0}, 0); + let Inst{24-17} = !if(P.EmitDst, vdst{7-0}, 0); + let Inst{30-25} = op; + let Inst{31} = 0x0; // encoding +} + +class VOP2_SDWA9Ae <bits<6> op, VOPProfile P> : VOP_SDWA9Ae <P> { + bits<8> vdst; + bits<9> src1; + let Inst{8-0} = 0xf9; // sdwa let Inst{16-9} = !if(P.HasSrc1, src1{7-0}, 0); let Inst{24-17} = !if(P.EmitDst, vdst{7-0}, 0); let Inst{30-25} = op; let Inst{31} = 0x0; // encoding + let Inst{63} = !if(P.HasSrc1, src1{8}, 0); // src1_sgpr } class VOP2_Pseudo <string opName, VOPProfile P, list<dag> pattern=[], string suffix = "_e32"> : @@ -93,6 +105,8 @@ class VOP2_Real <VOP2_Pseudo ps, int EncodingFamily> : let Constraints = ps.Constraints; let DisableEncoding = ps.DisableEncoding; let TSFlags = ps.TSFlags; + let UseNamedOperandTable = ps.UseNamedOperandTable; + let Uses = ps.Uses; } class VOP2_SDWA_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[]> : @@ -103,7 +117,10 @@ class VOP2_SDWA_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[]> : class getVOP2Pat64 <SDPatternOperator node, VOPProfile P> : LetDummies { list<dag> ret = !if(P.HasModifiers, [(set P.DstVT:$vdst, - (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)), + (node (P.Src0VT + !if(P.HasOMod, + (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod), + (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp))), (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers))))], [(set P.DstVT:$vdst, (node P.Src0VT:$src0, P.Src1VT:$src1))]); } @@ -119,11 +136,9 @@ multiclass VOP2Inst <string opName, def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>, Commutable_REV<revOp#"_e64", !eq(revOp, opName)>; - def _sdwa : VOP2_SDWA_Pseudo <opName, P>, - Commutable_REV<revOp#"_sdwa", !eq(revOp, opName)>; + def _sdwa : VOP2_SDWA_Pseudo <opName, P>; } -// TODO: add SDWA pseudo instructions for VOP2bInst and VOP2eInst multiclass VOP2bInst <string opName, VOPProfile P, SDPatternOperator node = null_frag, @@ -134,10 +149,12 @@ multiclass VOP2bInst <string opName, let Uses = !if(useSGPRInput, [VCC, EXEC], [EXEC]), Defs = [VCC] in { def _e32 : VOP2_Pseudo <opName, P>, Commutable_REV<revOp#"_e32", !eq(revOp, opName)>; - - def _sdwa : VOP2_SDWA_Pseudo <opName, P>, - Commutable_REV<revOp#"_sdwa", !eq(revOp, opName)>; + + def _sdwa : VOP2_SDWA_Pseudo <opName, P> { + let AsmMatchConverter = "cvtSdwaVOP2b"; + } } + def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>, Commutable_REV<revOp#"_e64", !eq(revOp, opName)>; } @@ -154,6 +171,7 @@ multiclass VOP2eInst <string opName, def _e32 : VOP2_Pseudo <opName, P>, Commutable_REV<revOp#"_e32", !eq(revOp, opName)>; } + def _e64 : VOP3_Pseudo <opName, P, getVOP2Pat64<node, P>.ret>, Commutable_REV<revOp#"_e64", !eq(revOp, opName)>; } @@ -162,8 +180,11 @@ multiclass VOP2eInst <string opName, class VOP_MADAK <ValueType vt> : VOPProfile <[vt, vt, vt, vt]> { field Operand ImmOpType = !if(!eq(vt.Size, 32), f32kimm, f16kimm); field dag Ins32 = (ins VCSrc_f32:$src0, VGPR_32:$src1, ImmOpType:$imm); - field string Asm32 = "$vdst, $src0, $src1, $imm"; field bit HasExt = 0; + + // Hack to stop printing _e64 + let DstRC = RegisterOperand<VGPR_32>; + field string Asm32 = " $vdst, $src0, $src1, $imm"; } def VOP_MADAK_F16 : VOP_MADAK <f16>; @@ -172,45 +193,55 @@ def VOP_MADAK_F32 : VOP_MADAK <f32>; class VOP_MADMK <ValueType vt> : VOPProfile <[vt, vt, vt, vt]> { field Operand ImmOpType = !if(!eq(vt.Size, 32), f32kimm, f16kimm); field dag Ins32 = (ins VCSrc_f32:$src0, ImmOpType:$imm, VGPR_32:$src1); - field string Asm32 = "$vdst, $src0, $imm, $src1"; field bit HasExt = 0; + + // Hack to stop printing _e64 + let DstRC = RegisterOperand<VGPR_32>; + field string Asm32 = " $vdst, $src0, $imm, $src1"; } def VOP_MADMK_F16 : VOP_MADMK <f16>; def VOP_MADMK_F32 : VOP_MADMK <f32>; +// FIXME: Remove src2_modifiers. It isn't used, so is wasting memory +// and processing time but it makes it easier to convert to mad. class VOP_MAC <ValueType vt> : VOPProfile <[vt, vt, vt, vt]> { let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1, VGPR_32:$src2); let Ins64 = getIns64<Src0RC64, Src1RC64, RegisterOperand<VGPR_32>, 3, - HasModifiers, Src0Mod, Src1Mod, Src2Mod>.ret; + HasModifiers, HasOMod, Src0Mod, Src1Mod, Src2Mod>.ret; let InsDPP = (ins Src0ModDPP:$src0_modifiers, Src0DPP:$src0, Src1ModDPP:$src1_modifiers, Src1DPP:$src1, VGPR_32:$src2, // stub argument dpp_ctrl:$dpp_ctrl, row_mask:$row_mask, bank_mask:$bank_mask, bound_ctrl:$bound_ctrl); + let InsSDWA = (ins Src0ModSDWA:$src0_modifiers, Src0SDWA:$src0, Src1ModSDWA:$src1_modifiers, Src1SDWA:$src1, VGPR_32:$src2, // stub argument - clampmod:$clamp, dst_sel:$dst_sel, dst_unused:$dst_unused, + clampmod:$clamp, omod:$omod, + dst_sel:$dst_sel, dst_unused:$dst_unused, src0_sel:$src0_sel, src1_sel:$src1_sel); let Asm32 = getAsm32<1, 2, vt>.ret; + let Asm64 = getAsm64<1, 2, HasModifiers, HasOMod, vt>.ret; let AsmDPP = getAsmDPP<1, 2, HasModifiers, vt>.ret; - let AsmSDWA = getAsmSDWA<1, 2, HasModifiers, vt>.ret; + let AsmSDWA = getAsmSDWA<1, 2, vt>.ret; + let AsmSDWA9 = getAsmSDWA9<1, 1, 2, vt>.ret; let HasSrc2 = 0; let HasSrc2Mods = 0; let HasExt = 1; + let HasSDWA9 = 0; } def VOP_MAC_F16 : VOP_MAC <f16> { // FIXME: Move 'Asm64' definition to VOP_MAC, and use 'vt'. Currently it gives // 'not a string initializer' error. - let Asm64 = getAsm64<1, 2, HasModifiers, f16>.ret; + let Asm64 = getAsm64<1, 2, HasModifiers, HasOMod, f16>.ret; } def VOP_MAC_F32 : VOP_MAC <f32> { // FIXME: Move 'Asm64' definition to VOP_MAC, and use 'vt'. Currently it gives // 'not a string initializer' error. - let Asm64 = getAsm64<1, 2, HasModifiers, f32>.ret; + let Asm64 = getAsm64<1, 2, HasModifiers, HasOMod, f32>.ret; } // Write out to vcc or arbitrary SGPR. @@ -218,6 +249,7 @@ def VOP2b_I32_I1_I32_I32 : VOPProfile<[i32, i32, i32, untyped]> { let Asm32 = "$vdst, vcc, $src0, $src1"; let Asm64 = "$vdst, $sdst, $src0, $src1"; let AsmSDWA = "$vdst, vcc, $src0_modifiers, $src1_modifiers$clamp $dst_sel $dst_unused $src0_sel $src1_sel"; + let AsmSDWA9 = "$vdst, vcc, $src0_modifiers, $src1_modifiers$clamp $dst_sel $dst_unused $src0_sel $src1_sel"; let AsmDPP = "$vdst, vcc, $src0, $src1 $dpp_ctrl$row_mask$bank_mask$bound_ctrl"; let Outs32 = (outs DstRC:$vdst); let Outs64 = (outs DstRC:$vdst, SReg_64:$sdst); @@ -235,6 +267,7 @@ def VOP2b_I32_I1_I32_I32_I1 : VOPProfile<[i32, i32, i32, i1]> { let Asm32 = "$vdst, vcc, $src0, $src1, vcc"; let Asm64 = "$vdst, $sdst, $src0, $src1, $src2"; let AsmSDWA = "$vdst, vcc, $src0_modifiers, $src1_modifiers, vcc $clamp $dst_sel $dst_unused $src0_sel $src1_sel"; + let AsmSDWA9 = "$vdst, vcc, $src0_modifiers, $src1_modifiers, vcc $clamp $dst_sel $dst_unused $src0_sel $src1_sel"; let AsmDPP = "$vdst, vcc, $src0, $src1, vcc $dpp_ctrl$row_mask$bank_mask$bound_ctrl"; let Outs32 = (outs DstRC:$vdst); let Outs64 = (outs DstRC:$vdst, SReg_64:$sdst); @@ -243,9 +276,10 @@ def VOP2b_I32_I1_I32_I32_I1 : VOPProfile<[i32, i32, i32, i1]> { // implicit VCC use. let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1); - let InsSDWA = (ins Src0Mod:$src0_modifiers, Src0SDWA:$src0, - Src1Mod:$src1_modifiers, Src1SDWA:$src1, - clampmod:$clamp, dst_sel:$dst_sel, dst_unused:$dst_unused, + let InsSDWA = (ins Src0ModSDWA:$src0_modifiers, Src0SDWA:$src0, + Src1ModSDWA:$src1_modifiers, Src1SDWA:$src1, + clampmod:$clamp, omod:$omod, + dst_sel:$dst_sel, dst_unused:$dst_unused, src0_sel:$src0_sel, src1_sel:$src1_sel); let InsDPP = (ins Src0Mod:$src0_modifiers, Src0DPP:$src0, @@ -253,6 +287,7 @@ def VOP2b_I32_I1_I32_I32_I1 : VOPProfile<[i32, i32, i32, i1]> { dpp_ctrl:$dpp_ctrl, row_mask:$row_mask, bank_mask:$bank_mask, bound_ctrl:$bound_ctrl); let HasExt = 1; + let HasSDWA9 = 1; } // Read in from vcc or arbitrary SGPR @@ -275,15 +310,19 @@ def VOP_READLANE : VOPProfile<[i32, i32, i32]> { let Ins64 = Ins32; let Asm32 = " $vdst, $src0, $src1"; let Asm64 = Asm32; + let HasExt = 0; + let HasSDWA9 = 0; } def VOP_WRITELANE : VOPProfile<[i32, i32, i32]> { let Outs32 = (outs VGPR_32:$vdst); let Outs64 = Outs32; - let Ins32 = (ins SReg_32:$src0, SCSrc_b32:$src1); + let Ins32 = (ins SCSrc_b32:$src0, SCSrc_b32:$src1); let Ins64 = Ins32; let Asm32 = " $vdst, $src0, $src1"; let Asm64 = Asm32; + let HasExt = 0; + let HasSDWA9 = 0; } //===----------------------------------------------------------------------===// @@ -293,7 +332,7 @@ def VOP_WRITELANE : VOPProfile<[i32, i32, i32]> { let SubtargetPredicate = isGCN in { defm V_CNDMASK_B32 : VOP2eInst <"v_cndmask_b32", VOP2e_I32_I32_I32_I1>; -def V_MADMK_F32 : VOP2_Pseudo <"v_madmk_f32", VOP_MADMK_F32>; +def V_MADMK_F32 : VOP2_Pseudo <"v_madmk_f32", VOP_MADMK_F32, [], "">; let isCommutable = 1 in { defm V_ADD_F32 : VOP2Inst <"v_add_f32", VOP_F32_F32_F32, fadd>; @@ -323,7 +362,7 @@ let Constraints = "$vdst = $src2", DisableEncoding="$src2", defm V_MAC_F32 : VOP2Inst <"v_mac_f32", VOP_MAC_F32>; } -def V_MADAK_F32 : VOP2_Pseudo <"v_madak_f32", VOP_MADAK_F32>; +def V_MADAK_F32 : VOP2_Pseudo <"v_madak_f32", VOP_MADAK_F32, [], "">; // No patterns so that the scalar instructions are always selected. // The scalar versions will be replaced with vector when needed later. @@ -346,20 +385,29 @@ def V_READLANE_B32 : VOP2_Pseudo<"v_readlane_b32", VOP_READLANE, def V_WRITELANE_B32 : VOP2_Pseudo<"v_writelane_b32", VOP_WRITELANE, [], "">; } // End isConvergent = 1 -defm V_BFM_B32 : VOP2Inst <"v_bfm_b32", VOP_I32_I32_I32>; -defm V_BCNT_U32_B32 : VOP2Inst <"v_bcnt_u32_b32", VOP_I32_I32_I32>; -defm V_MBCNT_LO_U32_B32 : VOP2Inst <"v_mbcnt_lo_u32_b32", VOP_I32_I32_I32, int_amdgcn_mbcnt_lo>; -defm V_MBCNT_HI_U32_B32 : VOP2Inst <"v_mbcnt_hi_u32_b32", VOP_I32_I32_I32, int_amdgcn_mbcnt_hi>; -defm V_LDEXP_F32 : VOP2Inst <"v_ldexp_f32", VOP_F32_F32_I32, AMDGPUldexp>; -defm V_CVT_PKACCUM_U8_F32 : VOP2Inst <"v_cvt_pkaccum_u8_f32", VOP_I32_F32_I32>; // TODO: set "Uses = dst" -defm V_CVT_PKNORM_I16_F32 : VOP2Inst <"v_cvt_pknorm_i16_f32", VOP_I32_F32_F32>; -defm V_CVT_PKNORM_U16_F32 : VOP2Inst <"v_cvt_pknorm_u16_f32", VOP_I32_F32_F32>; -defm V_CVT_PKRTZ_F16_F32 : VOP2Inst <"v_cvt_pkrtz_f16_f32", VOP_I32_F32_F32, int_SI_packf16>; -defm V_CVT_PK_U16_U32 : VOP2Inst <"v_cvt_pk_u16_u32", VOP_I32_I32_I32>; -defm V_CVT_PK_I16_I32 : VOP2Inst <"v_cvt_pk_i16_i32", VOP_I32_I32_I32>; +defm V_BFM_B32 : VOP2Inst <"v_bfm_b32", VOP_NO_EXT<VOP_I32_I32_I32>>; +defm V_BCNT_U32_B32 : VOP2Inst <"v_bcnt_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>>; +defm V_MBCNT_LO_U32_B32 : VOP2Inst <"v_mbcnt_lo_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>, int_amdgcn_mbcnt_lo>; +defm V_MBCNT_HI_U32_B32 : VOP2Inst <"v_mbcnt_hi_u32_b32", VOP_NO_EXT<VOP_I32_I32_I32>, int_amdgcn_mbcnt_hi>; +defm V_LDEXP_F32 : VOP2Inst <"v_ldexp_f32", VOP_NO_EXT<VOP_F32_F32_I32>, AMDGPUldexp>; +defm V_CVT_PKACCUM_U8_F32 : VOP2Inst <"v_cvt_pkaccum_u8_f32", VOP_NO_EXT<VOP_I32_F32_I32>>; // TODO: set "Uses = dst" +defm V_CVT_PKNORM_I16_F32 : VOP2Inst <"v_cvt_pknorm_i16_f32", VOP_NO_EXT<VOP_I32_F32_F32>>; +defm V_CVT_PKNORM_U16_F32 : VOP2Inst <"v_cvt_pknorm_u16_f32", VOP_NO_EXT<VOP_I32_F32_F32>>; +defm V_CVT_PKRTZ_F16_F32 : VOP2Inst <"v_cvt_pkrtz_f16_f32", VOP_NO_EXT<VOP_I32_F32_F32>, AMDGPUpkrtz_f16_f32>; +defm V_CVT_PK_U16_U32 : VOP2Inst <"v_cvt_pk_u16_u32", VOP_NO_EXT<VOP_I32_I32_I32>>; +defm V_CVT_PK_I16_I32 : VOP2Inst <"v_cvt_pk_i16_i32", VOP_NO_EXT<VOP_I32_I32_I32>>; } // End SubtargetPredicate = isGCN +def : Pat< + (AMDGPUadde i32:$src0, i32:$src1, i1:$src2), + (V_ADDC_U32_e64 $src0, $src1, $src2) +>; + +def : Pat< + (AMDGPUsube i32:$src0, i32:$src1, i1:$src2), + (V_SUBB_U32_e64 $src0, $src1, $src2) +>; // These instructions only exist on SI and CI let SubtargetPredicate = isSICI in { @@ -376,9 +424,9 @@ defm V_LSHL_B32 : VOP2Inst <"v_lshl_b32", VOP_I32_I32_I32>; } // End let SubtargetPredicate = SICI -let SubtargetPredicate = isVI in { +let SubtargetPredicate = Has16BitInsts in { -def V_MADMK_F16 : VOP2_Pseudo <"v_madmk_f16", VOP_MADMK_F16>; +def V_MADMK_F16 : VOP2_Pseudo <"v_madmk_f16", VOP_MADMK_F16, [], "">; defm V_LSHLREV_B16 : VOP2Inst <"v_lshlrev_b16", VOP_I16_I16_I16>; defm V_LSHRREV_B16 : VOP2Inst <"v_lshrrev_b16", VOP_I16_I16_I16>; defm V_ASHRREV_I16 : VOP2Inst <"v_ashrrev_i16", VOP_I16_I16_I16>; @@ -389,7 +437,7 @@ defm V_ADD_F16 : VOP2Inst <"v_add_f16", VOP_F16_F16_F16, fadd>; defm V_SUB_F16 : VOP2Inst <"v_sub_f16", VOP_F16_F16_F16, fsub>; defm V_SUBREV_F16 : VOP2Inst <"v_subrev_f16", VOP_F16_F16_F16, null_frag, "v_sub_f16">; defm V_MUL_F16 : VOP2Inst <"v_mul_f16", VOP_F16_F16_F16, fmul>; -def V_MADAK_F16 : VOP2_Pseudo <"v_madak_f16", VOP_MADAK_F16>; +def V_MADAK_F16 : VOP2_Pseudo <"v_madak_f16", VOP_MADAK_F16, [], "">; defm V_ADD_U16 : VOP2Inst <"v_add_u16", VOP_I16_I16_I16>; defm V_SUB_U16 : VOP2Inst <"v_sub_u16" , VOP_I16_I16_I16>; defm V_SUBREV_U16 : VOP2Inst <"v_subrev_u16", VOP_I16_I16_I16, null_frag, "v_sub_u16">; @@ -407,7 +455,7 @@ defm V_MAC_F16 : VOP2Inst <"v_mac_f16", VOP_MAC_F16>; } } // End isCommutable = 1 -} // End SubtargetPredicate = isVI +} // End SubtargetPredicate = Has16BitInsts // Note: 16-bit instructions produce a 0 result in the high 16-bits. multiclass Arithmetic_i16_Pats <SDPatternOperator op, Instruction inst> { @@ -457,7 +505,7 @@ class ZExt_i16_i1_Pat <SDNode ext> : Pat < (V_CNDMASK_B32_e64 (i32 0), (i32 1), $src) >; -let Predicates = [isVI] in { +let Predicates = [Has16BitInsts] in { defm : Arithmetic_i16_Pats<add, V_ADD_U16_e64>; defm : Arithmetic_i16_Pats<mul, V_MUL_LO_U16_e64>; @@ -494,7 +542,15 @@ def : Pat < (V_CNDMASK_B32_e64 (i32 0), (i32 -1), $src) >; -} // End Predicates = [isVI] +// Undo sub x, c -> add x, -c canonicalization since c is more likely +// an inline immediate than -c. +// TODO: Also do for 64-bit. +def : Pat< + (add i16:$src0, (i16 NegSubInlineConst16:$src1)), + (V_SUB_U16_e64 $src0, NegSubInlineConst16:$src1) +>; + +} // End Predicates = [Has16BitInsts] //===----------------------------------------------------------------------===// // SI @@ -566,7 +622,10 @@ defm V_SUBB_U32 : VOP2be_Real_e32e64_si <0x29>; defm V_SUBBREV_U32 : VOP2be_Real_e32e64_si <0x2a>; defm V_READLANE_B32 : VOP2_Real_si <0x01>; + +let InOperandList = (ins SSrc_b32:$src0, SCSrc_b32:$src1) in { defm V_WRITELANE_B32 : VOP2_Real_si <0x02>; +} defm V_MAC_LEGACY_F32 : VOP2_Real_e32e64_si <0x6>; defm V_MIN_LEGACY_F32 : VOP2_Real_e32e64_si <0xd>; @@ -635,6 +694,17 @@ multiclass VOP2_Real_e64_vi <bits<10> op> { VOP3e_vi <op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl>; } +multiclass VOP2_Real_e64only_vi <bits<10> op> { + def _e64_vi : + VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>, + VOP3e_vi <op, !cast<VOP3_Pseudo>(NAME#"_e64").Pfl> { + // Hack to stop printing _e64 + VOP3_Pseudo ps = !cast<VOP3_Pseudo>(NAME#"_e64"); + let OutOperandList = (outs VGPR_32:$vdst); + let AsmString = ps.Mnemonic # " " # ps.AsmOperands; + } +} + multiclass Base_VOP2be_Real_e32e64_vi <bits<6> op> : VOP2_Real_e32_vi<op> { def _e64_vi : VOP3_Real<!cast<VOP3_Pseudo>(NAME#"_e64"), SIEncodingFamily.VI>, @@ -646,22 +716,28 @@ multiclass Base_VOP2_Real_e32e64_vi <bits<6> op> : VOP2_Real_e64_vi<{0, 1, 0, 0, op{5-0}}>; } // End AssemblerPredicates = [isVI], DecoderNamespace = "VI" - + multiclass VOP2_SDWA_Real <bits<6> op> { def _sdwa_vi : VOP_SDWA_Real <!cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa")>, VOP2_SDWAe <op{5-0}, !cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa").Pfl>; } +multiclass VOP2_SDWA9_Real <bits<6> op> { + def _sdwa_gfx9 : + VOP_SDWA9_Real <!cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa")>, + VOP2_SDWA9Ae <op{5-0}, !cast<VOP2_SDWA_Pseudo>(NAME#"_sdwa").Pfl>; +} + multiclass VOP2be_Real_e32e64_vi <bits<6> op> : - Base_VOP2be_Real_e32e64_vi<op>, VOP2_SDWA_Real<op> { + Base_VOP2be_Real_e32e64_vi<op>, VOP2_SDWA_Real<op>, VOP2_SDWA9_Real<op> { // For now left dpp only for asm/dasm // TODO: add corresponding pseudo def _dpp : VOP2_DPP<op, !cast<VOP2_Pseudo>(NAME#"_e32")>; } multiclass VOP2_Real_e32e64_vi <bits<6> op> : - Base_VOP2_Real_e32e64_vi<op>, VOP2_SDWA_Real<op> { + Base_VOP2_Real_e32e64_vi<op>, VOP2_SDWA_Real<op>, VOP2_SDWA9_Real<op> { // For now left dpp only for asm/dasm // TODO: add corresponding pseudo def _dpp : VOP2_DPP<op, !cast<VOP2_Pseudo>(NAME#"_e32")>; @@ -702,17 +778,17 @@ defm V_SUBBREV_U32 : VOP2be_Real_e32e64_vi <0x1e>; defm V_READLANE_B32 : VOP32_Real_vi <0x289>; defm V_WRITELANE_B32 : VOP32_Real_vi <0x28a>; -defm V_BFM_B32 : VOP2_Real_e64_vi <0x293>; -defm V_BCNT_U32_B32 : VOP2_Real_e64_vi <0x28b>; -defm V_MBCNT_LO_U32_B32 : VOP2_Real_e64_vi <0x28c>; -defm V_MBCNT_HI_U32_B32 : VOP2_Real_e64_vi <0x28d>; -defm V_LDEXP_F32 : VOP2_Real_e64_vi <0x288>; -defm V_CVT_PKACCUM_U8_F32 : VOP2_Real_e64_vi <0x1f0>; -defm V_CVT_PKNORM_I16_F32 : VOP2_Real_e64_vi <0x294>; -defm V_CVT_PKNORM_U16_F32 : VOP2_Real_e64_vi <0x295>; -defm V_CVT_PKRTZ_F16_F32 : VOP2_Real_e64_vi <0x296>; -defm V_CVT_PK_U16_U32 : VOP2_Real_e64_vi <0x297>; -defm V_CVT_PK_I16_I32 : VOP2_Real_e64_vi <0x298>; +defm V_BFM_B32 : VOP2_Real_e64only_vi <0x293>; +defm V_BCNT_U32_B32 : VOP2_Real_e64only_vi <0x28b>; +defm V_MBCNT_LO_U32_B32 : VOP2_Real_e64only_vi <0x28c>; +defm V_MBCNT_HI_U32_B32 : VOP2_Real_e64only_vi <0x28d>; +defm V_LDEXP_F32 : VOP2_Real_e64only_vi <0x288>; +defm V_CVT_PKACCUM_U8_F32 : VOP2_Real_e64only_vi <0x1f0>; +defm V_CVT_PKNORM_I16_F32 : VOP2_Real_e64only_vi <0x294>; +defm V_CVT_PKNORM_U16_F32 : VOP2_Real_e64only_vi <0x295>; +defm V_CVT_PKRTZ_F16_F32 : VOP2_Real_e64only_vi <0x296>; +defm V_CVT_PK_U16_U32 : VOP2_Real_e64only_vi <0x297>; +defm V_CVT_PK_I16_I32 : VOP2_Real_e64only_vi <0x298>; defm V_ADD_F16 : VOP2_Real_e32e64_vi <0x1f>; defm V_SUB_F16 : VOP2_Real_e32e64_vi <0x20>; @@ -740,9 +816,11 @@ let SubtargetPredicate = isVI in { // Aliases to simplify matching of floating-point instructions that // are VOP2 on SI and VOP3 on VI. -class SI2_VI3Alias <string name, Instruction inst> : InstAlias < +class SI2_VI3Alias <string name, VOP3_Real inst> : InstAlias < name#" $dst, $src0, $src1", - (inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0, 0) + !if(inst.Pfl.HasOMod, + (inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0, 0), + (inst VGPR_32:$dst, 0, VCSrc_f32:$src0, 0, VCSrc_f32:$src1, 0)) >, PredicateControl { let UseInstAsmMatchConverter = 0; let AsmVariantName = AMDGPUAsmVariants.VOP3; |
