diff --git a/lib/Target/AMDGPU/SIInstrFormats.td b/lib/Target/AMDGPU/SIInstrFormats.td index 4957b660616..f1599bc3c09 100644 --- a/lib/Target/AMDGPU/SIInstrFormats.td +++ b/lib/Target/AMDGPU/SIInstrFormats.td @@ -157,14 +157,6 @@ class VOPAnyCommon pattern> : let VALU = 1; } -class VOPCCommon pattern> : - VOPAnyCommon <(outs), ins, asm, pattern> { - - let VOPC = 1; - let Size = 4; - let Defs = [VCC]; -} - class VOP1Common pattern> : VOPAnyCommon { @@ -284,14 +276,6 @@ class VOP3e op> : VOP3a { let Inst{7-0} = vdst; } -// Encoding used for VOPC instructions encoded as VOP3 -// Differs from VOP3e by destination name (sdst) as VOPC doesn't have vector dst -class VOP3ce op> : VOP3a { - bits<8> sdst; - - let Inst{7-0} = sdst; -} - class VOP3be op> : Enc64 { bits<8> vdst; bits<2> src0_modifiers; @@ -316,16 +300,6 @@ class VOP3be op> : Enc64 { let Inst{63} = src2_modifiers{0}; } -class VOPCe op> : Enc32 { - bits<9> src0; - bits<8> src1; - - let Inst{8-0} = src0; - let Inst{16-9} = src1; - let Inst{24-17} = op; - let Inst{31-25} = 0x3e; -} - class VINTRPe op> : Enc32 { bits<8> vdst; bits<8> vsrc; @@ -406,9 +380,6 @@ class VOP2 op, dag outs, dag ins, string asm, list pattern> : let isCodeGenOnly = 0; } -class VOPC op, dag ins, string asm, list pattern> : - VOPCCommon , VOPCe ; - class VINTRPCommon pattern> : InstSI { let mayLoad = 1; diff --git a/lib/Target/AMDGPU/SIInstrInfo.td b/lib/Target/AMDGPU/SIInstrInfo.td index d04ff6a86ec..bdd55093649 100644 --- a/lib/Target/AMDGPU/SIInstrInfo.td +++ b/lib/Target/AMDGPU/SIInstrInfo.td @@ -19,14 +19,6 @@ class vop { field bits<10> VI3; } -class vopc si, bits<8> vi = !add(0x40, si)> : vop { - field bits<8> SI = si; - field bits<8> VI = vi; - - field bits<9> SI3 = {0, si{7-0}}; - field bits<10> VI3 = {0, 0, vi{7-0}}; -} - class vop1 si, bits<8> vi = si> : vop { field bits<8> SI = si; field bits<8> VI = vi; @@ -654,6 +646,14 @@ class hasModifiers { 0)); } +class hasIntModifiers { + bit ret = + !if(!eq(SrcVT.Value, i32.Value), 1, + !if(!eq(SrcVT.Value, i64.Value), 1, + 0)); +} + + // Return type of input modifiers operand for specified input operand class getSrcMod { bit isFP = !if(!eq(VT.Value, f16.Value), 1, @@ -919,7 +919,21 @@ class VOPProfile _ArgVT> { field bit HasDst = !if(!eq(DstVT.Value, untyped.Value), 0, 1); field bit HasDst32 = HasDst; field int NumSrcArgs = getNumSrcArgs.ret; - field bit HasModifiers = hasModifiers.ret; + field bit HasSrc0 = !if(!eq(Src0VT.Value, untyped.Value), 0, 1); + field bit HasSrc1 = !if(!eq(Src1VT.Value, untyped.Value), 0, 1); + field bit HasSrc2 = !if(!eq(Src2VT.Value, untyped.Value), 0, 1); + field bit HasSrc0Mods = hasModifiers.ret; + field bit HasSrc1Mods = hasModifiers.ret; + field bit HasSrc2Mods = hasModifiers.ret; + + field bit HasSrc0IntMods = hasIntModifiers.ret; + field bit HasSrc1IntMods = hasIntModifiers.ret; + field bit HasSrc2IntMods = hasIntModifiers.ret; + + field bit HasModifiers = HasSrc0Mods; + field bit HasOMod = HasModifiers; + field bit HasClamp = HasModifiers; + field bit HasSDWAClamp = HasSrc0; field bit HasExt = getHasExt.ret; @@ -1073,34 +1087,6 @@ def VOP3b_F64_I1_F64_F64_F64 : VOP3b_Profile { let DstRC = RegisterOperand; } -// VOPC instructions are a special case because for the 32-bit -// encoding, we want to display the implicit vcc write as if it were -// an explicit $dst. -class VOPC_Profile : VOPProfile <[i1, vt0, vt1, untyped]> { - let Asm32 = "vcc, $src0, $src1"; - // The destination for 32-bit encoding is implicit. - let HasDst32 = 0; - let Outs64 = (outs DstRC:$sdst); -} - -class VOPC_Class_Profile : VOPC_Profile { - let Ins64 = (ins Src0Mod:$src0_modifiers, Src0RC64:$src0, Src1RC64:$src1); - let Asm64 = "$sdst, $src0_modifiers, $src1"; - let InsSDWA = (ins Src0Mod:$src0_fmodifiers, Src0RC64:$src0, - Int32InputMods:$src1_imodifiers, Src1RC64:$src1, - clampmod:$clamp, src0_sel:$src0_sel, src1_sel:$src1_sel); - let AsmSDWA = " vcc, $src0_fmodifiers, $src1_imodifiers$clamp $src0_sel $src1_sel"; - -} - -def VOPC_I1_F32_F32 : VOPC_Profile; -def VOPC_I1_F64_F64 : VOPC_Profile; -def VOPC_I1_I32_I32 : VOPC_Profile; -def VOPC_I1_I64_I64 : VOPC_Profile; - -def VOPC_I1_F32_I32 : VOPC_Class_Profile; -def VOPC_I1_F64_I32 : VOPC_Class_Profile; - def VOP_I64_I64_I32 : VOPProfile <[i64, i64, i32, untyped]>; def VOP_I64_I32_I64 : VOPProfile <[i64, i32, i64, untyped]>; def VOP_I64_I64_I64 : VOPProfile <[i64, i64, i64, untyped]>; @@ -1142,58 +1128,6 @@ def VOP_I32_F32_I32_I32 : VOPProfile <[i32, f32, i32, i32]>; def VOP_I64_I64_I32_I64 : VOPProfile <[i64, i64, i32, i64]>; def VOP_V4I32_I64_I32_V4I32 : VOPProfile <[v4i32, i64, i32, v4i32]>; -// This class is used only with VOPC instructions. Use $sdst for out operand -class SIInstAlias : - InstAlias , PredicateControl { - - field bit isCompare; - field bit isCommutable; - - let ResultInst = - !if (p.HasDst32, - !if (!eq(p.NumSrcArgs, 0), - // 1 dst, 0 src - (inst p.DstRC:$sdst), - !if (!eq(p.NumSrcArgs, 1), - // 1 dst, 1 src - (inst p.DstRC:$sdst, p.Src0RC32:$src0), - !if (!eq(p.NumSrcArgs, 2), - // 1 dst, 2 src - (inst p.DstRC:$sdst, p.Src0RC32:$src0, p.Src1RC32:$src1), - // else - unreachable - (inst)))), - // else - !if (!eq(p.NumSrcArgs, 2), - // 0 dst, 2 src - (inst p.Src0RC32:$src0, p.Src1RC32:$src1), - !if (!eq(p.NumSrcArgs, 1), - // 0 dst, 1 src - (inst p.Src0RC32:$src1), - // else - // 0 dst, 0 src - (inst)))); - - let AsmVariantName = VariantName; -} - -class SIInstAliasSI : - SIInstAlias (op_name#"_e32_si"), p, VariantName> { - let AssemblerPredicate = SIAssemblerPredicate; -} - -class SIInstAliasVI : - SIInstAlias (op_name#"_e32_vi"), p, VariantName> { - let AssemblerPredicates = [isVI]; -} - -multiclass SIInstAliasBuilder { - - def : SIInstAliasSI ; - - def : SIInstAliasVI ; -} - class VOP { string OpName = opName; } @@ -1426,26 +1360,6 @@ class VOP3_Real_vi op, dag outs, dag ins, string asm, string opName, let DisableDecoder = DisableVIDecoder; } -class VOP3_C_Real_si op, dag outs, dag ins, string asm, string opName, - bit HasMods = 0, bit VOP3Only = 0> : - VOP3Common , - VOP3ce , - SIMCInstr { - let AssemblerPredicates = [isSICI]; - let DecoderNamespace = "SICI"; - let DisableDecoder = DisableSIDecoder; -} - -class VOP3_C_Real_vi op, dag outs, dag ins, string asm, string opName, - bit HasMods = 0, bit VOP3Only = 0> : - VOP3Common , - VOP3ce_vi , - SIMCInstr { - let AssemblerPredicates = [isVI]; - let DecoderNamespace = "VI"; - let DisableDecoder = DisableVIDecoder; -} - class VOP3b_Real_si op, dag outs, dag ins, string asm, string opName, bit HasMods = 0, bit VOP3Only = 0> : VOP3Common , @@ -1577,29 +1491,6 @@ multiclass VOP3e_2_3_m ; } -multiclass VOP3_C_m pattern, string opName, - bit HasMods, bit defExec, - string revOp, list sched> { - - def "" : VOP3_Pseudo , - Commutable_REV { - let Defs = !if(defExec, [EXEC], []); - let SchedRW = sched; - } - - def _si : VOP3_C_Real_si , - VOP3DisableFields<1, 0, HasMods> { - let Defs = !if(defExec, [EXEC], []); - let SchedRW = sched; - } - - def _vi : VOP3_C_Real_vi , - VOP3DisableFields<1, 0, HasMods> { - let Defs = !if(defExec, [EXEC], []); - let SchedRW = sched; - } -} // An instruction that is VOP2 on SI and VOP3 on VI, no modifiers. multiclass VOP2SI_3VI_m pattern, string opName> : - VOPCCommon , - VOP , - SIMCInstr { - let isPseudo = 1; - let isCodeGenOnly = 1; -} - -class VOPC_SDWA : - VOPC_SDWAe , - VOP_SDWA , - SDWADisableFields

{ - let Defs = !if(DefExec, [VCC, EXEC], [VCC]); - let hasSideEffects = DefExec; - let AsmMatchConverter = "cvtSdwaVOPC"; - let AssemblerPredicates = !if(p.HasExt, [isVI], [DisableInst]); - let AsmVariantName = !if(p.HasExt, AMDGPUAsmVariants.SDWA, - AMDGPUAsmVariants.Disable); - let DecoderNamespace = "SDWA"; - let DisableDecoder = DisableVIDecoder; -} - -multiclass VOPC_m pattern, - string opName, bit DefExec, VOPProfile p, - list sched, - string revOpName = "", string asm = opName#"_e32 "#op_asm, - string alias_asm = opName#" "#op_asm> { - def "" : VOPC_Pseudo , - Commutable_REV { - let Defs = !if(DefExec, [VCC, EXEC], [VCC]); - let SchedRW = sched; - let isConvergent = DefExec; - } - - let AssemblerPredicates = [isSICI] in { - def _si : VOPC, - SIMCInstr { - let Defs = !if(DefExec, [VCC, EXEC], [VCC]); - let isConvergent = DefExec; - let SchedRW = sched; - let DecoderNamespace = "SICI"; - let DisableDecoder = DisableSIDecoder; - } - - } // End AssemblerPredicates = [isSICI] - - let AssemblerPredicates = [isVI] in { - def _vi : VOPC, - SIMCInstr { - let Defs = !if(DefExec, [VCC, EXEC], [VCC]); - let isConvergent = DefExec; - let SchedRW = sched; - let DecoderNamespace = "VI"; - let DisableDecoder = DisableVIDecoder; - } - - } // End AssemblerPredicates = [isVI] - - defm : SIInstAliasBuilder; -} - -multiclass VOPC_Helper pat32, - list pat64, bit DefExec, string revOp, - VOPProfile p, list sched> { - defm _e32 : VOPC_m ; - - defm _e64 : VOP3_C_m ; - - def _sdwa : VOPC_SDWA ; -} - -// Special case for class instructions which only have modifiers on -// the 1st source operand. -multiclass VOPC_Class_Helper pat64, bit DefExec, string revOp, - VOPProfile p, list sched> { - defm _e32 : VOPC_m ; - - defm _e64 : VOP3_C_m , - VOP3DisableModFields<1, 0, 0>; - - def _sdwa : VOPC_SDWA { - let src1_fmodifiers = 0; - let src1_imodifiers = ?; - } -} - -multiclass VOPCInst sched = [Write32Bit]> : - VOPC_Helper < - op, opName, [], - !if(P.HasModifiers, - [(set i1:$sdst, - (setcc (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, - i1:$clamp, i32:$omod)), - (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers)), - cond))], - [(set i1:$sdst, (setcc P.Src0VT:$src0, P.Src1VT:$src1, cond))]), - DefExec, revOp, P, sched ->; - -multiclass VOPCClassInst sched> : VOPC_Class_Helper < - op, opName, - [(set i1:$sdst, - (AMDGPUfp_class - (P.Src0VT (VOP3Mods0Clamp0OMod P.Src0VT:$src0, i32:$src0_modifiers)), - P.Src1VT:$src1))], - DefExec, opName, P, sched ->; - -multiclass VOPC_F32 : - VOPCInst ; - -multiclass VOPC_F64 : - VOPCInst ; - -multiclass VOPC_I32 : - VOPCInst ; - -multiclass VOPC_I64 : - VOPCInst ; - - -multiclass VOPCX sched, - string revOp = ""> - : VOPCInst ; - -multiclass VOPCX_F32 : - VOPCX ; - -multiclass VOPCX_F64 : - VOPCX ; - -multiclass VOPCX_I32 : - VOPCX ; - -multiclass VOPCX_I64 : - VOPCX ; - - -multiclass VOPC_CLASS_F32 : - VOPCClassInst ; - -multiclass VOPCX_CLASS_F32 : - VOPCClassInst ; - -multiclass VOPC_CLASS_F64 : - VOPCClassInst ; - -multiclass VOPCX_CLASS_F64 : - VOPCClassInst ; - - multiclass VOP3_Helper pat, int NumSrcArgs, bit HasMods, bit VOP3Only = 0> : VOP3_m < diff --git a/lib/Target/AMDGPU/SIInstructions.td b/lib/Target/AMDGPU/SIInstructions.td index 86c3fd6815e..fdea84982c8 100644 --- a/lib/Target/AMDGPU/SIInstructions.td +++ b/lib/Target/AMDGPU/SIInstructions.td @@ -21,6 +21,7 @@ def isSI : Predicate<"Subtarget->getGeneration() " def has16BankLDS : Predicate<"Subtarget->getLDSBankCount() == 16">; def has32BankLDS : Predicate<"Subtarget->getLDSBankCount() == 32">; +include "VOPInstructions.td" include "SOPInstructions.td" include "SMInstructions.td" include "FLATInstructions.td" @@ -34,244 +35,6 @@ let SubtargetPredicate = isGCN in { defm EXP : EXP_m; -//===----------------------------------------------------------------------===// -// VOPC Instructions -//===----------------------------------------------------------------------===// - -let isCompare = 1, isCommutable = 1 in { - -defm V_CMP_F_F32 : VOPC_F32 , "v_cmp_f_f32">; -defm V_CMP_LT_F32 : VOPC_F32 , "v_cmp_lt_f32", COND_OLT, "v_cmp_gt_f32">; -defm V_CMP_EQ_F32 : VOPC_F32 , "v_cmp_eq_f32", COND_OEQ>; -defm V_CMP_LE_F32 : VOPC_F32 , "v_cmp_le_f32", COND_OLE, "v_cmp_ge_f32">; -defm V_CMP_GT_F32 : VOPC_F32 , "v_cmp_gt_f32", COND_OGT>; -defm V_CMP_LG_F32 : VOPC_F32 , "v_cmp_lg_f32", COND_ONE>; -defm V_CMP_GE_F32 : VOPC_F32 , "v_cmp_ge_f32", COND_OGE>; -defm V_CMP_O_F32 : VOPC_F32 , "v_cmp_o_f32", COND_O>; -defm V_CMP_U_F32 : VOPC_F32 , "v_cmp_u_f32", COND_UO>; -defm V_CMP_NGE_F32 : VOPC_F32 , "v_cmp_nge_f32", COND_ULT, "v_cmp_nle_f32">; -defm V_CMP_NLG_F32 : VOPC_F32 , "v_cmp_nlg_f32", COND_UEQ>; -defm V_CMP_NGT_F32 : VOPC_F32 , "v_cmp_ngt_f32", COND_ULE, "v_cmp_nlt_f32">; -defm V_CMP_NLE_F32 : VOPC_F32 , "v_cmp_nle_f32", COND_UGT>; -defm V_CMP_NEQ_F32 : VOPC_F32 , "v_cmp_neq_f32", COND_UNE>; -defm V_CMP_NLT_F32 : VOPC_F32 , "v_cmp_nlt_f32", COND_UGE>; -defm V_CMP_TRU_F32 : VOPC_F32 , "v_cmp_tru_f32">; - - -defm V_CMPX_F_F32 : VOPCX_F32 , "v_cmpx_f_f32">; -defm V_CMPX_LT_F32 : VOPCX_F32 , "v_cmpx_lt_f32", "v_cmpx_gt_f32">; -defm V_CMPX_EQ_F32 : VOPCX_F32 , "v_cmpx_eq_f32">; -defm V_CMPX_LE_F32 : VOPCX_F32 , "v_cmpx_le_f32", "v_cmpx_ge_f32">; -defm V_CMPX_GT_F32 : VOPCX_F32 , "v_cmpx_gt_f32">; -defm V_CMPX_LG_F32 : VOPCX_F32 , "v_cmpx_lg_f32">; -defm V_CMPX_GE_F32 : VOPCX_F32 , "v_cmpx_ge_f32">; -defm V_CMPX_O_F32 : VOPCX_F32 , "v_cmpx_o_f32">; -defm V_CMPX_U_F32 : VOPCX_F32 , "v_cmpx_u_f32">; -defm V_CMPX_NGE_F32 : VOPCX_F32 , "v_cmpx_nge_f32">; -defm V_CMPX_NLG_F32 : VOPCX_F32 , "v_cmpx_nlg_f32">; -defm V_CMPX_NGT_F32 : VOPCX_F32 , "v_cmpx_ngt_f32">; -defm V_CMPX_NLE_F32 : VOPCX_F32 , "v_cmpx_nle_f32">; -defm V_CMPX_NEQ_F32 : VOPCX_F32 , "v_cmpx_neq_f32">; -defm V_CMPX_NLT_F32 : VOPCX_F32 , "v_cmpx_nlt_f32">; -defm V_CMPX_TRU_F32 : VOPCX_F32 , "v_cmpx_tru_f32">; - - -defm V_CMP_F_F64 : VOPC_F64 , "v_cmp_f_f64">; -defm V_CMP_LT_F64 : VOPC_F64 , "v_cmp_lt_f64", COND_OLT, "v_cmp_gt_f64">; -defm V_CMP_EQ_F64 : VOPC_F64 , "v_cmp_eq_f64", COND_OEQ>; -defm V_CMP_LE_F64 : VOPC_F64 , "v_cmp_le_f64", COND_OLE, "v_cmp_ge_f64">; -defm V_CMP_GT_F64 : VOPC_F64 , "v_cmp_gt_f64", COND_OGT>; -defm V_CMP_LG_F64 : VOPC_F64 , "v_cmp_lg_f64", COND_ONE>; -defm V_CMP_GE_F64 : VOPC_F64 , "v_cmp_ge_f64", COND_OGE>; -defm V_CMP_O_F64 : VOPC_F64 , "v_cmp_o_f64", COND_O>; -defm V_CMP_U_F64 : VOPC_F64 , "v_cmp_u_f64", COND_UO>; -defm V_CMP_NGE_F64 : VOPC_F64 , "v_cmp_nge_f64", COND_ULT, "v_cmp_nle_f64">; -defm V_CMP_NLG_F64 : VOPC_F64 , "v_cmp_nlg_f64", COND_UEQ>; -defm V_CMP_NGT_F64 : VOPC_F64 , "v_cmp_ngt_f64", COND_ULE, "v_cmp_nlt_f64">; -defm V_CMP_NLE_F64 : VOPC_F64 , "v_cmp_nle_f64", COND_UGT>; -defm V_CMP_NEQ_F64 : VOPC_F64 , "v_cmp_neq_f64", COND_UNE>; -defm V_CMP_NLT_F64 : VOPC_F64 , "v_cmp_nlt_f64", COND_UGE>; -defm V_CMP_TRU_F64 : VOPC_F64 , "v_cmp_tru_f64">; - - -defm V_CMPX_F_F64 : VOPCX_F64 , "v_cmpx_f_f64">; -defm V_CMPX_LT_F64 : VOPCX_F64 , "v_cmpx_lt_f64", "v_cmpx_gt_f64">; -defm V_CMPX_EQ_F64 : VOPCX_F64 , "v_cmpx_eq_f64">; -defm V_CMPX_LE_F64 : VOPCX_F64 , "v_cmpx_le_f64", "v_cmpx_ge_f64">; -defm V_CMPX_GT_F64 : VOPCX_F64 , "v_cmpx_gt_f64">; -defm V_CMPX_LG_F64 : VOPCX_F64 , "v_cmpx_lg_f64">; -defm V_CMPX_GE_F64 : VOPCX_F64 , "v_cmpx_ge_f64">; -defm V_CMPX_O_F64 : VOPCX_F64 , "v_cmpx_o_f64">; -defm V_CMPX_U_F64 : VOPCX_F64 , "v_cmpx_u_f64">; -defm V_CMPX_NGE_F64 : VOPCX_F64 , "v_cmpx_nge_f64", "v_cmpx_nle_f64">; -defm V_CMPX_NLG_F64 : VOPCX_F64 , "v_cmpx_nlg_f64">; -defm V_CMPX_NGT_F64 : VOPCX_F64 , "v_cmpx_ngt_f64", "v_cmpx_nlt_f64">; -defm V_CMPX_NLE_F64 : VOPCX_F64 , "v_cmpx_nle_f64">; -defm V_CMPX_NEQ_F64 : VOPCX_F64 , "v_cmpx_neq_f64">; -defm V_CMPX_NLT_F64 : VOPCX_F64 , "v_cmpx_nlt_f64">; -defm V_CMPX_TRU_F64 : VOPCX_F64 , "v_cmpx_tru_f64">; - - -let SubtargetPredicate = isSICI in { - -defm V_CMPS_F_F32 : VOPC_F32 , "v_cmps_f_f32">; -defm V_CMPS_LT_F32 : VOPC_F32 , "v_cmps_lt_f32", COND_NULL, "v_cmps_gt_f32">; -defm V_CMPS_EQ_F32 : VOPC_F32 , "v_cmps_eq_f32">; -defm V_CMPS_LE_F32 : VOPC_F32 , "v_cmps_le_f32", COND_NULL, "v_cmps_ge_f32">; -defm V_CMPS_GT_F32 : VOPC_F32 , "v_cmps_gt_f32">; -defm V_CMPS_LG_F32 : VOPC_F32 , "v_cmps_lg_f32">; -defm V_CMPS_GE_F32 : VOPC_F32 , "v_cmps_ge_f32">; -defm V_CMPS_O_F32 : VOPC_F32 , "v_cmps_o_f32">; -defm V_CMPS_U_F32 : VOPC_F32 , "v_cmps_u_f32">; -defm V_CMPS_NGE_F32 : VOPC_F32 , "v_cmps_nge_f32", COND_NULL, "v_cmps_nle_f32">; -defm V_CMPS_NLG_F32 : VOPC_F32 , "v_cmps_nlg_f32">; -defm V_CMPS_NGT_F32 : VOPC_F32 , "v_cmps_ngt_f32", COND_NULL, "v_cmps_nlt_f32">; -defm V_CMPS_NLE_F32 : VOPC_F32 , "v_cmps_nle_f32">; -defm V_CMPS_NEQ_F32 : VOPC_F32 , "v_cmps_neq_f32">; -defm V_CMPS_NLT_F32 : VOPC_F32 , "v_cmps_nlt_f32">; -defm V_CMPS_TRU_F32 : VOPC_F32 , "v_cmps_tru_f32">; - - -defm V_CMPSX_F_F32 : VOPCX_F32 , "v_cmpsx_f_f32">; -defm V_CMPSX_LT_F32 : VOPCX_F32 , "v_cmpsx_lt_f32", "v_cmpsx_gt_f32">; -defm V_CMPSX_EQ_F32 : VOPCX_F32 , "v_cmpsx_eq_f32">; -defm V_CMPSX_LE_F32 : VOPCX_F32 , "v_cmpsx_le_f32", "v_cmpsx_ge_f32">; -defm V_CMPSX_GT_F32 : VOPCX_F32 , "v_cmpsx_gt_f32">; -defm V_CMPSX_LG_F32 : VOPCX_F32 , "v_cmpsx_lg_f32">; -defm V_CMPSX_GE_F32 : VOPCX_F32 , "v_cmpsx_ge_f32">; -defm V_CMPSX_O_F32 : VOPCX_F32 , "v_cmpsx_o_f32">; -defm V_CMPSX_U_F32 : VOPCX_F32 , "v_cmpsx_u_f32">; -defm V_CMPSX_NGE_F32 : VOPCX_F32 , "v_cmpsx_nge_f32", "v_cmpsx_nle_f32">; -defm V_CMPSX_NLG_F32 : VOPCX_F32 , "v_cmpsx_nlg_f32">; -defm V_CMPSX_NGT_F32 : VOPCX_F32 , "v_cmpsx_ngt_f32", "v_cmpsx_nlt_f32">; -defm V_CMPSX_NLE_F32 : VOPCX_F32 , "v_cmpsx_nle_f32">; -defm V_CMPSX_NEQ_F32 : VOPCX_F32 , "v_cmpsx_neq_f32">; -defm V_CMPSX_NLT_F32 : VOPCX_F32 , "v_cmpsx_nlt_f32">; -defm V_CMPSX_TRU_F32 : VOPCX_F32 , "v_cmpsx_tru_f32">; - - -defm V_CMPS_F_F64 : VOPC_F64 , "v_cmps_f_f64">; -defm V_CMPS_LT_F64 : VOPC_F64 , "v_cmps_lt_f64", COND_NULL, "v_cmps_gt_f64">; -defm V_CMPS_EQ_F64 : VOPC_F64 , "v_cmps_eq_f64">; -defm V_CMPS_LE_F64 : VOPC_F64 , "v_cmps_le_f64", COND_NULL, "v_cmps_ge_f64">; -defm V_CMPS_GT_F64 : VOPC_F64 , "v_cmps_gt_f64">; -defm V_CMPS_LG_F64 : VOPC_F64 , "v_cmps_lg_f64">; -defm V_CMPS_GE_F64 : VOPC_F64 , "v_cmps_ge_f64">; -defm V_CMPS_O_F64 : VOPC_F64 , "v_cmps_o_f64">; -defm V_CMPS_U_F64 : VOPC_F64 , "v_cmps_u_f64">; -defm V_CMPS_NGE_F64 : VOPC_F64 , "v_cmps_nge_f64", COND_NULL, "v_cmps_nle_f64">; -defm V_CMPS_NLG_F64 : VOPC_F64 , "v_cmps_nlg_f64">; -defm V_CMPS_NGT_F64 : VOPC_F64 , "v_cmps_ngt_f64", COND_NULL, "v_cmps_nlt_f64">; -defm V_CMPS_NLE_F64 : VOPC_F64 , "v_cmps_nle_f64">; -defm V_CMPS_NEQ_F64 : VOPC_F64 , "v_cmps_neq_f64">; -defm V_CMPS_NLT_F64 : VOPC_F64 , "v_cmps_nlt_f64">; -defm V_CMPS_TRU_F64 : VOPC_F64 , "v_cmps_tru_f64">; - - -defm V_CMPSX_F_F64 : VOPCX_F64 , "v_cmpsx_f_f64">; -defm V_CMPSX_LT_F64 : VOPCX_F64 , "v_cmpsx_lt_f64", "v_cmpsx_gt_f64">; -defm V_CMPSX_EQ_F64 : VOPCX_F64 , "v_cmpsx_eq_f64">; -defm V_CMPSX_LE_F64 : VOPCX_F64 , "v_cmpsx_le_f64", "v_cmpsx_ge_f64">; -defm V_CMPSX_GT_F64 : VOPCX_F64 , "v_cmpsx_gt_f64">; -defm V_CMPSX_LG_F64 : VOPCX_F64 , "v_cmpsx_lg_f64">; -defm V_CMPSX_GE_F64 : VOPCX_F64 , "v_cmpsx_ge_f64">; -defm V_CMPSX_O_F64 : VOPCX_F64 , "v_cmpsx_o_f64">; -defm V_CMPSX_U_F64 : VOPCX_F64 , "v_cmpsx_u_f64">; -defm V_CMPSX_NGE_F64 : VOPCX_F64 , "v_cmpsx_nge_f64", "v_cmpsx_nle_f64">; -defm V_CMPSX_NLG_F64 : VOPCX_F64 , "v_cmpsx_nlg_f64">; -defm V_CMPSX_NGT_F64 : VOPCX_F64 , "v_cmpsx_ngt_f64", "v_cmpsx_nlt_f64">; -defm V_CMPSX_NLE_F64 : VOPCX_F64 , "v_cmpsx_nle_f64">; -defm V_CMPSX_NEQ_F64 : VOPCX_F64 , "v_cmpsx_neq_f64">; -defm V_CMPSX_NLT_F64 : VOPCX_F64 , "v_cmpsx_nlt_f64">; -defm V_CMPSX_TRU_F64 : VOPCX_F64 , "v_cmpsx_tru_f64">; - -} // End SubtargetPredicate = isSICI - -defm V_CMP_F_I32 : VOPC_I32 , "v_cmp_f_i32">; -defm V_CMP_LT_I32 : VOPC_I32 , "v_cmp_lt_i32", COND_SLT, "v_cmp_gt_i32">; -defm V_CMP_EQ_I32 : VOPC_I32 , "v_cmp_eq_i32", COND_EQ>; -defm V_CMP_LE_I32 : VOPC_I32 , "v_cmp_le_i32", COND_SLE, "v_cmp_ge_i32">; -defm V_CMP_GT_I32 : VOPC_I32 , "v_cmp_gt_i32", COND_SGT>; -defm V_CMP_NE_I32 : VOPC_I32 , "v_cmp_ne_i32", COND_NE>; -defm V_CMP_GE_I32 : VOPC_I32 , "v_cmp_ge_i32", COND_SGE>; -defm V_CMP_T_I32 : VOPC_I32 , "v_cmp_t_i32">; - - -defm V_CMPX_F_I32 : VOPCX_I32 , "v_cmpx_f_i32">; -defm V_CMPX_LT_I32 : VOPCX_I32 , "v_cmpx_lt_i32", "v_cmpx_gt_i32">; -defm V_CMPX_EQ_I32 : VOPCX_I32 , "v_cmpx_eq_i32">; -defm V_CMPX_LE_I32 : VOPCX_I32 , "v_cmpx_le_i32", "v_cmpx_ge_i32">; -defm V_CMPX_GT_I32 : VOPCX_I32 , "v_cmpx_gt_i32">; -defm V_CMPX_NE_I32 : VOPCX_I32 , "v_cmpx_ne_i32">; -defm V_CMPX_GE_I32 : VOPCX_I32 , "v_cmpx_ge_i32">; -defm V_CMPX_T_I32 : VOPCX_I32 , "v_cmpx_t_i32">; - - -defm V_CMP_F_I64 : VOPC_I64 , "v_cmp_f_i64">; -defm V_CMP_LT_I64 : VOPC_I64 , "v_cmp_lt_i64", COND_SLT, "v_cmp_gt_i64">; -defm V_CMP_EQ_I64 : VOPC_I64 , "v_cmp_eq_i64", COND_EQ>; -defm V_CMP_LE_I64 : VOPC_I64 , "v_cmp_le_i64", COND_SLE, "v_cmp_ge_i64">; -defm V_CMP_GT_I64 : VOPC_I64 , "v_cmp_gt_i64", COND_SGT>; -defm V_CMP_NE_I64 : VOPC_I64 , "v_cmp_ne_i64", COND_NE>; -defm V_CMP_GE_I64 : VOPC_I64 , "v_cmp_ge_i64", COND_SGE>; -defm V_CMP_T_I64 : VOPC_I64 , "v_cmp_t_i64">; - - -defm V_CMPX_F_I64 : VOPCX_I64 , "v_cmpx_f_i64">; -defm V_CMPX_LT_I64 : VOPCX_I64 , "v_cmpx_lt_i64", "v_cmpx_gt_i64">; -defm V_CMPX_EQ_I64 : VOPCX_I64 , "v_cmpx_eq_i64">; -defm V_CMPX_LE_I64 : VOPCX_I64 , "v_cmpx_le_i64", "v_cmpx_ge_i64">; -defm V_CMPX_GT_I64 : VOPCX_I64 , "v_cmpx_gt_i64">; -defm V_CMPX_NE_I64 : VOPCX_I64 , "v_cmpx_ne_i64">; -defm V_CMPX_GE_I64 : VOPCX_I64 , "v_cmpx_ge_i64">; -defm V_CMPX_T_I64 : VOPCX_I64 , "v_cmpx_t_i64">; - - -defm V_CMP_F_U32 : VOPC_I32 , "v_cmp_f_u32">; -defm V_CMP_LT_U32 : VOPC_I32 , "v_cmp_lt_u32", COND_ULT, "v_cmp_gt_u32">; -defm V_CMP_EQ_U32 : VOPC_I32 , "v_cmp_eq_u32", COND_EQ>; -defm V_CMP_LE_U32 : VOPC_I32 , "v_cmp_le_u32", COND_ULE, "v_cmp_ge_u32">; -defm V_CMP_GT_U32 : VOPC_I32 , "v_cmp_gt_u32", COND_UGT>; -defm V_CMP_NE_U32 : VOPC_I32 , "v_cmp_ne_u32", COND_NE>; -defm V_CMP_GE_U32 : VOPC_I32 , "v_cmp_ge_u32", COND_UGE>; -defm V_CMP_T_U32 : VOPC_I32 , "v_cmp_t_u32">; - - -defm V_CMPX_F_U32 : VOPCX_I32 , "v_cmpx_f_u32">; -defm V_CMPX_LT_U32 : VOPCX_I32 , "v_cmpx_lt_u32", "v_cmpx_gt_u32">; -defm V_CMPX_EQ_U32 : VOPCX_I32 , "v_cmpx_eq_u32">; -defm V_CMPX_LE_U32 : VOPCX_I32 , "v_cmpx_le_u32", "v_cmpx_le_u32">; -defm V_CMPX_GT_U32 : VOPCX_I32 , "v_cmpx_gt_u32">; -defm V_CMPX_NE_U32 : VOPCX_I32 , "v_cmpx_ne_u32">; -defm V_CMPX_GE_U32 : VOPCX_I32 , "v_cmpx_ge_u32">; -defm V_CMPX_T_U32 : VOPCX_I32 , "v_cmpx_t_u32">; - - -defm V_CMP_F_U64 : VOPC_I64 , "v_cmp_f_u64">; -defm V_CMP_LT_U64 : VOPC_I64 , "v_cmp_lt_u64", COND_ULT, "v_cmp_gt_u64">; -defm V_CMP_EQ_U64 : VOPC_I64 , "v_cmp_eq_u64", COND_EQ>; -defm V_CMP_LE_U64 : VOPC_I64 , "v_cmp_le_u64", COND_ULE, "v_cmp_ge_u64">; -defm V_CMP_GT_U64 : VOPC_I64 , "v_cmp_gt_u64", COND_UGT>; -defm V_CMP_NE_U64 : VOPC_I64 , "v_cmp_ne_u64", COND_NE>; -defm V_CMP_GE_U64 : VOPC_I64 , "v_cmp_ge_u64", COND_UGE>; -defm V_CMP_T_U64 : VOPC_I64 , "v_cmp_t_u64">; - -defm V_CMPX_F_U64 : VOPCX_I64 , "v_cmpx_f_u64">; -defm V_CMPX_LT_U64 : VOPCX_I64 , "v_cmpx_lt_u64", "v_cmpx_gt_u64">; -defm V_CMPX_EQ_U64 : VOPCX_I64 , "v_cmpx_eq_u64">; -defm V_CMPX_LE_U64 : VOPCX_I64 , "v_cmpx_le_u64", "v_cmpx_ge_u64">; -defm V_CMPX_GT_U64 : VOPCX_I64 , "v_cmpx_gt_u64">; -defm V_CMPX_NE_U64 : VOPCX_I64 , "v_cmpx_ne_u64">; -defm V_CMPX_GE_U64 : VOPCX_I64 , "v_cmpx_ge_u64">; -defm V_CMPX_T_U64 : VOPCX_I64 , "v_cmpx_t_u64">; - -} // End isCompare = 1, isCommutable = 1 - -defm V_CMP_CLASS_F32 : VOPC_CLASS_F32 , "v_cmp_class_f32">; -defm V_CMPX_CLASS_F32 : VOPCX_CLASS_F32 , "v_cmpx_class_f32">; -defm V_CMP_CLASS_F64 : VOPC_CLASS_F64 , "v_cmp_class_f64">; -defm V_CMPX_CLASS_F64 : VOPCX_CLASS_F64 , "v_cmpx_class_f64">; - //===----------------------------------------------------------------------===// // VOP1 Instructions //===----------------------------------------------------------------------===// @@ -1266,71 +1029,6 @@ def : Pat < $src0, $src1, $src2, $src3) >; -//===----------------------------------------------------------------------===// -// V_ICMPIntrinsic Pattern. -//===----------------------------------------------------------------------===// -class ICMP_Pattern : Pat < - (AMDGPUsetcc vt:$src0, vt:$src1, cond), - (inst $src0, $src1) ->; - -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; - -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; -def : ICMP_Pattern ; - -class FCMP_Pattern : Pat < - (i64 (AMDGPUsetcc (vt (VOP3Mods vt:$src0, i32:$src0_modifiers)), - (vt (VOP3Mods vt:$src1, i32:$src1_modifiers)), cond)), - (inst $src0_modifiers, $src0, $src1_modifiers, $src1, - DSTCLAMP.NONE, DSTOMOD.NONE) ->; - -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; - -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; - -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; - -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; -def : FCMP_Pattern ; - //===----------------------------------------------------------------------===// // VOP1 Patterns //===----------------------------------------------------------------------===// diff --git a/lib/Target/AMDGPU/VIInstrFormats.td b/lib/Target/AMDGPU/VIInstrFormats.td index 9196c105bb7..fd9f6e7a62e 100644 --- a/lib/Target/AMDGPU/VIInstrFormats.td +++ b/lib/Target/AMDGPU/VIInstrFormats.td @@ -42,14 +42,6 @@ class VOP3e_vi op> : VOP3a_vi { let Inst{7-0} = vdst; } -// Encoding used for VOPC instructions encoded as VOP3 -// Differs from VOP3e by destination name (sdst) as VOPC doesn't have vector dst -class VOP3ce_vi op> : VOP3a_vi { - bits<8> sdst; - - let Inst{7-0} = sdst; -} - class VOP3be_vi op> : Enc64 { bits<8> vdst; bits<2> src0_modifiers; @@ -176,19 +168,6 @@ class VOP2_SDWAe op> : VOP_SDWAe { let Inst{31} = 0x0; // encoding } -class VOPC_SDWAe op> : VOP_SDWAe { - bits<8> src1; - - let Inst{8-0} = 0xf9; // sdwa - let Inst{16-9} = src1; - let Inst{24-17} = op; - let Inst{31-25} = 0x3e; // encoding - - // VOPC disallows dst_sel and dst_unused as they have no effect on destination - let Inst{42-40} = 0x6; - let Inst{44-43} = 0x2; -} - class EXPe_vi : EXPe { let Inst{31-26} = 0x31; //encoding } diff --git a/lib/Target/AMDGPU/VOPCInstructions.td b/lib/Target/AMDGPU/VOPCInstructions.td new file mode 100644 index 00000000000..8c7738b6cc0 --- /dev/null +++ b/lib/Target/AMDGPU/VOPCInstructions.td @@ -0,0 +1,964 @@ +//===-- VOPCInstructions.td - Vector Instruction Defintions ---------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// Encodings +//===----------------------------------------------------------------------===// + +class VOPCe op> : Enc32 { + bits<9> src0; + bits<8> src1; + + let Inst{8-0} = src0; + let Inst{16-9} = src1; + let Inst{24-17} = op; + let Inst{31-25} = 0x3e; +} + +//===----------------------------------------------------------------------===// +// VOPC classes +//===----------------------------------------------------------------------===// + +// VOPC instructions are a special case because for the 32-bit +// encoding, we want to display the implicit vcc write as if it were +// an explicit $dst. +class VOPC_Profile sched, ValueType vt0, ValueType vt1 = vt0> : + VOPProfile <[i1, vt0, vt1, untyped]> { + let Asm32 = "vcc, $src0, $src1"; + // The destination for 32-bit encoding is implicit. + let HasDst32 = 0; + let Outs64 = (outs VOPDstS64:$sdst); + list Schedule = sched; +} + +class VOPC_Pseudo pattern=[]> : + InstSI<(outs), P.Ins32, "", pattern>, + VOP , + SIMCInstr { + + let isPseudo = 1; + let isCodeGenOnly = 1; + let UseNamedOperandTable = 1; + + string Mnemonic = opName; + string AsmOperands = P.Asm32; + + let Size = 4; + let mayLoad = 0; + let mayStore = 0; + let hasSideEffects = 0; + + let VALU = 1; + let VOPC = 1; + let Uses = [EXEC]; + let Defs = [VCC]; + + let SubtargetPredicate = isGCN; + + VOPProfile Pfl = P; +} + +class VOPC_Real : + InstSI , + SIMCInstr { + + let isPseudo = 0; + let isCodeGenOnly = 0; + + // copy relevant pseudo op flags + let SubtargetPredicate = ps.SubtargetPredicate; + let AsmMatchConverter = ps.AsmMatchConverter; + let Constraints = ps.Constraints; + let DisableEncoding = ps.DisableEncoding; + let TSFlags = ps.TSFlags; +} + +// This class is used only with VOPC instructions. Use $sdst for out operand +class VOPCInstAlias : + InstAlias , PredicateControl { + + field bit isCompare; + field bit isCommutable; + + let ResultInst = + !if (p.HasDst32, + !if (!eq(p.NumSrcArgs, 0), + // 1 dst, 0 src + (inst p.DstRC:$sdst), + !if (!eq(p.NumSrcArgs, 1), + // 1 dst, 1 src + (inst p.DstRC:$sdst, p.Src0RC32:$src0), + !if (!eq(p.NumSrcArgs, 2), + // 1 dst, 2 src + (inst p.DstRC:$sdst, p.Src0RC32:$src0, p.Src1RC32:$src1), + // else - unreachable + (inst)))), + // else + !if (!eq(p.NumSrcArgs, 2), + // 0 dst, 2 src + (inst p.Src0RC32:$src0, p.Src1RC32:$src1), + !if (!eq(p.NumSrcArgs, 1), + // 0 dst, 1 src + (inst p.Src0RC32:$src1), + // else + // 0 dst, 0 src + (inst)))); + + let AsmVariantName = AMDGPUAsmVariants.Default; + let SubtargetPredicate = AssemblerPredicate; +} + +multiclass VOPC_Pseudos { + + def _e32 : VOPC_Pseudo , + Commutable_REV { + let Defs = !if(DefExec, [VCC, EXEC], [VCC]); + let SchedRW = P.Schedule; + let isConvergent = DefExec; + let isCompare = 1; + let isCommutable = 1; + } + def _e64 : VOP3_PseudoNew, + Commutable_REV { + let Defs = !if(DefExec, [EXEC], []); + let SchedRW = P.Schedule; + let isCompare = 1; + let isCommutable = 1; + } +} + +def VOPC_I1_F32_F32 : VOPC_Profile<[Write32Bit], f32>; +def VOPC_I1_F64_F64 : VOPC_Profile<[WriteDoubleAdd], f64>; +def VOPC_I1_I32_I32 : VOPC_Profile<[Write32Bit], i32>; +def VOPC_I1_I64_I64 : VOPC_Profile<[Write64Bit], i64>; + +multiclass VOPC_F32 : + VOPC_Pseudos ; + +multiclass VOPC_F64 : + VOPC_Pseudos ; + +multiclass VOPC_I32 : + VOPC_Pseudos ; + +multiclass VOPC_I64 : + VOPC_Pseudos ; + +multiclass VOPCX_F32 : + VOPC_Pseudos ; + +multiclass VOPCX_F64 : + VOPC_Pseudos ; + +multiclass VOPCX_I32 : + VOPC_Pseudos ; + +multiclass VOPCX_I64 : + VOPC_Pseudos ; + + +//===----------------------------------------------------------------------===// +// Compare instructions +//===----------------------------------------------------------------------===// + +defm V_CMP_F_F32 : VOPC_F32 <"v_cmp_f_f32">; +defm V_CMP_LT_F32 : VOPC_F32 <"v_cmp_lt_f32", COND_OLT, "v_cmp_gt_f32">; +defm V_CMP_EQ_F32 : VOPC_F32 <"v_cmp_eq_f32", COND_OEQ>; +defm V_CMP_LE_F32 : VOPC_F32 <"v_cmp_le_f32", COND_OLE, "v_cmp_ge_f32">; +defm V_CMP_GT_F32 : VOPC_F32 <"v_cmp_gt_f32", COND_OGT>; +defm V_CMP_LG_F32 : VOPC_F32 <"v_cmp_lg_f32", COND_ONE>; +defm V_CMP_GE_F32 : VOPC_F32 <"v_cmp_ge_f32", COND_OGE>; +defm V_CMP_O_F32 : VOPC_F32 <"v_cmp_o_f32", COND_O>; +defm V_CMP_U_F32 : VOPC_F32 <"v_cmp_u_f32", COND_UO>; +defm V_CMP_NGE_F32 : VOPC_F32 <"v_cmp_nge_f32", COND_ULT, "v_cmp_nle_f32">; +defm V_CMP_NLG_F32 : VOPC_F32 <"v_cmp_nlg_f32", COND_UEQ>; +defm V_CMP_NGT_F32 : VOPC_F32 <"v_cmp_ngt_f32", COND_ULE, "v_cmp_nlt_f32">; +defm V_CMP_NLE_F32 : VOPC_F32 <"v_cmp_nle_f32", COND_UGT>; +defm V_CMP_NEQ_F32 : VOPC_F32 <"v_cmp_neq_f32", COND_UNE>; +defm V_CMP_NLT_F32 : VOPC_F32 <"v_cmp_nlt_f32", COND_UGE>; +defm V_CMP_TRU_F32 : VOPC_F32 <"v_cmp_tru_f32">; + +defm V_CMPX_F_F32 : VOPCX_F32 <"v_cmpx_f_f32">; +defm V_CMPX_LT_F32 : VOPCX_F32 <"v_cmpx_lt_f32", "v_cmpx_gt_f32">; +defm V_CMPX_EQ_F32 : VOPCX_F32 <"v_cmpx_eq_f32">; +defm V_CMPX_LE_F32 : VOPCX_F32 <"v_cmpx_le_f32", "v_cmpx_ge_f32">; +defm V_CMPX_GT_F32 : VOPCX_F32 <"v_cmpx_gt_f32">; +defm V_CMPX_LG_F32 : VOPCX_F32 <"v_cmpx_lg_f32">; +defm V_CMPX_GE_F32 : VOPCX_F32 <"v_cmpx_ge_f32">; +defm V_CMPX_O_F32 : VOPCX_F32 <"v_cmpx_o_f32">; +defm V_CMPX_U_F32 : VOPCX_F32 <"v_cmpx_u_f32">; +defm V_CMPX_NGE_F32 : VOPCX_F32 <"v_cmpx_nge_f32">; +defm V_CMPX_NLG_F32 : VOPCX_F32 <"v_cmpx_nlg_f32">; +defm V_CMPX_NGT_F32 : VOPCX_F32 <"v_cmpx_ngt_f32">; +defm V_CMPX_NLE_F32 : VOPCX_F32 <"v_cmpx_nle_f32">; +defm V_CMPX_NEQ_F32 : VOPCX_F32 <"v_cmpx_neq_f32">; +defm V_CMPX_NLT_F32 : VOPCX_F32 <"v_cmpx_nlt_f32">; +defm V_CMPX_TRU_F32 : VOPCX_F32 <"v_cmpx_tru_f32">; + +defm V_CMP_F_F64 : VOPC_F64 <"v_cmp_f_f64">; +defm V_CMP_LT_F64 : VOPC_F64 <"v_cmp_lt_f64", COND_OLT, "v_cmp_gt_f64">; +defm V_CMP_EQ_F64 : VOPC_F64 <"v_cmp_eq_f64", COND_OEQ>; +defm V_CMP_LE_F64 : VOPC_F64 <"v_cmp_le_f64", COND_OLE, "v_cmp_ge_f64">; +defm V_CMP_GT_F64 : VOPC_F64 <"v_cmp_gt_f64", COND_OGT>; +defm V_CMP_LG_F64 : VOPC_F64 <"v_cmp_lg_f64", COND_ONE>; +defm V_CMP_GE_F64 : VOPC_F64 <"v_cmp_ge_f64", COND_OGE>; +defm V_CMP_O_F64 : VOPC_F64 <"v_cmp_o_f64", COND_O>; +defm V_CMP_U_F64 : VOPC_F64 <"v_cmp_u_f64", COND_UO>; +defm V_CMP_NGE_F64 : VOPC_F64 <"v_cmp_nge_f64", COND_ULT, "v_cmp_nle_f64">; +defm V_CMP_NLG_F64 : VOPC_F64 <"v_cmp_nlg_f64", COND_UEQ>; +defm V_CMP_NGT_F64 : VOPC_F64 <"v_cmp_ngt_f64", COND_ULE, "v_cmp_nlt_f64">; +defm V_CMP_NLE_F64 : VOPC_F64 <"v_cmp_nle_f64", COND_UGT>; +defm V_CMP_NEQ_F64 : VOPC_F64 <"v_cmp_neq_f64", COND_UNE>; +defm V_CMP_NLT_F64 : VOPC_F64 <"v_cmp_nlt_f64", COND_UGE>; +defm V_CMP_TRU_F64 : VOPC_F64 <"v_cmp_tru_f64">; + +defm V_CMPX_F_F64 : VOPCX_F64 <"v_cmpx_f_f64">; +defm V_CMPX_LT_F64 : VOPCX_F64 <"v_cmpx_lt_f64", "v_cmpx_gt_f64">; +defm V_CMPX_EQ_F64 : VOPCX_F64 <"v_cmpx_eq_f64">; +defm V_CMPX_LE_F64 : VOPCX_F64 <"v_cmpx_le_f64", "v_cmpx_ge_f64">; +defm V_CMPX_GT_F64 : VOPCX_F64 <"v_cmpx_gt_f64">; +defm V_CMPX_LG_F64 : VOPCX_F64 <"v_cmpx_lg_f64">; +defm V_CMPX_GE_F64 : VOPCX_F64 <"v_cmpx_ge_f64">; +defm V_CMPX_O_F64 : VOPCX_F64 <"v_cmpx_o_f64">; +defm V_CMPX_U_F64 : VOPCX_F64 <"v_cmpx_u_f64">; +defm V_CMPX_NGE_F64 : VOPCX_F64 <"v_cmpx_nge_f64", "v_cmpx_nle_f64">; +defm V_CMPX_NLG_F64 : VOPCX_F64 <"v_cmpx_nlg_f64">; +defm V_CMPX_NGT_F64 : VOPCX_F64 <"v_cmpx_ngt_f64", "v_cmpx_nlt_f64">; +defm V_CMPX_NLE_F64 : VOPCX_F64 <"v_cmpx_nle_f64">; +defm V_CMPX_NEQ_F64 : VOPCX_F64 <"v_cmpx_neq_f64">; +defm V_CMPX_NLT_F64 : VOPCX_F64 <"v_cmpx_nlt_f64">; +defm V_CMPX_TRU_F64 : VOPCX_F64 <"v_cmpx_tru_f64">; + +let SubtargetPredicate = isSICI in { + +defm V_CMPS_F_F32 : VOPC_F32 <"v_cmps_f_f32">; +defm V_CMPS_LT_F32 : VOPC_F32 <"v_cmps_lt_f32", COND_NULL, "v_cmps_gt_f32">; +defm V_CMPS_EQ_F32 : VOPC_F32 <"v_cmps_eq_f32">; +defm V_CMPS_LE_F32 : VOPC_F32 <"v_cmps_le_f32", COND_NULL, "v_cmps_ge_f32">; +defm V_CMPS_GT_F32 : VOPC_F32 <"v_cmps_gt_f32">; +defm V_CMPS_LG_F32 : VOPC_F32 <"v_cmps_lg_f32">; +defm V_CMPS_GE_F32 : VOPC_F32 <"v_cmps_ge_f32">; +defm V_CMPS_O_F32 : VOPC_F32 <"v_cmps_o_f32">; +defm V_CMPS_U_F32 : VOPC_F32 <"v_cmps_u_f32">; +defm V_CMPS_NGE_F32 : VOPC_F32 <"v_cmps_nge_f32", COND_NULL, "v_cmps_nle_f32">; +defm V_CMPS_NLG_F32 : VOPC_F32 <"v_cmps_nlg_f32">; +defm V_CMPS_NGT_F32 : VOPC_F32 <"v_cmps_ngt_f32", COND_NULL, "v_cmps_nlt_f32">; +defm V_CMPS_NLE_F32 : VOPC_F32 <"v_cmps_nle_f32">; +defm V_CMPS_NEQ_F32 : VOPC_F32 <"v_cmps_neq_f32">; +defm V_CMPS_NLT_F32 : VOPC_F32 <"v_cmps_nlt_f32">; +defm V_CMPS_TRU_F32 : VOPC_F32 <"v_cmps_tru_f32">; + +defm V_CMPSX_F_F32 : VOPCX_F32 <"v_cmpsx_f_f32">; +defm V_CMPSX_LT_F32 : VOPCX_F32 <"v_cmpsx_lt_f32", "v_cmpsx_gt_f32">; +defm V_CMPSX_EQ_F32 : VOPCX_F32 <"v_cmpsx_eq_f32">; +defm V_CMPSX_LE_F32 : VOPCX_F32 <"v_cmpsx_le_f32", "v_cmpsx_ge_f32">; +defm V_CMPSX_GT_F32 : VOPCX_F32 <"v_cmpsx_gt_f32">; +defm V_CMPSX_LG_F32 : VOPCX_F32 <"v_cmpsx_lg_f32">; +defm V_CMPSX_GE_F32 : VOPCX_F32 <"v_cmpsx_ge_f32">; +defm V_CMPSX_O_F32 : VOPCX_F32 <"v_cmpsx_o_f32">; +defm V_CMPSX_U_F32 : VOPCX_F32 <"v_cmpsx_u_f32">; +defm V_CMPSX_NGE_F32 : VOPCX_F32 <"v_cmpsx_nge_f32", "v_cmpsx_nle_f32">; +defm V_CMPSX_NLG_F32 : VOPCX_F32 <"v_cmpsx_nlg_f32">; +defm V_CMPSX_NGT_F32 : VOPCX_F32 <"v_cmpsx_ngt_f32", "v_cmpsx_nlt_f32">; +defm V_CMPSX_NLE_F32 : VOPCX_F32 <"v_cmpsx_nle_f32">; +defm V_CMPSX_NEQ_F32 : VOPCX_F32 <"v_cmpsx_neq_f32">; +defm V_CMPSX_NLT_F32 : VOPCX_F32 <"v_cmpsx_nlt_f32">; +defm V_CMPSX_TRU_F32 : VOPCX_F32 <"v_cmpsx_tru_f32">; + +defm V_CMPS_F_F64 : VOPC_F64 <"v_cmps_f_f64">; +defm V_CMPS_LT_F64 : VOPC_F64 <"v_cmps_lt_f64", COND_NULL, "v_cmps_gt_f64">; +defm V_CMPS_EQ_F64 : VOPC_F64 <"v_cmps_eq_f64">; +defm V_CMPS_LE_F64 : VOPC_F64 <"v_cmps_le_f64", COND_NULL, "v_cmps_ge_f64">; +defm V_CMPS_GT_F64 : VOPC_F64 <"v_cmps_gt_f64">; +defm V_CMPS_LG_F64 : VOPC_F64 <"v_cmps_lg_f64">; +defm V_CMPS_GE_F64 : VOPC_F64 <"v_cmps_ge_f64">; +defm V_CMPS_O_F64 : VOPC_F64 <"v_cmps_o_f64">; +defm V_CMPS_U_F64 : VOPC_F64 <"v_cmps_u_f64">; +defm V_CMPS_NGE_F64 : VOPC_F64 <"v_cmps_nge_f64", COND_NULL, "v_cmps_nle_f64">; +defm V_CMPS_NLG_F64 : VOPC_F64 <"v_cmps_nlg_f64">; +defm V_CMPS_NGT_F64 : VOPC_F64 <"v_cmps_ngt_f64", COND_NULL, "v_cmps_nlt_f64">; +defm V_CMPS_NLE_F64 : VOPC_F64 <"v_cmps_nle_f64">; +defm V_CMPS_NEQ_F64 : VOPC_F64 <"v_cmps_neq_f64">; +defm V_CMPS_NLT_F64 : VOPC_F64 <"v_cmps_nlt_f64">; +defm V_CMPS_TRU_F64 : VOPC_F64 <"v_cmps_tru_f64">; + +defm V_CMPSX_F_F64 : VOPCX_F64 <"v_cmpsx_f_f64">; +defm V_CMPSX_LT_F64 : VOPCX_F64 <"v_cmpsx_lt_f64", "v_cmpsx_gt_f64">; +defm V_CMPSX_EQ_F64 : VOPCX_F64 <"v_cmpsx_eq_f64">; +defm V_CMPSX_LE_F64 : VOPCX_F64 <"v_cmpsx_le_f64", "v_cmpsx_ge_f64">; +defm V_CMPSX_GT_F64 : VOPCX_F64 <"v_cmpsx_gt_f64">; +defm V_CMPSX_LG_F64 : VOPCX_F64 <"v_cmpsx_lg_f64">; +defm V_CMPSX_GE_F64 : VOPCX_F64 <"v_cmpsx_ge_f64">; +defm V_CMPSX_O_F64 : VOPCX_F64 <"v_cmpsx_o_f64">; +defm V_CMPSX_U_F64 : VOPCX_F64 <"v_cmpsx_u_f64">; +defm V_CMPSX_NGE_F64 : VOPCX_F64 <"v_cmpsx_nge_f64", "v_cmpsx_nle_f64">; +defm V_CMPSX_NLG_F64 : VOPCX_F64 <"v_cmpsx_nlg_f64">; +defm V_CMPSX_NGT_F64 : VOPCX_F64 <"v_cmpsx_ngt_f64", "v_cmpsx_nlt_f64">; +defm V_CMPSX_NLE_F64 : VOPCX_F64 <"v_cmpsx_nle_f64">; +defm V_CMPSX_NEQ_F64 : VOPCX_F64 <"v_cmpsx_neq_f64">; +defm V_CMPSX_NLT_F64 : VOPCX_F64 <"v_cmpsx_nlt_f64">; +defm V_CMPSX_TRU_F64 : VOPCX_F64 <"v_cmpsx_tru_f64">; + +} // End SubtargetPredicate = isSICI + +defm V_CMP_F_I32 : VOPC_I32 <"v_cmp_f_i32">; +defm V_CMP_LT_I32 : VOPC_I32 <"v_cmp_lt_i32", COND_SLT, "v_cmp_gt_i32">; +defm V_CMP_EQ_I32 : VOPC_I32 <"v_cmp_eq_i32", COND_EQ>; +defm V_CMP_LE_I32 : VOPC_I32 <"v_cmp_le_i32", COND_SLE, "v_cmp_ge_i32">; +defm V_CMP_GT_I32 : VOPC_I32 <"v_cmp_gt_i32", COND_SGT>; +defm V_CMP_NE_I32 : VOPC_I32 <"v_cmp_ne_i32", COND_NE>; +defm V_CMP_GE_I32 : VOPC_I32 <"v_cmp_ge_i32", COND_SGE>; +defm V_CMP_T_I32 : VOPC_I32 <"v_cmp_t_i32">; + +defm V_CMPX_F_I32 : VOPCX_I32 <"v_cmpx_f_i32">; +defm V_CMPX_LT_I32 : VOPCX_I32 <"v_cmpx_lt_i32", "v_cmpx_gt_i32">; +defm V_CMPX_EQ_I32 : VOPCX_I32 <"v_cmpx_eq_i32">; +defm V_CMPX_LE_I32 : VOPCX_I32 <"v_cmpx_le_i32", "v_cmpx_ge_i32">; +defm V_CMPX_GT_I32 : VOPCX_I32 <"v_cmpx_gt_i32">; +defm V_CMPX_NE_I32 : VOPCX_I32 <"v_cmpx_ne_i32">; +defm V_CMPX_GE_I32 : VOPCX_I32 <"v_cmpx_ge_i32">; +defm V_CMPX_T_I32 : VOPCX_I32 <"v_cmpx_t_i32">; + +defm V_CMP_F_I64 : VOPC_I64 <"v_cmp_f_i64">; +defm V_CMP_LT_I64 : VOPC_I64 <"v_cmp_lt_i64", COND_SLT, "v_cmp_gt_i64">; +defm V_CMP_EQ_I64 : VOPC_I64 <"v_cmp_eq_i64", COND_EQ>; +defm V_CMP_LE_I64 : VOPC_I64 <"v_cmp_le_i64", COND_SLE, "v_cmp_ge_i64">; +defm V_CMP_GT_I64 : VOPC_I64 <"v_cmp_gt_i64", COND_SGT>; +defm V_CMP_NE_I64 : VOPC_I64 <"v_cmp_ne_i64", COND_NE>; +defm V_CMP_GE_I64 : VOPC_I64 <"v_cmp_ge_i64", COND_SGE>; +defm V_CMP_T_I64 : VOPC_I64 <"v_cmp_t_i64">; + +defm V_CMPX_F_I64 : VOPCX_I64 <"v_cmpx_f_i64">; +defm V_CMPX_LT_I64 : VOPCX_I64 <"v_cmpx_lt_i64", "v_cmpx_gt_i64">; +defm V_CMPX_EQ_I64 : VOPCX_I64 <"v_cmpx_eq_i64">; +defm V_CMPX_LE_I64 : VOPCX_I64 <"v_cmpx_le_i64", "v_cmpx_ge_i64">; +defm V_CMPX_GT_I64 : VOPCX_I64 <"v_cmpx_gt_i64">; +defm V_CMPX_NE_I64 : VOPCX_I64 <"v_cmpx_ne_i64">; +defm V_CMPX_GE_I64 : VOPCX_I64 <"v_cmpx_ge_i64">; +defm V_CMPX_T_I64 : VOPCX_I64 <"v_cmpx_t_i64">; + +defm V_CMP_F_U32 : VOPC_I32 <"v_cmp_f_u32">; +defm V_CMP_LT_U32 : VOPC_I32 <"v_cmp_lt_u32", COND_ULT, "v_cmp_gt_u32">; +defm V_CMP_EQ_U32 : VOPC_I32 <"v_cmp_eq_u32", COND_EQ>; +defm V_CMP_LE_U32 : VOPC_I32 <"v_cmp_le_u32", COND_ULE, "v_cmp_ge_u32">; +defm V_CMP_GT_U32 : VOPC_I32 <"v_cmp_gt_u32", COND_UGT>; +defm V_CMP_NE_U32 : VOPC_I32 <"v_cmp_ne_u32", COND_NE>; +defm V_CMP_GE_U32 : VOPC_I32 <"v_cmp_ge_u32", COND_UGE>; +defm V_CMP_T_U32 : VOPC_I32 <"v_cmp_t_u32">; + +defm V_CMPX_F_U32 : VOPCX_I32 <"v_cmpx_f_u32">; +defm V_CMPX_LT_U32 : VOPCX_I32 <"v_cmpx_lt_u32", "v_cmpx_gt_u32">; +defm V_CMPX_EQ_U32 : VOPCX_I32 <"v_cmpx_eq_u32">; +defm V_CMPX_LE_U32 : VOPCX_I32 <"v_cmpx_le_u32", "v_cmpx_le_u32">; +defm V_CMPX_GT_U32 : VOPCX_I32 <"v_cmpx_gt_u32">; +defm V_CMPX_NE_U32 : VOPCX_I32 <"v_cmpx_ne_u32">; +defm V_CMPX_GE_U32 : VOPCX_I32 <"v_cmpx_ge_u32">; +defm V_CMPX_T_U32 : VOPCX_I32 <"v_cmpx_t_u32">; + +defm V_CMP_F_U64 : VOPC_I64 <"v_cmp_f_u64">; +defm V_CMP_LT_U64 : VOPC_I64 <"v_cmp_lt_u64", COND_ULT, "v_cmp_gt_u64">; +defm V_CMP_EQ_U64 : VOPC_I64 <"v_cmp_eq_u64", COND_EQ>; +defm V_CMP_LE_U64 : VOPC_I64 <"v_cmp_le_u64", COND_ULE, "v_cmp_ge_u64">; +defm V_CMP_GT_U64 : VOPC_I64 <"v_cmp_gt_u64", COND_UGT>; +defm V_CMP_NE_U64 : VOPC_I64 <"v_cmp_ne_u64", COND_NE>; +defm V_CMP_GE_U64 : VOPC_I64 <"v_cmp_ge_u64", COND_UGE>; +defm V_CMP_T_U64 : VOPC_I64 <"v_cmp_t_u64">; + +defm V_CMPX_F_U64 : VOPCX_I64 <"v_cmpx_f_u64">; +defm V_CMPX_LT_U64 : VOPCX_I64 <"v_cmpx_lt_u64", "v_cmpx_gt_u64">; +defm V_CMPX_EQ_U64 : VOPCX_I64 <"v_cmpx_eq_u64">; +defm V_CMPX_LE_U64 : VOPCX_I64 <"v_cmpx_le_u64", "v_cmpx_ge_u64">; +defm V_CMPX_GT_U64 : VOPCX_I64 <"v_cmpx_gt_u64">; +defm V_CMPX_NE_U64 : VOPCX_I64 <"v_cmpx_ne_u64">; +defm V_CMPX_GE_U64 : VOPCX_I64 <"v_cmpx_ge_u64">; +defm V_CMPX_T_U64 : VOPCX_I64 <"v_cmpx_t_u64">; + +//===----------------------------------------------------------------------===// +// Class instructions +//===----------------------------------------------------------------------===// + +class VOPC_Class_Profile sched, ValueType vt> : + VOPC_Profile { + let Ins64 = (ins Src0Mod:$src0_modifiers, Src0RC64:$src0, Src1RC64:$src1); + let Asm64 = "$sdst, $src0_modifiers, $src1"; + let InsSDWA = (ins Src0Mod:$src0_fmodifiers, Src0RC64:$src0, + Int32InputMods:$src1_imodifiers, Src1RC64:$src1, + clampmod:$clamp, src0_sel:$src0_sel, src1_sel:$src1_sel); + let AsmSDWA = " vcc, $src0_fmodifiers, $src1_imodifiers$clamp $src0_sel $src1_sel"; + let HasClamp = 0; + let HasOMod = 0; +} + +class getVOPCClassPat64 { + list ret = + [(set i1:$sdst, + (AMDGPUfp_class + (P.Src0VT (VOP3Mods0Clamp0OMod P.Src0VT:$src0, i32:$src0_modifiers)), + P.Src1VT:$src1))]; +} + +// Special case for class instructions which only have modifiers on +// the 1st source operand. +multiclass VOPC_Class_Pseudos { + def _e32 : VOPC_Pseudo { + let Defs = !if(DefExec, [VCC, EXEC], [VCC]); + let SchedRW = p.Schedule; + let isConvergent = DefExec; + } + def _e64 : VOP3_PseudoNew.ret> { + let Defs = !if(DefExec, [EXEC], []); + let SchedRW = p.Schedule; + } +} + +def VOPC_I1_F32_I32 : VOPC_Class_Profile<[Write32Bit], f32>; +def VOPC_I1_F64_I32 : VOPC_Class_Profile<[WriteDoubleAdd], f64>; + +multiclass VOPC_CLASS_F32 : + VOPC_Class_Pseudos ; + +multiclass VOPCX_CLASS_F32 : + VOPC_Class_Pseudos ; + +multiclass VOPC_CLASS_F64 : + VOPC_Class_Pseudos ; + +multiclass VOPCX_CLASS_F64 : + VOPC_Class_Pseudos ; + +defm V_CMP_CLASS_F32 : VOPC_CLASS_F32 <"v_cmp_class_f32">; +defm V_CMPX_CLASS_F32 : VOPCX_CLASS_F32 <"v_cmpx_class_f32">; +defm V_CMP_CLASS_F64 : VOPC_CLASS_F64 <"v_cmp_class_f64">; +defm V_CMPX_CLASS_F64 : VOPCX_CLASS_F64 <"v_cmpx_class_f64">; + +//===----------------------------------------------------------------------===// +// V_ICMPIntrinsic Pattern. +//===----------------------------------------------------------------------===// + +let Predicates = [isGCN] in { + +class ICMP_Pattern : Pat < + (AMDGPUsetcc vt:$src0, vt:$src1, cond), + (inst $src0, $src1) +>; + +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; + +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; +def : ICMP_Pattern ; + +class FCMP_Pattern : Pat < + (i64 (AMDGPUsetcc (vt (VOP3Mods vt:$src0, i32:$src0_modifiers)), + (vt (VOP3Mods vt:$src1, i32:$src1_modifiers)), cond)), + (inst $src0_modifiers, $src0, $src1_modifiers, $src1, + DSTCLAMP.NONE, DSTOMOD.NONE) +>; + +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; + +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; + +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; + +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; +def : FCMP_Pattern ; + +} // End Predicates = [isGCN] + +//===----------------------------------------------------------------------===// +// Target +//===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// SI +//===----------------------------------------------------------------------===// + +multiclass VOPC_Real_si op> { + let AssemblerPredicates = [isSICI], DecoderNamespace = "SICI" in { + def _e32_si : + VOPC_Real(NAME#"_e32"), SIEncodingFamily.SI>, + VOPCe; + + def _e64_si : + VOP3_Real(NAME#"_e64"), SIEncodingFamily.SI>, + VOP3a_siNew (NAME#"_e64").Pfl> { + // Encoding used for VOPC instructions encoded as VOP3 + // Differs from VOP3e by destination name (sdst) as VOPC doesn't have vector dst + bits<8> sdst; + let Inst{7-0} = sdst; + } + } + def : VOPCInstAlias (NAME#"_e64"), + !cast(NAME#"_e32_si")> { + let AssemblerPredicate = isSICI; + } +} + +defm V_CMP_F_F32 : VOPC_Real_si <0x0>; +defm V_CMP_LT_F32 : VOPC_Real_si <0x1>; +defm V_CMP_EQ_F32 : VOPC_Real_si <0x2>; +defm V_CMP_LE_F32 : VOPC_Real_si <0x3>; +defm V_CMP_GT_F32 : VOPC_Real_si <0x4>; +defm V_CMP_LG_F32 : VOPC_Real_si <0x5>; +defm V_CMP_GE_F32 : VOPC_Real_si <0x6>; +defm V_CMP_O_F32 : VOPC_Real_si <0x7>; +defm V_CMP_U_F32 : VOPC_Real_si <0x8>; +defm V_CMP_NGE_F32 : VOPC_Real_si <0x9>; +defm V_CMP_NLG_F32 : VOPC_Real_si <0xa>; +defm V_CMP_NGT_F32 : VOPC_Real_si <0xb>; +defm V_CMP_NLE_F32 : VOPC_Real_si <0xc>; +defm V_CMP_NEQ_F32 : VOPC_Real_si <0xd>; +defm V_CMP_NLT_F32 : VOPC_Real_si <0xe>; +defm V_CMP_TRU_F32 : VOPC_Real_si <0xf>; + +defm V_CMPX_F_F32 : VOPC_Real_si <0x10>; +defm V_CMPX_LT_F32 : VOPC_Real_si <0x11>; +defm V_CMPX_EQ_F32 : VOPC_Real_si <0x12>; +defm V_CMPX_LE_F32 : VOPC_Real_si <0x13>; +defm V_CMPX_GT_F32 : VOPC_Real_si <0x14>; +defm V_CMPX_LG_F32 : VOPC_Real_si <0x15>; +defm V_CMPX_GE_F32 : VOPC_Real_si <0x16>; +defm V_CMPX_O_F32 : VOPC_Real_si <0x17>; +defm V_CMPX_U_F32 : VOPC_Real_si <0x18>; +defm V_CMPX_NGE_F32 : VOPC_Real_si <0x19>; +defm V_CMPX_NLG_F32 : VOPC_Real_si <0x1a>; +defm V_CMPX_NGT_F32 : VOPC_Real_si <0x1b>; +defm V_CMPX_NLE_F32 : VOPC_Real_si <0x1c>; +defm V_CMPX_NEQ_F32 : VOPC_Real_si <0x1d>; +defm V_CMPX_NLT_F32 : VOPC_Real_si <0x1e>; +defm V_CMPX_TRU_F32 : VOPC_Real_si <0x1f>; + +defm V_CMP_F_F64 : VOPC_Real_si <0x20>; +defm V_CMP_LT_F64 : VOPC_Real_si <0x21>; +defm V_CMP_EQ_F64 : VOPC_Real_si <0x22>; +defm V_CMP_LE_F64 : VOPC_Real_si <0x23>; +defm V_CMP_GT_F64 : VOPC_Real_si <0x24>; +defm V_CMP_LG_F64 : VOPC_Real_si <0x25>; +defm V_CMP_GE_F64 : VOPC_Real_si <0x26>; +defm V_CMP_O_F64 : VOPC_Real_si <0x27>; +defm V_CMP_U_F64 : VOPC_Real_si <0x28>; +defm V_CMP_NGE_F64 : VOPC_Real_si <0x29>; +defm V_CMP_NLG_F64 : VOPC_Real_si <0x2a>; +defm V_CMP_NGT_F64 : VOPC_Real_si <0x2b>; +defm V_CMP_NLE_F64 : VOPC_Real_si <0x2c>; +defm V_CMP_NEQ_F64 : VOPC_Real_si <0x2d>; +defm V_CMP_NLT_F64 : VOPC_Real_si <0x2e>; +defm V_CMP_TRU_F64 : VOPC_Real_si <0x2f>; + +defm V_CMPX_F_F64 : VOPC_Real_si <0x30>; +defm V_CMPX_LT_F64 : VOPC_Real_si <0x31>; +defm V_CMPX_EQ_F64 : VOPC_Real_si <0x32>; +defm V_CMPX_LE_F64 : VOPC_Real_si <0x33>; +defm V_CMPX_GT_F64 : VOPC_Real_si <0x34>; +defm V_CMPX_LG_F64 : VOPC_Real_si <0x35>; +defm V_CMPX_GE_F64 : VOPC_Real_si <0x36>; +defm V_CMPX_O_F64 : VOPC_Real_si <0x37>; +defm V_CMPX_U_F64 : VOPC_Real_si <0x38>; +defm V_CMPX_NGE_F64 : VOPC_Real_si <0x39>; +defm V_CMPX_NLG_F64 : VOPC_Real_si <0x3a>; +defm V_CMPX_NGT_F64 : VOPC_Real_si <0x3b>; +defm V_CMPX_NLE_F64 : VOPC_Real_si <0x3c>; +defm V_CMPX_NEQ_F64 : VOPC_Real_si <0x3d>; +defm V_CMPX_NLT_F64 : VOPC_Real_si <0x3e>; +defm V_CMPX_TRU_F64 : VOPC_Real_si <0x3f>; + +defm V_CMPS_F_F32 : VOPC_Real_si <0x40>; +defm V_CMPS_LT_F32 : VOPC_Real_si <0x41>; +defm V_CMPS_EQ_F32 : VOPC_Real_si <0x42>; +defm V_CMPS_LE_F32 : VOPC_Real_si <0x43>; +defm V_CMPS_GT_F32 : VOPC_Real_si <0x44>; +defm V_CMPS_LG_F32 : VOPC_Real_si <0x45>; +defm V_CMPS_GE_F32 : VOPC_Real_si <0x46>; +defm V_CMPS_O_F32 : VOPC_Real_si <0x47>; +defm V_CMPS_U_F32 : VOPC_Real_si <0x48>; +defm V_CMPS_NGE_F32 : VOPC_Real_si <0x49>; +defm V_CMPS_NLG_F32 : VOPC_Real_si <0x4a>; +defm V_CMPS_NGT_F32 : VOPC_Real_si <0x4b>; +defm V_CMPS_NLE_F32 : VOPC_Real_si <0x4c>; +defm V_CMPS_NEQ_F32 : VOPC_Real_si <0x4d>; +defm V_CMPS_NLT_F32 : VOPC_Real_si <0x4e>; +defm V_CMPS_TRU_F32 : VOPC_Real_si <0x4f>; + +defm V_CMPSX_F_F32 : VOPC_Real_si <0x50>; +defm V_CMPSX_LT_F32 : VOPC_Real_si <0x51>; +defm V_CMPSX_EQ_F32 : VOPC_Real_si <0x52>; +defm V_CMPSX_LE_F32 : VOPC_Real_si <0x53>; +defm V_CMPSX_GT_F32 : VOPC_Real_si <0x54>; +defm V_CMPSX_LG_F32 : VOPC_Real_si <0x55>; +defm V_CMPSX_GE_F32 : VOPC_Real_si <0x56>; +defm V_CMPSX_O_F32 : VOPC_Real_si <0x57>; +defm V_CMPSX_U_F32 : VOPC_Real_si <0x58>; +defm V_CMPSX_NGE_F32 : VOPC_Real_si <0x59>; +defm V_CMPSX_NLG_F32 : VOPC_Real_si <0x5a>; +defm V_CMPSX_NGT_F32 : VOPC_Real_si <0x5b>; +defm V_CMPSX_NLE_F32 : VOPC_Real_si <0x5c>; +defm V_CMPSX_NEQ_F32 : VOPC_Real_si <0x5d>; +defm V_CMPSX_NLT_F32 : VOPC_Real_si <0x5e>; +defm V_CMPSX_TRU_F32 : VOPC_Real_si <0x5f>; + +defm V_CMPS_F_F64 : VOPC_Real_si <0x60>; +defm V_CMPS_LT_F64 : VOPC_Real_si <0x61>; +defm V_CMPS_EQ_F64 : VOPC_Real_si <0x62>; +defm V_CMPS_LE_F64 : VOPC_Real_si <0x63>; +defm V_CMPS_GT_F64 : VOPC_Real_si <0x64>; +defm V_CMPS_LG_F64 : VOPC_Real_si <0x65>; +defm V_CMPS_GE_F64 : VOPC_Real_si <0x66>; +defm V_CMPS_O_F64 : VOPC_Real_si <0x67>; +defm V_CMPS_U_F64 : VOPC_Real_si <0x68>; +defm V_CMPS_NGE_F64 : VOPC_Real_si <0x69>; +defm V_CMPS_NLG_F64 : VOPC_Real_si <0x6a>; +defm V_CMPS_NGT_F64 : VOPC_Real_si <0x6b>; +defm V_CMPS_NLE_F64 : VOPC_Real_si <0x6c>; +defm V_CMPS_NEQ_F64 : VOPC_Real_si <0x6d>; +defm V_CMPS_NLT_F64 : VOPC_Real_si <0x6e>; +defm V_CMPS_TRU_F64 : VOPC_Real_si <0x6f>; + +defm V_CMPSX_F_F64 : VOPC_Real_si <0x70>; +defm V_CMPSX_LT_F64 : VOPC_Real_si <0x71>; +defm V_CMPSX_EQ_F64 : VOPC_Real_si <0x72>; +defm V_CMPSX_LE_F64 : VOPC_Real_si <0x73>; +defm V_CMPSX_GT_F64 : VOPC_Real_si <0x74>; +defm V_CMPSX_LG_F64 : VOPC_Real_si <0x75>; +defm V_CMPSX_GE_F64 : VOPC_Real_si <0x76>; +defm V_CMPSX_O_F64 : VOPC_Real_si <0x77>; +defm V_CMPSX_U_F64 : VOPC_Real_si <0x78>; +defm V_CMPSX_NGE_F64 : VOPC_Real_si <0x79>; +defm V_CMPSX_NLG_F64 : VOPC_Real_si <0x7a>; +defm V_CMPSX_NGT_F64 : VOPC_Real_si <0x7b>; +defm V_CMPSX_NLE_F64 : VOPC_Real_si <0x7c>; +defm V_CMPSX_NEQ_F64 : VOPC_Real_si <0x7d>; +defm V_CMPSX_NLT_F64 : VOPC_Real_si <0x7e>; +defm V_CMPSX_TRU_F64 : VOPC_Real_si <0x7f>; + +defm V_CMP_F_I32 : VOPC_Real_si <0x80>; +defm V_CMP_LT_I32 : VOPC_Real_si <0x81>; +defm V_CMP_EQ_I32 : VOPC_Real_si <0x82>; +defm V_CMP_LE_I32 : VOPC_Real_si <0x83>; +defm V_CMP_GT_I32 : VOPC_Real_si <0x84>; +defm V_CMP_NE_I32 : VOPC_Real_si <0x85>; +defm V_CMP_GE_I32 : VOPC_Real_si <0x86>; +defm V_CMP_T_I32 : VOPC_Real_si <0x87>; + +defm V_CMPX_F_I32 : VOPC_Real_si <0x90>; +defm V_CMPX_LT_I32 : VOPC_Real_si <0x91>; +defm V_CMPX_EQ_I32 : VOPC_Real_si <0x92>; +defm V_CMPX_LE_I32 : VOPC_Real_si <0x93>; +defm V_CMPX_GT_I32 : VOPC_Real_si <0x94>; +defm V_CMPX_NE_I32 : VOPC_Real_si <0x95>; +defm V_CMPX_GE_I32 : VOPC_Real_si <0x96>; +defm V_CMPX_T_I32 : VOPC_Real_si <0x97>; + +defm V_CMP_F_I64 : VOPC_Real_si <0xa0>; +defm V_CMP_LT_I64 : VOPC_Real_si <0xa1>; +defm V_CMP_EQ_I64 : VOPC_Real_si <0xa2>; +defm V_CMP_LE_I64 : VOPC_Real_si <0xa3>; +defm V_CMP_GT_I64 : VOPC_Real_si <0xa4>; +defm V_CMP_NE_I64 : VOPC_Real_si <0xa5>; +defm V_CMP_GE_I64 : VOPC_Real_si <0xa6>; +defm V_CMP_T_I64 : VOPC_Real_si <0xa7>; + +defm V_CMPX_F_I64 : VOPC_Real_si <0xb0>; +defm V_CMPX_LT_I64 : VOPC_Real_si <0xb1>; +defm V_CMPX_EQ_I64 : VOPC_Real_si <0xb2>; +defm V_CMPX_LE_I64 : VOPC_Real_si <0xb3>; +defm V_CMPX_GT_I64 : VOPC_Real_si <0xb4>; +defm V_CMPX_NE_I64 : VOPC_Real_si <0xb5>; +defm V_CMPX_GE_I64 : VOPC_Real_si <0xb6>; +defm V_CMPX_T_I64 : VOPC_Real_si <0xb7>; + +defm V_CMP_F_U32 : VOPC_Real_si <0xc0>; +defm V_CMP_LT_U32 : VOPC_Real_si <0xc1>; +defm V_CMP_EQ_U32 : VOPC_Real_si <0xc2>; +defm V_CMP_LE_U32 : VOPC_Real_si <0xc3>; +defm V_CMP_GT_U32 : VOPC_Real_si <0xc4>; +defm V_CMP_NE_U32 : VOPC_Real_si <0xc5>; +defm V_CMP_GE_U32 : VOPC_Real_si <0xc6>; +defm V_CMP_T_U32 : VOPC_Real_si <0xc7>; + +defm V_CMPX_F_U32 : VOPC_Real_si <0xd0>; +defm V_CMPX_LT_U32 : VOPC_Real_si <0xd1>; +defm V_CMPX_EQ_U32 : VOPC_Real_si <0xd2>; +defm V_CMPX_LE_U32 : VOPC_Real_si <0xd3>; +defm V_CMPX_GT_U32 : VOPC_Real_si <0xd4>; +defm V_CMPX_NE_U32 : VOPC_Real_si <0xd5>; +defm V_CMPX_GE_U32 : VOPC_Real_si <0xd6>; +defm V_CMPX_T_U32 : VOPC_Real_si <0xd7>; + +defm V_CMP_F_U64 : VOPC_Real_si <0xe0>; +defm V_CMP_LT_U64 : VOPC_Real_si <0xe1>; +defm V_CMP_EQ_U64 : VOPC_Real_si <0xe2>; +defm V_CMP_LE_U64 : VOPC_Real_si <0xe3>; +defm V_CMP_GT_U64 : VOPC_Real_si <0xe4>; +defm V_CMP_NE_U64 : VOPC_Real_si <0xe5>; +defm V_CMP_GE_U64 : VOPC_Real_si <0xe6>; +defm V_CMP_T_U64 : VOPC_Real_si <0xe7>; + +defm V_CMPX_F_U64 : VOPC_Real_si <0xf0>; +defm V_CMPX_LT_U64 : VOPC_Real_si <0xf1>; +defm V_CMPX_EQ_U64 : VOPC_Real_si <0xf2>; +defm V_CMPX_LE_U64 : VOPC_Real_si <0xf3>; +defm V_CMPX_GT_U64 : VOPC_Real_si <0xf4>; +defm V_CMPX_NE_U64 : VOPC_Real_si <0xf5>; +defm V_CMPX_GE_U64 : VOPC_Real_si <0xf6>; +defm V_CMPX_T_U64 : VOPC_Real_si <0xf7>; + +defm V_CMP_CLASS_F32 : VOPC_Real_si <0x88>; +defm V_CMPX_CLASS_F32 : VOPC_Real_si <0x98>; +defm V_CMP_CLASS_F64 : VOPC_Real_si <0xa8>; +defm V_CMPX_CLASS_F64 : VOPC_Real_si <0xb8>; + +//===----------------------------------------------------------------------===// +// VI +//===----------------------------------------------------------------------===// + +class VOPC_SDWAe op, VOPProfile P> : VOP_SDWAeNew

{ + bits<8> src1; + + let Inst{8-0} = 0xf9; // sdwa + let Inst{16-9} = !if(P.HasSrc1, src1{7-0}, 0); + let Inst{24-17} = op; + let Inst{31-25} = 0x3e; // encoding + + // VOPC disallows dst_sel and dst_unused as they have no effect on destination + let Inst{42-40} = SDWA_DWORD; + let Inst{44-43} = SDWA_UNUSED_PRESERVE; +} + +class VOPC_SDWA op, VOPC_Pseudo ps, VOPProfile p = ps.Pfl> : + VOP_SDWA , + VOPC_SDWAe { + let Defs = ps.Defs; + let hasSideEffects = ps.hasSideEffects; + let AsmMatchConverter = "cvtSdwaVOPC"; + let SubtargetPredicate = isVI; + let AssemblerPredicate = !if(p.HasExt, isVI, DisableInst); + let AsmVariantName = !if(p.HasExt, AMDGPUAsmVariants.SDWA, + AMDGPUAsmVariants.Disable); + let DecoderNamespace = "SDWA"; + let isCompare = ps.isCompare; + let isCommutable = ps.isCommutable; +} + +multiclass VOPC_Real_vi op> { + let AssemblerPredicates = [isVI], DecoderNamespace = "VI" in { + def _e32_vi : + VOPC_Real(NAME#"_e32"), SIEncodingFamily.VI>, + VOPCe; + + def _e64_vi : + VOP3_Real(NAME#"_e64"), SIEncodingFamily.VI>, + VOP3a_viNew (NAME#"_e64").Pfl> { + // Encoding used for VOPC instructions encoded as VOP3 + // Differs from VOP3e by destination name (sdst) as VOPC doesn't have vector dst + bits<8> sdst; + let Inst{7-0} = sdst; + } + } + + // for now left sdwa only for asm/dasm + // TODO: add corresponding pseudo + def _sdwa : VOPC_SDWA(NAME#"_e32")>; + + def : VOPCInstAlias (NAME#"_e64"), + !cast(NAME#"_e32_vi")> { + let AssemblerPredicate = isVI; + } +} + +defm V_CMP_F_F32 : VOPC_Real_vi <0x40>; +defm V_CMP_LT_F32 : VOPC_Real_vi <0x41>; +defm V_CMP_EQ_F32 : VOPC_Real_vi <0x42>; +defm V_CMP_LE_F32 : VOPC_Real_vi <0x43>; +defm V_CMP_GT_F32 : VOPC_Real_vi <0x44>; +defm V_CMP_LG_F32 : VOPC_Real_vi <0x45>; +defm V_CMP_GE_F32 : VOPC_Real_vi <0x46>; +defm V_CMP_O_F32 : VOPC_Real_vi <0x47>; +defm V_CMP_U_F32 : VOPC_Real_vi <0x48>; +defm V_CMP_NGE_F32 : VOPC_Real_vi <0x49>; +defm V_CMP_NLG_F32 : VOPC_Real_vi <0x4a>; +defm V_CMP_NGT_F32 : VOPC_Real_vi <0x4b>; +defm V_CMP_NLE_F32 : VOPC_Real_vi <0x4c>; +defm V_CMP_NEQ_F32 : VOPC_Real_vi <0x4d>; +defm V_CMP_NLT_F32 : VOPC_Real_vi <0x4e>; +defm V_CMP_TRU_F32 : VOPC_Real_vi <0x4f>; + +defm V_CMPX_F_F32 : VOPC_Real_vi <0x50>; +defm V_CMPX_LT_F32 : VOPC_Real_vi <0x51>; +defm V_CMPX_EQ_F32 : VOPC_Real_vi <0x52>; +defm V_CMPX_LE_F32 : VOPC_Real_vi <0x53>; +defm V_CMPX_GT_F32 : VOPC_Real_vi <0x54>; +defm V_CMPX_LG_F32 : VOPC_Real_vi <0x55>; +defm V_CMPX_GE_F32 : VOPC_Real_vi <0x56>; +defm V_CMPX_O_F32 : VOPC_Real_vi <0x57>; +defm V_CMPX_U_F32 : VOPC_Real_vi <0x58>; +defm V_CMPX_NGE_F32 : VOPC_Real_vi <0x59>; +defm V_CMPX_NLG_F32 : VOPC_Real_vi <0x5a>; +defm V_CMPX_NGT_F32 : VOPC_Real_vi <0x5b>; +defm V_CMPX_NLE_F32 : VOPC_Real_vi <0x5c>; +defm V_CMPX_NEQ_F32 : VOPC_Real_vi <0x5d>; +defm V_CMPX_NLT_F32 : VOPC_Real_vi <0x5e>; +defm V_CMPX_TRU_F32 : VOPC_Real_vi <0x5f>; + +defm V_CMP_F_F64 : VOPC_Real_vi <0x60>; +defm V_CMP_LT_F64 : VOPC_Real_vi <0x61>; +defm V_CMP_EQ_F64 : VOPC_Real_vi <0x62>; +defm V_CMP_LE_F64 : VOPC_Real_vi <0x63>; +defm V_CMP_GT_F64 : VOPC_Real_vi <0x64>; +defm V_CMP_LG_F64 : VOPC_Real_vi <0x65>; +defm V_CMP_GE_F64 : VOPC_Real_vi <0x66>; +defm V_CMP_O_F64 : VOPC_Real_vi <0x67>; +defm V_CMP_U_F64 : VOPC_Real_vi <0x68>; +defm V_CMP_NGE_F64 : VOPC_Real_vi <0x69>; +defm V_CMP_NLG_F64 : VOPC_Real_vi <0x6a>; +defm V_CMP_NGT_F64 : VOPC_Real_vi <0x6b>; +defm V_CMP_NLE_F64 : VOPC_Real_vi <0x6c>; +defm V_CMP_NEQ_F64 : VOPC_Real_vi <0x6d>; +defm V_CMP_NLT_F64 : VOPC_Real_vi <0x6e>; +defm V_CMP_TRU_F64 : VOPC_Real_vi <0x6f>; + +defm V_CMPX_F_F64 : VOPC_Real_vi <0x70>; +defm V_CMPX_LT_F64 : VOPC_Real_vi <0x71>; +defm V_CMPX_EQ_F64 : VOPC_Real_vi <0x72>; +defm V_CMPX_LE_F64 : VOPC_Real_vi <0x73>; +defm V_CMPX_GT_F64 : VOPC_Real_vi <0x74>; +defm V_CMPX_LG_F64 : VOPC_Real_vi <0x75>; +defm V_CMPX_GE_F64 : VOPC_Real_vi <0x76>; +defm V_CMPX_O_F64 : VOPC_Real_vi <0x77>; +defm V_CMPX_U_F64 : VOPC_Real_vi <0x78>; +defm V_CMPX_NGE_F64 : VOPC_Real_vi <0x79>; +defm V_CMPX_NLG_F64 : VOPC_Real_vi <0x7a>; +defm V_CMPX_NGT_F64 : VOPC_Real_vi <0x7b>; +defm V_CMPX_NLE_F64 : VOPC_Real_vi <0x7c>; +defm V_CMPX_NEQ_F64 : VOPC_Real_vi <0x7d>; +defm V_CMPX_NLT_F64 : VOPC_Real_vi <0x7e>; +defm V_CMPX_TRU_F64 : VOPC_Real_vi <0x7f>; + +defm V_CMP_F_I32 : VOPC_Real_vi <0xc0>; +defm V_CMP_LT_I32 : VOPC_Real_vi <0xc1>; +defm V_CMP_EQ_I32 : VOPC_Real_vi <0xc2>; +defm V_CMP_LE_I32 : VOPC_Real_vi <0xc3>; +defm V_CMP_GT_I32 : VOPC_Real_vi <0xc4>; +defm V_CMP_NE_I32 : VOPC_Real_vi <0xc5>; +defm V_CMP_GE_I32 : VOPC_Real_vi <0xc6>; +defm V_CMP_T_I32 : VOPC_Real_vi <0xc7>; + +defm V_CMPX_F_I32 : VOPC_Real_vi <0xd0>; +defm V_CMPX_LT_I32 : VOPC_Real_vi <0xd1>; +defm V_CMPX_EQ_I32 : VOPC_Real_vi <0xd2>; +defm V_CMPX_LE_I32 : VOPC_Real_vi <0xd3>; +defm V_CMPX_GT_I32 : VOPC_Real_vi <0xd4>; +defm V_CMPX_NE_I32 : VOPC_Real_vi <0xd5>; +defm V_CMPX_GE_I32 : VOPC_Real_vi <0xd6>; +defm V_CMPX_T_I32 : VOPC_Real_vi <0xd7>; + +defm V_CMP_F_I64 : VOPC_Real_vi <0xe0>; +defm V_CMP_LT_I64 : VOPC_Real_vi <0xe1>; +defm V_CMP_EQ_I64 : VOPC_Real_vi <0xe2>; +defm V_CMP_LE_I64 : VOPC_Real_vi <0xe3>; +defm V_CMP_GT_I64 : VOPC_Real_vi <0xe4>; +defm V_CMP_NE_I64 : VOPC_Real_vi <0xe5>; +defm V_CMP_GE_I64 : VOPC_Real_vi <0xe6>; +defm V_CMP_T_I64 : VOPC_Real_vi <0xe7>; + +defm V_CMPX_F_I64 : VOPC_Real_vi <0xf0>; +defm V_CMPX_LT_I64 : VOPC_Real_vi <0xf1>; +defm V_CMPX_EQ_I64 : VOPC_Real_vi <0xf2>; +defm V_CMPX_LE_I64 : VOPC_Real_vi <0xf3>; +defm V_CMPX_GT_I64 : VOPC_Real_vi <0xf4>; +defm V_CMPX_NE_I64 : VOPC_Real_vi <0xf5>; +defm V_CMPX_GE_I64 : VOPC_Real_vi <0xf6>; +defm V_CMPX_T_I64 : VOPC_Real_vi <0xf7>; + +defm V_CMP_F_U32 : VOPC_Real_vi <0xc8>; +defm V_CMP_LT_U32 : VOPC_Real_vi <0xc9>; +defm V_CMP_EQ_U32 : VOPC_Real_vi <0xca>; +defm V_CMP_LE_U32 : VOPC_Real_vi <0xcb>; +defm V_CMP_GT_U32 : VOPC_Real_vi <0xcc>; +defm V_CMP_NE_U32 : VOPC_Real_vi <0xcd>; +defm V_CMP_GE_U32 : VOPC_Real_vi <0xce>; +defm V_CMP_T_U32 : VOPC_Real_vi <0xcf>; + +defm V_CMPX_F_U32 : VOPC_Real_vi <0xd8>; +defm V_CMPX_LT_U32 : VOPC_Real_vi <0xd9>; +defm V_CMPX_EQ_U32 : VOPC_Real_vi <0xda>; +defm V_CMPX_LE_U32 : VOPC_Real_vi <0xdb>; +defm V_CMPX_GT_U32 : VOPC_Real_vi <0xdc>; +defm V_CMPX_NE_U32 : VOPC_Real_vi <0xdd>; +defm V_CMPX_GE_U32 : VOPC_Real_vi <0xde>; +defm V_CMPX_T_U32 : VOPC_Real_vi <0xdf>; + +defm V_CMP_F_U64 : VOPC_Real_vi <0xe8>; +defm V_CMP_LT_U64 : VOPC_Real_vi <0xe9>; +defm V_CMP_EQ_U64 : VOPC_Real_vi <0xea>; +defm V_CMP_LE_U64 : VOPC_Real_vi <0xeb>; +defm V_CMP_GT_U64 : VOPC_Real_vi <0xec>; +defm V_CMP_NE_U64 : VOPC_Real_vi <0xed>; +defm V_CMP_GE_U64 : VOPC_Real_vi <0xee>; +defm V_CMP_T_U64 : VOPC_Real_vi <0xef>; + +defm V_CMPX_F_U64 : VOPC_Real_vi <0xf8>; +defm V_CMPX_LT_U64 : VOPC_Real_vi <0xf9>; +defm V_CMPX_EQ_U64 : VOPC_Real_vi <0xfa>; +defm V_CMPX_LE_U64 : VOPC_Real_vi <0xfb>; +defm V_CMPX_GT_U64 : VOPC_Real_vi <0xfc>; +defm V_CMPX_NE_U64 : VOPC_Real_vi <0xfd>; +defm V_CMPX_GE_U64 : VOPC_Real_vi <0xfe>; +defm V_CMPX_T_U64 : VOPC_Real_vi <0xff>; + +defm V_CMP_CLASS_F32 : VOPC_Real_vi <0x10>; +defm V_CMPX_CLASS_F32 : VOPC_Real_vi <0x11>; +defm V_CMP_CLASS_F64 : VOPC_Real_vi <0x12>; +defm V_CMPX_CLASS_F64 : VOPC_Real_vi <0x13>; diff --git a/lib/Target/AMDGPU/VOPInstructions.td b/lib/Target/AMDGPU/VOPInstructions.td new file mode 100644 index 00000000000..863089f1038 --- /dev/null +++ b/lib/Target/AMDGPU/VOPInstructions.td @@ -0,0 +1,130 @@ +//===-- VOPInstructions.td - Vector Instruction Defintions ----------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +class VOP3_PseudoNew pattern, bit VOP3Only = 0> : + InstSI , + VOP , + SIMCInstr, + MnemonicAlias { + + let isPseudo = 1; + let isCodeGenOnly = 1; + let UseNamedOperandTable = 1; + + string Mnemonic = opName; + string AsmOperands = P.Asm64; + + let Size = 8; + let mayLoad = 0; + let mayStore = 0; + let hasSideEffects = 0; + let SubtargetPredicate = isGCN; + + // Because SGPRs may be allowed if there are multiple operands, we + // need a post-isel hook to insert copies in order to avoid + // violating constant bus requirements. + let hasPostISelHook = 1; + + // Using complex patterns gives VOP3 patterns a very high complexity rating, + // but standalone patterns are almost always prefered, so we need to adjust the + // priority lower. The goal is to use a high number to reduce complexity to + // zero (or less than zero). + let AddedComplexity = -1000; + + let VOP3 = 1; + let VALU = 1; + let Uses = [EXEC]; + + let AsmVariantName = AMDGPUAsmVariants.VOP3; + let AsmMatchConverter = + !if(!eq(VOP3Only,1), + "cvtVOP3", + !if(!eq(P.HasModifiers, 1), "cvtVOP3_2_mod", "")); + + VOPProfile Pfl = P; +} + +class VOP3_Real : + InstSI , + SIMCInstr { + + let isPseudo = 0; + let isCodeGenOnly = 0; + + // copy relevant pseudo op flags + let SubtargetPredicate = ps.SubtargetPredicate; + let AsmMatchConverter = ps.AsmMatchConverter; + let AsmVariantName = ps.AsmVariantName; + let Constraints = ps.Constraints; + let DisableEncoding = ps.DisableEncoding; + let TSFlags = ps.TSFlags; +} + +class VOP3aNew : Enc64 { + bits<2> src0_modifiers; + bits<9> src0; + bits<2> src1_modifiers; + bits<9> src1; + bits<2> src2_modifiers; + bits<9> src2; + bits<1> clamp; + bits<2> omod; + + let Inst{8} = !if(P.HasSrc0Mods, src0_modifiers{1}, 0); + let Inst{9} = !if(P.HasSrc1Mods, src1_modifiers{1}, 0); + let Inst{10} = !if(P.HasSrc2Mods, src2_modifiers{1}, 0); + + let Inst{31-26} = 0x34; //encoding + let Inst{40-32} = src0; + let Inst{49-41} = !if(P.HasSrc1, src1, 0); + let Inst{58-50} = !if(P.HasSrc2, src2, 0); + let Inst{60-59} = !if(P.HasOMod, omod, 0); + let Inst{61} = !if(P.HasSrc0Mods, src0_modifiers{0}, 0); + let Inst{62} = !if(P.HasSrc1Mods, src1_modifiers{0}, 0); + let Inst{63} = !if(P.HasSrc2Mods, src2_modifiers{0}, 0); +} + +class VOP3a_siNew op, VOPProfile P> : VOP3aNew

{ + let Inst{25-17} = op; + let Inst{11} = !if(P.HasClamp, clamp, 0); +} + +class VOP3a_viNew op, VOPProfile P> : VOP3aNew

{ + let Inst{25-16} = op; + let Inst{15} = !if(P.HasClamp, clamp, 0); +} + +class VOP_SDWAeNew : Enc64 { + bits<8> src0; + bits<3> src0_sel; + bits<2> src0_fmodifiers; // {abs,neg} + bits<1> src0_imodifiers; // sext + bits<3> src1_sel; + bits<2> src1_fmodifiers; + bits<1> src1_imodifiers; + bits<3> dst_sel; + bits<2> dst_unused; + bits<1> clamp; + + bits<3> SDWA_DWORD = 6; + bits<2> SDWA_UNUSED_PRESERVE = 2; + + let Inst{39-32} = !if(P.HasSrc0, src0{7-0}, 0); + let Inst{42-40} = !if(P.HasDst, dst_sel{2-0}, SDWA_DWORD{2-0}); + let Inst{44-43} = !if(P.HasDst, dst_unused{1-0}, SDWA_UNUSED_PRESERVE{1-0}); + let Inst{45} = !if(P.HasSDWAClamp, clamp{0}, 0); + let Inst{50-48} = !if(P.HasSrc0, src0_sel{2-0}, SDWA_DWORD{2-0}); + let Inst{53-52} = !if(P.HasSrc0Mods, src0_fmodifiers{1-0}, 0); + let Inst{51} = !if(P.HasSrc0IntMods, src0_imodifiers{0}, 0); + let Inst{58-56} = !if(P.HasSrc1, src1_sel{2-0}, SDWA_DWORD{2-0}); + let Inst{61-60} = !if(P.HasSrc1Mods, src1_fmodifiers{1-0}, 0); + let Inst{59} = !if(P.HasSrc1IntMods, src1_imodifiers{0}, 0); +} + +include "VOPCInstructions.td"