Index: lib/Target/AMDGPU/SIInstrInfo.td =================================================================== --- lib/Target/AMDGPU/SIInstrInfo.td +++ lib/Target/AMDGPU/SIInstrInfo.td @@ -1160,6 +1160,21 @@ let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1); } +class VOP3b_Profile : VOPProfile<[vt, vt, vt, vt]> { + let Outs64 = (outs DstRC:$vdst, SReg_64:$sdst); + let Asm64 = "$vdst, $sdst, $src0_modifiers, $src1_modifiers, $src2_modifiers"#"$clamp"#"$omod"; +} + +def VOP3b_F32_I1_F32_F32_F32 : VOP3b_Profile { + // FIXME: Hack to stop printing _e64 + let DstRC = RegisterOperand; +} + +def VOP3b_F64_I1_F64_F64_F64 : VOP3b_Profile { + // FIXME: Hack to stop printing _e64 + 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 $dst. @@ -1848,25 +1863,13 @@ 3, 1 >; -multiclass VOP3b_Helper pattern> : +multiclass VOP3bInst pattern = []> : VOP3b_2_3_m < - op, (outs vrc:$vdst, SReg_64:$sdst), - (ins InputModsNoDefault:$src0_modifiers, arc:$src0, - InputModsNoDefault:$src1_modifiers, arc:$src1, - InputModsNoDefault:$src2_modifiers, arc:$src2, - ClampMod:$clamp, omod:$omod), - opName#" $vdst, $sdst, $src0_modifiers, $src1_modifiers, $src2_modifiers"#"$clamp"#"$omod", pattern, - opName, opName, 1, 1 + op, P.Outs64, P.Ins64, + opName#" "#P.Asm64, pattern, + opName, "", 1, 1 >; -multiclass VOP3b_64 pattern> : - VOP3b_Helper ; - -multiclass VOP3b_32 pattern> : - VOP3b_Helper ; - - class Vop3ModPat : Pat< (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)), (P.Src1VT (VOP3Mods P.Src1VT:$src1, i32:$src1_modifiers)), Index: lib/Target/AMDGPU/SIInstructions.td =================================================================== --- lib/Target/AMDGPU/SIInstructions.td +++ lib/Target/AMDGPU/SIInstructions.td @@ -1761,12 +1761,16 @@ } // isCommutable = 1, SchedRW = [WriteQuarterRate32] let SchedRW = [WriteFloatFMA, WriteSALU] in { -defm V_DIV_SCALE_F32 : VOP3b_32 , "v_div_scale_f32", []>; +defm V_DIV_SCALE_F32 : VOP3bInst , "v_div_scale_f32", + VOP3b_F32_I1_F32_F32_F32 +>; } let SchedRW = [WriteDouble, WriteSALU] in { // Double precision division pre-scale. -defm V_DIV_SCALE_F64 : VOP3b_64 , "v_div_scale_f64", []>; +defm V_DIV_SCALE_F64 : VOP3bInst , "v_div_scale_f64", + VOP3b_F64_I1_F64_F64_F64 +>; } // let SchedRW = [WriteDouble] let isCommutable = 1, Uses = [VCC, EXEC] in {