@@ -870,25 +870,40 @@ class VOPC_Class_Profile<list<SchedReadWrite> sched, ValueType src0VT, ValueType
870870
871871multiclass VOPC_Class_Profile_t16<list<SchedReadWrite> sched> {
872872 def NAME : VOPC_Class_Profile<sched, f16>;
873- def _t16 : VOPC_Class_Profile <sched, f16, i16 > {
873+ def _t16 : VOPC_Class_Profile_Base <sched, f16, f16 > {
874874 let IsTrue16 = 1;
875875 let IsRealTrue16 = 1;
876- let Src1RC32 = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
877- let Src1RC64 = VSrc_b32;
878- let Src0DPP = getVregSrcForVT<Src0VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
879- let Src1DPP = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
880- let Src2DPP = getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
881- let Src0ModDPP = getSrcModDPP_t16<Src0VT>.ret;
882- let Src1ModDPP = getSrcModDPP_t16<Src1VT>.ret;
883- let Src2ModDPP = getSrcModDPP_t16<Src2VT>.ret;
876+ let HasOpSel = 1;
877+ let HasModifiers = 1; // All instructions at least have OpSel
878+ let DstRC = getVALUDstForVT<DstVT, 1 /*IsTrue16*/, 0 /*IsVOP3Encoding*/>.ret;
879+ let Src0RC32 = getVOPSrc0ForVT<Src0VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
880+ let Src1RC32 = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
881+ let Src0DPP = getVregSrcForVT<Src0VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
882+ let Src1DPP = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
883+ let Src2DPP = getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
884+ let Src0ModDPP = getSrcModDPP_t16<Src0VT, 0/*IsFake16*/>.ret;
885+ let Src1ModDPP = getSrcModDPP_t16<Src1VT, 0/*IsFake16*/>.ret;
886+ let Src2ModDPP = getSrcModDPP_t16<Src2VT, 0/*IsFake16*/>.ret;
887+ let Src0VOP3DPP = VGPRSrc_16;
888+ let Src1VOP3DPP = getVOP3DPPSrcForVT<Src1VT, 0/*IsFake16*/>.ret;
889+ let Src2VOP3DPP = getVOP3DPPSrcForVT<Src2VT, 0/*IsFake16*/>.ret;
890+
891+ let DstRC64 = getVALUDstForVT<DstVT, 1/*IsTrue16*/, 1/*IsVOP3Encoding*/>.ret;
892+ let Src0RC64 = getVOP3SrcForVT<Src0VT, 1/*IsTrue16*/>.ret;
893+ let Src1RC64 = getVOP3SrcForVT<Src1VT, 1/*IsTrue16*/>.ret;
894+ let Src2RC64 = getVOP3SrcForVT<Src2VT, 1/*IsTrue16*/>.ret;
895+ let Src0Mod = getSrc0Mod<Src0VT, DstVT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
896+ let Src1Mod = getSrcMod<Src1VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
897+ let Src2Mod = getSrcMod<Src2VT, 1/*IsTrue16*/, 0/*IsFake16*/>.ret;
898+ let Src0ModVOP3DPP = getSrc0ModVOP3DPP<Src0VT, DstVT, 0/*IsFake16*/>.ret;
899+ let Src1ModVOP3DPP = getSrcModVOP3DPP<Src1VT, 0/*IsFake16*/>.ret;
900+ let Src2ModVOP3DPP = getSrcModVOP3DPP<Src2VT, 0/*IsFake16*/>.ret;
884901 }
885902 def _fake16 : VOPC_Class_Profile_Base<sched, f16, f16> {
886903 let IsTrue16 = 1;
887904 let DstRC = getVALUDstForVT_fake16<DstVT>.ret;
888- let DstRC64 = getVALUDstForVT<DstVT>.ret;
889905 let Src0RC32 = getVOPSrc0ForVT<Src0VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
890906 let Src1RC32 = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
891- let Src1RC64 = VSrc_b32;
892907 let Src0DPP = getVregSrcForVT<Src0VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
893908 let Src1DPP = getVregSrcForVT<Src1VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
894909 let Src2DPP = getVregSrcForVT<Src2VT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
@@ -898,6 +913,14 @@ multiclass VOPC_Class_Profile_t16<list<SchedReadWrite> sched> {
898913 let Src0VOP3DPP = VGPRSrc_32;
899914 let Src1VOP3DPP = getVOP3DPPSrcForVT<Src1VT, 1/*IsFake16*/>.ret;
900915 let Src2VOP3DPP = getVOP3DPPSrcForVT<Src2VT, 1/*IsFake16*/>.ret;
916+
917+ let DstRC64 = getVALUDstForVT<DstVT>.ret;
918+ let Src0RC64 = getVOP3SrcForVT<Src0VT, 0/*IsTrue16*/>.ret;
919+ let Src1RC64 = getVOP3SrcForVT<Src1VT, 0/*IsTrue16*/>.ret;
920+ let Src2RC64 = getVOP3SrcForVT<Src2VT, 0/*IsTrue16*/>.ret;
921+ let Src0Mod = getSrc0Mod<Src0VT, DstVT, 1/*IsTrue16*/, 1/*IsFake16*/>.ret;
922+ let Src1Mod = getSrcMod<Src1VT, 0/*IsTrue16*/, 1/*IsFake16*/>.ret;
923+ let Src2Mod = getSrcMod<Src2VT, 0/*IsTrue16*/, 1/*IsFake16*/>.ret;
901924 let Src0ModVOP3DPP = getSrc0ModVOP3DPP<Src0VT, DstVT, 1/*IsFake16*/>.ret;
902925 let Src1ModVOP3DPP = getSrcModVOP3DPP<Src1VT, 1/*IsFake16*/>.ret;
903926 let Src2ModVOP3DPP = getSrcModVOP3DPP<Src2VT, 1/*IsFake16*/>.ret;
@@ -1843,7 +1866,7 @@ defm V_CMP_NE_U64 : VOPC_Real_gfx11_gfx12<0x05d>;
18431866defm V_CMP_GE_U64 : VOPC_Real_gfx11_gfx12<0x05e>;
18441867defm V_CMP_T_U64 : VOPC_Real_gfx11<0x05f>;
18451868
1846- defm V_CMP_CLASS_F16_fake16 : VOPC_Real_t16_gfx11_gfx12 <0x07d, "v_cmp_class_f16">;
1869+ defm V_CMP_CLASS_F16 : VOPC_Real_t16_and_fake16_gfx11_gfx12 <0x07d, "v_cmp_class_f16">;
18471870defm V_CMP_CLASS_F32 : VOPC_Real_gfx11_gfx12<0x07e>;
18481871defm V_CMP_CLASS_F64 : VOPC_Real_gfx11_gfx12<0x07f>;
18491872
0 commit comments