aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/Target/AMDGPU/VOP3Instructions.td
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/AMDGPU/VOP3Instructions.td')
-rw-r--r--llvm/lib/Target/AMDGPU/VOP3Instructions.td31
1 files changed, 22 insertions, 9 deletions
diff --git a/llvm/lib/Target/AMDGPU/VOP3Instructions.td b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
index 4a2b54d..42ec8ba 100644
--- a/llvm/lib/Target/AMDGPU/VOP3Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3Instructions.td
@@ -97,6 +97,7 @@ class VOP3Interp<string OpName, VOPProfile P, list<dag> pattern = []> :
VOP3_Pseudo<OpName, P, pattern> {
let AsmMatchConverter = "cvtVOP3Interp";
let mayRaiseFPException = 0;
+ let VOP3_OPSEL = P.HasOpSel;
}
def VOP3_INTERP : VOPProfile<[f32, f32, i32, untyped]> {
@@ -119,16 +120,17 @@ def VOP3_INTERP_MOV : VOPProfile<[f32, i32, i32, untyped]> {
let HasSrc0Mods = 0;
}
-class getInterp16Asm <bit HasSrc2, bit HasOMod> {
+class getInterp16Asm <bit HasSrc2, bit HasOMod, bit OpSel> {
string src2 = !if(HasSrc2, ", $src2_modifiers", "");
string omod = !if(HasOMod, "$omod", "");
+ string opsel = !if(OpSel, "$op_sel", "");
string ret =
- " $vdst, $src0_modifiers, $attr$attrchan"#src2#"$high$clamp"#omod;
+ " $vdst, $src0_modifiers, $attr$attrchan"#src2#"$high$clamp"#omod#opsel;
}
class getInterp16Ins <bit HasSrc2, bit HasOMod,
- Operand Src0Mod, Operand Src2Mod> {
- dag ret = !if(HasSrc2,
+ Operand Src0Mod, Operand Src2Mod, bit OpSel> {
+ dag ret1 = !if(HasSrc2,
!if(HasOMod,
(ins Src0Mod:$src0_modifiers, VRegSrc_32:$src0,
InterpAttr:$attr, InterpAttrChan:$attrchan,
@@ -143,19 +145,22 @@ class getInterp16Ins <bit HasSrc2, bit HasOMod,
InterpAttr:$attr, InterpAttrChan:$attrchan,
highmod:$high, Clamp0:$clamp, omod0:$omod)
);
+ dag ret2 = !if(OpSel, (ins op_sel0:$op_sel), (ins));
+ dag ret = !con(ret1, ret2);
}
-class VOP3_INTERP16 <list<ValueType> ArgVT> : VOPProfile<ArgVT> {
+class VOP3_INTERP16 <list<ValueType> ArgVT, bit OpSel = 0> : VOPProfile<ArgVT> {
let IsSingle = 1;
let HasOMod = !ne(DstVT.Value, f16.Value);
let HasHigh = 1;
+ let HasOpSel = OpSel;
let Src0Mod = FPVRegInputMods;
let Src2Mod = FPVRegInputMods;
let Outs64 = (outs DstRC.RegClass:$vdst);
- let Ins64 = getInterp16Ins<HasSrc2, HasOMod, Src0Mod, Src2Mod>.ret;
- let Asm64 = getInterp16Asm<HasSrc2, HasOMod>.ret;
+ let Ins64 = getInterp16Ins<HasSrc2, HasOMod, Src0Mod, Src2Mod, OpSel>.ret;
+ let Asm64 = getInterp16Asm<HasSrc2, HasOMod, OpSel>.ret;
}
//===----------------------------------------------------------------------===//
@@ -480,7 +485,7 @@ let SubtargetPredicate = isGFX9Plus in {
defm V_MAD_U16_gfx9 : VOP3Inst_t16 <"v_mad_u16_gfx9", VOP_I16_I16_I16_I16>;
defm V_MAD_I16_gfx9 : VOP3Inst_t16 <"v_mad_i16_gfx9", VOP_I16_I16_I16_I16>;
let OtherPredicates = [isNotGFX90APlus] in
-def V_INTERP_P2_F16_gfx9 : VOP3Interp <"v_interp_p2_f16_gfx9", VOP3_INTERP16<[f16, f32, i32, f32]>>;
+def V_INTERP_P2_F16_opsel : VOP3Interp <"v_interp_p2_f16_opsel", VOP3_INTERP16<[f16, f32, i32, f32], /*OpSel*/ 1>>;
} // End SubtargetPredicate = isGFX9Plus
// This predicate should only apply to the selection pattern. The
@@ -2676,6 +2681,14 @@ multiclass VOP3Interp_F16_Real_gfx9<bits<10> op, string OpName, string AsmName>
}
}
+multiclass VOP3Interp_F16_OpSel_Real_gfx9<bits<10> op, string OpName, string AsmName> {
+ def _gfx9 : VOP3_Real<!cast<VOP3_Pseudo>(OpName), SIEncodingFamily.GFX9>,
+ VOP3Interp_OpSel_gfx9 <op, !cast<VOP3_Pseudo>(OpName).Pfl> {
+ VOP3_Pseudo ps = !cast<VOP3_Pseudo>(OpName);
+ let AsmString = AsmName # ps.AsmOperands;
+ }
+}
+
multiclass VOP3_Real_gfx9<bits<10> op, string AsmName> {
def _gfx9 : VOP3_Real<!cast<VOP_Pseudo>(NAME#"_e64"), SIEncodingFamily.GFX9>,
VOP3e_vi <op, !cast<VOP_Pseudo>(NAME#"_e64").Pfl> {
@@ -2788,7 +2801,7 @@ defm V_MAD_U16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x204, "v_mad_u16">;
defm V_MAD_I16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x205, "v_mad_i16">;
defm V_FMA_F16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x206, "v_fma_f16">;
defm V_DIV_FIXUP_F16_gfx9 : VOP3OpSel_F16_Real_gfx9 <0x207, "v_div_fixup_f16">;
-defm V_INTERP_P2_F16_gfx9 : VOP3Interp_F16_Real_gfx9 <0x277, "V_INTERP_P2_F16_gfx9", "v_interp_p2_f16">;
+defm V_INTERP_P2_F16_opsel : VOP3Interp_F16_OpSel_Real_gfx9 <0x277, "V_INTERP_P2_F16_opsel", "v_interp_p2_f16">;
defm V_ADD_I32 : VOP3_Real_vi <0x29c>;
defm V_SUB_I32 : VOP3_Real_vi <0x29d>;