Skip to content

Commit 15e2a21

Browse files
RKSimonRKSimon
committed
[CostModel][X86] Treat lrint/llrint as fptosi calls
X86 can use the CVTP2SI instructions to lower lrint/llrint calls, which have the same costs as the CVTTP2SI (fptosi) instructions Followup to llvm#90065
1 parent e4bb663 commit 15e2a21

File tree

6 files changed

+423
-566
lines changed

6 files changed

+423
-566
lines changed

‎llvm/lib/Target/X86/X86TargetTransformInfo.cpp‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4152,6 +4152,14 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
41524152
}
41534153
}
41544154
break;
4155+
case Intrinsic::lrint:
4156+
case Intrinsic::llrint:
4157+
if (!ICA.isTypeBasedOnly()) {
4158+
const SmallVectorImpl<Type *> &ArgTys = ICA.getArgTypes();
4159+
return getCastInstrCost(Instruction::FPToSI, RetTy, ArgTys[0],
4160+
TTI::CastContextHint::None, CostKind);
4161+
}
4162+
break;
41554163
case Intrinsic::maxnum:
41564164
case Intrinsic::minnum:
41574165
// FMINNUM has same costs so don't duplicate.

‎llvm/test/Analysis/CostModel/X86/arith-fp-codesize.ll‎

Lines changed: 20 additions & 64 deletions
Original file line numberDiff line numberDiff line change
@@ -664,38 +664,16 @@ define i32 @rint(i32 %arg) {
664664
}
665665

666666
define i32 @lrint(i32 %arg) {
667-
; SSE1-LABEL: 'lrint'
668-
; SSE1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i32 @llvm.lrint.i32.f32(float undef)
669-
; SSE1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F32 = call <4 x i32> @llvm.lrint.v4i32.v4f32(<4 x float> undef)
670-
; SSE1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8F32 = call <8 x i32> @llvm.lrint.v8i32.v8f32(<8 x float> undef)
671-
; SSE1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16F32 = call <16 x i32> @llvm.lrint.v16i32.v16f32(<16 x float> undef)
672-
; SSE1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i32 @llvm.lrint.i32.f64(double undef)
673-
; SSE1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F64 = call <2 x i32> @llvm.lrint.v2i32.v2f64(<2 x double> undef)
674-
; SSE1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F64 = call <4 x i32> @llvm.lrint.v4i32.v4f64(<4 x double> undef)
675-
; SSE1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8F64 = call <8 x i32> @llvm.lrint.v8i32.v8f64(<8 x double> undef)
676-
; SSE1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
677-
;
678-
; AVX-LABEL: 'lrint'
679-
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i32 @llvm.lrint.i32.f32(float undef)
680-
; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F32 = call <4 x i32> @llvm.lrint.v4i32.v4f32(<4 x float> undef)
681-
; AVX-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8F32 = call <8 x i32> @llvm.lrint.v8i32.v8f32(<8 x float> undef)
682-
; AVX-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16F32 = call <16 x i32> @llvm.lrint.v16i32.v16f32(<16 x float> undef)
683-
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i32 @llvm.lrint.i32.f64(double undef)
684-
; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F64 = call <2 x i32> @llvm.lrint.v2i32.v2f64(<2 x double> undef)
685-
; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F64 = call <4 x i32> @llvm.lrint.v4i32.v4f64(<4 x double> undef)
686-
; AVX-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8F64 = call <8 x i32> @llvm.lrint.v8i32.v8f64(<8 x double> undef)
687-
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
688-
;
689-
; AVX512-LABEL: 'lrint'
690-
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i32 @llvm.lrint.i32.f32(float undef)
691-
; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F32 = call <4 x i32> @llvm.lrint.v4i32.v4f32(<4 x float> undef)
692-
; AVX512-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8F32 = call <8 x i32> @llvm.lrint.v8i32.v8f32(<8 x float> undef)
693-
; AVX512-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V16F32 = call <16 x i32> @llvm.lrint.v16i32.v16f32(<16 x float> undef)
694-
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i32 @llvm.lrint.i32.f64(double undef)
695-
; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F64 = call <2 x i32> @llvm.lrint.v2i32.v2f64(<2 x double> undef)
696-
; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F64 = call <4 x i32> @llvm.lrint.v4i32.v4f64(<4 x double> undef)
697-
; AVX512-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8F64 = call <8 x i32> @llvm.lrint.v8i32.v8f64(<8 x double> undef)
698-
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
667+
; CHECK-LABEL: 'lrint'
668+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i32 @llvm.lrint.i32.f32(float undef)
669+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call <4 x i32> @llvm.lrint.v4i32.v4f32(<4 x float> undef)
670+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call <8 x i32> @llvm.lrint.v8i32.v8f32(<8 x float> undef)
671+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x i32> @llvm.lrint.v16i32.v16f32(<16 x float> undef)
672+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i32 @llvm.lrint.i32.f64(double undef)
673+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = call <2 x i32> @llvm.lrint.v2i32.v2f64(<2 x double> undef)
674+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = call <4 x i32> @llvm.lrint.v4i32.v4f64(<4 x double> undef)
675+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x i32> @llvm.lrint.v8i32.v8f64(<8 x double> undef)
676+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
699677
;
700678
%F32 = call i32 @llvm.lrint.i32.f32(float undef)
701679
%V4F32 = call <4 x i32> @llvm.lrint.v4i32.v4f32(<4 x float> undef)
@@ -711,38 +689,16 @@ define i32 @lrint(i32 %arg) {
711689
}
712690

713691
define i32 @llrint(i32 %arg) {
714-
; SSE1-LABEL: 'llrint'
715-
; SSE1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
716-
; SSE1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
717-
; SSE1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
718-
; SSE1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
719-
; SSE1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
720-
; SSE1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
721-
; SSE1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
722-
; SSE1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
723-
; SSE1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
724-
;
725-
; AVX-LABEL: 'llrint'
726-
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
727-
; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
728-
; AVX-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
729-
; AVX-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
730-
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
731-
; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
732-
; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
733-
; AVX-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
734-
; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
735-
;
736-
; AVX512-LABEL: 'llrint'
737-
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
738-
; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
739-
; AVX512-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
740-
; AVX512-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
741-
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
742-
; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
743-
; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
744-
; AVX512-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
745-
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
692+
; CHECK-LABEL: 'llrint'
693+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
694+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
695+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
696+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
697+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
698+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
699+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
700+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
701+
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
746702
;
747703
%F32 = call i64 @llvm.llrint.i64.f32(float undef)
748704
%V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)

0 commit comments

Comments
 (0)