This is an archive of the discontinued LLVM Phabricator instance.

Differential D90613

[x86] add AVX2 cost model entries for maxnum of 256-bit vectors
ClosedPublic

Authored by spatel on Nov 2 2020, 7:46 AM.

Details

Reviewers
RKSimon
pengfei
craig.topper
Commits
rG35fa3c474f4e: [x86] add AVX2 cost model entries for maxnum of 256-bit vectors
Summary

As noticed in D90554 , the AVX2 costs for 256-bit vectors did not include FMAXNUM entries, so we fell back to AVX1 which assumes those ops will be split into 128-bit halves.

Diff Detail

Event Timeline

spatel created this revision.Nov 2 2020, 7:46 AM
Herald added a project: Restricted Project. · View Herald TranscriptNov 2 2020, 7:46 AM
Herald added subscribers: hiraditya, mcrosier. · View Herald Transcript
spatel requested review of this revision.Nov 2 2020, 7:46 AM
RKSimon accepted this revision.Nov 2 2020, 8:07 AM

LGTM - cheers

This revision is now accepted and ready to land.Nov 2 2020, 8:07 AM
Closed by commit rG35fa3c474f4e: [x86] add AVX2 cost model entries for maxnum of 256-bit vectors (authored by spatel). · Explain WhyNov 2 2020, 9:20 AM
This revision was automatically updated to reflect the committed changes.
spatel added a commit: rG35fa3c474f4e: [x86] add AVX2 cost model entries for maxnum of 256-bit vectors.
spatel mentioned this in D90554: [CostModel] remove cost-kind predicate for intrinsics in basic TTI implementation.Nov 2 2020, 12:02 PM
pengfei added inline comments.Nov 2 2020, 5:08 PM
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
2531

Nit: the comments should be MAXSS + CMPUNORDSS.
The same below.

spatel mentioned this in rG9af561ec9935: [x86] update cost table comments for maxnum; NFC.Nov 3 2020, 5:10 AM
spatel added inline comments.Nov 3 2020, 5:15 AM
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
2531

Thanks - updated with 9af561e.

Revision Contents

PathSize
llvm/
lib/
Target/
X86/
6 lines
test/
Analysis/
CostModel/
X86/
60 lines
60 lines

Diff 302308

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

Show First 20 Lines • Show All 2,467 Lines • ▼ Show 20 Lines static const CostTblEntry AVX2CostTbl[] = {
{ ISD::UMAX, MVT::v16i16, 1 }, { ISD::UMAX, MVT::v16i16, 1 },
{ ISD::UMAX, MVT::v32i8, 1 }, { ISD::UMAX, MVT::v32i8, 1 },
{ ISD::UMIN, MVT::v8i32, 1 }, { ISD::UMIN, MVT::v8i32, 1 },
{ ISD::UMIN, MVT::v16i16, 1 }, { ISD::UMIN, MVT::v16i16, 1 },
{ ISD::UMIN, MVT::v32i8, 1 }, { ISD::UMIN, MVT::v32i8, 1 },
{ ISD::USUBSAT, MVT::v16i16, 1 }, { ISD::USUBSAT, MVT::v16i16, 1 },
{ ISD::USUBSAT, MVT::v32i8, 1 }, { ISD::USUBSAT, MVT::v32i8, 1 },
{ ISD::USUBSAT, MVT::v8i32, 2 }, // pmaxud + psubd { ISD::USUBSAT, MVT::v8i32, 2 }, // pmaxud + psubd
{ ISD::FMAXNUM, MVT::v8f32, 3 }, // MAXPS + CMPUNORDPS + BLENDVPS
{ ISD::FMAXNUM, MVT::v4f64, 3 }, // MAXPD + CMPUNORDPD + BLENDVPD
{ ISD::FSQRT, MVT::f32, 7 }, // Haswell from http://www.agner.org/ { ISD::FSQRT, MVT::f32, 7 }, // Haswell from http://www.agner.org/
{ ISD::FSQRT, MVT::v4f32, 7 }, // Haswell from http://www.agner.org/ { ISD::FSQRT, MVT::v4f32, 7 }, // Haswell from http://www.agner.org/
{ ISD::FSQRT, MVT::v8f32, 14 }, // Haswell from http://www.agner.org/ { ISD::FSQRT, MVT::v8f32, 14 }, // Haswell from http://www.agner.org/
{ ISD::FSQRT, MVT::f64, 14 }, // Haswell from http://www.agner.org/ { ISD::FSQRT, MVT::f64, 14 }, // Haswell from http://www.agner.org/
{ ISD::FSQRT, MVT::v2f64, 14 }, // Haswell from http://www.agner.org/ { ISD::FSQRT, MVT::v2f64, 14 }, // Haswell from http://www.agner.org/
{ ISD::FSQRT, MVT::v4f64, 28 }, // Haswell from http://www.agner.org/ { ISD::FSQRT, MVT::v4f64, 28 }, // Haswell from http://www.agner.org/
}; };
static const CostTblEntry AVX1CostTbl[] = { static const CostTblEntry AVX1CostTbl[] = {
Show All 37 Lines static const CostTblEntry AVX1CostTbl[] = {
{ ISD::UMAX, MVT::v16i16, 4 }, // 2 x 128-bit Op + extract/insert { ISD::UMAX, MVT::v16i16, 4 }, // 2 x 128-bit Op + extract/insert
{ ISD::UMAX, MVT::v32i8, 4 }, // 2 x 128-bit Op + extract/insert { ISD::UMAX, MVT::v32i8, 4 }, // 2 x 128-bit Op + extract/insert
{ ISD::UMIN, MVT::v8i32, 4 }, // 2 x 128-bit Op + extract/insert { ISD::UMIN, MVT::v8i32, 4 }, // 2 x 128-bit Op + extract/insert
{ ISD::UMIN, MVT::v16i16, 4 }, // 2 x 128-bit Op + extract/insert { ISD::UMIN, MVT::v16i16, 4 }, // 2 x 128-bit Op + extract/insert
{ ISD::UMIN, MVT::v32i8, 4 }, // 2 x 128-bit Op + extract/insert { ISD::UMIN, MVT::v32i8, 4 }, // 2 x 128-bit Op + extract/insert
{ ISD::USUBSAT, MVT::v16i16, 4 }, // 2 x 128-bit Op + extract/insert { ISD::USUBSAT, MVT::v16i16, 4 }, // 2 x 128-bit Op + extract/insert
{ ISD::USUBSAT, MVT::v32i8, 4 }, // 2 x 128-bit Op + extract/insert { ISD::USUBSAT, MVT::v32i8, 4 }, // 2 x 128-bit Op + extract/insert
{ ISD::USUBSAT, MVT::v8i32, 6 }, // 2 x 128-bit Op + extract/insert { ISD::USUBSAT, MVT::v8i32, 6 }, // 2 x 128-bit Op + extract/insert
{ ISD::FMAXNUM, MVT::f32, 3 }, { ISD::FMAXNUM, MVT::f32, 3 }, // MAXPS + CMPUNORDPS + BLENDVPS
pengfeiUnsubmitted
Not Done
ReplyInline Actions

Nit: the comments should be MAXSS + CMPUNORDSS.
The same below.

pengfei: Nit: the comments should be MAXSS + CMPUNORDSS. The same below.
spatelAuthorUnsubmitted
Done
ReplyInline Actions

Thanks - updated with 9af561e.

spatel: Thanks - updated with 9af561e.
{ ISD::FMAXNUM, MVT::v4f32, 3 }, { ISD::FMAXNUM, MVT::v4f32, 3 },
{ ISD::FMAXNUM, MVT::v8f32, 5 }, { ISD::FMAXNUM, MVT::v8f32, 5 },
{ ISD::FMAXNUM, MVT::f64, 3 }, { ISD::FMAXNUM, MVT::f64, 3 }, // MAXPD + CMPUNORDPD + BLENDVPD
{ ISD::FMAXNUM, MVT::v2f64, 3 }, { ISD::FMAXNUM, MVT::v2f64, 3 },
{ ISD::FMAXNUM, MVT::v4f64, 5 }, { ISD::FMAXNUM, MVT::v4f64, 5 },
{ ISD::FSQRT, MVT::f32, 14 }, // SNB from http://www.agner.org/ { ISD::FSQRT, MVT::f32, 14 }, // SNB from http://www.agner.org/
{ ISD::FSQRT, MVT::v4f32, 14 }, // SNB from http://www.agner.org/ { ISD::FSQRT, MVT::v4f32, 14 }, // SNB from http://www.agner.org/
{ ISD::FSQRT, MVT::v8f32, 28 }, // SNB from http://www.agner.org/ { ISD::FSQRT, MVT::v8f32, 28 }, // SNB from http://www.agner.org/
{ ISD::FSQRT, MVT::f64, 21 }, // SNB from http://www.agner.org/ { ISD::FSQRT, MVT::f64, 21 }, // SNB from http://www.agner.org/
{ ISD::FSQRT, MVT::v2f64, 21 }, // SNB from http://www.agner.org/ { ISD::FSQRT, MVT::v2f64, 21 }, // SNB from http://www.agner.org/
{ ISD::FSQRT, MVT::v4f64, 43 }, // SNB from http://www.agner.org/ { ISD::FSQRT, MVT::v4f64, 43 }, // SNB from http://www.agner.org/
▲ Show 20 Lines • Show All 2,181 Lines • Show Last 20 Lines

llvm/test/Analysis/CostModel/X86/fmaxnum.ll

Show All 13 Lines
; SSE-LABEL: 'f32' ; SSE-LABEL: 'f32'
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX-LABEL: 'f32' ; AVX1-LABEL: 'f32'
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX2-LABEL: 'f32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX512-LABEL: 'f32' ; AVX512-LABEL: 'f32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
Show All 9 Lines
; GLM-LABEL: 'f32' ; GLM-LABEL: 'f32'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; BTVER2-LABEL: 'f32'
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
%F32 = call float @llvm.maxnum.f32(float undef, float undef) %F32 = call float @llvm.maxnum.f32(float undef, float undef)
%V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
%V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
%V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
%V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
ret i32 undef ret i32 undef
} }
define i32 @f64(i32 %arg) { define i32 @f64(i32 %arg) {
; SSE-LABEL: 'f64' ; SSE-LABEL: 'f64'
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX-LABEL: 'f64' ; AVX1-LABEL: 'f64'
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX2-LABEL: 'f64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX512-LABEL: 'f64' ; AVX512-LABEL: 'f64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
Show All 9 Lines
; GLM-LABEL: 'f64' ; GLM-LABEL: 'f64'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; BTVER2-LABEL: 'f64'
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
%f64 = call double @llvm.maxnum.f64(double undef, double undef) %f64 = call double @llvm.maxnum.f64(double undef, double undef)
%V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
%V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
%V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
%V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
ret i32 undef ret i32 undef
} }
Show All 11 Lines

llvm/test/Analysis/CostModel/X86/fminnum.ll

Show All 13 Lines
; SSE-LABEL: 'f32' ; SSE-LABEL: 'f32'
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX-LABEL: 'f32' ; AVX1-LABEL: 'f32'
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX2-LABEL: 'f32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX512-LABEL: 'f32' ; AVX512-LABEL: 'f32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
Show All 9 Lines
; GLM-LABEL: 'f32' ; GLM-LABEL: 'f32'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; BTVER2-LABEL: 'f32'
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
%F32 = call float @llvm.minnum.f32(float undef, float undef) %F32 = call float @llvm.minnum.f32(float undef, float undef)
%V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
%V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
%V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
%V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
ret i32 undef ret i32 undef
} }
define i32 @f64(i32 %arg) { define i32 @f64(i32 %arg) {
; SSE-LABEL: 'f64' ; SSE-LABEL: 'f64'
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX-LABEL: 'f64' ; AVX1-LABEL: 'f64'
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) ; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX2-LABEL: 'f64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; AVX512-LABEL: 'f64' ; AVX512-LABEL: 'f64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
Show All 9 Lines
; GLM-LABEL: 'f64' ; GLM-LABEL: 'f64'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) ; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
; ;
; BTVER2-LABEL: 'f64'
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
%f64 = call double @llvm.minnum.f64(double undef, double undef) %f64 = call double @llvm.minnum.f64(double undef, double undef)
%V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
%V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
%V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
%V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
ret i32 undef ret i32 undef
} }
Show All 11 Lines