Skip to content

[CostModel][X86] Update SK_Reverse based on cost kinds #150650

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jul 26, 2025
Merged

[CostModel][X86] Update SK_Reverse based on cost kinds #150650

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ab96552
[CostModel][X86] Update SK_Reverse based on cost kinds
RKSimon Jul 25, 2025
7f587df
Merge branch 'main' into x86-costs-sk_reverse
RKSimon Jul 26, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
55 changes: 28 additions & 27 deletions llvm/lib/Target/X86/X86TargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1842,10 +1842,11 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
{ TTI::SK_Broadcast, MVT::v32f16, { 1, 3, 1, 1 } }, // vpbroadcastw
{ TTI::SK_Broadcast, MVT::v64i8, { 1, 3, 1, 1 } }, // vpbroadcastb

{ TTI::SK_Reverse, MVT::v32i16, { 2, 2, 2, 2 } }, // vpermw
{ TTI::SK_Reverse, MVT::v32f16, { 2, 2, 2, 2 } }, // vpermw
{ TTI::SK_Reverse, MVT::v32i16, { 2, 6, 2, 4 } }, // vpermw
{ TTI::SK_Reverse, MVT::v32f16, { 2, 6, 2, 4 } }, // vpermw
{ TTI::SK_Reverse, MVT::v16i16, { 2, 2, 2, 2 } }, // vpermw
{ TTI::SK_Reverse, MVT::v64i8, { 2, 2, 2, 2 } }, // pshufb + vshufi64x2
{ TTI::SK_Reverse, MVT::v16f16, { 2, 2, 2, 2 } }, // vpermw
{ TTI::SK_Reverse, MVT::v64i8, { 2, 9, 2, 3 } }, // pshufb + vshufi64x2

{ TTI::SK_PermuteSingleSrc, MVT::v32i16, { 2, 2, 2, 2 } }, // vpermw
{ TTI::SK_PermuteSingleSrc, MVT::v32f16, { 2, 2, 2, 2 } }, // vpermw
Expand Down Expand Up @@ -1889,10 +1890,10 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
{TTI::SK_Broadcast, MVT::v64i8, { 1, 3, 1, 1 } }, // vpbroadcastb
{TTI::SK_Broadcast, MVT::v32i8, { 1, 3, 1, 1 }}, // vpbroadcastb

{TTI::SK_Reverse, MVT::v8f64, { 1, 3, 1, 1 } }, // vpermpd
{TTI::SK_Reverse, MVT::v16f32, { 1, 3, 1, 1 } }, // vpermps
{TTI::SK_Reverse, MVT::v8i64, { 1, 3, 1, 1 } }, // vpermq
{TTI::SK_Reverse, MVT::v16i32, { 1, 3, 1, 1 } }, // vpermd
{TTI::SK_Reverse, MVT::v8f64, { 1, 5, 2, 3 } }, // vpermpd
{TTI::SK_Reverse, MVT::v16f32, { 1, 3, 2, 3 } }, // vpermps
{TTI::SK_Reverse, MVT::v8i64, { 1, 5, 2, 3 } }, // vpermq
{TTI::SK_Reverse, MVT::v16i32, { 1, 3, 2, 3 } }, // vpermd
{TTI::SK_Reverse, MVT::v32i16, { 7, 7, 7, 7 } }, // per mca
{TTI::SK_Reverse, MVT::v32f16, { 7, 7, 7, 7 } }, // per mca
{TTI::SK_Reverse, MVT::v64i8, { 7, 7, 7, 7 } }, // per mca
Expand Down Expand Up @@ -1991,13 +1992,13 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
{ TTI::SK_Broadcast, MVT::v32i8, { 1, 3, 1, 2 } }, // vpbroadcastb
{ TTI::SK_Broadcast, MVT::v16i8, { 1, 3, 1, 1 } }, // vpbroadcastb

{ TTI::SK_Reverse, MVT::v4f64, { 1, 1, 1, 1 } }, // vpermpd
{ TTI::SK_Reverse, MVT::v8f32, { 1, 1, 1, 1 } }, // vpermps
{ TTI::SK_Reverse, MVT::v4i64, { 1, 1, 1, 1 } }, // vpermq
{ TTI::SK_Reverse, MVT::v8i32, { 1, 1, 1, 1 } }, // vpermd
{ TTI::SK_Reverse, MVT::v16i16, { 2, 2, 2, 2 } }, // vperm2i128 + pshufb
{ TTI::SK_Reverse, MVT::v16f16, { 2, 2, 2, 2 } }, // vperm2i128 + pshufb
{ TTI::SK_Reverse, MVT::v32i8, { 2, 2, 2, 2 } }, // vperm2i128 + pshufb
{ TTI::SK_Reverse, MVT::v4f64, { 1, 6, 1, 2 } }, // vpermpd
{ TTI::SK_Reverse, MVT::v8f32, { 2, 7, 2, 4 } }, // vpermps
{ TTI::SK_Reverse, MVT::v4i64, { 1, 6, 1, 2 } }, // vpermq
{ TTI::SK_Reverse, MVT::v8i32, { 2, 7, 2, 4 } }, // vpermd
{ TTI::SK_Reverse, MVT::v16i16, { 2, 9, 2, 4 } }, // vperm2i128 + pshufb
{ TTI::SK_Reverse, MVT::v16f16, { 2, 9, 2, 4 } }, // vperm2i128 + pshufb
{ TTI::SK_Reverse, MVT::v32i8, { 2, 9, 2, 4 } }, // vperm2i128 + pshufb

{ TTI::SK_Select, MVT::v16i16, { 1, 1, 1, 1 } }, // vpblendvb
{ TTI::SK_Select, MVT::v16f16, { 1, 1, 1, 1 } }, // vpblendvb
Expand Down Expand Up @@ -2095,15 +2096,15 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
{TTI::SK_Broadcast, MVT::v16f16, {2,3,3,4}}, // vpshuflw + vpshufd + vinsertf128
{TTI::SK_Broadcast, MVT::v32i8, {3,4,3,6}}, // vpshufb + vinsertf128

{TTI::SK_Reverse, MVT::v4f64, {2,2,2,2}}, // vperm2f128 + vpermilpd
{TTI::SK_Reverse, MVT::v8f32, {2,2,2,2}}, // vperm2f128 + vpermilps
{TTI::SK_Reverse, MVT::v4i64, {2,2,2,2}}, // vperm2f128 + vpermilpd
{TTI::SK_Reverse, MVT::v8i32, {2,2,2,2}}, // vperm2f128 + vpermilps
{TTI::SK_Reverse, MVT::v16i16, {4,4,4,4}}, // vextractf128 + 2*pshufb
{TTI::SK_Reverse, MVT::v4f64, {2,6,2,2}}, // vperm2f128 + vpermilpd
{TTI::SK_Reverse, MVT::v8f32, {2,7,2,4}}, // vperm2f128 + vpermilps
{TTI::SK_Reverse, MVT::v4i64, {2,6,2,2}}, // vperm2f128 + vpermilpd
{TTI::SK_Reverse, MVT::v8i32, {2,7,2,4}}, // vperm2f128 + vpermilps
{TTI::SK_Reverse, MVT::v16i16, {2,9,5,5}}, // vextractf128 + 2*pshufb
// + vinsertf128
{TTI::SK_Reverse, MVT::v16f16, {4,4,4,4}}, // vextractf128 + 2*pshufb
{TTI::SK_Reverse, MVT::v16f16, {2,9,5,5}}, // vextractf128 + 2*pshufb
// + vinsertf128
{TTI::SK_Reverse, MVT::v32i8, {4,4,4,4}}, // vextractf128 + 2*pshufb
{TTI::SK_Reverse, MVT::v32i8, {2,9,5,5}}, // vextractf128 + 2*pshufb
// + vinsertf128

{TTI::SK_Select, MVT::v4i64, {1,1,1,1}}, // vblendpd
Expand Down Expand Up @@ -2170,9 +2171,9 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
{TTI::SK_Broadcast, MVT::v8f16, {1, 3, 2, 2}}, // pshufb
{TTI::SK_Broadcast, MVT::v16i8, {1, 3, 2, 2}}, // pshufb

{TTI::SK_Reverse, MVT::v8i16, {1, 1, 1, 1}}, // pshufb
{TTI::SK_Reverse, MVT::v8f16, {1, 1, 1, 1}}, // pshufb
{TTI::SK_Reverse, MVT::v16i8, {1, 1, 1, 1}}, // pshufb
{TTI::SK_Reverse, MVT::v8i16, {1, 2, 1, 2}}, // pshufb
{TTI::SK_Reverse, MVT::v8f16, {1, 2, 1, 2}}, // pshufb
{TTI::SK_Reverse, MVT::v16i8, {1, 2, 1, 2}}, // pshufb

{TTI::SK_Select, MVT::v8i16, {3, 3, 3, 3}}, // 2*pshufb + por
{TTI::SK_Select, MVT::v8f16, {3, 3, 3, 3}}, // 2*pshufb + por
Expand Down Expand Up @@ -2209,9 +2210,9 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
{TTI::SK_Reverse, MVT::v2f64, {1, 1, 1, 1}}, // shufpd
{TTI::SK_Reverse, MVT::v2i64, {1, 1, 1, 1}}, // pshufd
{TTI::SK_Reverse, MVT::v4i32, {1, 1, 1, 1}}, // pshufd
{TTI::SK_Reverse, MVT::v8i16, {3, 3, 3, 3}}, // pshuflw + pshufhw + pshufd
{TTI::SK_Reverse, MVT::v8f16, {3, 3, 3, 3}}, // pshuflw + pshufhw + pshufd
{TTI::SK_Reverse, MVT::v16i8, {9, 9, 9, 9}}, // 2*pshuflw + 2*pshufhw
{TTI::SK_Reverse, MVT::v8i16, {2, 3, 3, 3}}, // pshuflw + pshufhw + pshufd
{TTI::SK_Reverse, MVT::v8f16, {2, 3, 3, 3}}, // pshuflw + pshufhw + pshufd
{TTI::SK_Reverse, MVT::v16i8, {5, 6,11,11}}, // 2*pshuflw + 2*pshufhw
// + 2*pshufd + 2*unpck + packus

{TTI::SK_Select, MVT::v2i64, {1, 1, 1, 1}}, // movsd
Expand Down
8 changes: 4 additions & 4 deletions llvm/test/Analysis/CostModel/X86/shuffle-reverse-fp16.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,11 +3,11 @@

define void @test_vXf16(<2 x half> %src32, <4 x half> %src64, <8 x half> %src128, <16 x half> %src256, <32 x half> %src512) {
; CHECK-LABEL: 'test_vXf16'
; CHECK-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> <i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> <i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V64 = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:4 for: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
%V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> <i32 1, i32 0>
Expand Down
Loading
Loading