diff options
author | Simon Pilgrim <llvm-dev@redking.me.uk> | 2025-07-26 13:52:47 +0100 |
---|---|---|
committer | GitHub <noreply@github.com> | 2025-07-26 13:52:47 +0100 |
commit | 0fa0ce1f3af316f21e61286da732ef5c96da495b (patch) | |
tree | 8bd8021c1d9b507d06ed3293707e8e74f252847b /llvm/lib/Target | |
parent | 6a85f7cef384a8f4682e4091f34759602d715ed4 (diff) | |
download | llvm-0fa0ce1f3af316f21e61286da732ef5c96da495b.zip llvm-0fa0ce1f3af316f21e61286da732ef5c96da495b.tar.gz llvm-0fa0ce1f3af316f21e61286da732ef5c96da495b.tar.bz2 |
[CostModel][X86] Update SK_Broadcast based on cost kinds (#150620)
When these were converted to CostKindTblEntry the throughput was mainly copied to all cost kinds
Regenerated with my check_cost_tables.py helper script
Diffstat (limited to 'llvm/lib/Target')
-rw-r--r-- | llvm/lib/Target/X86/X86TargetTransformInfo.cpp | 70 |
1 files changed, 40 insertions, 30 deletions
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp index 37a7b37..6d2c7b4 100644 --- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp +++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp @@ -1838,9 +1838,9 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, return LT.first * *KindCost; static const CostKindTblEntry AVX512BWShuffleTbl[] = { - { TTI::SK_Broadcast, MVT::v32i16, { 1, 1, 1, 1 } }, // vpbroadcastw - { TTI::SK_Broadcast, MVT::v32f16, { 1, 1, 1, 1 } }, // vpbroadcastw - { TTI::SK_Broadcast, MVT::v64i8, { 1, 1, 1, 1 } }, // vpbroadcastb + { TTI::SK_Broadcast, MVT::v32i16, { 1, 3, 1, 1 } }, // vpbroadcastw + { 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 @@ -1874,13 +1874,20 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, return LT.first * *KindCost; static const CostKindTblEntry AVX512ShuffleTbl[] = { - {TTI::SK_Broadcast, MVT::v8f64, { 1, 1, 1, 1 } }, // vbroadcastsd - {TTI::SK_Broadcast, MVT::v16f32, { 1, 1, 1, 1 } }, // vbroadcastss - {TTI::SK_Broadcast, MVT::v8i64, { 1, 1, 1, 1 } }, // vpbroadcastq - {TTI::SK_Broadcast, MVT::v16i32, { 1, 1, 1, 1 } }, // vpbroadcastd - {TTI::SK_Broadcast, MVT::v32i16, { 1, 1, 1, 1 } }, // vpbroadcastw - {TTI::SK_Broadcast, MVT::v32f16, { 1, 1, 1, 1 } }, // vpbroadcastw - {TTI::SK_Broadcast, MVT::v64i8, { 1, 1, 1, 1 } }, // vpbroadcastb + {TTI::SK_Broadcast, MVT::v8f64, { 1, 3, 1, 1 } }, // vbroadcastsd + {TTI::SK_Broadcast, MVT::v4f64, { 1, 3, 1, 1 } }, // vbroadcastsd + {TTI::SK_Broadcast, MVT::v16f32, { 1, 3, 1, 1 } }, // vbroadcastss + {TTI::SK_Broadcast, MVT::v8f32, { 1, 3, 1, 1 } }, // vbroadcastss + {TTI::SK_Broadcast, MVT::v8i64, { 1, 3, 1, 1 } }, // vpbroadcastq + {TTI::SK_Broadcast, MVT::v4i64, { 1, 3, 1, 1 } }, // vpbroadcastq + {TTI::SK_Broadcast, MVT::v16i32, { 1, 3, 1, 1 } }, // vpbroadcastd + {TTI::SK_Broadcast, MVT::v8i32, { 1, 3, 1, 1 } }, // vpbroadcastd + {TTI::SK_Broadcast, MVT::v32i16, { 1, 3, 1, 1 } }, // vpbroadcastw + {TTI::SK_Broadcast, MVT::v16i16, { 1, 3, 1, 1 } }, // vpbroadcastw + {TTI::SK_Broadcast, MVT::v32f16, { 1, 3, 1, 1 } }, // vpbroadcastw + {TTI::SK_Broadcast, MVT::v16f16, { 1, 3, 1, 1 } }, // vpbroadcastw + {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 @@ -1973,13 +1980,16 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, return LT.first * *KindCost; static const CostKindTblEntry AVX2ShuffleTbl[] = { - { TTI::SK_Broadcast, MVT::v4f64, { 1, 1, 1, 1 } }, // vbroadcastpd - { TTI::SK_Broadcast, MVT::v8f32, { 1, 1, 1, 1 } }, // vbroadcastps - { TTI::SK_Broadcast, MVT::v4i64, { 1, 1, 1, 1 } }, // vpbroadcastq - { TTI::SK_Broadcast, MVT::v8i32, { 1, 1, 1, 1 } }, // vpbroadcastd - { TTI::SK_Broadcast, MVT::v16i16, { 1, 1, 1, 1 } }, // vpbroadcastw - { TTI::SK_Broadcast, MVT::v16f16, { 1, 1, 1, 1 } }, // vpbroadcastw - { TTI::SK_Broadcast, MVT::v32i8, { 1, 1, 1, 1 } }, // vpbroadcastb + { TTI::SK_Broadcast, MVT::v4f64, { 1, 3, 1, 2 } }, // vbroadcastpd + { TTI::SK_Broadcast, MVT::v8f32, { 1, 3, 1, 2 } }, // vbroadcastps + { TTI::SK_Broadcast, MVT::v4i64, { 1, 3, 1, 2 } }, // vpbroadcastq + { TTI::SK_Broadcast, MVT::v8i32, { 1, 3, 1, 2 } }, // vpbroadcastd + { TTI::SK_Broadcast, MVT::v16i16, { 1, 3, 1, 2 } }, // vpbroadcastw + { TTI::SK_Broadcast, MVT::v8i16, { 1, 3, 1, 1 } }, // vpbroadcastw + { TTI::SK_Broadcast, MVT::v16f16, { 1, 3, 1, 2 } }, // vpbroadcastw + { TTI::SK_Broadcast, MVT::v8f16, { 1, 3, 1, 1 } }, // vpbroadcastw + { 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 @@ -2077,13 +2087,13 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, return LT.first * *KindCost; static const CostKindTblEntry AVX1ShuffleTbl[] = { - {TTI::SK_Broadcast, MVT::v4f64, {2,2,2,2}}, // vperm2f128 + vpermilpd - {TTI::SK_Broadcast, MVT::v8f32, {2,2,2,2}}, // vperm2f128 + vpermilps - {TTI::SK_Broadcast, MVT::v4i64, {2,2,2,2}}, // vperm2f128 + vpermilpd - {TTI::SK_Broadcast, MVT::v8i32, {2,2,2,2}}, // vperm2f128 + vpermilps - {TTI::SK_Broadcast, MVT::v16i16, {3,3,3,3}}, // vpshuflw + vpshufd + vinsertf128 - {TTI::SK_Broadcast, MVT::v16f16, {3,3,3,3}}, // vpshuflw + vpshufd + vinsertf128 - {TTI::SK_Broadcast, MVT::v32i8, {2,2,2,2}}, // vpshufb + vinsertf128 + {TTI::SK_Broadcast, MVT::v4f64, {2,3,2,3}}, // vperm2f128 + vpermilpd + {TTI::SK_Broadcast, MVT::v8f32, {2,3,2,3}}, // vperm2f128 + vpermilps + {TTI::SK_Broadcast, MVT::v4i64, {2,3,2,3}}, // vperm2f128 + vpermilpd + {TTI::SK_Broadcast, MVT::v8i32, {2,3,2,3}}, // vperm2f128 + vpermilps + {TTI::SK_Broadcast, MVT::v16i16, {2,3,3,4}}, // vpshuflw + vpshufd + vinsertf128 + {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 @@ -2156,9 +2166,9 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, return LT.first * *KindCost; static const CostKindTblEntry SSSE3ShuffleTbl[] = { - {TTI::SK_Broadcast, MVT::v8i16, {1, 1, 1, 1}}, // pshufb - {TTI::SK_Broadcast, MVT::v8f16, {1, 1, 1, 1}}, // pshufb - {TTI::SK_Broadcast, MVT::v16i8, {1, 1, 1, 1}}, // pshufb + {TTI::SK_Broadcast, MVT::v8i16, {1, 3, 2, 2}}, // pshufb + {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 @@ -2192,9 +2202,9 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, {TTI::SK_Broadcast, MVT::v2f64, {1, 1, 1, 1}}, // shufpd {TTI::SK_Broadcast, MVT::v2i64, {1, 1, 1, 1}}, // pshufd {TTI::SK_Broadcast, MVT::v4i32, {1, 1, 1, 1}}, // pshufd - {TTI::SK_Broadcast, MVT::v8i16, {2, 2, 2, 2}}, // pshuflw + pshufd - {TTI::SK_Broadcast, MVT::v8f16, {2, 2, 2, 2}}, // pshuflw + pshufd - {TTI::SK_Broadcast, MVT::v16i8, {3, 3, 3, 3}}, // unpck + pshuflw + pshufd + {TTI::SK_Broadcast, MVT::v8i16, {1, 2, 2, 2}}, // pshuflw + pshufd + {TTI::SK_Broadcast, MVT::v8f16, {1, 2, 2, 2}}, // pshuflw + pshufd + {TTI::SK_Broadcast, MVT::v16i8, {2, 3, 3, 4}}, // unpck + pshuflw + pshufd {TTI::SK_Reverse, MVT::v2f64, {1, 1, 1, 1}}, // shufpd {TTI::SK_Reverse, MVT::v2i64, {1, 1, 1, 1}}, // pshufd |