From: Craig Topper Date: Tue, 14 Apr 2020 03:43:43 +0000 (-0700) Subject: [X86] Use a more realisitic cost for truncate v16i64->v16i8 with avx512f. X-Git-Tag: llvmorg-12-init~9147 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=2f60fbce6cdac9ef72339dde32946b883b6a96a8;p=platform%2Fupstream%2Fllvm.git [X86] Use a more realisitic cost for truncate v16i64->v16i8 with avx512f. Still not great and we could probably codegen this better, but 11 was clearly ridiculous. --- diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp index da776be..c7df5ae 100644 --- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp +++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp @@ -1356,6 +1356,7 @@ int X86TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src, { ISD::TRUNCATE, MVT::v8i8, MVT::v8i64, 2 }, { ISD::TRUNCATE, MVT::v8i16, MVT::v8i64, 1 }, { ISD::TRUNCATE, MVT::v8i32, MVT::v8i64, 1 }, + { ISD::TRUNCATE, MVT::v16i8, MVT::v16i64, 7 },// 2*vpmovqd+concat+vpmovdb // v16i1 -> v16i32 - load + broadcast { ISD::SIGN_EXTEND, MVT::v16i32, MVT::v16i1, 2 }, diff --git a/llvm/test/Analysis/CostModel/X86/trunc.ll b/llvm/test/Analysis/CostModel/X86/trunc.ll index 792310f..9b8b762 100644 --- a/llvm/test/Analysis/CostModel/X86/trunc.ll +++ b/llvm/test/Analysis/CostModel/X86/trunc.ll @@ -283,7 +283,7 @@ define i32 @trunc_vXi8() { ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = trunc <2 x i64> undef to <2 x i8> ; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4i64 = trunc <4 x i64> undef to <4 x i8> ; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i8> -; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8> +; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8> ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8> ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8> ; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8> @@ -304,7 +304,7 @@ define i32 @trunc_vXi8() { ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = trunc <2 x i64> undef to <2 x i8> ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4i64 = trunc <4 x i64> undef to <4 x i8> ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i8> -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8> +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8> ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8> ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8> ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8>