Index: llvm/lib/Target/ARM/ARMISelLowering.cpp =================================================================== --- llvm/lib/Target/ARM/ARMISelLowering.cpp +++ llvm/lib/Target/ARM/ARMISelLowering.cpp @@ -21,6 +21,7 @@ #include "ARMRegisterInfo.h" #include "ARMSelectionDAGInfo.h" #include "ARMSubtarget.h" +#include "ARMTargetTransformInfo.h" #include "MCTargetDesc/ARMAddressingModes.h" #include "MCTargetDesc/ARMBaseInfo.h" #include "Utils/ARMBaseInfo.h" @@ -6965,35 +6966,6 @@ return true; } -/// isVREVMask - Check if a vector shuffle corresponds to a VREV -/// instruction with the specified blocksize. (The order of the elements -/// within each block of the vector is reversed.) -static bool isVREVMask(ArrayRef M, EVT VT, unsigned BlockSize) { - assert((BlockSize==16 || BlockSize==32 || BlockSize==64) && - "Only possible block sizes for VREV are: 16, 32, 64"); - - unsigned EltSz = VT.getScalarSizeInBits(); - if (EltSz == 64) - return false; - - unsigned NumElts = VT.getVectorNumElements(); - unsigned BlockElts = M[0] + 1; - // If the first shuffle index is UNDEF, be optimistic. - if (M[0] < 0) - BlockElts = BlockSize / EltSz; - - if (BlockSize <= EltSz || BlockSize != BlockElts * EltSz) - return false; - - for (unsigned i = 0; i < NumElts; ++i) { - if (M[i] < 0) continue; // ignore UNDEF indices - if ((unsigned) M[i] != (i - i%BlockElts) + (BlockElts - 1 - i%BlockElts)) - return false; - } - - return true; -} - static bool isVTBLMask(ArrayRef M, EVT VT) { // We can handle <8 x i8> vector shuffles. If the index in the mask is out of // range, then 0 is placed into the resulting vector. So pretty much any mask Index: llvm/lib/Target/ARM/ARMTargetTransformInfo.h =================================================================== --- llvm/lib/Target/ARM/ARMTargetTransformInfo.h +++ llvm/lib/Target/ARM/ARMTargetTransformInfo.h @@ -282,6 +282,34 @@ /// @} }; +/// isVREVMask - Check if a vector shuffle corresponds to a VREV +/// instruction with the specified blocksize. (The order of the elements +/// within each block of the vector is reversed.) +inline bool isVREVMask(ArrayRef M, EVT VT, unsigned BlockSize) { + assert((BlockSize==16 || BlockSize==32 || BlockSize==64) && + "Only possible block sizes for VREV are: 16, 32, 64"); + + unsigned EltSz = VT.getScalarSizeInBits(); + if (EltSz != 8 && EltSz != 16 && EltSz != 32) + return false; + + unsigned BlockElts = M[0] + 1; + // If the first shuffle index is UNDEF, be optimistic. + if (M[0] < 0) + BlockElts = BlockSize / EltSz; + + if (BlockSize <= EltSz || BlockSize != BlockElts * EltSz) + return false; + + for (unsigned i = 0; i < M.size(); ++i) { + if (M[i] < 0) continue; // ignore UNDEF indices + if ((unsigned) M[i] != (i - i%BlockElts) + (BlockElts - 1 - i%BlockElts)) + return false; + } + + return true; +} + } // end namespace llvm #endif // LLVM_LIB_TARGET_ARM_ARMTARGETTRANSFORMINFO_H Index: llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp =================================================================== --- llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp +++ llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp @@ -1218,7 +1218,15 @@ return LT.first * Entry->Cost * ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput); } + + if (!Mask.empty()) { + std::pair LT = TLI->getTypeLegalizationCost(DL, Tp); + if (isVREVMask(Mask, LT.second, 16) || isVREVMask(Mask, LT.second, 32) || + isVREVMask(Mask, LT.second, 64)) + return ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput) * LT.first; + } } + int BaseCost = ST->hasMVEIntegerOps() && Tp->isVectorTy() ? ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput) : 1; Index: llvm/test/Analysis/CostModel/ARM/shuffle.ll =================================================================== --- llvm/test/Analysis/CostModel/ARM/shuffle.ll +++ llvm/test/Analysis/CostModel/ARM/shuffle.ll @@ -101,11 +101,11 @@ ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v2i64 = shufflevector <2 x i64> undef, <2 x i64> undef, <2 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v4i64 = shufflevector <4 x i64> undef, <4 x i64> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v2f16 = shufflevector <2 x half> undef, <2 x half> undef, <2 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f16 = shufflevector <4 x half> undef, <4 x half> undef, <4 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f16 = shufflevector <2 x half> undef, <2 x half> undef, <2 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4f16 = shufflevector <4 x half> undef, <4 x half> undef, <4 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f16 = shufflevector <8 x half> undef, <8 x half> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v2f32 = shufflevector <2 x float> undef, <2 x float> undef, <2 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f32 = shufflevector <2 x float> undef, <2 x float> undef, <2 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f32 = shufflevector <4 x float> undef, <4 x float> undef, <4 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f32 = shufflevector <8 x float> undef, <8 x float> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v2f64 = shufflevector <2 x double> undef, <2 x double> undef, <2 x i32> @@ -246,11 +246,11 @@ ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v2i64 = shufflevector <2 x i64> undef, <2 x i64> undef, <2 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v4i64 = shufflevector <4 x i64> undef, <4 x i64> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v2f16 = shufflevector <2 x half> undef, <2 x half> undef, <2 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f16 = shufflevector <2 x half> undef, <2 x half> undef, <2 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f16 = shufflevector <4 x half> undef, <4 x half> undef, <4 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f16 = shufflevector <8 x half> undef, <8 x half> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v2f32 = shufflevector <2 x float> undef, <2 x float> undef, <2 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f32 = shufflevector <2 x float> undef, <2 x float> undef, <2 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f32 = shufflevector <4 x float> undef, <4 x float> undef, <4 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f32 = shufflevector <8 x float> undef, <8 x float> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v2f64 = shufflevector <2 x double> undef, <2 x double> undef, <2 x i32> @@ -316,20 +316,20 @@ define void @vrev2() { ; CHECK-MVE-LABEL: 'vrev2' -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v4i8 = shufflevector <4 x i8> undef, <4 x i8> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <8 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 256 for instruction: %v16i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v4i16 = shufflevector <4 x i16> undef, <4 x i16> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 256 for instruction: %v16i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v4i32 = shufflevector <4 x i32> undef, <4 x i32> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8 = shufflevector <4 x i8> undef, <4 x i8> undef, <4 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i16 = shufflevector <4 x i16> undef, <4 x i16> undef, <4 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i32 = shufflevector <4 x i32> undef, <4 x i32> undef, <4 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v4i64 = shufflevector <4 x i64> undef, <4 x i64> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f16 = shufflevector <4 x half> undef, <4 x half> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f16 = shufflevector <8 x half> undef, <8 x half> undef, <8 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f32 = shufflevector <4 x float> undef, <4 x float> undef, <4 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f32 = shufflevector <8 x float> undef, <8 x float> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4f16 = shufflevector <4 x half> undef, <4 x half> undef, <4 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8f16 = shufflevector <8 x half> undef, <8 x half> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4f32 = shufflevector <4 x float> undef, <4 x float> undef, <4 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v8f32 = shufflevector <8 x float> undef, <8 x float> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f64 = shufflevector <4 x double> undef, <4 x double> undef, <4 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; @@ -378,13 +378,13 @@ define void @vrev4() { ; CHECK-MVE-LABEL: 'vrev4' -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <8 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 256 for instruction: %v16i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 256 for instruction: %v16i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i8 = shufflevector <8 x i8> undef, <8 x i8> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i8 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i16 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i16 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v8i32 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f16 = shufflevector <8 x half> undef, <8 x half> undef, <8 x i32> -; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8f16 = shufflevector <8 x half> undef, <8 x half> undef, <8 x i32> +; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %v8f32 = shufflevector <8 x float> undef, <8 x float> undef, <8 x i32> ; CHECK-MVE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; Index: llvm/test/Transforms/PhaseOrdering/ARM/mve-floatreduce.ll =================================================================== --- llvm/test/Transforms/PhaseOrdering/ARM/mve-floatreduce.ll +++ llvm/test/Transforms/PhaseOrdering/ARM/mve-floatreduce.ll @@ -10,11 +10,13 @@ ; CHECK-NEXT: entry: ; CHECK-NEXT: [[TMP0:%.*]] = shufflevector <8 x half> [[IN:%.*]], <8 x half> poison, <8 x i32> ; CHECK-NEXT: [[TMP1:%.*]] = fadd fast <8 x half> [[TMP0]], [[IN]] -; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <8 x half> [[TMP1]], <8 x half> poison, <8 x i32> -; CHECK-NEXT: [[TMP3:%.*]] = fadd fast <8 x half> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[TMP4:%.*]] = extractelement <8 x half> [[TMP3]], i32 0 -; CHECK-NEXT: [[TMP5:%.*]] = extractelement <8 x half> [[TMP3]], i32 4 -; CHECK-NEXT: [[ADD:%.*]] = fadd fast half [[TMP4]], [[TMP5]] +; CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x half> [[TMP1]] to <4 x i32> +; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <4 x i32> +; CHECK-NEXT: [[TMP4:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x half> +; CHECK-NEXT: [[TMP5:%.*]] = fadd fast <8 x half> [[TMP1]], [[TMP4]] +; CHECK-NEXT: [[TMP6:%.*]] = extractelement <8 x half> [[TMP5]], i32 0 +; CHECK-NEXT: [[TMP7:%.*]] = extractelement <8 x half> [[TMP5]], i32 4 +; CHECK-NEXT: [[ADD:%.*]] = fadd fast half [[TMP6]], [[TMP7]] ; CHECK-NEXT: ret half [[ADD]] ; entry: