Index: llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp =================================================================== --- llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp +++ llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp @@ -1994,6 +1994,18 @@ case Intrinsic::fshl: case Intrinsic::fshr: { + // Canonicalize a shift amount constant operand to be modulo the bit-width. + unsigned BitWidth = II->getType()->getScalarSizeInBits(); + Constant *ShAmtC; + if (match(II->getArgOperand(2), m_Constant(ShAmtC))) { + Constant *WidthC = ConstantInt::get(II->getType(), BitWidth); + Constant *ModuloC = ConstantExpr::get(Instruction::URem, ShAmtC, WidthC); + if (ModuloC != ShAmtC) { + II->setArgOperand(2, ModuloC); + return II; + } + } + const APInt *SA; if (match(II->getArgOperand(2), m_APInt(SA))) { Value *Op0 = II->getArgOperand(0), *Op1 = II->getArgOperand(1); @@ -2020,7 +2032,6 @@ // The shift amount (operand 2) of a funnel shift is modulo the bitwidth, // so only the low bits of the shift amount are demanded if the bitwidth is // a power-of-2. - unsigned BitWidth = II->getType()->getScalarSizeInBits(); if (!isPowerOf2_32(BitWidth)) break; APInt Op2Demanded = APInt::getLowBitsSet(BitWidth, Log2_32_Ceil(BitWidth)); Index: llvm/test/Transforms/InstCombine/fsh.ll =================================================================== --- llvm/test/Transforms/InstCombine/fsh.ll +++ llvm/test/Transforms/InstCombine/fsh.ll @@ -310,7 +310,7 @@ define i32 @fshl_constant_shift_amount_modulo_bitwidth(i32 %x, i32 %y) { ; CHECK-LABEL: @fshl_constant_shift_amount_modulo_bitwidth( -; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.fshl.i32(i32 [[X:%.*]], i32 [[Y:%.*]], i32 33) +; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.fshl.i32(i32 [[X:%.*]], i32 [[Y:%.*]], i32 1) ; CHECK-NEXT: ret i32 [[R]] ; %r = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 33) @@ -319,7 +319,7 @@ define i33 @fshr_constant_shift_amount_modulo_bitwidth(i33 %x, i33 %y) { ; CHECK-LABEL: @fshr_constant_shift_amount_modulo_bitwidth( -; CHECK-NEXT: [[R:%.*]] = call i33 @llvm.fshr.i33(i33 [[X:%.*]], i33 [[Y:%.*]], i33 34) +; CHECK-NEXT: [[R:%.*]] = call i33 @llvm.fshr.i33(i33 [[X:%.*]], i33 [[Y:%.*]], i33 1) ; CHECK-NEXT: ret i33 [[R]] ; %r = call i33 @llvm.fshr.i33(i33 %x, i33 %y, i33 34) @@ -328,7 +328,7 @@ define <2 x i32> @fshr_constant_shift_amount_modulo_bitwidth_vec(<2 x i32> %x, <2 x i32> %y) { ; CHECK-LABEL: @fshr_constant_shift_amount_modulo_bitwidth_vec( -; CHECK-NEXT: [[R:%.*]] = call <2 x i32> @llvm.fshr.v2i32(<2 x i32> [[X:%.*]], <2 x i32> [[Y:%.*]], <2 x i32> ) +; CHECK-NEXT: [[R:%.*]] = call <2 x i32> @llvm.fshr.v2i32(<2 x i32> [[X:%.*]], <2 x i32> [[Y:%.*]], <2 x i32> ) ; CHECK-NEXT: ret <2 x i32> [[R]] ; %r = call <2 x i32> @llvm.fshr.v2i32(<2 x i32> %x, <2 x i32> %y, <2 x i32> ) @@ -337,17 +337,18 @@ define <2 x i31> @fshl_constant_shift_amount_modulo_bitwidth_vec(<2 x i31> %x, <2 x i31> %y) { ; CHECK-LABEL: @fshl_constant_shift_amount_modulo_bitwidth_vec( -; CHECK-NEXT: [[R:%.*]] = call <2 x i31> @llvm.fshl.v2i31(<2 x i31> [[X:%.*]], <2 x i31> [[Y:%.*]], <2 x i31> ) +; CHECK-NEXT: [[R:%.*]] = call <2 x i31> @llvm.fshl.v2i31(<2 x i31> [[X:%.*]], <2 x i31> [[Y:%.*]], <2 x i31> ) ; CHECK-NEXT: ret <2 x i31> [[R]] ; %r = call <2 x i31> @llvm.fshl.v2i31(<2 x i31> %x, <2 x i31> %y, <2 x i31> ) ret <2 x i31> %r } -; The shift modulo bitwidth is the same for all vector elements, but this is not simplified yet. +; The shift modulo bitwidth is the same for all vector elements. + define <2 x i31> @fshl_only_op1_demanded_vec_nonsplat(<2 x i31> %x, <2 x i31> %y) { ; CHECK-LABEL: @fshl_only_op1_demanded_vec_nonsplat( -; CHECK-NEXT: [[Z:%.*]] = call <2 x i31> @llvm.fshl.v2i31(<2 x i31> [[X:%.*]], <2 x i31> [[Y:%.*]], <2 x i31> ) +; CHECK-NEXT: [[Z:%.*]] = lshr <2 x i31> [[Y:%.*]], ; CHECK-NEXT: [[R:%.*]] = and <2 x i31> [[Z]], ; CHECK-NEXT: ret <2 x i31> [[R]] ;