Index: lib/Transforms/InstCombine/InstCombineSelect.cpp =================================================================== --- lib/Transforms/InstCombine/InstCombineSelect.cpp +++ lib/Transforms/InstCombine/InstCombineSelect.cpp @@ -912,6 +912,37 @@ return nullptr; } +/// If one of the operands is a sext/zext from i1 and the other is a constant, +/// we may be able to create an i1 select which can be further folded to +/// logical ops. +static Instruction *foldSelectExtConst(InstCombiner::BuilderTy &Builder, + SelectInst &SI, Instruction *EI, + ConstantInt *C, bool isExtTrueVal, + bool isSigned) { + Value *SmallVal = EI->getOperand(0); + Type *SmallType = SmallVal->getType(); + + if (SmallType->getIntegerBitWidth() != 1) + return nullptr; + + const APInt &CInt = C->getValue(); + + if (CInt != 0 && (isSigned || CInt != 1) && + (!isSigned || !CInt.isAllOnesValue())) + return nullptr; + + Value *SmallConst = ConstantInt::get(SmallType, CInt.trunc(1)); + Value *TrueVal = isExtTrueVal ? SmallVal : SmallConst; + Value *FalseVal = isExtTrueVal ? SmallConst : SmallVal; + Value *Select = Builder.CreateSelect(SI.getOperand(0), TrueVal, FalseVal, + "fold." + SI.getName()); + + if (isSigned) + return new SExtInst(Select, SI.getType()); + + return new ZExtInst(Select, SI.getType()); +} + Instruction *InstCombiner::visitSelectInst(SelectInst &SI) { Value *CondVal = SI.getCondition(); Value *TrueVal = SI.getTrueValue(); @@ -1098,6 +1129,35 @@ if (Instruction *IV = FoldSelectOpOp(SI, TI, FI)) return IV; + // (select C, (sext X), const) -> (sext (select C, X, const')) and + // variations thereof when extending from i1, as that allows further folding + // into logic ops. When the sext is from a larger type, we prefer to have it + // as an operand. + if (TI) { + if (ConstantInt *C = dyn_cast(FalseVal)) { + if (isa(TI)) + if (Instruction *IV = + foldSelectExtConst(*Builder, SI, TI, C, true, true)) + return IV; + if (isa(TI)) + if (Instruction *IV = + foldSelectExtConst(*Builder, SI, TI, C, true, false)) + return IV; + } + } + if (FI) { + if (ConstantInt *C = dyn_cast(TrueVal)) { + if (isa(FI)) + if (Instruction *IV = + foldSelectExtConst(*Builder, SI, FI, C, false, true)) + return IV; + if (isa(FI)) + if (Instruction *IV = + foldSelectExtConst(*Builder, SI, FI, C, false, false)) + return IV; + } + } + // See if we can fold the select into one of our operands. if (SelType->isIntOrIntVectorTy() || SelType->isFPOrFPVectorTy()) { if (Instruction *FoldI = FoldSelectIntoOp(SI, TrueVal, FalseVal)) Index: test/Transforms/InstCombine/select-bitext.ll =================================================================== --- /dev/null +++ test/Transforms/InstCombine/select-bitext.ll @@ -0,0 +1,168 @@ +; RUN: opt < %s -instcombine -S | FileCheck %s + +; CHECK-LABEL: @test_sext1 +; CHECK: and +; CHECK: sext +define i32 @test_sext1(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = sext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 %ccax, i32 0 + ret i32 %r +} + +; CHECK-LABEL: @test_sext2 +; CHECK: or +; CHECK: sext +define i32 @test_sext2(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = sext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 -1, i32 %ccax + ret i32 %r +} + +; CHECK-LABEL: @test_sext3 +; CHECK: icmp sgt +; CHECK: icmp slt +; CHECK: and +; CHECK: sext +define i32 @test_sext3(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = sext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 0, i32 %ccax + ret i32 %r +} + +; CHECK-LABEL: @test_sext4 +; CHECK: icmp sgt +; CHECK: icmp slt +; CHECK: or +; CHECK: sext +define i32 @test_sext4(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = sext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 %ccax, i32 -1 + ret i32 %r +} + +; CHECK-LABEL: @test_zext1 +; CHECK: and +; CHECK: zext +define i32 @test_zext1(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = zext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 %ccax, i32 0 + ret i32 %r +} + +; CHECK-LABEL: @test_zext2 +; CHECK: or +; CHECK: zext +define i32 @test_zext2(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = zext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 1, i32 %ccax + ret i32 %r +} + +; CHECK-LABEL: @test_zext3 +; CHECK: icmp sgt +; CHECK: icmp slt +; CHECK: and +; CHECK: zext +define i32 @test_zext3(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = zext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 0, i32 %ccax + ret i32 %r +} + +; CHECK-LABEL: @test_zext4 +; CHECK: icmp sgt +; CHECK: icmp slt +; CHECK: or +; CHECK: zext +define i32 @test_zext4(i32 %a, i32 %b) { + %cca = icmp sgt i32 %a, 0 + %ccax = zext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %r = select i1 %ccb, i32 %ccax, i32 1 + ret i32 %r +} + +; CHECK-LABEL: @test_negative_sext +; CHECK: sext +; CHECK: select +define i32 @test_negative_sext(i1 %a, i1 %cc) { + %a.ext = sext i1 %a to i32 + %r = select i1 %cc, i32 %a.ext, i32 1 + ret i32 %r +} + +; CHECK-LABEL: @test_negative_zext +; CHECK: zext +; CHECK: select +define i32 @test_negative_zext(i1 %a, i1 %cc) { + %a.ext = zext i1 %a to i32 + %r = select i1 %cc, i32 %a.ext, i32 -1 + ret i32 %r +} + +; CHECK-LABEL: @test_bits_sext +; CHECK: sext +; CHECK: select +define i32 @test_bits_sext(i8 %a, i1 %cc) { + %a.ext = sext i8 %a to i32 + %r = select i1 %cc, i32 %a.ext, i32 -128 + ret i32 %r +} + +; CHECK-LABEL: @test_bits_zext +; CHECK: zext +; CHECK: select +define i32 @test_bits_zext(i8 %a, i1 %cc) { + %a.ext = zext i8 %a to i32 + %r = select i1 %cc, i32 %a.ext, i32 255 + ret i32 %r +} + +; CHECK-LABEL: @test_op_op +; CHECK: select +; CHECK: sext +define i32 @test_op_op(i32 %a, i32 %b, i32 %c) { + %cca = icmp sgt i32 %a, 0 + %ccax = sext i1 %cca to i32 + %ccb = icmp sgt i32 %b, 0 + %ccbx = sext i1 %ccb to i32 + %ccc = icmp sgt i32 %c, 0 + %r = select i1 %ccc, i32 %ccax, i32 %ccbx + ret i32 %r +} + +; CHECK-LABEL: @test_no_vectors1 +; CHECK: sext +; CHECK: select +define <2 x i32> @test_no_vectors1(<2 x i32> %a, <2 x i32> %b, <2 x i32> %c) { + %cca = icmp sgt <2 x i32> %a, %b + %ccax = sext <2 x i1> %cca to <2 x i32> + %ccb = icmp sgt <2 x i32> %b, %c + %r = select <2 x i1> %ccb, <2 x i32> %ccax, <2 x i32> + ret <2 x i32> %r +} + +; CHECK-LABEL: @test_no_vectors2 +; CHECK: sext +; CHECK: select +define <2 x i32> @test_no_vectors2(<2 x i32> %a, <2 x i32> %b, i32 %c) { + %cca = icmp sgt <2 x i32> %a, %b + %ccax = sext <2 x i1> %cca to <2 x i32> + %ccb = icmp sgt i32 %c, 0 + %r = select i1 %ccb, <2 x i32> %ccax, <2 x i32> + ret <2 x i32> %r +}