From 7159a307b9a0a53f9a33e416d24b9308ef547d97 Mon Sep 17 00:00:00 2001 From: Benjamin Kramer Date: Fri, 10 May 2013 16:26:37 +0000 Subject: [PATCH] InstCombine: Don't claim to be able to evaluate any shl in a zexted type. The shift amount may be larger than the type leading to undefined behavior. Limit the transform to constant shift amounts. While there update the bits to clear in the result which may enable additional optimizations. PR15959. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181604 91177308-0d34-0410-b5e6-96231b3b80d8 --- .../InstCombine/InstCombineCasts.cpp | 12 +++++++- test/Transforms/InstCombine/cast.ll | 28 +++++++++++++++++++ 2 files changed, 39 insertions(+), 1 deletion(-) diff --git a/lib/Transforms/InstCombine/InstCombineCasts.cpp b/lib/Transforms/InstCombine/InstCombineCasts.cpp index 2ee1278d23d..361acdde81f 100644 --- a/lib/Transforms/InstCombine/InstCombineCasts.cpp +++ b/lib/Transforms/InstCombine/InstCombineCasts.cpp @@ -677,7 +677,6 @@ static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear) { case Instruction::Add: case Instruction::Sub: case Instruction::Mul: - case Instruction::Shl: if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear) || !CanEvaluateZExtd(I->getOperand(1), Ty, Tmp)) return false; @@ -701,6 +700,17 @@ static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear) { // Otherwise, we don't know how to analyze this BitsToClear case yet. return false; + case Instruction::Shl: + // We can promote shl(x, cst) if we can promote x. Since shl overwrites the + // upper bits we can reduce BitsToClear by the shift amount. + if (ConstantInt *Amt = dyn_cast(I->getOperand(1))) { + if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear)) + return false; + uint64_t ShiftAmt = Amt->getZExtValue(); + BitsToClear = ShiftAmt < BitsToClear ? BitsToClear - ShiftAmt : 0; + return true; + } + return false; case Instruction::LShr: // We can promote lshr(x, cst) if we can promote x. This requires the // ultimate 'and' to clear out the high zero bits we're clearing out though. diff --git a/test/Transforms/InstCombine/cast.ll b/test/Transforms/InstCombine/cast.ll index de738bb7c06..ff2c0a92898 100644 --- a/test/Transforms/InstCombine/cast.ll +++ b/test/Transforms/InstCombine/cast.ll @@ -898,3 +898,31 @@ define double @test81(double *%p, float %f) { %l = load double* %r ret double %l } + +define i64 @test82(i64 %A) nounwind { + %B = trunc i64 %A to i32 + %C = lshr i32 %B, 8 + %D = shl i32 %C, 9 + %E = zext i32 %D to i64 + ret i64 %E + +; CHECK: @test82 +; CHECK-NEXT: [[REG:%[0-9]*]] = shl i64 %A, 1 +; CHECK-NEXT: %E = and i64 [[REG]], 4294966784 +; CHECK-NEXT: ret i64 %E +} + +; PR15959 +define i64 @test83(i16 %a, i64 %k) { + %conv = sext i16 %a to i32 + %sub = add nsw i64 %k, -1 + %sh_prom = trunc i64 %sub to i32 + %shl = shl i32 %conv, %sh_prom + %sh_prom1 = zext i32 %shl to i64 + ret i64 %sh_prom1 + +; CHECK: @test83 +; CHECK: %sub = add nsw i64 %k, 4294967295 +; CHECK: %sh_prom = trunc i64 %sub to i32 +; CHECK: %shl = shl i32 %conv, %sh_prom +}