diff options
| author | Sanjay Patel <spatel@rotateright.com> | 2018-08-09 18:42:12 +0000 |
|---|---|---|
| committer | Sanjay Patel <spatel@rotateright.com> | 2018-08-09 18:42:12 +0000 |
| commit | 55accd7dd38c2e18ba16d63cad232f447b4c8d4c (patch) | |
| tree | d70c34563d967bfed66026ee89d21e38de3ea1a0 | |
| parent | 89005c33c3e5fd1877d83477c565278ceac88c4d (diff) | |
| download | bcm5719-llvm-55accd7dd38c2e18ba16d63cad232f447b4c8d4c.tar.gz bcm5719-llvm-55accd7dd38c2e18ba16d63cad232f447b4c8d4c.zip | |
[InstCombine] allow fsub+fmul FMF folds for vectors
llvm-svn: 339368
| -rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp | 11 | ||||
| -rw-r--r-- | llvm/test/Transforms/InstCombine/fast-math.ll | 14 |
2 files changed, 17 insertions, 8 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp index 16e3d742ce0..e3391345ac0 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp @@ -1914,6 +1914,17 @@ Instruction *InstCombiner::visitFSub(BinaryOperator &I) { if (match(Op1, m_c_FAdd(m_Specific(Op0), m_Value(X)))) return BinaryOperator::CreateFNegFMF(X, &I); + // (X * C) - X --> X * (C - 1.0) + if (match(Op0, m_FMul(m_Specific(Op1), m_Constant(C)))) { + Constant *CSubOne = ConstantExpr::getFSub(C, ConstantFP::get(Ty, 1.0)); + return BinaryOperator::CreateFMulFMF(Op1, CSubOne, &I); + } + // X - (X * C) --> X * (1.0 - C) + if (match(Op1, m_FMul(m_Specific(Op0), m_Constant(C)))) { + Constant *OneSubC = ConstantExpr::getFSub(ConstantFP::get(Ty, 1.0), C); + return BinaryOperator::CreateFMulFMF(Op0, OneSubC, &I); + } + // TODO: This performs reassociative folds for FP ops. Some fraction of the // functionality has been subsumed by simple pattern matching here and in // InstSimplify. We should let a dedicated reassociation pass handle more diff --git a/llvm/test/Transforms/InstCombine/fast-math.ll b/llvm/test/Transforms/InstCombine/fast-math.ll index 7b5b34ed612..8a0766d1756 100644 --- a/llvm/test/Transforms/InstCombine/fast-math.ll +++ b/llvm/test/Transforms/InstCombine/fast-math.ll @@ -442,8 +442,8 @@ define double @fail2(double %f1, double %f2) { define float @fsub_op0_fmul_const(float %x) { ; CHECK-LABEL: @fsub_op0_fmul_const( -; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz float [[X:%.*]], 6.000000e+00 -; CHECK-NEXT: ret float [[TMP1]] +; CHECK-NEXT: [[SUB:%.*]] = fmul reassoc nsz float [[X:%.*]], 6.000000e+00 +; CHECK-NEXT: ret float [[SUB]] ; %mul = fmul float %x, 7.0 %sub = fsub reassoc nsz float %mul, %x @@ -454,8 +454,7 @@ define float @fsub_op0_fmul_const(float %x) { define <2 x float> @fsub_op0_fmul_const_vec(<2 x float> %x) { ; CHECK-LABEL: @fsub_op0_fmul_const_vec( -; CHECK-NEXT: [[MUL:%.*]] = fmul <2 x float> [[X:%.*]], <float 7.000000e+00, float -4.200000e+01> -; CHECK-NEXT: [[SUB:%.*]] = fsub reassoc nsz <2 x float> [[MUL]], [[X]] +; CHECK-NEXT: [[SUB:%.*]] = fmul reassoc nsz <2 x float> [[X:%.*]], <float 6.000000e+00, float -4.300000e+01> ; CHECK-NEXT: ret <2 x float> [[SUB]] ; %mul = fmul <2 x float> %x, <float 7.0, float -42.0> @@ -467,8 +466,8 @@ define <2 x float> @fsub_op0_fmul_const_vec(<2 x float> %x) { define float @fsub_op1_fmul_const(float %x) { ; CHECK-LABEL: @fsub_op1_fmul_const( -; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz float [[X:%.*]], -6.000000e+00 -; CHECK-NEXT: ret float [[TMP1]] +; CHECK-NEXT: [[SUB:%.*]] = fmul reassoc nsz float [[X:%.*]], -6.000000e+00 +; CHECK-NEXT: ret float [[SUB]] ; %mul = fmul float %x, 7.0 %sub = fsub reassoc nsz float %x, %mul @@ -479,8 +478,7 @@ define float @fsub_op1_fmul_const(float %x) { define <2 x float> @fsub_op1_fmul_const_vec(<2 x float> %x) { ; CHECK-LABEL: @fsub_op1_fmul_const_vec( -; CHECK-NEXT: [[MUL:%.*]] = fmul <2 x float> [[X:%.*]], <float 7.000000e+00, float 0.000000e+00> -; CHECK-NEXT: [[SUB:%.*]] = fsub reassoc nsz <2 x float> [[X]], [[MUL]] +; CHECK-NEXT: [[SUB:%.*]] = fmul reassoc nsz <2 x float> [[X:%.*]], <float -6.000000e+00, float 1.000000e+00> ; CHECK-NEXT: ret <2 x float> [[SUB]] ; %mul = fmul <2 x float> %x, <float 7.0, float 0.0> |
