diff options
author | Nadav Rotem <nrotem@apple.com> | 2013-06-26 16:54:53 +0000 |
---|---|---|
committer | Nadav Rotem <nrotem@apple.com> | 2013-06-26 16:54:53 +0000 |
commit | 29acf7e03af9b5524daa1e7523e0296cc766ff24 (patch) | |
tree | 97c116b8abc4ca1848f3fa2d4a0c5580042f196c | |
parent | c19bd321362166805194cbaf170e06a4790d2da9 (diff) | |
download | external_llvm-29acf7e03af9b5524daa1e7523e0296cc766ff24.zip external_llvm-29acf7e03af9b5524daa1e7523e0296cc766ff24.tar.gz external_llvm-29acf7e03af9b5524daa1e7523e0296cc766ff24.tar.bz2 |
Do not add cse-ed instructions into the visited map because we dont want to consider them as a candidate for replacement of instructions to be visited.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184966 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r-- | lib/Transforms/Vectorize/SLPVectorizer.cpp | 14 | ||||
-rw-r--r-- | test/Transforms/SLPVectorizer/X86/cse.ll | 84 |
2 files changed, 93 insertions, 5 deletions
diff --git a/lib/Transforms/Vectorize/SLPVectorizer.cpp b/lib/Transforms/Vectorize/SLPVectorizer.cpp index 9c8244b..bb37994 100644 --- a/lib/Transforms/Vectorize/SLPVectorizer.cpp +++ b/lib/Transforms/Vectorize/SLPVectorizer.cpp @@ -1258,6 +1258,8 @@ Value *FuncSLP::vectorizeArith(ArrayRef<Value *> Operands) { for (unsigned i = 0, e = Operands.size(); i != e; ++i) { Value *S = Builder.CreateExtractElement(Vec, Builder.getInt32(i)); Operands[i]->replaceAllUsesWith(S); + Instruction *I = cast<Instruction>(Operands[i]); + I->eraseFromParent(); } return Vec; @@ -1280,7 +1282,7 @@ void FuncSLP::optimizeGatherSequence() { // Check if it has a preheader. BasicBlock *PreHeader = L->getLoopPreheader(); if (!PreHeader) - return; + continue; // If the vector or the element that we insert into it are // instructions that are defined in this basic block then we can't @@ -1310,17 +1312,19 @@ void FuncSLP::optimizeGatherSequence() { if (!Insert || !GatherSeq.count(Insert)) continue; - // Check if we can replace this instruction with any of the - // visited instructions. + // Check if we can replace this instruction with any of the + // visited instructions. for (SmallPtrSet<Instruction*, 16>::iterator v = Visited.begin(), ve = Visited.end(); v != ve; ++v) { if (Insert->isIdenticalTo(*v) && - DT->dominates((*v)->getParent(), Insert->getParent())) { + DT->dominates((*v)->getParent(), Insert->getParent())) { Insert->replaceAllUsesWith(*v); + Insert = 0; break; } } - Visited.insert(Insert); + if (Insert) + Visited.insert(Insert); } } } diff --git a/test/Transforms/SLPVectorizer/X86/cse.ll b/test/Transforms/SLPVectorizer/X86/cse.ll index d286798..1c1e4b5 100644 --- a/test/Transforms/SLPVectorizer/X86/cse.ll +++ b/test/Transforms/SLPVectorizer/X86/cse.ll @@ -134,3 +134,87 @@ define i32 @test2(double* nocapture %G, i32 %k) { ret i32 undef } + +;int foo(double *A, int n) { +; A[0] = A[0] * 7.9 * n + 6.0; +; A[1] = A[1] * 7.9 * n + 6.0; +; A[2] = A[2] * 7.9 * n + 6.0; +; A[3] = A[3] * 7.9 * n + 6.0; +;} +;CHECK: @foo4 +;CHECK: insertelement <2 x double> +;CHECK: insertelement <2 x double> +;CHECK-NOT: insertelement <2 x double> +;CHECK: ret +define i32 @foo4(double* nocapture %A, i32 %n) { +entry: + %0 = load double* %A, align 8 + %mul = fmul double %0, 7.900000e+00 + %conv = sitofp i32 %n to double + %mul1 = fmul double %conv, %mul + %add = fadd double %mul1, 6.000000e+00 + store double %add, double* %A, align 8 + %arrayidx3 = getelementptr inbounds double* %A, i64 1 + %1 = load double* %arrayidx3, align 8 + %mul4 = fmul double %1, 7.900000e+00 + %mul6 = fmul double %conv, %mul4 + %add7 = fadd double %mul6, 6.000000e+00 + store double %add7, double* %arrayidx3, align 8 + %arrayidx9 = getelementptr inbounds double* %A, i64 2 + %2 = load double* %arrayidx9, align 8 + %mul10 = fmul double %2, 7.900000e+00 + %mul12 = fmul double %conv, %mul10 + %add13 = fadd double %mul12, 6.000000e+00 + store double %add13, double* %arrayidx9, align 8 + %arrayidx15 = getelementptr inbounds double* %A, i64 3 + %3 = load double* %arrayidx15, align 8 + %mul16 = fmul double %3, 7.900000e+00 + %mul18 = fmul double %conv, %mul16 + %add19 = fadd double %mul18, 6.000000e+00 + store double %add19, double* %arrayidx15, align 8 + ret i32 undef +} + +;int partial_mrg(double *A, int n) { +; A[0] = A[0] * n; +; A[1] = A[1] * n; +; if (n < 4) return 0; +; A[2] = A[2] * n; +; A[3] = A[3] * (n+4); +;} +;CHECK: @partial_mrg +;CHECK: insertelement <2 x double> +;CHECK: insertelement <2 x double> +;CHECK: insertelement <2 x double> +;CHECK-NOT: insertelement <2 x double> +;CHECK: ret +define i32 @partial_mrg(double* nocapture %A, i32 %n) { +entry: + %0 = load double* %A, align 8 + %conv = sitofp i32 %n to double + %mul = fmul double %conv, %0 + store double %mul, double* %A, align 8 + %arrayidx2 = getelementptr inbounds double* %A, i64 1 + %1 = load double* %arrayidx2, align 8 + %mul4 = fmul double %conv, %1 + store double %mul4, double* %arrayidx2, align 8 + %cmp = icmp slt i32 %n, 4 + br i1 %cmp, label %return, label %if.end + +if.end: ; preds = %entry + %arrayidx7 = getelementptr inbounds double* %A, i64 2 + %2 = load double* %arrayidx7, align 8 + %mul9 = fmul double %conv, %2 + store double %mul9, double* %arrayidx7, align 8 + %arrayidx11 = getelementptr inbounds double* %A, i64 3 + %3 = load double* %arrayidx11, align 8 + %add = add nsw i32 %n, 4 + %conv12 = sitofp i32 %add to double + %mul13 = fmul double %conv12, %3 + store double %mul13, double* %arrayidx11, align 8 + br label %return + +return: ; preds = %entry, %if.end + ret i32 0 +} + |