diff options
-rw-r--r-- | lib/Transforms/Vectorize/LoopVectorize.cpp | 22 | ||||
-rw-r--r-- | test/Transforms/LoopVectorize/scalar-select.ll | 37 |
2 files changed, 53 insertions, 6 deletions
diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp index 0fbf3e0831..59e8e5e5fd 100644 --- a/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -725,12 +725,22 @@ SingleBlockLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) { } case Instruction::Select: { // Widen selects. - // TODO: If the selector is loop invariant we can issue a select - // instruction with a scalar condition. - Value *A = getVectorValue(Inst->getOperand(0)); - Value *B = getVectorValue(Inst->getOperand(1)); - Value *C = getVectorValue(Inst->getOperand(2)); - WidenMap[Inst] = Builder.CreateSelect(A, B, C); + // If the selector is loop invariant we can create a select + // instruction with a scalar condition. Otherwise, use vector-select. + Value *Cond = Inst->getOperand(0); + bool InvariantCond = SE->isLoopInvariant(SE->getSCEV(Cond), Orig); + + // The condition can be loop invariant but still defined inside the + // loop. This means that we can't just use the original 'cond' value. + // We have to take the 'vectorized' value and pick the first lane. + // Instcombine will make this a no-op. + Cond = getVectorValue(Cond); + if (InvariantCond) + Cond = Builder.CreateExtractElement(Cond, Builder.getInt32(0)); + + Value *Op0 = getVectorValue(Inst->getOperand(1)); + Value *Op1 = getVectorValue(Inst->getOperand(2)); + WidenMap[Inst] = Builder.CreateSelect(Cond, Op0, Op1); break; } diff --git a/test/Transforms/LoopVectorize/scalar-select.ll b/test/Transforms/LoopVectorize/scalar-select.ll new file mode 100644 index 0000000000..8d5b6fd8af --- /dev/null +++ b/test/Transforms/LoopVectorize/scalar-select.ll @@ -0,0 +1,37 @@ +; RUN: opt < %s -loop-vectorize -dce -instcombine -licm -S | FileCheck %s + +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx10.8.0" + +@a = common global [2048 x i32] zeroinitializer, align 16 +@b = common global [2048 x i32] zeroinitializer, align 16 +@c = common global [2048 x i32] zeroinitializer, align 16 + +;CHECK: @example1 +;CHECK: load <4 x i32> +; make sure that we have a scalar condition and a vector operand. +;CHECK: select i1 %cond, <4 x i32> +;CHECK: store <4 x i32> +;CHECK: ret void +define void @example1(i1 %cond) nounwind uwtable ssp { + br label %1 + +; <label>:1 ; preds = %1, %0 + %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ] + %2 = getelementptr inbounds [2048 x i32]* @b, i64 0, i64 %indvars.iv + %3 = load i32* %2, align 4 + %4 = getelementptr inbounds [2048 x i32]* @c, i64 0, i64 %indvars.iv + %5 = load i32* %4, align 4 + %6 = add nsw i32 %5, %3 + %7 = getelementptr inbounds [2048 x i32]* @a, i64 0, i64 %indvars.iv + %sel = select i1 %cond, i32 %6, i32 zeroinitializer + store i32 %sel, i32* %7, align 4 + %indvars.iv.next = add i64 %indvars.iv, 1 + %lftr.wideiv = trunc i64 %indvars.iv.next to i32 + %exitcond = icmp eq i32 %lftr.wideiv, 256 + br i1 %exitcond, label %8, label %1 + +; <label>:8 ; preds = %1 + ret void +} + |