From 8611d4449a77ca05e808823bc966573a85da00cb Mon Sep 17 00:00:00 2001 From: Benjamin Kramer Date: Thu, 28 Feb 2013 19:09:33 +0000 Subject: Cost model support for lowered math builtins. We make the cost for calling libm functions extremely high as emitting the calls is expensive and causes spills (on x86) so performance suffers. We still vectorize important calls like ceilf and friends on SSE4.1. and fabs. Differential Revision: http://llvm-reviews.chandlerc.com/D466 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176287 91177308-0d34-0410-b5e6-96231b3b80d8 --- test/Analysis/CostModel/X86/intrinsic-cost.ll | 32 +++++++++++++++++++++++++++ 1 file changed, 32 insertions(+) create mode 100644 test/Analysis/CostModel/X86/intrinsic-cost.ll (limited to 'test/Analysis/CostModel') diff --git a/test/Analysis/CostModel/X86/intrinsic-cost.ll b/test/Analysis/CostModel/X86/intrinsic-cost.ll new file mode 100644 index 0000000..e235a36 --- /dev/null +++ b/test/Analysis/CostModel/X86/intrinsic-cost.ll @@ -0,0 +1,32 @@ +; RUN: opt -S -mtriple=x86_64-apple-darwin -mcpu=core2 -cost-model -analyze < %s | FileCheck %s -check-prefix=CORE2 +; RUN: opt -S -mtriple=x86_64-apple-darwin -mcpu=corei7 -cost-model -analyze < %s | FileCheck %s -check-prefix=COREI7 + +; If SSE4.1 roundps instruction is available it is cheap to lower, otherwise +; it'll be scalarized into calls which are expensive. +define void @test1(float* nocapture %f) nounwind { +vector.ph: + br label %vector.body + +vector.body: ; preds = %vector.body, %vector.ph + %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ] + %0 = getelementptr inbounds float* %f, i64 %index + %1 = bitcast float* %0 to <4 x float>* + %wide.load = load <4 x float>* %1, align 4 + %2 = call <4 x float> @llvm.ceil.v4f32(<4 x float> %wide.load) + store <4 x float> %2, <4 x float>* %1, align 4 + %index.next = add i64 %index, 4 + %3 = icmp eq i64 %index.next, 1024 + br i1 %3, label %for.end, label %vector.body + +for.end: ; preds = %vector.body + ret void + +; CORE2: Printing analysis 'Cost Model Analysis' for function 'test1': +; CORE2: Cost Model: Found an estimated cost of 400 for instruction: %2 = call <4 x float> @llvm.ceil.v4f32(<4 x float> %wide.load) + +; COREI7: Printing analysis 'Cost Model Analysis' for function 'test1': +; COREI7: Cost Model: Found an estimated cost of 1 for instruction: %2 = call <4 x float> @llvm.ceil.v4f32(<4 x float> %wide.load) + +} + +declare <4 x float> @llvm.ceil.v4f32(<4 x float>) nounwind readnone -- cgit v1.1