File: 2006-10-07-ScalarSSEMiscompile.ll

package info (click to toggle)
llvm-3.1 3.1-1
  • links: PTS, VCS
  • area: main
  • in suites: wheezy
  • size: 80,224 kB
  • sloc: cpp: 491,014; asm: 110,971; ansic: 14,579; sh: 13,120; python: 6,152; ml: 4,719; makefile: 1,830; pascal: 1,553; perl: 874; xml: 283; lisp: 187; csh: 117; exp: 4
file content (15 lines) | stat: -rw-r--r-- 608 bytes parent folder | download | duplicates (7)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
; RUN: llc < %s -march=x86 -mattr=sse | grep movaps
; Test that the load is NOT folded into the intrinsic, which would zero the top
; elts of the loaded vector.

target datalayout = "e-p:32:32"
target triple = "i686-apple-darwin8.7.2"

define <4 x float> @test(<4 x float> %A, <4 x float>* %B) nounwind {
        %BV = load <4 x float>* %B              ; <<4 x float>> [#uses=1]
        %tmp28 = tail call <4 x float> @llvm.x86.sse.sub.ss( <4 x float> %A, <4 x float> %BV )       ; <<4 x float>> [#uses=1]
        ret <4 x float> %tmp28
}

declare <4 x float> @llvm.x86.sse.sub.ss(<4 x float>, <4 x float>)