File: 2006-10-07-ScalarSSEMiscompile.ll

package info (click to toggle)
llvm-2.7 2.7-6
  • links: PTS
  • area: main
  • in suites: squeeze
  • size: 59,992 kB
  • ctags: 49,499
  • sloc: cpp: 373,792; ansic: 16,885; sh: 12,614; asm: 6,809; ada: 3,083; ml: 2,942; python: 2,671; perl: 2,404; makefile: 1,691; pascal: 1,235; exp: 399; objc: 291; lisp: 184; csh: 117; xml: 38; f90: 36
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>)