• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; Tests for SSE1 and below, without SSE2+.
2; RUN: llc < %s -mtriple=i386-unknown-unknown -march=x86 -mcpu=pentium3 -O3 | FileCheck %s
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -march=x86-64 -mattr=-sse2,+sse -O3 | FileCheck %s
4
5; PR7993
6;define <4 x i32> @test3(<4 x i16> %a) nounwind {
7;  %c = sext <4 x i16> %a to <4 x i32>             ; <<4 x i32>> [#uses=1]
8;  ret <4 x i32> %c
9;}
10
11; This should not emit shuffles to populate the top 2 elements of the 4-element
12; vector that this ends up returning.
13; rdar://8368414
14define <2 x float> @test4(<2 x float> %A, <2 x float> %B) nounwind {
15; CHECK-LABEL: test4:
16; CHECK:       # BB#0: # %entry
17; CHECK-NEXT:    movaps %xmm0, %xmm2
18; CHECK-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,1,2,3]
19; CHECK-NEXT:    addss %xmm1, %xmm0
20; CHECK-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1,2,3]
21; CHECK-NEXT:    subss %xmm1, %xmm2
22; CHECK-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
23; CHECK-NEXT:    ret
24entry:
25  %tmp7 = extractelement <2 x float> %A, i32 0
26  %tmp5 = extractelement <2 x float> %A, i32 1
27  %tmp3 = extractelement <2 x float> %B, i32 0
28  %tmp1 = extractelement <2 x float> %B, i32 1
29  %add.r = fadd float %tmp7, %tmp3
30  %add.i = fsub float %tmp5, %tmp1
31  %tmp11 = insertelement <2 x float> undef, float %add.r, i32 0
32  %tmp9 = insertelement <2 x float> %tmp11, float %add.i, i32 1
33  ret <2 x float> %tmp9
34}
35
36; We used to get stuck in type legalization for this example when lowering the
37; vselect. With SSE1 v4f32 is a legal type but v4i1 (or any vector integer type)
38; is not. We used to ping pong between splitting the vselect for the v4i
39; condition operand and widening the resulting vselect for the v4f32 result.
40; PR18036
41
42define <4 x float> @vselect(<4 x float>*%p, <4 x i32> %q) {
43; CHECK-LABEL: vselect:
44; CHECK:         ret
45entry:
46  %a1 = icmp eq <4 x i32> %q, zeroinitializer
47  %a14 = select <4 x i1> %a1, <4 x float> <float 1.000000e+00, float 2.000000e+00, float 3.000000e+00, float 4.000000e+0> , <4 x float> zeroinitializer
48  ret <4 x float> %a14
49}
50
51; v4i32 isn't legal for SSE1, but this should be cmpps.
52
53define <4 x float> @PR28044(<4 x float> %a0, <4 x float> %a1) nounwind {
54; CHECK-LABEL: PR28044:
55; CHECK:       # BB#0:
56; CHECK-NEXT:    cmpeqps %xmm1, %xmm0
57; CHECK-NEXT:    ret
58;
59  %cmp = fcmp oeq <4 x float> %a0, %a1
60  %sext = sext <4 x i1> %cmp to <4 x i32>
61  %res = bitcast <4 x i32> %sext to <4 x float>
62  ret <4 x float> %res
63}
64
65