1; RUN: llc -mtriple=x86_64-linux -mcpu=nehalem < %s | FileCheck %s --check-prefix=LIN
2; RUN: llc -mtriple=x86_64-win32 -mcpu=nehalem < %s | FileCheck %s --check-prefix=WIN
3; rdar://7398554
4
5; When doing vector gather-scatter index calculation with 32-bit indices,
6; bounce the vector off of cache rather than shuffling each individual
7; element out of the index vector.
8
9; CHECK: foo:
10; LIN: movaps	(%rsi), %xmm0
11; LIN: andps	(%rdx), %xmm0
12; LIN: movaps	%xmm0, -24(%rsp)
13; LIN: movslq	-24(%rsp), %[[REG1:r.+]]
14; LIN: movslq	-20(%rsp), %[[REG2:r.+]]
15; LIN: movslq	-16(%rsp), %[[REG3:r.+]]
16; LIN: movslq	-12(%rsp), %[[REG4:r.+]]
17; LIN: movsd	(%rdi,%[[REG1]],8), %xmm0
18; LIN: movhpd	(%rdi,%[[REG2]],8), %xmm0
19; LIN: movsd	(%rdi,%[[REG3]],8), %xmm1
20; LIN: movhpd	(%rdi,%[[REG4]],8), %xmm1
21
22; WIN: movaps	(%rdx), %xmm0
23; WIN: andps	(%r8), %xmm0
24; WIN: movaps	%xmm0, (%rsp)
25; WIN: movslq	(%rsp), %[[REG1:r.+]]
26; WIN: movslq	4(%rsp), %[[REG2:r.+]]
27; WIN: movslq	8(%rsp), %[[REG3:r.+]]
28; WIN: movslq	12(%rsp), %[[REG4:r.+]]
29; WIN: movsd	(%rcx,%[[REG1]],8), %xmm0
30; WIN: movhpd	(%rcx,%[[REG2]],8), %xmm0
31; WIN: movsd	(%rcx,%[[REG3]],8), %xmm1
32; WIN: movhpd	(%rcx,%[[REG4]],8), %xmm1
33
34define <4 x double> @foo(double* %p, <4 x i32>* %i, <4 x i32>* %h) nounwind {
35  %a = load <4 x i32>* %i
36  %b = load <4 x i32>* %h
37  %j = and <4 x i32> %a, %b
38  %d0 = extractelement <4 x i32> %j, i32 0
39  %d1 = extractelement <4 x i32> %j, i32 1
40  %d2 = extractelement <4 x i32> %j, i32 2
41  %d3 = extractelement <4 x i32> %j, i32 3
42  %q0 = getelementptr double* %p, i32 %d0
43  %q1 = getelementptr double* %p, i32 %d1
44  %q2 = getelementptr double* %p, i32 %d2
45  %q3 = getelementptr double* %p, i32 %d3
46  %r0 = load double* %q0
47  %r1 = load double* %q1
48  %r2 = load double* %q2
49  %r3 = load double* %q3
50  %v0 = insertelement <4 x double> undef, double %r0, i32 0
51  %v1 = insertelement <4 x double> %v0, double %r1, i32 1
52  %v2 = insertelement <4 x double> %v1, double %r2, i32 2
53  %v3 = insertelement <4 x double> %v2, double %r3, i32 3
54  ret <4 x double> %v3
55}
56