1; RUN: llc < %s -march=x86 -mtriple=i386-linux-gnu -mcpu=penryn -mattr=sse4.1 | FileCheck %s --check-prefix=X32
2; RUN: llc < %s -mtriple=x86_64-linux -mcpu=penryn -mattr=sse4.1 | FileCheck %s --check-prefix=X64
3; RUN: llc < %s -mtriple=x86_64-win32 -mcpu=penryn -mattr=sse4.1 | FileCheck %s --check-prefix=X64
4
5define i32 @test1() nounwind readonly {
6; X32-LABEL: test1:
7; X32:       # BB#0: # %entry
8; X32-NEXT:    movl %gs:196, %eax
9; X32-NEXT:    movl (%eax), %eax
10; X32-NEXT:    retl
11;
12; X64-LABEL: test1:
13; X64:       # BB#0: # %entry
14; X64-NEXT:    movq %gs:320, %rax
15; X64-NEXT:    movl (%rax), %eax
16; X64-NEXT:    retq
17entry:
18	%tmp = load i32*, i32* addrspace(256)* getelementptr (i32*, i32* addrspace(256)* inttoptr (i32 72 to i32* addrspace(256)*), i32 31)		; <i32*> [#uses=1]
19	%tmp1 = load i32, i32* %tmp		; <i32> [#uses=1]
20	ret i32 %tmp1
21}
22
23define i64 @test2(void (i8*)* addrspace(256)* %tmp8) nounwind {
24; X32-LABEL: test2:
25; X32:       # BB#0: # %entry
26; X32-NEXT:    subl $12, %esp
27; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
28; X32-NEXT:    calll *%gs:(%eax)
29; X32-NEXT:    xorl %eax, %eax
30; X32-NEXT:    xorl %edx, %edx
31; X32-NEXT:    addl $12, %esp
32; X32-NEXT:    retl
33;
34; X64-LABEL: test2:
35; X64:       # BB#0: # %entry
36; X64-NEXT:    {{(subq.*%rsp|pushq)}}
37; X64-NEXT:    callq *%gs:(%{{(rcx|rdi)}})
38; X64-NEXT:    xorl %eax, %eax
39; X64-NEXT:    {{(addq.*%rsp|popq)}}
40; X64-NEXT:    retq
41entry:
42  %tmp9 = load void (i8*)*, void (i8*)* addrspace(256)* %tmp8, align 8
43  tail call void %tmp9(i8* undef) nounwind optsize
44  ret i64 0
45}
46
47define <2 x i64> @pmovsxwd_1(i64 addrspace(256)* %p) nounwind readonly {
48; X32-LABEL: pmovsxwd_1:
49; X32:       # BB#0: # %entry
50; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
51; X32-NEXT:    pmovsxwd %gs:(%eax), %xmm0
52; X32-NEXT:    retl
53;
54; X64-LABEL: pmovsxwd_1:
55; X64:       # BB#0: # %entry
56; X64-NEXT:    pmovsxwd %gs:(%{{(rcx|rdi)}}), %xmm0
57; X64-NEXT:    retq
58entry:
59  %0 = load i64, i64 addrspace(256)* %p
60  %tmp2 = insertelement <2 x i64> zeroinitializer, i64 %0, i32 0
61  %1 = bitcast <2 x i64> %tmp2 to <8 x i16>
62  %2 = tail call <4 x i32> @llvm.x86.sse41.pmovsxwd(<8 x i16> %1) nounwind readnone
63  %3 = bitcast <4 x i32> %2 to <2 x i64>
64  ret <2 x i64> %3
65}
66
67; The two loads here both look identical to selection DAG, except for their
68; address spaces.  Make sure they aren't CSE'd.
69define i32 @test_no_cse() nounwind readonly {
70; X32-LABEL: test_no_cse:
71; X32:       # BB#0: # %entry
72; X32-NEXT:    movl %gs:196, %eax
73; X32-NEXT:    movl (%eax), %eax
74; X32-NEXT:    movl %fs:196, %ecx
75; X32-NEXT:    addl (%ecx), %eax
76; X32-NEXT:    retl
77;
78; X64-LABEL: test_no_cse:
79; X64:       # BB#0: # %entry
80; X64-NEXT:    movq %gs:320, %rax
81; X64-NEXT:    movl (%rax), %eax
82; X64-NEXT:    movq %fs:320, %rcx
83; X64-NEXT:    addl (%rcx), %eax
84; X64-NEXT:    retq
85entry:
86	%tmp = load i32*, i32* addrspace(256)* getelementptr (i32*, i32* addrspace(256)* inttoptr (i32 72 to i32* addrspace(256)*), i32 31)		; <i32*> [#uses=1]
87	%tmp1 = load i32, i32* %tmp		; <i32> [#uses=1]
88	%tmp2 = load i32*, i32* addrspace(257)* getelementptr (i32*, i32* addrspace(257)* inttoptr (i32 72 to i32* addrspace(257)*), i32 31)		; <i32*> [#uses=1]
89	%tmp3 = load i32, i32* %tmp2		; <i32> [#uses=1]
90	%tmp4 = add i32 %tmp1, %tmp3
91	ret i32 %tmp4
92}
93
94declare <4 x i32> @llvm.x86.sse41.pmovsxwd(<8 x i16>) nounwind readnone
95