[x86] Rename avx-{s,z}ext.ll to vector-{s,z}ext.ll.

[oota-llvm.git] / test / CodeGen / X86 / movgs.ll

diff --git a/test/CodeGen/X86/movgs.ll b/test/CodeGen/X86/movgs.ll
index 00190e802fc94e1c74fcf38ee268c0965dcf915b..71b0723c429e0a17e32c16d442a18d2e817da2c0 100644 (file)
--- a/test/CodeGen/X86/movgs.ll
+++ b/test/CodeGen/X86/movgs.ll
@@ -1,5 +1,6 @@
-; RUN: llc < %s -march=x86 -mattr=sse41 | FileCheck %s --check-prefix=X32
-; RUN: llc < %s -march=x86-64 -mattr=sse41 | FileCheck %s --check-prefix=X64
+; RUN: llc < %s -march=x86 -mtriple=i386-linux-gnu -mcpu=penryn -mattr=sse4.1 | FileCheck %s --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-linux -mcpu=penryn -mattr=sse4.1 | FileCheck %s --check-prefix=X64
+; RUN: llc < %s -mtriple=x86_64-win32 -mcpu=penryn -mattr=sse4.1 | FileCheck %s --check-prefix=X64
 
 define i32 @test1() nounwind readonly {
 entry:
@@ -7,12 +8,12 @@ entry:
        %tmp1 = load i32* %tmp          ; <i32> [#uses=1]
        ret i32 %tmp1
 }
-; X32: test1:
+; X32-LABEL: test1:
 ; X32:         movl    %gs:196, %eax
 ; X32:         movl    (%eax), %eax
 ; X32:         ret
 
-; X64: test1:
+; X64-LABEL: test1:
 ; X64:         movq    %gs:320, %rax
 ; X64:         movl    (%rax), %eax
 ; X64:         ret
@@ -25,12 +26,12 @@ entry:
 }
 
 ; rdar://8453210
-; X32: test2:
+; X32-LABEL: test2:
 ; X32: movl    {{.*}}(%esp), %eax
 ; X32: calll   *%gs:(%eax)
 
-; X64: test2:
-; X64: callq   *%gs:(%rdi)
+; X64-LABEL: test2:
+; X64: callq   *%gs:([[A0:%rdi|%rcx]])
 
 
 
@@ -44,14 +45,30 @@ entry:
   %3 = bitcast <4 x i32> %2 to <2 x i64>
   ret <2 x i64> %3
   
-; X32: pmovsxwd_1:
+; X32-LABEL: pmovsxwd_1:
 ; X32:         movl    4(%esp), %eax
 ; X32:         pmovsxwd        %gs:(%eax), %xmm0
 ; X32:         ret
 
-; X64: pmovsxwd_1:
-; X64: pmovsxwd        %gs:(%rdi), %xmm0
+; X64-LABEL: pmovsxwd_1:
+; X64: pmovsxwd        %gs:([[A0]]), %xmm0
 ; X64: ret
 }
 
+; The two loads here both look identical to selection DAG, except for their
+; address spaces.  Make sure they aren't CSE'd.
+define i32 @test_no_cse() nounwind readonly {
+entry:
+       %tmp = load i32* addrspace(256)* getelementptr (i32* addrspace(256)* inttoptr (i32 72 to i32* addrspace(256)*), i32 31)         ; <i32*> [#uses=1]
+       %tmp1 = load i32* %tmp          ; <i32> [#uses=1]
+       %tmp2 = load i32* addrspace(257)* getelementptr (i32* addrspace(257)* inttoptr (i32 72 to i32* addrspace(257)*), i32 31)                ; <i32*> [#uses=1]
+       %tmp3 = load i32* %tmp2         ; <i32> [#uses=1]
+       %tmp4 = add i32 %tmp1, %tmp3
+       ret i32 %tmp4
+}
+; X32-LABEL: test_no_cse:
+; X32:         movl    %gs:196
+; X32:         movl    %fs:196
+; X32:         ret
+
 declare <4 x i32> @llvm.x86.sse41.pmovsxwd(<8 x i16>) nounwind readnone