[x86] Remove some windows line endings that snuck into the tests here.

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

diff --git a/test/CodeGen/X86/vec_extract-avx.ll b/test/CodeGen/X86/vec_extract-avx.ll
index e5e79375fff81ea30a43679f7dc07b09c4901d4f..fbb84170dc83baedd422da4cda52967bf5cc131e 100644 (file)
--- a/test/CodeGen/X86/vec_extract-avx.ll
+++ b/test/CodeGen/X86/vec_extract-avx.ll
@@ -1,82 +1,82 @@
-target triple = "x86_64-unknown-unknown"\r
-\r
-; RUN: llc < %s -march=x86-64 -mattr=+avx | FileCheck %s\r
-\r
-; When extracting multiple consecutive elements from a larger\r
-; vector into a smaller one, do it efficiently. We should use\r
-; an EXTRACT_SUBVECTOR node internally rather than a bunch of\r
-; single element extractions. \r
-\r
-; Extracting the low elements only requires using the right kind of store.\r
-define void @low_v8f32_to_v4f32(<8 x float> %v, <4 x float>* %ptr) {\r
-  %ext0 = extractelement <8 x float> %v, i32 0\r
-  %ext1 = extractelement <8 x float> %v, i32 1\r
-  %ext2 = extractelement <8 x float> %v, i32 2\r
-  %ext3 = extractelement <8 x float> %v, i32 3\r
-  %ins0 = insertelement <4 x float> undef, float %ext0, i32 0\r
-  %ins1 = insertelement <4 x float> %ins0, float %ext1, i32 1\r
-  %ins2 = insertelement <4 x float> %ins1, float %ext2, i32 2\r
-  %ins3 = insertelement <4 x float> %ins2, float %ext3, i32 3\r
-  store <4 x float> %ins3, <4 x float>* %ptr, align 16\r
-  ret void\r
-\r
-; CHECK-LABEL: low_v8f32_to_v4f32\r
-; CHECK: vmovaps\r
-; CHECK-NEXT: vzeroupper\r
-; CHECK-NEXT: retq\r
-}\r
-\r
-; Extracting the high elements requires just one AVX instruction. \r
-define void @high_v8f32_to_v4f32(<8 x float> %v, <4 x float>* %ptr) {\r
-  %ext0 = extractelement <8 x float> %v, i32 4\r
-  %ext1 = extractelement <8 x float> %v, i32 5\r
-  %ext2 = extractelement <8 x float> %v, i32 6\r
-  %ext3 = extractelement <8 x float> %v, i32 7\r
-  %ins0 = insertelement <4 x float> undef, float %ext0, i32 0\r
-  %ins1 = insertelement <4 x float> %ins0, float %ext1, i32 1\r
-  %ins2 = insertelement <4 x float> %ins1, float %ext2, i32 2\r
-  %ins3 = insertelement <4 x float> %ins2, float %ext3, i32 3\r
-  store <4 x float> %ins3, <4 x float>* %ptr, align 16\r
-  ret void\r
-\r
-; CHECK-LABEL: high_v8f32_to_v4f32\r
-; CHECK: vextractf128\r
-; CHECK-NEXT: vzeroupper\r
-; CHECK-NEXT: retq\r
-}\r
-\r
-; Make sure element type doesn't alter the codegen. Note that\r
-; if we were actually using the vector in this function and\r
-; have AVX2, we should generate vextracti128 (the int version).\r
-define void @high_v8i32_to_v4i32(<8 x i32> %v, <4 x i32>* %ptr) {\r
-  %ext0 = extractelement <8 x i32> %v, i32 4\r
-  %ext1 = extractelement <8 x i32> %v, i32 5\r
-  %ext2 = extractelement <8 x i32> %v, i32 6\r
-  %ext3 = extractelement <8 x i32> %v, i32 7\r
-  %ins0 = insertelement <4 x i32> undef, i32 %ext0, i32 0\r
-  %ins1 = insertelement <4 x i32> %ins0, i32 %ext1, i32 1\r
-  %ins2 = insertelement <4 x i32> %ins1, i32 %ext2, i32 2\r
-  %ins3 = insertelement <4 x i32> %ins2, i32 %ext3, i32 3\r
-  store <4 x i32> %ins3, <4 x i32>* %ptr, align 16\r
-  ret void\r
-\r
-; CHECK-LABEL: high_v8i32_to_v4i32\r
-; CHECK: vextractf128\r
-; CHECK-NEXT: vzeroupper\r
-; CHECK-NEXT: retq\r
-}\r
-\r
-; Make sure that element size doesn't alter the codegen.\r
-define void @high_v4f64_to_v2f64(<4 x double> %v, <2 x double>* %ptr) {\r
-  %ext0 = extractelement <4 x double> %v, i32 2\r
-  %ext1 = extractelement <4 x double> %v, i32 3\r
-  %ins0 = insertelement <2 x double> undef, double %ext0, i32 0\r
-  %ins1 = insertelement <2 x double> %ins0, double %ext1, i32 1\r
-  store <2 x double> %ins1, <2 x double>* %ptr, align 16\r
-  ret void\r
-\r
-; CHECK-LABEL: high_v4f64_to_v2f64\r
-; CHECK: vextractf128\r
-; CHECK-NEXT: vzeroupper\r
-; CHECK-NEXT: retq\r
-}\r
+target triple = "x86_64-unknown-unknown"
+
+; RUN: llc < %s -march=x86-64 -mattr=+avx | FileCheck %s
+
+; When extracting multiple consecutive elements from a larger
+; vector into a smaller one, do it efficiently. We should use
+; an EXTRACT_SUBVECTOR node internally rather than a bunch of
+; single element extractions. 
+
+; Extracting the low elements only requires using the right kind of store.
+define void @low_v8f32_to_v4f32(<8 x float> %v, <4 x float>* %ptr) {
+  %ext0 = extractelement <8 x float> %v, i32 0
+  %ext1 = extractelement <8 x float> %v, i32 1
+  %ext2 = extractelement <8 x float> %v, i32 2
+  %ext3 = extractelement <8 x float> %v, i32 3
+  %ins0 = insertelement <4 x float> undef, float %ext0, i32 0
+  %ins1 = insertelement <4 x float> %ins0, float %ext1, i32 1
+  %ins2 = insertelement <4 x float> %ins1, float %ext2, i32 2
+  %ins3 = insertelement <4 x float> %ins2, float %ext3, i32 3
+  store <4 x float> %ins3, <4 x float>* %ptr, align 16
+  ret void
+
+; CHECK-LABEL: low_v8f32_to_v4f32
+; CHECK: vmovaps
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+}
+
+; Extracting the high elements requires just one AVX instruction. 
+define void @high_v8f32_to_v4f32(<8 x float> %v, <4 x float>* %ptr) {
+  %ext0 = extractelement <8 x float> %v, i32 4
+  %ext1 = extractelement <8 x float> %v, i32 5
+  %ext2 = extractelement <8 x float> %v, i32 6
+  %ext3 = extractelement <8 x float> %v, i32 7
+  %ins0 = insertelement <4 x float> undef, float %ext0, i32 0
+  %ins1 = insertelement <4 x float> %ins0, float %ext1, i32 1
+  %ins2 = insertelement <4 x float> %ins1, float %ext2, i32 2
+  %ins3 = insertelement <4 x float> %ins2, float %ext3, i32 3
+  store <4 x float> %ins3, <4 x float>* %ptr, align 16
+  ret void
+
+; CHECK-LABEL: high_v8f32_to_v4f32
+; CHECK: vextractf128
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+}
+
+; Make sure element type doesn't alter the codegen. Note that
+; if we were actually using the vector in this function and
+; have AVX2, we should generate vextracti128 (the int version).
+define void @high_v8i32_to_v4i32(<8 x i32> %v, <4 x i32>* %ptr) {
+  %ext0 = extractelement <8 x i32> %v, i32 4
+  %ext1 = extractelement <8 x i32> %v, i32 5
+  %ext2 = extractelement <8 x i32> %v, i32 6
+  %ext3 = extractelement <8 x i32> %v, i32 7
+  %ins0 = insertelement <4 x i32> undef, i32 %ext0, i32 0
+  %ins1 = insertelement <4 x i32> %ins0, i32 %ext1, i32 1
+  %ins2 = insertelement <4 x i32> %ins1, i32 %ext2, i32 2
+  %ins3 = insertelement <4 x i32> %ins2, i32 %ext3, i32 3
+  store <4 x i32> %ins3, <4 x i32>* %ptr, align 16
+  ret void
+
+; CHECK-LABEL: high_v8i32_to_v4i32
+; CHECK: vextractf128
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+}
+
+; Make sure that element size doesn't alter the codegen.
+define void @high_v4f64_to_v2f64(<4 x double> %v, <2 x double>* %ptr) {
+  %ext0 = extractelement <4 x double> %v, i32 2
+  %ext1 = extractelement <4 x double> %v, i32 3
+  %ins0 = insertelement <2 x double> undef, double %ext0, i32 0
+  %ins1 = insertelement <2 x double> %ins0, double %ext1, i32 1
+  store <2 x double> %ins1, <2 x double>* %ptr, align 16
+  ret void
+
+; CHECK-LABEL: high_v4f64_to_v2f64
+; CHECK: vextractf128
+; CHECK-NEXT: vzeroupper
+; CHECK-NEXT: retq
+}