5 files changed, 44 insertions, 4 deletions
diff --git a/test/Analysis/CostModel/X86/cast.ll b/test/Analysis/CostModel/X86/cast.ll
index b69b3bf..f3c1283 100644
--- a/test/Analysis/CostModel/X86/cast.ll
+++ b/test/Analysis/CostModel/X86/cast.ll
@@ -38,6 +38,10 @@ define i32 @zext_sext(<8 x i1> %in) {
   ;CHECK: cost of 9 {{.*}} sext
   %S = sext <8 x i1> %in to <8 x i32>
 
+  ;CHECK: cost of 1 {{.*}} zext
+  %A1 = zext <16 x i8> undef to <16 x i16>
+  ;CHECK: cost of 1 {{.*}} sext
+  %A2 = sext <16 x i8> undef to <16 x i16>
   ;CHECK: cost of 1 {{.*}} sext
   %A = sext <8 x i16> undef to <8 x i32>
   ;CHECK: cost of 1 {{.*}} zext
@@ -51,11 +55,13 @@ define i32 @zext_sext(<8 x i1> %in) {
 
   ;CHECK: cost of 1 {{.*}} zext
   %D = zext <4 x i32> undef to <4 x i64>
-  ;CHECK: cost of 1 {{.*}} trunc
 
+  ;CHECK: cost of 1 {{.*}} trunc
   %E = trunc <4 x i64> undef to <4 x i32>
   ;CHECK: cost of 1 {{.*}} trunc
   %F = trunc <8 x i32> undef to <8 x i16>
+  ;CHECK: cost of 2 {{.*}} trunc
+  %F1 = trunc <16 x i16> undef to <16 x i8>
 
   ;CHECK: cost of 3 {{.*}} trunc
   %G = trunc <8 x i64> undef to <8 x i32>
diff --git a/test/CodeGen/X86/avx-sext.ll b/test/CodeGen/X86/avx-sext.ll
index b9c7000..fb2287f 100644
--- a/test/CodeGen/X86/avx-sext.ll
+++ b/test/CodeGen/X86/avx-sext.ll
@@ -154,6 +154,17 @@ define <4 x i64> @sext_4i1_to_4i64(<4 x i1> %mask) {
   ret <4 x i64> %extmask
 }
 
+; AVX-LABEL: sext_16i8_to_16i16
+; AVX: vpmovsxbw
+; AVX: vmovhlps
+; AVX: vpmovsxbw
+; AVX: ret
+define <16 x i16> @sext_16i8_to_16i16(<16 x i8> *%ptr) {
+ %X = load <16 x i8>* %ptr
+ %Y = sext <16 x i8> %X to <16 x i16>
+ ret <16 x i16> %Y
+}
+
 ; AVX: sext_4i8_to_4i64
 ; AVX: vpslld  $24
 ; AVX: vpsrad  $24
diff --git a/test/CodeGen/X86/avx-trunc.ll b/test/CodeGen/X86/avx-trunc.ll
index d007736..58d0a35 100644
--- a/test/CodeGen/X86/avx-trunc.ll
+++ b/test/CodeGen/X86/avx-trunc.ll
@@ -12,4 +12,9 @@ define <8 x i16> @trunc_32_16(<8 x i32> %A) nounwind uwtable readnone ssp{
   %B = trunc <8 x i32> %A to <8 x i16>
   ret <8 x i16>%B
 }
-
+define <16 x i8> @trunc_16_8(<16 x i16> %A) nounwind uwtable readnone ssp{
+; CHECK-LABEL: trunc_16_8
+; CHECK: pshufb
+  %B = trunc <16 x i16> %A to <16 x i8>
+  ret <16 x i8> %B
+}
diff --git a/test/CodeGen/X86/avx2-conversions.ll b/test/CodeGen/X86/avx2-conversions.ll
index 0143f18..f49718e 100644
--- a/test/CodeGen/X86/avx2-conversions.ll
+++ b/test/CodeGen/X86/avx2-conversions.ll
@@ -72,6 +72,25 @@ define <16 x i16> @zext_16i8_16i16(<16 x i8> %z) {
   ret <16 x i16> %t
 }
 
+; CHECK-LABEL: sext_16i8_16i16:
+; CHECK: vpmovsxbw
+; CHECK-NOT: vinsert
+; CHECK: ret
+define <16 x i16> @sext_16i8_16i16(<16 x i8> %z) {
+  %t = sext <16 x i8> %z to <16 x i16>
+  ret <16 x i16> %t
+}
+
+; CHECK-LABEL: trunc_16i16_16i8:
+; CHECK: vpshufb
+; CHECK: vpshufb
+; CHECK: vpor
+; CHECK: ret
+define <16 x i8> @trunc_16i16_16i8(<16 x i16> %z) {
+  %t = trunc <16 x i16> %z to <16 x i8>
+  ret <16 x i8> %t
+}
+
 ; CHECK: load_sext_test1
 ; CHECK: vpmovsxdq (%r{{[^,]*}}), %ymm{{.*}}
 ; CHECK: ret 
diff --git a/test/CodeGen/X86/pmovsx-inreg.ll b/test/CodeGen/X86/pmovsx-inreg.ll
index d30d7d0..07979f6 100644
--- a/test/CodeGen/X86/pmovsx-inreg.ll
+++ b/test/CodeGen/X86/pmovsx-inreg.ll
@@ -86,8 +86,7 @@ define void @test6(<16 x i8>* %in, <16 x i16>* %out) nounwind {
   ret void
 
 ; AVX2-LABEL: test6:
-; FIXME: v16i8 -> v16i16 is scalarized.
-; AVX2-NOT: pmovsx
+; AVX2: vpmovsxbw
 }
 
 define void @test7(<2 x i16>* %in, <2 x i64>* %out) nounwind {