Vendor import of llvm trunk r300422:vendor/llvm/llvm-trunk-r300422

https://llvm.org/svn/llvm-project/llvm/trunk@300422

1 files changed, 32 insertions, 32 deletions

diff --git a/test/CodeGen/AMDGPU/global-extload-i16.ll b/test/CodeGen/AMDGPU/global-extload-i16.ll
index 2c7c02de1673..19e592f50bea 100644
--- a/test/CodeGen/AMDGPU/global-extload-i16.ll
+++ b/test/CodeGen/AMDGPU/global-extload-i16.ll
@@ -7,7 +7,7 @@
 ; SI: buffer_load_ushort
 ; SI: buffer_store_dword
 ; SI: s_endpgm
-define void @zextload_global_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
+define amdgpu_kernel void @zextload_global_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
   %a = load i16, i16 addrspace(1)* %in
   %ext = zext i16 %a to i32
   store i32 %ext, i32 addrspace(1)* %out
@@ -18,7 +18,7 @@ define void @zextload_global_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)
 ; SI: buffer_load_sshort
 ; SI: buffer_store_dword
 ; SI: s_endpgm
-define void @sextload_global_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
+define amdgpu_kernel void @sextload_global_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
   %a = load i16, i16 addrspace(1)* %in
   %ext = sext i16 %a to i32
   store i32 %ext, i32 addrspace(1)* %out
@@ -28,7 +28,7 @@ define void @sextload_global_i16_to_i32(i32 addrspace(1)* %out, i16 addrspace(1)
 ; FUNC-LABEL: {{^}}zextload_global_v1i16_to_v1i32:
 ; SI: buffer_load_ushort
 ; SI: s_endpgm
-define void @zextload_global_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <1 x i16>, <1 x i16> addrspace(1)* %in
   %ext = zext <1 x i16> %load to <1 x i32>
   store <1 x i32> %ext, <1 x i32> addrspace(1)* %out
@@ -38,7 +38,7 @@ define void @zextload_global_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i
 ; FUNC-LABEL: {{^}}sextload_global_v1i16_to_v1i32:
 ; SI: buffer_load_sshort
 ; SI: s_endpgm
-define void @sextload_global_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <1 x i16>, <1 x i16> addrspace(1)* %in
   %ext = sext <1 x i16> %load to <1 x i32>
   store <1 x i32> %ext, <1 x i32> addrspace(1)* %out
@@ -47,7 +47,7 @@ define void @sextload_global_v1i16_to_v1i32(<1 x i32> addrspace(1)* %out, <1 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v2i16_to_v2i32:
 ; SI: s_endpgm
-define void @zextload_global_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <2 x i16>, <2 x i16> addrspace(1)* %in
   %ext = zext <2 x i16> %load to <2 x i32>
   store <2 x i32> %ext, <2 x i32> addrspace(1)* %out
@@ -56,7 +56,7 @@ define void @zextload_global_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i
 
 ; FUNC-LABEL: {{^}}sextload_global_v2i16_to_v2i32:
 ; SI: s_endpgm
-define void @sextload_global_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <2 x i16>, <2 x i16> addrspace(1)* %in
   %ext = sext <2 x i16> %load to <2 x i32>
   store <2 x i32> %ext, <2 x i32> addrspace(1)* %out
@@ -65,7 +65,7 @@ define void @sextload_global_v2i16_to_v2i32(<2 x i32> addrspace(1)* %out, <2 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v4i16_to_v4i32:
 ; SI: s_endpgm
-define void @zextload_global_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <4 x i16>, <4 x i16> addrspace(1)* %in
   %ext = zext <4 x i16> %load to <4 x i32>
   store <4 x i32> %ext, <4 x i32> addrspace(1)* %out
@@ -74,7 +74,7 @@ define void @zextload_global_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i
 
 ; FUNC-LABEL: {{^}}sextload_global_v4i16_to_v4i32:
 ; SI: s_endpgm
-define void @sextload_global_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <4 x i16>, <4 x i16> addrspace(1)* %in
   %ext = sext <4 x i16> %load to <4 x i32>
   store <4 x i32> %ext, <4 x i32> addrspace(1)* %out
@@ -83,7 +83,7 @@ define void @sextload_global_v4i16_to_v4i32(<4 x i32> addrspace(1)* %out, <4 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v8i16_to_v8i32:
 ; SI: s_endpgm
-define void @zextload_global_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <8 x i16>, <8 x i16> addrspace(1)* %in
   %ext = zext <8 x i16> %load to <8 x i32>
   store <8 x i32> %ext, <8 x i32> addrspace(1)* %out
@@ -92,7 +92,7 @@ define void @zextload_global_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i
 
 ; FUNC-LABEL: {{^}}sextload_global_v8i16_to_v8i32:
 ; SI: s_endpgm
-define void @sextload_global_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <8 x i16>, <8 x i16> addrspace(1)* %in
   %ext = sext <8 x i16> %load to <8 x i32>
   store <8 x i32> %ext, <8 x i32> addrspace(1)* %out
@@ -101,7 +101,7 @@ define void @sextload_global_v8i16_to_v8i32(<8 x i32> addrspace(1)* %out, <8 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v16i16_to_v16i32:
 ; SI: s_endpgm
-define void @zextload_global_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <16 x i16>, <16 x i16> addrspace(1)* %in
   %ext = zext <16 x i16> %load to <16 x i32>
   store <16 x i32> %ext, <16 x i32> addrspace(1)* %out
@@ -110,7 +110,7 @@ define void @zextload_global_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16
 
 ; FUNC-LABEL: {{^}}sextload_global_v16i16_to_v16i32:
 ; SI: s_endpgm
-define void @sextload_global_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <16 x i16>, <16 x i16> addrspace(1)* %in
   %ext = sext <16 x i16> %load to <16 x i32>
   store <16 x i32> %ext, <16 x i32> addrspace(1)* %out
@@ -119,7 +119,7 @@ define void @sextload_global_v16i16_to_v16i32(<16 x i32> addrspace(1)* %out, <16
 
 ; FUNC-LABEL: {{^}}zextload_global_v32i16_to_v32i32:
 ; SI: s_endpgm
-define void @zextload_global_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <32 x i16>, <32 x i16> addrspace(1)* %in
   %ext = zext <32 x i16> %load to <32 x i32>
   store <32 x i32> %ext, <32 x i32> addrspace(1)* %out
@@ -128,7 +128,7 @@ define void @zextload_global_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32
 
 ; FUNC-LABEL: {{^}}sextload_global_v32i16_to_v32i32:
 ; SI: s_endpgm
-define void @sextload_global_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <32 x i16>, <32 x i16> addrspace(1)* %in
   %ext = sext <32 x i16> %load to <32 x i32>
   store <32 x i32> %ext, <32 x i32> addrspace(1)* %out
@@ -137,7 +137,7 @@ define void @sextload_global_v32i16_to_v32i32(<32 x i32> addrspace(1)* %out, <32
 
 ; FUNC-LABEL: {{^}}zextload_global_v64i16_to_v64i32:
 ; SI: s_endpgm
-define void @zextload_global_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <64 x i16>, <64 x i16> addrspace(1)* %in
   %ext = zext <64 x i16> %load to <64 x i32>
   store <64 x i32> %ext, <64 x i32> addrspace(1)* %out
@@ -146,7 +146,7 @@ define void @zextload_global_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64
 
 ; FUNC-LABEL: {{^}}sextload_global_v64i16_to_v64i32:
 ; SI: s_endpgm
-define void @sextload_global_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <64 x i16>, <64 x i16> addrspace(1)* %in
   %ext = sext <64 x i16> %load to <64 x i32>
   store <64 x i32> %ext, <64 x i32> addrspace(1)* %out
@@ -157,7 +157,7 @@ define void @sextload_global_v64i16_to_v64i32(<64 x i32> addrspace(1)* %out, <64
 ; SI-DAG: buffer_load_ushort v[[LO:[0-9]+]],
 ; SI-DAG: v_mov_b32_e32 v[[HI:[0-9]+]], 0{{$}}
 ; SI: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]]
-define void @zextload_global_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
+define amdgpu_kernel void @zextload_global_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
   %a = load i16, i16 addrspace(1)* %in
   %ext = zext i16 %a to i64
   store i64 %ext, i64 addrspace(1)* %out
@@ -168,7 +168,7 @@ define void @zextload_global_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)
 ; VI: buffer_load_ushort [[LOAD:v[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0
 ; VI: v_ashrrev_i32_e32 v{{[0-9]+}}, 31, [[LOAD]]
 ; VI: buffer_store_dwordx2 v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], 0
-define void @sextload_global_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
+define amdgpu_kernel void @sextload_global_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)* %in) nounwind {
   %a = load i16, i16 addrspace(1)* %in
   %ext = sext i16 %a to i64
   store i64 %ext, i64 addrspace(1)* %out
@@ -177,7 +177,7 @@ define void @sextload_global_i16_to_i64(i64 addrspace(1)* %out, i16 addrspace(1)
 
 ; FUNC-LABEL: {{^}}zextload_global_v1i16_to_v1i64:
 ; SI: s_endpgm
-define void @zextload_global_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <1 x i16>, <1 x i16> addrspace(1)* %in
   %ext = zext <1 x i16> %load to <1 x i64>
   store <1 x i64> %ext, <1 x i64> addrspace(1)* %out
@@ -186,7 +186,7 @@ define void @zextload_global_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i
 
 ; FUNC-LABEL: {{^}}sextload_global_v1i16_to_v1i64:
 ; SI: s_endpgm
-define void @sextload_global_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <1 x i16>, <1 x i16> addrspace(1)* %in
   %ext = sext <1 x i16> %load to <1 x i64>
   store <1 x i64> %ext, <1 x i64> addrspace(1)* %out
@@ -195,7 +195,7 @@ define void @sextload_global_v1i16_to_v1i64(<1 x i64> addrspace(1)* %out, <1 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v2i16_to_v2i64:
 ; SI: s_endpgm
-define void @zextload_global_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <2 x i16>, <2 x i16> addrspace(1)* %in
   %ext = zext <2 x i16> %load to <2 x i64>
   store <2 x i64> %ext, <2 x i64> addrspace(1)* %out
@@ -204,7 +204,7 @@ define void @zextload_global_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i
 
 ; FUNC-LABEL: {{^}}sextload_global_v2i16_to_v2i64:
 ; SI: s_endpgm
-define void @sextload_global_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <2 x i16>, <2 x i16> addrspace(1)* %in
   %ext = sext <2 x i16> %load to <2 x i64>
   store <2 x i64> %ext, <2 x i64> addrspace(1)* %out
@@ -213,7 +213,7 @@ define void @sextload_global_v2i16_to_v2i64(<2 x i64> addrspace(1)* %out, <2 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v4i16_to_v4i64:
 ; SI: s_endpgm
-define void @zextload_global_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <4 x i16>, <4 x i16> addrspace(1)* %in
   %ext = zext <4 x i16> %load to <4 x i64>
   store <4 x i64> %ext, <4 x i64> addrspace(1)* %out
@@ -222,7 +222,7 @@ define void @zextload_global_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i
 
 ; FUNC-LABEL: {{^}}sextload_global_v4i16_to_v4i64:
 ; SI: s_endpgm
-define void @sextload_global_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <4 x i16>, <4 x i16> addrspace(1)* %in
   %ext = sext <4 x i16> %load to <4 x i64>
   store <4 x i64> %ext, <4 x i64> addrspace(1)* %out
@@ -231,7 +231,7 @@ define void @sextload_global_v4i16_to_v4i64(<4 x i64> addrspace(1)* %out, <4 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v8i16_to_v8i64:
 ; SI: s_endpgm
-define void @zextload_global_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <8 x i16>, <8 x i16> addrspace(1)* %in
   %ext = zext <8 x i16> %load to <8 x i64>
   store <8 x i64> %ext, <8 x i64> addrspace(1)* %out
@@ -240,7 +240,7 @@ define void @zextload_global_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i
 
 ; FUNC-LABEL: {{^}}sextload_global_v8i16_to_v8i64:
 ; SI: s_endpgm
-define void @sextload_global_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <8 x i16>, <8 x i16> addrspace(1)* %in
   %ext = sext <8 x i16> %load to <8 x i64>
   store <8 x i64> %ext, <8 x i64> addrspace(1)* %out
@@ -249,7 +249,7 @@ define void @sextload_global_v8i16_to_v8i64(<8 x i64> addrspace(1)* %out, <8 x i
 
 ; FUNC-LABEL: {{^}}zextload_global_v16i16_to_v16i64:
 ; SI: s_endpgm
-define void @zextload_global_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <16 x i16>, <16 x i16> addrspace(1)* %in
   %ext = zext <16 x i16> %load to <16 x i64>
   store <16 x i64> %ext, <16 x i64> addrspace(1)* %out
@@ -258,7 +258,7 @@ define void @zextload_global_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16
 
 ; FUNC-LABEL: {{^}}sextload_global_v16i16_to_v16i64:
 ; SI: s_endpgm
-define void @sextload_global_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <16 x i16>, <16 x i16> addrspace(1)* %in
   %ext = sext <16 x i16> %load to <16 x i64>
   store <16 x i64> %ext, <16 x i64> addrspace(1)* %out
@@ -267,7 +267,7 @@ define void @sextload_global_v16i16_to_v16i64(<16 x i64> addrspace(1)* %out, <16
 
 ; FUNC-LABEL: {{^}}zextload_global_v32i16_to_v32i64:
 ; SI: s_endpgm
-define void @zextload_global_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <32 x i16>, <32 x i16> addrspace(1)* %in
   %ext = zext <32 x i16> %load to <32 x i64>
   store <32 x i64> %ext, <32 x i64> addrspace(1)* %out
@@ -276,7 +276,7 @@ define void @zextload_global_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32
 
 ; FUNC-LABEL: {{^}}sextload_global_v32i16_to_v32i64:
 ; SI: s_endpgm
-define void @sextload_global_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <32 x i16>, <32 x i16> addrspace(1)* %in
   %ext = sext <32 x i16> %load to <32 x i64>
   store <32 x i64> %ext, <32 x i64> addrspace(1)* %out
@@ -285,7 +285,7 @@ define void @sextload_global_v32i16_to_v32i64(<32 x i64> addrspace(1)* %out, <32
 
 ; FUNC-LABEL: {{^}}zextload_global_v64i16_to_v64i64:
 ; SI: s_endpgm
-define void @zextload_global_v64i16_to_v64i64(<64 x i64> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @zextload_global_v64i16_to_v64i64(<64 x i64> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <64 x i16>, <64 x i16> addrspace(1)* %in
   %ext = zext <64 x i16> %load to <64 x i64>
   store <64 x i64> %ext, <64 x i64> addrspace(1)* %out
@@ -294,7 +294,7 @@ define void @zextload_global_v64i16_to_v64i64(<64 x i64> addrspace(1)* %out, <64
 
 ; FUNC-LABEL: {{^}}sextload_global_v64i16_to_v64i64:
 ; SI: s_endpgm
-define void @sextload_global_v64i16_to_v64i64(<64 x i64> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
+define amdgpu_kernel void @sextload_global_v64i16_to_v64i64(<64 x i64> addrspace(1)* %out, <64 x i16> addrspace(1)* nocapture %in) nounwind {
   %load = load <64 x i16>, <64 x i16> addrspace(1)* %in
   %ext = sext <64 x i16> %load to <64 x i64>
   store <64 x i64> %ext, <64 x i64> addrspace(1)* %out