summaryrefslogtreecommitdiffstats
path: root/libclc/generic/lib/shared/vstore_impl.ll
diff options
context:
space:
mode:
authorAaron Watry <awatry@gmail.com>2014-08-20 13:58:57 +0000
committerAaron Watry <awatry@gmail.com>2014-08-20 13:58:57 +0000
commitf991505d02dae4c2e9cd9ae02372a3116da2a14b (patch)
tree99138f8a26c76632c9db66bf8622e97c38a6ddda /libclc/generic/lib/shared/vstore_impl.ll
parent474972585a1f4328c9f3138c88365a43c11556c0 (diff)
downloadbcm5719-llvm-f991505d02dae4c2e9cd9ae02372a3116da2a14b.tar.gz
bcm5719-llvm-f991505d02dae4c2e9cd9ae02372a3116da2a14b.zip
vload/vstore: Use casts instead of scalarizing everything in CLC version
This generates bitcode which is indistinguishable from what was hand-written for int32 types in v[load|store]_impl.ll. v4: Use vec2+scalar for vec3 load/stores to prevent corruption (per Tom) v3: Also remove unused generic/lib/shared/v[load|store]_impl.ll v2: (Per Matt Arsenault) Fix alignment issues with vector load stores Signed-off-by: Aaron Watry <awatry@gmail.com> Reviewed-by: Tom Stellard <thomas.stellard@amd.com> CC: Matt Arsenault <Matthew.Arsenault@amd.com> CC: Tom Stellard <thomas.stellard@amd.com> llvm-svn: 216069
Diffstat (limited to 'libclc/generic/lib/shared/vstore_impl.ll')
-rw-r--r--libclc/generic/lib/shared/vstore_impl.ll40
1 files changed, 0 insertions, 40 deletions
diff --git a/libclc/generic/lib/shared/vstore_impl.ll b/libclc/generic/lib/shared/vstore_impl.ll
deleted file mode 100644
index 9e2a37b2970..00000000000
--- a/libclc/generic/lib/shared/vstore_impl.ll
+++ /dev/null
@@ -1,40 +0,0 @@
-; This provides optimized implementations of vstore2/3/4/8/16 for 32-bit int/uint
-; The address spaces get mapped to data types in target-specific usages
-
-define void @__clc_vstore2_i32__addr1(<2 x i32> %vec, i32 addrspace(1)* nocapture %addr) nounwind alwaysinline {
- %1 = bitcast i32 addrspace(1)* %addr to <2 x i32> addrspace(1)*
- store <2 x i32> %vec, <2 x i32> addrspace(1)* %1, align 4, !tbaa !3
- ret void
-}
-
-define void @__clc_vstore3_i32__addr1(<3 x i32> %vec, i32 addrspace(1)* nocapture %addr) nounwind alwaysinline {
- %1 = bitcast i32 addrspace(1)* %addr to <3 x i32> addrspace(1)*
- store <3 x i32> %vec, <3 x i32> addrspace(1)* %1, align 4, !tbaa !3
- ret void
-}
-
-define void @__clc_vstore4_i32__addr1(<4 x i32> %vec, i32 addrspace(1)* nocapture %addr) nounwind alwaysinline {
- %1 = bitcast i32 addrspace(1)* %addr to <4 x i32> addrspace(1)*
- store <4 x i32> %vec, <4 x i32> addrspace(1)* %1, align 4, !tbaa !3
- ret void
-}
-
-define void @__clc_vstore8_i32__addr1(<8 x i32> %vec, i32 addrspace(1)* nocapture %addr) nounwind alwaysinline {
- %1 = bitcast i32 addrspace(1)* %addr to <8 x i32> addrspace(1)*
- store <8 x i32> %vec, <8 x i32> addrspace(1)* %1, align 4, !tbaa !3
- ret void
-}
-
-define void @__clc_vstore16_i32__addr1(<16 x i32> %vec, i32 addrspace(1)* nocapture %addr) nounwind alwaysinline {
- %1 = bitcast i32 addrspace(1)* %addr to <16 x i32> addrspace(1)*
- store <16 x i32> %vec, <16 x i32> addrspace(1)* %1, align 4, !tbaa !3
- ret void
-}
-
-!1 = metadata !{metadata !"char", metadata !5}
-!2 = metadata !{metadata !"short", metadata !5}
-!3 = metadata !{metadata !"int", metadata !5}
-!4 = metadata !{metadata !"long", metadata !5}
-!5 = metadata !{metadata !"omnipotent char", metadata !6}
-!6 = metadata !{metadata !"Simple C/C++ TBAA"}
-
OpenPOWER on IntegriCloud