diff options
| author | Aaron Watry <awatry@gmail.com> | 2013-07-16 14:29:01 +0000 |
|---|---|---|
| committer | Aaron Watry <awatry@gmail.com> | 2013-07-16 14:29:01 +0000 |
| commit | 99a2f3b27475fd717618ce464646e0dcd35e2b14 (patch) | |
| tree | 754eef3bd26da0f0d4a71e9ef5b2ab0e238d1280 /libclc/generic/lib/shared/vload_impl.ll | |
| parent | 4cb7cf276df36a15e9b416f3d0b55d5604615112 (diff) | |
| download | bcm5719-llvm-99a2f3b27475fd717618ce464646e0dcd35e2b14.tar.gz bcm5719-llvm-99a2f3b27475fd717618ce464646e0dcd35e2b14.zip | |
Fix and re-enable R600 vload/vstore assembly
The assembly optimizations were making unsafe assumptions about which address
spaces had which identifiers.
Also, fix vload/vstore with 64-bit pointers. This was broken previously on
Radeon SI.
This version still only has assembly versions of int/uint 2/4/8/16 for global
loads and stores on R600, but it does it in a way that would be very easily
extended to private/local/constant and could also be handled easily on other
architectures.
v2: 1) Leave v[load|store]_impl.ll in generic/lib
2) Remove vload_if.ll and vstore_if.ll interfaces
3) Fix address+offset calculations
3) Remove offset from assembly arg list
llvm-svn: 186416
Diffstat (limited to 'libclc/generic/lib/shared/vload_impl.ll')
| -rw-r--r-- | libclc/generic/lib/shared/vload_impl.ll | 50 |
1 files changed, 20 insertions, 30 deletions
diff --git a/libclc/generic/lib/shared/vload_impl.ll b/libclc/generic/lib/shared/vload_impl.ll index ae719e0ae6d..2e70e5f2029 100644 --- a/libclc/generic/lib/shared/vload_impl.ll +++ b/libclc/generic/lib/shared/vload_impl.ll @@ -1,43 +1,33 @@ ; This provides optimized implementations of vload4/8/16 for 32-bit int/uint -define <2 x i32> @__clc_vload2_impl_i32__global(i32 %offset, i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { - %1 = ptrtoint i32 addrspace(1)* %addr to i32 - %2 = add i32 %1, %offset - %3 = inttoptr i32 %2 to <2 x i32> addrspace(1)* - %4 = load <2 x i32> addrspace(1)* %3, align 4, !tbaa !3 - ret <2 x i32> %4 +define <2 x i32> @__clc_vload2_i32__addr1(i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { + %1 = bitcast i32 addrspace(1)* %addr to <2 x i32> addrspace(1)* + %2 = load <2 x i32> addrspace(1)* %1, align 4, !tbaa !3 + ret <2 x i32> %2 } -define <3 x i32> @__clc_vload3_impl_i32__global(i32 %offset, i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { - %1 = ptrtoint i32 addrspace(1)* %addr to i32 - %2 = add i32 %1, %offset - %3 = inttoptr i32 %2 to <3 x i32> addrspace(1)* - %4 = load <3 x i32> addrspace(1)* %3, align 4, !tbaa !3 - ret <3 x i32> %4 +define <3 x i32> @__clc_vload3_i32__addr1(i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { + %1 = bitcast i32 addrspace(1)* %addr to <3 x i32> addrspace(1)* + %2 = load <3 x i32> addrspace(1)* %1, align 4, !tbaa !3 + ret <3 x i32> %2 } -define <4 x i32> @__clc_vload4_impl_i32__global(i32 %offset, i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { - %1 = ptrtoint i32 addrspace(1)* %addr to i32 - %2 = add i32 %1, %offset - %3 = inttoptr i32 %2 to <4 x i32> addrspace(1)* - %4 = load <4 x i32> addrspace(1)* %3, align 4, !tbaa !3 - ret <4 x i32> %4 +define <4 x i32> @__clc_vload4_i32__addr1(i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { + %1 = bitcast i32 addrspace(1)* %addr to <4 x i32> addrspace(1)* + %2 = load <4 x i32> addrspace(1)* %1, align 4, !tbaa !3 + ret <4 x i32> %2 } -define <8 x i32> @__clc_vload8_impl_i32__global(i32 %offset, i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { - %1 = ptrtoint i32 addrspace(1)* %addr to i32 - %2 = add i32 %1, %offset - %3 = inttoptr i32 %2 to <8 x i32> addrspace(1)* - %4 = load <8 x i32> addrspace(1)* %3, align 4, !tbaa !3 - ret <8 x i32> %4 +define <8 x i32> @__clc_vload8_i32__addr1(i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { + %1 = bitcast i32 addrspace(1)* %addr to <8 x i32> addrspace(1)* + %2 = load <8 x i32> addrspace(1)* %1, align 4, !tbaa !3 + ret <8 x i32> %2 } -define <16 x i32> @__clc_vload16_impl_i32__global(i32 %offset, i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { - %1 = ptrtoint i32 addrspace(1)* %addr to i32 - %2 = add i32 %1, %offset - %3 = inttoptr i32 %2 to <16 x i32> addrspace(1)* - %4 = load <16 x i32> addrspace(1)* %3, align 4, !tbaa !3 - ret <16 x i32> %4 +define <16 x i32> @__clc_vload16_i32__addr1(i32 addrspace(1)* nocapture %addr) nounwind readonly alwaysinline { + %1 = bitcast i32 addrspace(1)* %addr to <16 x i32> addrspace(1)* + %2 = load <16 x i32> addrspace(1)* %1, align 4, !tbaa !3 + ret <16 x i32> %2 } !1 = metadata !{metadata !"char", metadata !5} |

