mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-24 03:33:20 +01:00
262321d1ff
This patch lets the llvm tools handle the new HVX target features that
are added by frontend (clang). The target-features are of the form
"hvx-length64b" for 64 Byte HVX mode, "hvx-length128b" for 128 Byte mode HVX.
"hvx-double" is an alias to "hvx-length128b" and is soon will be deprecated.
The hvx version target feature is upgated form "+hvx" to "+hvxv{version_number}.
Eg: "+hvxv62"
For the correct HVX code generation, the user must use the following
target features.
For 64B mode: "+hvxv62" "+hvx-length64b"
For 128B mode: "+hvxv62" "+hvx-length128b"
Clang picks a default length if none is specified. If for some reason,
no hvx-length is specified to llvm, the compilation will bail out.
There is a corresponding clang patch.
Differential Revision: https://reviews.llvm.org/D38851
llvm-svn: 316101
49 lines
2.0 KiB
LLVM
49 lines
2.0 KiB
LLVM
; RUN: llc -march=hexagon < %s | FileCheck %s
|
|
|
|
@v0 = global <16 x i32> zeroinitializer, align 64
|
|
@v1 = global <16 x i32> zeroinitializer, align 64
|
|
|
|
; CHECK-LABEL: danny:
|
|
; CHECK-NOT: vcombine
|
|
|
|
define void @danny() #0 {
|
|
%t0 = load <16 x i32>, <16 x i32>* @v0, align 64
|
|
%t1 = load <16 x i32>, <16 x i32>* @v1, align 64
|
|
%t2 = call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %t0, <16 x i32> %t1)
|
|
%t3 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %t2)
|
|
%t4 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %t2)
|
|
store <16 x i32> %t3, <16 x i32>* @v0, align 64
|
|
store <16 x i32> %t4, <16 x i32>* @v1, align 64
|
|
ret void
|
|
}
|
|
|
|
@w0 = global <32 x i32> zeroinitializer, align 128
|
|
@w1 = global <32 x i32> zeroinitializer, align 128
|
|
|
|
; CHECK-LABEL: sammy:
|
|
; CHECK-NOT: vcombine
|
|
|
|
define void @sammy() #1 {
|
|
%t0 = load <32 x i32>, <32 x i32>* @w0, align 128
|
|
%t1 = load <32 x i32>, <32 x i32>* @w1, align 128
|
|
%t2 = call <64 x i32> @llvm.hexagon.V6.vcombine.128B(<32 x i32> %t0, <32 x i32> %t1)
|
|
%t3 = tail call <32 x i32> @llvm.hexagon.V6.lo.128B(<64 x i32> %t2)
|
|
%t4 = tail call <32 x i32> @llvm.hexagon.V6.hi.128B(<64 x i32> %t2)
|
|
store <32 x i32> %t3, <32 x i32>* @w0, align 128
|
|
store <32 x i32> %t4, <32 x i32>* @w1, align 128
|
|
ret void
|
|
}
|
|
|
|
declare <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32>, <16 x i32>) #2
|
|
declare <16 x i32> @llvm.hexagon.V6.lo(<32 x i32>) #2
|
|
declare <16 x i32> @llvm.hexagon.V6.hi(<32 x i32>) #2
|
|
|
|
declare <64 x i32> @llvm.hexagon.V6.vcombine.128B(<32 x i32>, <32 x i32>) #3
|
|
declare <32 x i32> @llvm.hexagon.V6.lo.128B(<64 x i32>) #3
|
|
declare <32 x i32> @llvm.hexagon.V6.hi.128B(<64 x i32>) #3
|
|
|
|
attributes #0 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
|
|
attributes #1 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length128b" }
|
|
attributes #2 = { nounwind readnone "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
|
|
attributes #3 = { nounwind readnone "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length128b" }
|