maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/rewrite-out-arguments-addre...

38 lines
2.0 KiB
LLVM
Raw Normal View History

AMDGPU: Add pass to replace out arguments It is better to return arguments directly in registers if we are making a call rather than introducing expensive stack usage. In one of sample compile from one of Blender's many kernel variants, this fires on about ~20 different functions. Future improvements may be to recognize simple cases where the pointer is indexing a small array. This also fails when the store to the out argument is in a separate block from the return, which happens in a few of the Blender functions. This should also probably be using MemorySSA which might help with that. I'm not sure this is correct as a FunctionPass, but MemoryDependenceAnalysis seems to not work with a ModulePass. I'm also not sure where it should run.I think it should run before DeadArgumentElimination, so maybe either EP_CGSCCOptimizerLate or EP_ScalarOptimizerLate. llvm-svn: 309416
2017-07-29 02:40:05 +08:00
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-any-address-space-out-arguments -amdgpu-rewrite-out-arguments < %s | FileCheck %s
; CHECK: %void_one_out_non_private_arg_i32_1_use = type { i32 }
AMDGPU: Look through a bitcast user of an out argument This allows handling of a lot more of the interesting cases in Blender. Most of the large functions unlikely to be inlined have this pattern. This is a special case for what clang emits for OpenCL 3 element vectors. Annoyingly, these are emitted as <3 x elt>* pointers, but accessed as <4 x elt>* operations. This also needs to handle cases where a struct containing a single vector is used. llvm-svn: 309419
2017-07-29 03:06:16 +08:00
; CHECK: %bitcast_pointer_as1 = type { <3 x i32> }
AMDGPU: Add pass to replace out arguments It is better to return arguments directly in registers if we are making a call rather than introducing expensive stack usage. In one of sample compile from one of Blender's many kernel variants, this fires on about ~20 different functions. Future improvements may be to recognize simple cases where the pointer is indexing a small array. This also fails when the store to the out argument is in a separate block from the return, which happens in a few of the Blender functions. This should also probably be using MemorySSA which might help with that. I'm not sure this is correct as a FunctionPass, but MemoryDependenceAnalysis seems to not work with a ModulePass. I'm also not sure where it should run.I think it should run before DeadArgumentElimination, so maybe either EP_CGSCCOptimizerLate or EP_ScalarOptimizerLate. llvm-svn: 309416
2017-07-29 02:40:05 +08:00
; CHECK-LABEL: define private %void_one_out_non_private_arg_i32_1_use @void_one_out_non_private_arg_i32_1_use.body(i32 addrspace(1)* %val) #0 {
; CHECK-NEXT: ret %void_one_out_non_private_arg_i32_1_use zeroinitializer
; CHECK-LABEL: define void @void_one_out_non_private_arg_i32_1_use(i32 addrspace(1)*) #1 {
; CHECK-NEXT: %2 = call %void_one_out_non_private_arg_i32_1_use @void_one_out_non_private_arg_i32_1_use.body(i32 addrspace(1)* undef)
; CHECK-NEXT: %3 = extractvalue %void_one_out_non_private_arg_i32_1_use %2, 0
; CHECK-NEXT: store i32 %3, i32 addrspace(1)* %0, align 4
; CHECK-NEXT: ret void
define void @void_one_out_non_private_arg_i32_1_use(i32 addrspace(1)* %val) #0 {
store i32 0, i32 addrspace(1)* %val
ret void
}
AMDGPU: Look through a bitcast user of an out argument This allows handling of a lot more of the interesting cases in Blender. Most of the large functions unlikely to be inlined have this pattern. This is a special case for what clang emits for OpenCL 3 element vectors. Annoyingly, these are emitted as <3 x elt>* pointers, but accessed as <4 x elt>* operations. This also needs to handle cases where a struct containing a single vector is used. llvm-svn: 309419
2017-07-29 03:06:16 +08:00
; CHECK-LABEL: define private %bitcast_pointer_as1 @bitcast_pointer_as1.body(<3 x i32> addrspace(1)* %out) #0 {
; CHECK-NEXT: %load = load volatile <4 x i32>, <4 x i32> addrspace(1)* undef
; CHECK-NEXT: %bitcast = bitcast <3 x i32> addrspace(1)* %out to <4 x i32> addrspace(1)*
; CHECK-NEXT: %1 = shufflevector <4 x i32> %load, <4 x i32> undef, <3 x i32> <i32 0, i32 1, i32 2>
; CHECK-NEXT: %2 = insertvalue %bitcast_pointer_as1 undef, <3 x i32> %1, 0
; CHECK-NEXT: ret %bitcast_pointer_as1 %2
; CHECK-LABEL: define void @bitcast_pointer_as1(<3 x i32> addrspace(1)*) #1 {
; CHECK-NEXT: %2 = call %bitcast_pointer_as1 @bitcast_pointer_as1.body(<3 x i32> addrspace(1)* undef)
define void @bitcast_pointer_as1(<3 x i32> addrspace(1)* %out) #0 {
%load = load volatile <4 x i32>, <4 x i32> addrspace(1)* undef
%bitcast = bitcast <3 x i32> addrspace(1)* %out to <4 x i32> addrspace(1)*
store <4 x i32> %load, <4 x i32> addrspace(1)* %bitcast
ret void
}
AMDGPU: Add pass to replace out arguments It is better to return arguments directly in registers if we are making a call rather than introducing expensive stack usage. In one of sample compile from one of Blender's many kernel variants, this fires on about ~20 different functions. Future improvements may be to recognize simple cases where the pointer is indexing a small array. This also fails when the store to the out argument is in a separate block from the return, which happens in a few of the Blender functions. This should also probably be using MemorySSA which might help with that. I'm not sure this is correct as a FunctionPass, but MemoryDependenceAnalysis seems to not work with a ModulePass. I'm also not sure where it should run.I think it should run before DeadArgumentElimination, so maybe either EP_CGSCCOptimizerLate or EP_ScalarOptimizerLate. llvm-svn: 309416
2017-07-29 02:40:05 +08:00
; CHECK: attributes #0 = { nounwind }
; CHECK: attributes #1 = { alwaysinline nounwind }
attributes #0 = { nounwind }