shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity

[AMDGPU] allow multi-dword flat scratch access since GFX9

Browse Source 
This is supported starting with GFX9.

Differential Revision: https://reviews.llvm.org/D72865
This commit is contained in:
Stanislav Mekhanoshin 2020-01-16 12:43:22 -08:00
parent f35b5a7297
commit eebdd85e7d
3 changed files with 18 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
View File

@ -724,6 +724,10 @@ public:
return ScalarFlatScratchInsts;
}
bool hasMultiDwordFlatScratchAddressing() const {
return getGeneration() >= GFX9;
}
bool hasFlatSegmentOffsetBug() const {
return HasFlatSegmentOffsetBug;
}

6
llvm/lib/Target/AMDGPU/SIISelLowering.cpp
View File

@ -7383,7 +7383,8 @@ SDValue SITargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
// If there is a possibilty that flat instruction access scratch memory
// then we need to use the same legalization rules we use for private.
if (AS == AMDGPUAS::FLAT_ADDRESS)
if (AS == AMDGPUAS::FLAT_ADDRESS &&
!Subtarget->hasMultiDwordFlatScratchAddressing())
AS = MFI->hasFlatScratchInit() ?
AMDGPUAS::PRIVATE_ADDRESS : AMDGPUAS::GLOBAL_ADDRESS;
@ -7886,7 +7887,8 @@ SDValue SITargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
// If there is a possibilty that flat instruction access scratch memory
// then we need to use the same legalization rules we use for private.
if (AS == AMDGPUAS::FLAT_ADDRESS)
if (AS == AMDGPUAS::FLAT_ADDRESS &&
!Subtarget->hasMultiDwordFlatScratchAddressing())
AS = MFI->hasFlatScratchInit() ?
AMDGPUAS::PRIVATE_ADDRESS : AMDGPUAS::GLOBAL_ADDRESS;

15
llvm/test/CodeGen/AMDGPU/flat-address-space.ll
View File

@ -1,7 +1,8 @@
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=bonaire < %s | FileCheck -check-prefixes=CHECK,CIVI %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefixes=CHECK,CIVI %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -mattr=-flat-for-global < %s | FileCheck -check-prefixes=CHECK,CIVI,HSA %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -mattr=-flat-for-global < %s | FileCheck -check-prefixes=CHECK,CIVI,HSA,CIVI-HSA %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -mattr=-flat-for-global < %s | FileCheck -check-prefixes=CHECK,HSA,GFX9 %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=-flat-for-global < %s | FileCheck -check-prefixes=CHECK,HSA,GFX10 %s
; CHECK-LABEL: {{^}}store_flat_i32:
; CHECK-DAG: s_load_dwordx2 s{{\[}}[[LO_SREG:[0-9]+]]:[[HI_SREG:[0-9]+]]],
@ -145,8 +146,10 @@ define amdgpu_kernel void @flat_scratch_unaligned_store() {
}
; CHECK-LABEL: flat_scratch_multidword_load:
; HSA: flat_load_dword
; HSA: flat_load_dword
; CIVI-HSA: flat_load_dword v
; CIVI-HSA: flat_load_dword v
; GFX9: flat_load_dwordx2
; GFX10: flat_load_dwordx2
; FIXME: These tests are broken for os = mesa3d, becasue it doesn't initialize flat_scr
define amdgpu_kernel void @flat_scratch_multidword_load() {
%scratch = alloca <2 x i32>, addrspace(5)
@ -156,8 +159,10 @@ define amdgpu_kernel void @flat_scratch_multidword_load() {
}
; CHECK-LABEL: flat_scratch_multidword_store:
; HSA: flat_store_dword
; HSA: flat_store_dword
; CIVI-HSA: flat_store_dword v
; CIVI-HSA: flat_store_dword v
; GFX9: flat_store_dwordx2
; GFX10: flat_store_dwordx2
; FIXME: These tests are broken for os = mesa3d, becasue it doesn't initialize flat_scr
define amdgpu_kernel void @flat_scratch_multidword_store() {
%scratch = alloca <2 x i32>, addrspace(5)