mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2025-02-01 05:01:59 +01:00
b9148f5d85
Summary: Extend cachepolicy operand in the new VMEM buffer intrinsics to supply information whether the buffer data is swizzled. Also, propagate this information to MIR. Intrinsics updated: int_amdgcn_raw_buffer_load int_amdgcn_raw_buffer_load_format int_amdgcn_raw_buffer_store int_amdgcn_raw_buffer_store_format int_amdgcn_raw_tbuffer_load int_amdgcn_raw_tbuffer_store int_amdgcn_struct_buffer_load int_amdgcn_struct_buffer_load_format int_amdgcn_struct_buffer_store int_amdgcn_struct_buffer_store_format int_amdgcn_struct_tbuffer_load int_amdgcn_struct_tbuffer_store Furthermore, disable merging of VMEM buffer instructions in SI Load/Store optimizer, if the "swizzled" bit on the instruction is on. The default value of the bit is 0, meaning that data in buffer is linear and buffer instructions can be merged. There is no difference in the generated code with this commit. However, in the future it will be expected that front-ends use buffer intrinsics with correct "swizzled" bit set. Reviewers: arsenm, nhaehnle, tpr Reviewed By: nhaehnle Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, arphaman, jfb, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D68200 llvm-svn: 373491
172 lines
5.6 KiB
YAML
172 lines
5.6 KiB
YAML
# RUN: llc -run-pass si-insert-waitcnts -march=amdgcn -mcpu=tahiti -o - %s | FileCheck %s
|
|
--- |
|
|
|
|
define amdgpu_kernel void @vccz_corrupt_workaround(float %cond, i32 addrspace(1)* %out) #0 {
|
|
entry:
|
|
%cmp0 = fcmp oeq float %cond, 0.000000e+00
|
|
br i1 %cmp0, label %if, label %else, !structurizecfg.uniform !0, !amdgpu.uniform !0
|
|
|
|
else: ; preds = %entry
|
|
store volatile i32 100, i32 addrspace(1)* undef
|
|
br label %done, !structurizecfg.uniform !0
|
|
|
|
if: ; preds = %entry
|
|
store volatile i32 9, i32 addrspace(1)* undef
|
|
br label %done, !structurizecfg.uniform !0
|
|
|
|
done: ; preds = %if, %else
|
|
%value = phi i32 [ 0, %if ], [ 1, %else ]
|
|
store i32 %value, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @vccz_corrupt_undef_vcc(float %cond, i32 addrspace(1)* %out) #0 {
|
|
entry:
|
|
br i1 undef, label %if, label %else, !structurizecfg.uniform !0, !amdgpu.uniform !0
|
|
|
|
else: ; preds = %entry
|
|
store volatile i32 100, i32 addrspace(1)* undef
|
|
br label %done, !structurizecfg.uniform !0
|
|
|
|
if: ; preds = %entry
|
|
store volatile i32 9, i32 addrspace(1)* undef
|
|
br label %done, !structurizecfg.uniform !0
|
|
|
|
done: ; preds = %if, %else
|
|
%value = phi i32 [ 0, %if ], [ 1, %else ]
|
|
store i32 %value, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = { nounwind }
|
|
attributes #1 = { readnone }
|
|
|
|
!0 = !{}
|
|
|
|
...
|
|
---
|
|
# CHECK-LABEL: name: vccz_corrupt_workaround
|
|
# CHECK: $vcc = V_CMP_EQ_F32
|
|
# CHECK-NEXT: S_WAITCNT 127
|
|
# CHECK-NEXT: $vcc = S_MOV_B64 $vcc
|
|
# CHECK-NEXT: S_CBRANCH_VCCZ %bb.2, implicit killed $vcc
|
|
|
|
name: vccz_corrupt_workaround
|
|
alignment: 1
|
|
exposesReturnsTwice: false
|
|
legalized: false
|
|
regBankSelected: false
|
|
selected: false
|
|
tracksRegLiveness: true
|
|
liveins:
|
|
- { reg: '$sgpr0_sgpr1' }
|
|
frameInfo:
|
|
isFrameAddressTaken: false
|
|
isReturnAddressTaken: false
|
|
hasStackMap: false
|
|
hasPatchPoint: false
|
|
stackSize: 0
|
|
offsetAdjustment: 0
|
|
maxAlignment: 0
|
|
adjustsStack: false
|
|
hasCalls: false
|
|
maxCallFrameSize: 0
|
|
hasOpaqueSPAdjustment: false
|
|
hasVAStart: false
|
|
hasMustTailInVarArgFunc: false
|
|
body: |
|
|
bb.0.entry:
|
|
liveins: $sgpr0_sgpr1, $vcc
|
|
|
|
$sgpr0_sgpr1 = S_LOAD_DWORDX2_IMM killed $sgpr0_sgpr1, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
|
|
$sgpr7 = S_MOV_B32 61440
|
|
$sgpr6 = S_MOV_B32 -1
|
|
$vcc = V_CMP_EQ_F32_e64 0, 0, 0, undef $sgpr2, 0, implicit $exec
|
|
S_CBRANCH_VCCZ %bb.1, implicit killed $vcc
|
|
|
|
bb.2.if:
|
|
liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
|
|
|
|
$vgpr0 = V_MOV_B32_e32 9, implicit $exec
|
|
BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
|
|
$vgpr0 = V_MOV_B32_e32 0, implicit $exec
|
|
S_BRANCH %bb.3
|
|
|
|
bb.1.else:
|
|
liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
|
|
|
|
$vgpr0 = V_MOV_B32_e32 100, implicit $exec
|
|
BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
|
|
$vgpr0 = V_MOV_B32_e32 1, implicit $exec
|
|
|
|
bb.3.done:
|
|
liveins: $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
|
|
|
|
$sgpr3 = S_MOV_B32 61440
|
|
$sgpr2 = S_MOV_B32 -1
|
|
BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (store 4 into %ir.out)
|
|
S_ENDPGM 0
|
|
|
|
...
|
|
---
|
|
# CHECK-LABEL: name: vccz_corrupt_undef_vcc
|
|
# CHECK: S_WAITCNT 3855
|
|
# CHECK-NEXT: $vgpr0 = V_MOV_B32_e32
|
|
|
|
name: vccz_corrupt_undef_vcc
|
|
alignment: 1
|
|
exposesReturnsTwice: false
|
|
legalized: false
|
|
regBankSelected: false
|
|
selected: false
|
|
tracksRegLiveness: true
|
|
liveins:
|
|
- { reg: '$sgpr0_sgpr1' }
|
|
frameInfo:
|
|
isFrameAddressTaken: false
|
|
isReturnAddressTaken: false
|
|
hasStackMap: false
|
|
hasPatchPoint: false
|
|
stackSize: 0
|
|
offsetAdjustment: 0
|
|
maxAlignment: 0
|
|
adjustsStack: false
|
|
hasCalls: false
|
|
maxCallFrameSize: 0
|
|
hasOpaqueSPAdjustment: false
|
|
hasVAStart: false
|
|
hasMustTailInVarArgFunc: false
|
|
body: |
|
|
bb.0.entry:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
$sgpr0_sgpr1 = S_LOAD_DWORDX2_IMM killed $sgpr0_sgpr1, 11, 0, 0 :: (non-temporal dereferenceable invariant load 8 from `i64 addrspace(4)* undef`)
|
|
$sgpr7 = S_MOV_B32 61440
|
|
$sgpr6 = S_MOV_B32 -1
|
|
S_CBRANCH_VCCZ %bb.1, implicit undef $vcc
|
|
|
|
bb.2.if:
|
|
liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
|
|
|
|
$vgpr0 = V_MOV_B32_e32 9, implicit $exec
|
|
BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
|
|
$vgpr0 = V_MOV_B32_e32 0, implicit $exec
|
|
S_BRANCH %bb.3
|
|
|
|
bb.1.else:
|
|
liveins: $sgpr6, $sgpr7, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
|
|
|
|
$vgpr0 = V_MOV_B32_e32 100, implicit $exec
|
|
BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (volatile store 4 into `i32 addrspace(1)* undef`)
|
|
$vgpr0 = V_MOV_B32_e32 1, implicit $exec
|
|
|
|
bb.3.done:
|
|
liveins: $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3:0x00000003
|
|
|
|
$sgpr3 = S_MOV_B32 61440
|
|
$sgpr2 = S_MOV_B32 -1
|
|
BUFFER_STORE_DWORD_OFFSET killed $vgpr0, killed $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (store 4 into %ir.out)
|
|
S_ENDPGM 0
|
|
|
|
...
|