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/sched-handleMoveUp-subreg-d...

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

135 lines
9.3 KiB
Plaintext
Raw Normal View History

LiveIntervals: Fix handleMoveUp with subreg def moving across a def If a subregister def was moved across another subregister def and another use, the main range was not correctly updated. The end point of the moved interval ended too early and missed the use from theh other lanes in the subreg def. llvm-svn: 375300
2019-10-19 07:24:25 +08:00
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -verify-machineinstrs -verify-misched -run-pass=machine-scheduler -o - %s | FileCheck %s
---
name: handleMoveUp_incorrect_interval
tracksRegLiveness: true
liveins:
- { reg: '$sgpr4_sgpr5', virtual-reg: '%0' }
frameInfo:
maxAlignment: 1
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: '$sgpr96_sgpr97_sgpr98_sgpr99'
scratchWaveOffsetReg: '$sgpr101'
frameOffsetReg: '$sgpr101'
stackPtrOffsetReg: '$sgpr101'
argumentInfo:
privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }
kernargSegmentPtr: { reg: '$sgpr4_sgpr5' }
workGroupIDX: { reg: '$sgpr6' }
privateSegmentWaveByteOffset: { reg: '$sgpr7' }
workItemIDX: { reg: '$vgpr0' }
body: |
; CHECK-LABEL: name: handleMoveUp_incorrect_interval
; CHECK: bb.0:
; CHECK: successors: %bb.1(0x80000000)
; CHECK: liveins: $sgpr4_sgpr5
; CHECK: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr4_sgpr5
; CHECK: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN undef %2:vgpr_32, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr101, 0, 0, 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 5)
; CHECK: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sgpr_64 = S_LOAD_DWORDX2_IMM [[COPY]](p4), 0, 0, 0 :: (dereferenceable invariant load 8, align 16, addrspace 4)
; CHECK: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 5329
; CHECK: undef %5.sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
; CHECK: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -4, implicit $exec
; CHECK: [[DEF:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
; CHECK: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK: [[DEF2:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; CHECK: bb.1:
; CHECK: successors: %bb.1(0x80000000)
; CHECK: INLINEASM &"", 1, 851978, def dead %11
; CHECK: GLOBAL_STORE_DWORD undef %12:vreg_64, [[BUFFER_LOAD_DWORD_OFFEN]], 0, 0, 0, 0, implicit $exec :: (store 4, addrspace 1)
[AMDGPU] Remove dubious logic in bidirectional list scheduler Summary: pickNodeBidirectional tried to compare the best top candidate and the best bottom candidate by examining TopCand.Reason and BotCand.Reason. This is unsound because, after calling pickNodeFromQueue, Cand.Reason does not reflect the most important reason why Cand was chosen. Rather it reflects the most recent reason why it beat some other potential candidate, which could have been for some low priority tie breaker reason. I have seen this cause problems where TopCand is a good candidate, but because TopCand.Reason is ORDER (which is very low priority) it is repeatedly ignored in favour of a mediocre BotCand. This is not how bidirectional scheduling is supposed to work. To fix this I changed the code to always compare TopCand and BotCand directly, like the generic implementation of pickNodeBidirectional does. This removes some uncommented AMDGPU-specific logic; if this logic turns out to be important then perhaps it could be moved into an override of tryCandidate instead. Graphics shader benchmarking on gfx10 shows a lot more positive than negative effects from this change. Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
; CHECK: [[V_MOV_B32_e32_2:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK: [[V_MOV_B32_e32_3:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
LiveIntervals: Fix handleMoveUp with subreg def moving across a def If a subregister def was moved across another subregister def and another use, the main range was not correctly updated. The end point of the moved interval ended too early and missed the use from theh other lanes in the subreg def. llvm-svn: 375300
2019-10-19 07:24:25 +08:00
; CHECK: [[DS_READ_B64_gfx9_:%[0-9]+]]:vreg_64 = DS_READ_B64_gfx9 undef %14:vgpr_32, 0, 0, implicit $exec :: (load 8, addrspace 3)
; CHECK: INLINEASM &"def $0 $1", 1, 851978, def %15, 851978, def %16
; CHECK: [[DS_READ_B32_gfx9_:%[0-9]+]]:vgpr_32 = DS_READ_B32_gfx9 [[V_MOV_B32_e32_]], 0, 0, implicit $exec
; CHECK: [[DS_READ_B32_gfx9_1:%[0-9]+]]:vgpr_32 = DS_READ_B32_gfx9 [[V_MOV_B32_e32_1]], 0, 0, implicit $exec
; CHECK: [[DS_READ_B32_gfx9_2:%[0-9]+]]:vgpr_32 = DS_READ_B32_gfx9 undef %20:vgpr_32, 0, 0, implicit $exec
; CHECK: INLINEASM &"def $0 $1", 1, 851978, def %21, 851978, def %22
; CHECK: [[DS_READ_B32_gfx9_3:%[0-9]+]]:vgpr_32 = DS_READ_B32_gfx9 [[V_MOV_B32_e32_1]], 0, 0, implicit $exec
; CHECK: INLINEASM &"", 1, 851978, def dead [[V_MOV_B32_e32_2]], 851978, def dead [[V_MOV_B32_e32_3]], 851977, [[DS_READ_B64_gfx9_]].sub0, 2147483657, [[V_MOV_B32_e32_2]](tied-def 3), 2147549193, [[V_MOV_B32_e32_3]](tied-def 5), 851977, %15, 851977, %16, 851977, [[DS_READ_B32_gfx9_1]], 851977, [[DS_READ_B32_gfx9_]], 851977, [[DS_READ_B32_gfx9_3]], 851977, [[DS_READ_B32_gfx9_2]]
[AMDGPU] Remove dubious logic in bidirectional list scheduler Summary: pickNodeBidirectional tried to compare the best top candidate and the best bottom candidate by examining TopCand.Reason and BotCand.Reason. This is unsound because, after calling pickNodeFromQueue, Cand.Reason does not reflect the most important reason why Cand was chosen. Rather it reflects the most recent reason why it beat some other potential candidate, which could have been for some low priority tie breaker reason. I have seen this cause problems where TopCand is a good candidate, but because TopCand.Reason is ORDER (which is very low priority) it is repeatedly ignored in favour of a mediocre BotCand. This is not how bidirectional scheduling is supposed to work. To fix this I changed the code to always compare TopCand and BotCand directly, like the generic implementation of pickNodeBidirectional does. This removes some uncommented AMDGPU-specific logic; if this logic turns out to be important then perhaps it could be moved into an override of tryCandidate instead. Graphics shader benchmarking on gfx10 shows a lot more positive than negative effects from this change. Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
; CHECK: %5.sub1:vreg_64 = COPY [[V_MOV_B32_e32_]]
LiveIntervals: Fix handleMoveUp with subreg def moving across a def If a subregister def was moved across another subregister def and another use, the main range was not correctly updated. The end point of the moved interval ended too early and missed the use from theh other lanes in the subreg def. llvm-svn: 375300
2019-10-19 07:24:25 +08:00
; CHECK: DS_WRITE_B32_gfx9 undef %28:vgpr_32, %21, 0, 0, implicit $exec :: (store 4, addrspace 3)
; CHECK: DS_WRITE_B32_gfx9 undef %29:vgpr_32, %22, 0, 0, implicit $exec :: (store 4, addrspace 3)
; CHECK: DS_WRITE_B64_gfx9 undef %30:vgpr_32, %5, 0, 0, implicit $exec :: (store 8, addrspace 3)
; CHECK: undef %31.sub1:vreg_64 = FLAT_LOAD_DWORD undef %32:vreg_64, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
[AMDGPU] Remove dubious logic in bidirectional list scheduler Summary: pickNodeBidirectional tried to compare the best top candidate and the best bottom candidate by examining TopCand.Reason and BotCand.Reason. This is unsound because, after calling pickNodeFromQueue, Cand.Reason does not reflect the most important reason why Cand was chosen. Rather it reflects the most recent reason why it beat some other potential candidate, which could have been for some low priority tie breaker reason. I have seen this cause problems where TopCand is a good candidate, but because TopCand.Reason is ORDER (which is very low priority) it is repeatedly ignored in favour of a mediocre BotCand. This is not how bidirectional scheduling is supposed to work. To fix this I changed the code to always compare TopCand and BotCand directly, like the generic implementation of pickNodeBidirectional does. This removes some uncommented AMDGPU-specific logic; if this logic turns out to be important then perhaps it could be moved into an override of tryCandidate instead. Graphics shader benchmarking on gfx10 shows a lot more positive than negative effects from this change. Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
; CHECK: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 1, [[DEF2]], implicit $exec
; CHECK: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
; CHECK: [[DEF]].sub1:vreg_64 = COPY [[V_MOV_B32_e32_]]
LiveIntervals: Fix handleMoveUp with subreg def moving across a def If a subregister def was moved across another subregister def and another use, the main range was not correctly updated. The end point of the moved interval ended too early and missed the use from theh other lanes in the subreg def. llvm-svn: 375300
2019-10-19 07:24:25 +08:00
; CHECK: [[V_MUL_LO_U32_:%[0-9]+]]:vgpr_32 = V_MUL_LO_U32 [[V_ADD_U32_e32_]], [[S_MOV_B32_]], implicit $exec
[AMDGPU] Remove dubious logic in bidirectional list scheduler Summary: pickNodeBidirectional tried to compare the best top candidate and the best bottom candidate by examining TopCand.Reason and BotCand.Reason. This is unsound because, after calling pickNodeFromQueue, Cand.Reason does not reflect the most important reason why Cand was chosen. Rather it reflects the most recent reason why it beat some other potential candidate, which could have been for some low priority tie breaker reason. I have seen this cause problems where TopCand is a good candidate, but because TopCand.Reason is ORDER (which is very low priority) it is repeatedly ignored in favour of a mediocre BotCand. This is not how bidirectional scheduling is supposed to work. To fix this I changed the code to always compare TopCand and BotCand directly, like the generic implementation of pickNodeBidirectional does. This removes some uncommented AMDGPU-specific logic; if this logic turns out to be important then perhaps it could be moved into an override of tryCandidate instead. Graphics shader benchmarking on gfx10 shows a lot more positive than negative effects from this change. Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D68338
2019-10-07 22:33:59 +08:00
; CHECK: [[V_CMP_GT_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_GT_U32_e64 64, [[V_ADD_U32_e32_]], implicit $exec
LiveIntervals: Fix handleMoveUp with subreg def moving across a def If a subregister def was moved across another subregister def and another use, the main range was not correctly updated. The end point of the moved interval ended too early and missed the use from theh other lanes in the subreg def. llvm-svn: 375300
2019-10-19 07:24:25 +08:00
; CHECK: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, [[V_ADD_U32_e32_]], [[V_CMP_GT_U32_e64_]], implicit $exec
; CHECK: [[V_SUB_U32_e32_:%[0-9]+]]:vgpr_32 = V_SUB_U32_e32 [[V_MUL_LO_U32_]], [[DEF1]], implicit $exec
; CHECK: [[V_MUL_LO_U32_1:%[0-9]+]]:vgpr_32 = V_MUL_LO_U32 [[V_CNDMASK_B32_e64_]], [[S_MOV_B32_]], implicit $exec
; CHECK: [[V_ADD_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[V_SUB_U32_e32_]], [[DEF]].sub0, implicit $exec
; CHECK: [[V_SUB_U32_e32_1:%[0-9]+]]:vgpr_32 = V_SUB_U32_e32 [[V_MUL_LO_U32_1]], [[V_MUL_LO_U32_]], implicit $exec
; CHECK: [[DEF]].sub0:vreg_64 = V_ADD_U32_e32 [[V_SUB_U32_e32_1]], [[V_ADD_U32_e32_1]], implicit $exec
; CHECK: undef %38.sub0:vreg_64, %39:sreg_64_xexec = V_ADD_I32_e64 [[S_LOAD_DWORDX2_IMM]].sub0, [[DEF]].sub0, 0, implicit $exec
; CHECK: undef %40.sub1:vreg_64, dead %41:sreg_64_xexec = V_ADDC_U32_e64 [[COPY1]], [[DEF]].sub1, %39, 0, implicit $exec
; CHECK: undef %42.sub0:sgpr_64 = V_READFIRSTLANE_B32 %38.sub0, implicit $exec
; CHECK: %42.sub1:sgpr_64 = V_READFIRSTLANE_B32 %40.sub1, implicit $exec
; CHECK: [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %42, 0, 0, 0 :: (load 4, addrspace 1)
; CHECK: INLINEASM &"", 1
; CHECK: [[DS_READ_B32_gfx9_4:%[0-9]+]]:vgpr_32 = DS_READ_B32_gfx9 undef %45:vgpr_32, 0, 0, implicit $exec :: (load 4, addrspace 3)
; CHECK: GLOBAL_STORE_DWORD undef %46:vreg_64, [[DS_READ_B32_gfx9_4]], 0, 0, 0, 0, implicit $exec :: (store 4, addrspace 1)
; CHECK: %31.sub0:vreg_64 = COPY [[S_LOAD_DWORD_IMM]], implicit $exec
; CHECK: DS_WRITE_B64_gfx9 undef %47:vgpr_32, %31, 0, 0, implicit $exec :: (store 8, addrspace 3)
; CHECK: S_BRANCH %bb.1
bb.0:
liveins: $sgpr4_sgpr5
%0:sgpr_64(p4) = COPY $sgpr4_sgpr5
%1:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN undef %2:vgpr_32, $sgpr96_sgpr97_sgpr98_sgpr99, $sgpr101, 0, 0, 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 5)
%3:sgpr_64 = S_LOAD_DWORDX2_IMM %0(p4), 0, 0, 0 :: (dereferenceable invariant load 8, align 16, addrspace 4)
%4:sreg_32_xm0 = S_MOV_B32 5329
undef %5.sub0:vreg_64 = V_MOV_B32_e32 0, implicit $exec
%6:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
%7:vgpr_32 = V_MOV_B32_e32 -4, implicit $exec
%8:vreg_64 = IMPLICIT_DEF
%9:vgpr_32 = IMPLICIT_DEF
%10:vgpr_32 = IMPLICIT_DEF
bb.1:
INLINEASM &"", 1, 851978, def %11:vgpr_32
GLOBAL_STORE_DWORD undef %12:vreg_64, %1, 0, 0, 0, 0, implicit $exec :: (store 4, addrspace 1)
%13:vreg_64 = DS_READ_B64_gfx9 undef %14:vgpr_32, 0, 0, implicit $exec :: (load 8, addrspace 3)
INLINEASM &"def $0 $1", 1, 851978, def %15:vgpr_32, 851978, def %16:vgpr_32
%17:vgpr_32 = DS_READ_B32_gfx9 %6, 0, 0, implicit $exec
%18:vgpr_32 = DS_READ_B32_gfx9 %7, 0, 0, implicit $exec
%19:vgpr_32 = DS_READ_B32_gfx9 undef %20:vgpr_32, 0, 0, implicit $exec
INLINEASM &"def $0 $1", 1, 851978, def %21:vgpr_32, 851978, def %22:vgpr_32
%23:vgpr_32 = DS_READ_B32_gfx9 %7, 0, 0, implicit $exec
%24:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
%5.sub1:vreg_64 = COPY %6
%25:vgpr_32 = V_ADD_U32_e32 1, %10, implicit $exec
%26:sreg_64_xexec = V_CMP_GT_U32_e64 64, %25, implicit $exec
%27:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
INLINEASM &"", 1, 851978, def dead %24, 851978, def dead %27, 851977, %13.sub0, 2147483657, %24(tied-def 3), 2147549193, %27(tied-def 5), 851977, %15, 851977, %16, 851977, %18, 851977, %17, 851977, %23, 851977, %19
DS_WRITE_B32_gfx9 undef %28:vgpr_32, %21, 0, 0, implicit $exec :: (store 4, addrspace 3)
DS_WRITE_B32_gfx9 undef %29:vgpr_32, %22, 0, 0, implicit $exec :: (store 4, addrspace 3)
DS_WRITE_B64_gfx9 undef %30:vgpr_32, %5, 0, 0, implicit $exec :: (store 8, addrspace 3)
undef %31.sub1:vreg_64 = FLAT_LOAD_DWORD undef %32:vreg_64, 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
%33:vgpr_32 = V_MUL_LO_U32 %25, %4, implicit $exec
%10:vgpr_32 = V_CNDMASK_B32_e64 0, 0, 0, %25, %26, implicit $exec
%34:vgpr_32 = V_SUB_U32_e32 %33, %9, implicit $exec
%9:vgpr_32 = V_MUL_LO_U32 %10, %4, implicit $exec
%35:vgpr_32 = V_ADD_U32_e32 %34, %8.sub0, implicit $exec
%36:vgpr_32 = V_SUB_U32_e32 %9, %33, implicit $exec
%37:vgpr_32 = COPY %3.sub1
undef %8.sub0:vreg_64 = V_ADD_U32_e32 %36, %35, implicit $exec
%8.sub1:vreg_64 = COPY %6
undef %38.sub0:vreg_64, %39:sreg_64_xexec = V_ADD_I32_e64 %3.sub0, %8.sub0, 0, implicit $exec
undef %40.sub1:vreg_64, dead %41:sreg_64_xexec = V_ADDC_U32_e64 %37, %8.sub1, %39, 0, implicit $exec
undef %42.sub0:sgpr_64 = V_READFIRSTLANE_B32 %38.sub0, implicit $exec
%42.sub1:sgpr_64 = V_READFIRSTLANE_B32 %40.sub1, implicit $exec
%43:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %42, 0, 0, 0 :: (load 4, addrspace 1)
INLINEASM &"", 1
%44:vgpr_32 = DS_READ_B32_gfx9 undef %45:vgpr_32, 0, 0, implicit $exec :: (load 4, addrspace 3)
GLOBAL_STORE_DWORD undef %46:vreg_64, %44, 0, 0, 0, 0, implicit $exec :: (store 4, addrspace 1)
%31.sub0:vreg_64 = COPY %43, implicit $exec
DS_WRITE_B64_gfx9 undef %47:vgpr_32, %31, 0, 0, implicit $exec :: (store 8, addrspace 3)
S_BRANCH %bb.1
...