replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Don't count debug instructions towards neighborhood count 

In computeRegisterLiveness, the max instructions to search
was counting dbg_value instructions, which could potentially
cause an observable codegen change from the presence of debug
info.

llvm-svn: 341028
This commit is contained in:
Matt Arsenault 2018-08-30 07:18:19 +00:00
parent 015a147c9f
commit f2edba8e43
2 changed files with 196 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
llvm/lib/CodeGen/MachineBasicBlock.cpp
View File

@ -1379,7 +1379,12 @@ MachineBasicBlock::computeRegisterLiveness(const TargetRegisterInfo *TRI,
const_iterator I(Before);
// If this is the last insn in the block, don't search forwards.
if (I != end()) {
for (++I; I != end() && N > 0; ++I, --N) {
for (++I; I != end() && N > 0; ++I) {
if (I->isDebugInstr())
continue;
--N;
MachineOperandIteratorBase::PhysRegInfo Info =
ConstMIOperands(*I).analyzePhysReg(Reg, TRI);
@ -1416,6 +1421,11 @@ MachineBasicBlock::computeRegisterLiveness(const TargetRegisterInfo *TRI,
do {
--I;
if (I->isDebugInstr())
continue;
--N;
MachineOperandIteratorBase::PhysRegInfo Info =
ConstMIOperands(*I).analyzePhysReg(Reg, TRI);
@ -1440,7 +1450,8 @@ MachineBasicBlock::computeRegisterLiveness(const TargetRegisterInfo *TRI,
// Register must be live if we read it.
if (Info.Read)
return LQR_Live;
} while (I != begin() && --N > 0);
} while (I != begin() && N > 0);
}
// Did we get to the start of the block?
@ -1454,7 +1465,6 @@ MachineBasicBlock::computeRegisterLiveness(const TargetRegisterInfo *TRI,
return LQR_Dead;
}
// At this point we have no idea of the liveness of the register.
return LQR_Unknown;
}

183
llvm/test/CodeGen/AMDGPU/fold-immediate-operand-shrink.mir
View File

@ -407,3 +407,186 @@ body: |
S_ENDPGM implicit %2
...
---
# This requires searching through many DBG_VALUE instructions before the insert poitn, which
# should not count against the search limit.
name: vcc_liveness_dbg_value_search_before
tracksRegLiveness: true
body: |
bb.0:
; GCN-LABEL: name: vcc_liveness_dbg_value_search_before
; GCN: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 12345
; GCN: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: [[V_ADD_I32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_I32_e32 [[S_MOV_B32_]], [[DEF]], implicit-def $vcc, implicit $exec
; GCN: S_ENDPGM implicit [[V_ADD_I32_e32_]]
%0:sreg_32_xm0 = S_MOV_B32 12345
%1:vgpr_32 = IMPLICIT_DEF
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
%2:vgpr_32, %3:sreg_64 = V_ADD_I32_e64 %0, %1, implicit $exec
S_ENDPGM implicit %2
...
---
# This requires searching through many DBG_VALUE instructions after the insert point, which
# should not count against the search limit.
name: vcc_liveness_dbg_value_search_after
tracksRegLiveness: true
body: |
bb.0:
; GCN-LABEL: name: vcc_liveness_dbg_value_search_after
; GCN: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 12345
; GCN: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
; GCN: [[V_ADD_I32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_I32_e32 [[S_MOV_B32_]], [[DEF]], implicit-def $vcc, implicit $exec
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: DBG_VALUE $noreg, 0
; GCN: S_ENDPGM implicit [[V_ADD_I32_e32_]]
%0:sreg_32_xm0 = S_MOV_B32 12345
%1:vgpr_32 = IMPLICIT_DEF
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
S_NOP 0
%2:vgpr_32, %3:sreg_64 = V_ADD_I32_e64 %0, %1, implicit $exec
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
DBG_VALUE $noreg, 0
$vcc = S_MOV_B64 0
S_ENDPGM implicit %2
...