[BranchFolding] Kill common hoisted debug instructions #149999
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-debuginfo Author: Orlando Cazalet-Hyams (OCHyams) ChangesMirror of #140091 with an assertion failure fixed in the latest commit as a DBG_PHI was tripping a checking in Reproducer shared here - #140091 (comment) Full diff: https://github.com/llvm/llvm-project/pull/149999.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/BranchFolding.cpp b/llvm/lib/CodeGen/BranchFolding.cpp
index 3b3e7a418feb5..1cb0a239c3d48 100644
--- a/llvm/lib/CodeGen/BranchFolding.cpp
+++ b/llvm/lib/CodeGen/BranchFolding.cpp
@@ -2083,22 +2083,59 @@ bool BranchFolder::HoistCommonCodeInSuccs(MachineBasicBlock *MBB) {
if (TBB == FBB) {
MBB->splice(Loc, TBB, TBB->begin(), TIB);
} else {
+ // Merge the debug locations, and hoist and kill the debug instructions from
+ // both branches. FIXME: We could probably try harder to preserve some debug
+ // instructions (but at least this isn't producing wrong locations).
+ MachineInstrBuilder MIRBuilder(*MBB->getParent(), Loc);
+ auto HoistAndKillDbgInstr = [MBB, Loc](MachineBasicBlock::iterator DI) {
+ assert(DI->isDebugInstr() && "Expected a debug instruction");
+ if (DI->isDebugRef()) {
+ const TargetInstrInfo *TII =
+ MBB->getParent()->getSubtarget().getInstrInfo();
+ const MCInstrDesc &DBGV = TII->get(TargetOpcode::DBG_VALUE);
+ DI = BuildMI(*MBB->getParent(), DI->getDebugLoc(), DBGV, false, 0,
+ DI->getDebugVariable(), DI->getDebugExpression());
+ MBB->insert(Loc, &*DI);
+ return;
+ }
+ // Deleting a DBG_PHI results in an undef at the referenced DBG_INSTR_REF.
+ if (DI->isDebugPHI()) {
+ DI->eraseFromParent();
+ return;
+ }
+
+ DI->setDebugValueUndef();
+ DI->moveBefore(&*Loc);
+ };
+
// TIB and FIB point to the end of the regions to hoist/merge in TBB and
// FBB.
MachineBasicBlock::iterator FE = FIB;
MachineBasicBlock::iterator FI = FBB->begin();
for (MachineBasicBlock::iterator TI :
make_early_inc_range(make_range(TBB->begin(), TIB))) {
- // Move debug instructions and pseudo probes without modifying them.
- // FIXME: This is the wrong thing to do for debug locations, which
- // should at least be killed (and hoisted from BOTH blocks).
- if (TI->isDebugOrPseudoInstr()) {
- TI->moveBefore(&*Loc);
+ // Hoist and kill debug instructions from FBB. After this loop FI points
+ // to the next non-debug instruction to hoist (checked in assert after the
+ // TBB debug instruction handling code).
+ while (FI->isDebugInstr()) {
+ assert(FI != FE && "Unexpected end of FBB range");
+ MachineBasicBlock::iterator FINext = std::next(FI);
+ HoistAndKillDbgInstr(FI);
+ FI = FINext;
+ }
+
+ // Kill debug instructions before moving.
+ if (TI->isDebugInstr()) {
+ HoistAndKillDbgInstr(TI);
continue;
}
- // Get the next non-meta instruction in FBB.
+ // If FI is a debug instruction, skip forward to the next non-debug
+ // instruction.
FI = skipDebugInstructionsForward(FI, FE, false);
+ // Pseudo probes are excluded from the range when identifying foldable
+ // instructions, so we don't expect to see one now.
+ assert(!TI->isPseudoProbe() && "Unexpected pseudo probe in range");
// NOTE: The loop above checks CheckKillDead but we can't do that here as
// it modifies some kill markers after the check.
assert(TI->isIdenticalTo(*FI, MachineInstr::CheckDefs) &&
@@ -2111,6 +2148,7 @@ bool BranchFolder::HoistCommonCodeInSuccs(MachineBasicBlock *MBB) {
++FI;
}
}
+
FBB->erase(FBB->begin(), FIB);
if (UpdateLiveIns)
diff --git a/llvm/test/DebugInfo/X86/branch-folder-dbg.mir b/llvm/test/DebugInfo/X86/branch-folder-dbg.mir
index 5c38fd2a43829..717bfc28b4065 100644
--- a/llvm/test/DebugInfo/X86/branch-folder-dbg.mir
+++ b/llvm/test/DebugInfo/X86/branch-folder-dbg.mir
@@ -1,16 +1,23 @@
# RUN: llc %s --start-before=branch-folder --stop-after=branch-folder -o - | FileCheck %s
## Check that common instructions hoisted from `if.then` and `if.else` into
-## common pred `entry` get merged debug locations.
-
-## FIXME: The debug instructions handling here is wrong.
+## common pred `entry` get merged debug locations. The debug instructions from
+## both branches should get hoisted and killed.
+##
+## The MIR debug instructions have been modified by hand in order to check they
+## can be killed.
+##
+## Check DBG_PHIs are deleted rather than hoisted.
# CHECK: bb.0
# CHECK: CALL64pcrel32 @f, csr_64, implicit $rsp, implicit $ssp, implicit-def $rsp, implicit-def $ssp, implicit-def $rax
-## --- Start splice from bb.2.if.else ---
-# CHECK-NEXT: DBG_VALUE 2, $noreg, ![[#]], !DIExpression(), debug-location ![[#]]
-# CHECK-NEXT: $edi = MOV32r0 implicit-def dead $eflags, debug-location !DILocation(line: 0, scope: ![[#]])
-## --- End splice --------------
+## --- Start splice from bb.2.if.else (and debug instructions from bb.1.if.then) ---
+# CHECK-NEXT: DBG_VALUE $noreg, $noreg, ![[#]], !DIExpression(), debug-location ![[#]]
+# CHECK-NEXT: DBG_VALUE $noreg, $noreg, ![[#]], !DIExpression(), debug-location ![[#]]
+# CHECK-NEXT: $edi = MOV32r0 implicit-def dead $eflags, debug-instr-number 2, debug-location !DILocation(line: 0, scope: ![[#]])
+# CHECK-NEXT: DBG_VALUE $noreg, $noreg, ![[#]], !DIExpression(DW_OP_LLVM_arg, 0), debug-location ![[#]]
+# CHECK-NEXT: DBG_VALUE $noreg, $noreg, ![[#]], !DIExpression(DW_OP_LLVM_arg, 0), debug-location ![[#]]
+## --- End splice ------------------------------------------------------------------
# CHECK-NEXT: TEST64rr killed renamable $rax, renamable $rax, implicit-def $eflags
# CHECK-NEXT: JCC_1 %bb.2, 9, implicit killed $eflags
# CHECK: bb.1
@@ -73,6 +80,8 @@
...
---
name: g
+tracksRegLiveness: true
+isSSA: false
body: |
bb.0 (%ir-block.0):
successors: %bb.1(0x40000000), %bb.2(0x40000000)
@@ -87,21 +96,23 @@ body: |
bb.1.if.then:
successors: %bb.3(0x80000000)
- DBG_VALUE 0, $noreg, !11, !DIExpression(), debug-location !13
- $edi = MOV32r0 implicit-def dead $eflags, debug-location !14
+ DBG_PHI $esp, 3
+ DBG_VALUE $esi, $noreg, !11, !DIExpression(), debug-location !13
+ $edi = MOV32r0 implicit-def dead $eflags, debug-instr-number 1, debug-location !14
+ DBG_INSTR_REF !11, !DIExpression(DW_OP_LLVM_arg, 0), dbg-instr-ref(1, 0), debug-location !13
$esi = MOV32r0 implicit-def dead $eflags, debug-location !14
CALL64pcrel32 target-flags(x86-plt) @_Z3fooii, csr_64, implicit $rsp, implicit $ssp, implicit killed $edi, implicit killed $esi, implicit-def $rsp, implicit-def $ssp, debug-location !14
- DBG_VALUE 1, $noreg, !11, !DIExpression(), debug-location !13
JMP_1 %bb.3, debug-location !15
bb.2.if.else:
successors: %bb.3(0x80000000)
- DBG_VALUE 2, $noreg, !11, !DIExpression(), debug-location !13
- $edi = MOV32r0 implicit-def dead $eflags, debug-location !16
+ DBG_PHI $esp, 4
+ DBG_VALUE $esp, $noreg, !11, !DIExpression(), debug-location !13
+ $edi = MOV32r0 implicit-def dead $eflags, debug-instr-number 2, debug-location !16
+ DBG_INSTR_REF !11, !DIExpression(DW_OP_LLVM_arg, 0), dbg-instr-ref(2, 0), debug-location !13
$esi = MOV32ri 1, debug-location !16
CALL64pcrel32 target-flags(x86-plt) @_Z3barii, csr_64, implicit $rsp, implicit $ssp, implicit killed $edi, implicit killed $esi, implicit-def $rsp, implicit-def $ssp, debug-location !16
- DBG_VALUE 3, $noreg, !11, !DIExpression(), debug-location !13
bb.3.if.end:
$eax = MOV32ri 2
|
jmorse
approved these changes
Jul 25, 2025
branch-folder hoists common instructions from TBB and FBB into their pred. Without this patch it achieves this by splicing the instructions from TBB and deleting the common ones in FBB. That moves the debug locations and debug instructions from TBB into the pred without modification, which is not ideal. Debug locations are handled in llvm#140063. This patch handles debug instructions - in the simplest way possible, which is to just kill (undef) them. We kill and hoist the ones in FBB as well as TBB because otherwise the fact there's an assignment on the code path is deleted (which might lead to a prior location extending further than it should). There's possibly something we could do to preserve some variable locations in some cases, but this is the easiest not-incorrect thing to do. Note I had to replace the constant DBG_VALUEs to use registers in the test- it turns out setDebugValueUndef doesn't undef constant DBG_VALUEs... which feels wrong to me, but isn't something I want to touch right now.
OCHyams
force-pushed
the
branch-fold-dbg-instr2
branch
from
337dfbe to
f1aaf4d
Compare
|
Added an implicit-check-not for good measure |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/11/builds/20364 Here is the relevant piece of the build log for the reference |
This seems unrelated |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/139/builds/17622 Here is the relevant piece of the build log for the reference |
OCHyams
added a commit
that referenced
this pull request
Jul 25, 2025
|
That's more condemning. Reverted for now #150632. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Mirror of #140091 with an assertion failure fixed in the latest commit as a DBG_PHI was tripping a checking in
setDebugValueUndef- @jmorse does the last commit look alright to you?Reproducer shared here - #140091 (comment)