[KeyInstr][LoopUnroll] Remap atoms while unrolling #133489
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
This was referenced Mar 28, 2025
This was referenced Mar 28, 2025
|
@llvm/pr-subscribers-debuginfo @llvm/pr-subscribers-llvm-transforms Author: Orlando Cazalet-Hyams (OCHyams) ChangesFull diff: https://github.com/llvm/llvm-project/pull/133489.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/LoopUnroll.cpp b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
index 45b49671dd3b6..1fa4eddc459c0 100644
--- a/llvm/lib/Transforms/Utils/LoopUnroll.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
@@ -752,6 +752,14 @@ llvm::UnrollLoop(Loop *L, UnrollLoopOptions ULO, LoopInfo *LI,
}
}
+ // Remap source location atom instance. Do this now, rather than
+ // when we remap instructions, because remap is called once we've
+ // cloned all blocks (all the clones would get the same atom
+ // number).
+ if (!VMap.AtomMap.empty())
+ for (Instruction &I : *New)
+ RemapSourceAtom(&I, VMap);
+
// Update our running map of newest clones
LastValueMap[*BB] = New;
for (ValueToValueMapTy::iterator VI = VMap.begin(), VE = VMap.end();
@@ -802,7 +810,8 @@ llvm::UnrollLoop(Loop *L, UnrollLoopOptions ULO, LoopInfo *LI,
}
}
- // Remap all instructions in the most recent iteration
+ // Remap all instructions in the most recent iteration.
+ // Key Instructions: Nothing to do - we've already remapped the atoms.
remapInstructionsInBlocks(NewBlocks, LastValueMap);
for (BasicBlock *NewBlock : NewBlocks)
for (Instruction &I : *NewBlock)
diff --git a/llvm/test/DebugInfo/KeyInstructions/Generic/loop-unroll-runtime.ll b/llvm/test/DebugInfo/KeyInstructions/Generic/loop-unroll-runtime.ll
new file mode 100644
index 0000000000000..6deb04cc00f8d
--- /dev/null
+++ b/llvm/test/DebugInfo/KeyInstructions/Generic/loop-unroll-runtime.ll
@@ -0,0 +1,69 @@
+
+; RUN: opt %s -S --passes=loop-unroll -unroll-runtime=true -unroll-count=4 -unroll-remainder -o - \
+; RUN: | FileCheck %s
+
+;; Check atoms are remapped for runtime unrolling.
+
+; CHECK: for.body.epil:
+; CHECK-NEXT: store i64 %indvars.iv.unr, ptr %p, align 4, !dbg [[G2R1:!.*]]
+
+; CHECK: for.body.epil.1:
+; CHECK-NEXT: store i64 %indvars.iv.next.epil, ptr %p, align 4, !dbg [[G3R1:!.*]]
+
+; CHECK: for.body.epil.2:
+; CHECK-NEXT: store i64 %indvars.iv.next.epil.1, ptr %p, align 4, !dbg [[G4R1:!.*]]
+
+; CHECK: for.body:
+; CHECK-NEXT: %indvars.iv = phi i64 [ 0, %for.body.lr.ph.new ], [ %indvars.iv.next.3, %for.body ]
+; CHECK-NEXT: %niter = phi i64 [ 0, %for.body.lr.ph.new ], [ %niter.next.3, %for.body ]
+; CHECK-NEXT: store i64 %indvars.iv, ptr %p, align 4, !dbg [[G1R1:!.*]]
+; CHECK-NEXT: %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+; CHECK-NEXT: store i64 %indvars.iv.next, ptr %p, align 4, !dbg [[G5R1:!.*]]
+; CHECK-NEXT: %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
+; CHECK-NEXT: store i64 %indvars.iv.next.1, ptr %p, align 4, !dbg [[G6R1:!.*]]
+; CHECK-NEXT: %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv, 3
+; CHECK-NEXT: store i64 %indvars.iv.next.2, ptr %p, align 4, !dbg [[G7R1:!.*]]
+; CHECK-NEXT: %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv, 4
+
+; CHECK: [[G2R1]] = !DILocation({{.*}}, atomGroup: 2, atomRank: 1)
+; CHECK: [[G3R1]] = !DILocation({{.*}}, atomGroup: 3, atomRank: 1)
+; CHECK: [[G4R1]] = !DILocation({{.*}}, atomGroup: 4, atomRank: 1)
+; CHECK: [[G1R1]] = !DILocation({{.*}}, atomGroup: 1, atomRank: 1)
+; CHECK: [[G5R1]] = !DILocation({{.*}}, atomGroup: 5, atomRank: 1)
+; CHECK: [[G6R1]] = !DILocation({{.*}}, atomGroup: 6, atomRank: 1)
+; CHECK: [[G7R1]] = !DILocation({{.*}}, atomGroup: 7, atomRank: 1)
+
+define i32 @unroll(ptr %p, i32 %N) local_unnamed_addr !dbg !5 {
+entry:
+ %cmp9 = icmp eq i32 %N, 0
+ br i1 %cmp9, label %for.cond.cleanup, label %for.body.lr.ph
+
+for.body.lr.ph: ; preds = %entry
+ %wide.trip.count = zext i32 %N to i64
+ br label %for.body
+
+for.cond.cleanup: ; preds = %for.body, %entry
+ %r = phi i32 [ 0, %entry ], [ 1, %for.body ]
+ ret i32 %r
+
+for.body: ; preds = %for.body, %for.body.lr.ph
+ %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
+ store i64 %indvars.iv, ptr %p, !dbg !8
+ %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+ %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count
+ br i1 %exitcond, label %for.cond.cleanup, label %for.body
+}
+
+!llvm.dbg.cu = !{!0}
+!llvm.debugify = !{!2, !3}
+!llvm.module.flags = !{!4}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "debugify", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug)
+!1 = !DIFile(filename: "test.ll", directory: "/")
+!2 = !{i32 17}
+!3 = !{i32 0}
+!4 = !{i32 2, !"Debug Info Version", i32 3}
+!5 = distinct !DISubprogram(name: "unroll", linkageName: "unroll", scope: null, file: !1, line: 1, type: !6, scopeLine: 1, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!6 = !DISubroutineType(types: !7)
+!7 = !{}
+!8 = !DILocation(line: 1, column: 1, scope: !5, atomGroup: 1, atomRank: 1)
diff --git a/llvm/test/DebugInfo/KeyInstructions/Generic/loop-unroll.ll b/llvm/test/DebugInfo/KeyInstructions/Generic/loop-unroll.ll
new file mode 100644
index 0000000000000..557e8387e5bc5
--- /dev/null
+++ b/llvm/test/DebugInfo/KeyInstructions/Generic/loop-unroll.ll
@@ -0,0 +1,48 @@
+; RUN: opt %s --passes=loop-unroll -S -o - -S | FileCheck %s
+
+; CHECK: store i32 %a, ptr %p, align 4, !dbg [[G1R1:!.*]]
+; CHECK: %v.1 = add i32 1, %a, !dbg [[G2R2:!.*]]
+; CHECK: store i32 %v.1, ptr %p, align 4, !dbg [[G2R1:!.*]]
+; CHECK: %v.2 = add i32 2, %a, !dbg [[G3R2:!.*]]
+; CHECK: store i32 %v.2, ptr %p, align 4, !dbg [[G3R1:!.*]]
+
+; CHECK: [[G1R1]] = !DILocation({{.*}}, atomGroup: 1, atomRank: 1)
+; CHECK: [[G2R2]] = !DILocation({{.*}}, atomGroup: 2, atomRank: 2)
+; CHECK: [[G2R1]] = !DILocation({{.*}}, atomGroup: 2, atomRank: 1)
+; CHECK: [[G3R2]] = !DILocation({{.*}}, atomGroup: 3, atomRank: 2)
+; CHECK: [[G3R1]] = !DILocation({{.*}}, atomGroup: 3, atomRank: 1)
+
+define void @f(ptr %p, i32 %a) local_unnamed_addr #0 !dbg !5 {
+entry:
+ br label %for.body
+
+for.cond.cleanup: ; preds = %for.body
+ ret void
+
+for.body: ; preds = %for.body, %entry
+ %i = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+ %v = add i32 %i, %a, !dbg !11
+ store i32 %v, ptr %p, align 4, !dbg !12
+ %inc = add i32 %i, 1
+ %cmp = icmp slt i32 %inc, 3
+ br i1 %cmp, label %for.body, label %for.cond.cleanup
+}
+
+declare i32 @bar(...) local_unnamed_addr
+
+attributes #0 = { nounwind ssp uwtable }
+
+!llvm.dbg.cu = !{!0}
+!llvm.debugify = !{!2, !3}
+!llvm.module.flags = !{!4}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "debugify", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug)
+!1 = !DIFile(filename: "/home/och/scratch/test.ll", directory: "/")
+!2 = !{i32 8}
+!3 = !{i32 0}
+!4 = !{i32 2, !"Debug Info Version", i32 3}
+!5 = distinct !DISubprogram(name: "f", linkageName: "f", scope: null, file: !1, line: 1, type: !6, scopeLine: 1, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!6 = !DISubroutineType(types: !7)
+!7 = !{}
+!11 = !DILocation(line: 4, column: 1, scope: !5, atomGroup: 1, atomRank: 2)
+!12 = !DILocation(line: 5, column: 1, scope: !5, atomGroup: 1, atomRank: 1)
|
This was referenced Apr 7, 2025
This was referenced Apr 7, 2025
jmorse
reviewed
Apr 11, 2025
Comment on lines
+755
to
+758
| // Remap source location atom instance. Do this now, rather than | ||
| // when we remap instructions, because remap is called once we've | ||
| // cloned all blocks (all the clones would get the same atom | ||
| // number). |
There was a problem hiding this comment.
I get why all the clones need to get distinct atom numbers; what is it here that ensures that's the case, don't we need to reset something in VMap to ensure that's the case? i.e., if we called RemapSourceAtom later what would be different?
(I get the overall picture, but not how it's achieved here.)
There was a problem hiding this comment.
CloneBasicBlock maps the atoms from the cloned block so new ones can be remapped in the duplicated block (the next unroll iteration). It stores that info in VMap. We need to update those instructions now using that info.
remapInstructionsInBlocks (which does also call RemapSourceAtom) which is called later for all cloned is passed LastValueMap. That map just has the "running map" of new cloned instructions, i.e., values are overwritten after each unrolled iteration. If we used that infra to remap source atoms, we'd just remap all atoms x to y, in all the unrolled blocks. But what we want is x -> y1, x->y2, with a unique value in each unrolled iteration.
Does that make any sense? I fear I'm mincing this explanation.
There was a problem hiding this comment.
Ah right -- it's the fact that Something Happens (TM) during CloneBasicBlock that I think I was missing.
OCHyams
force-pushed
the
users/OCHyams/ki-llvm-jumpthreading3
branch
2 times, most recently
from
4dda51f to
3c02939
Compare
OCHyams
force-pushed
the
users/OCHyams/ki-llvm-loopunroll
branch
from
fc36dec to
520151b
Compare
jmorse
approved these changes
May 9, 2025
Comment on lines
+755
to
+758
| // Remap source location atom instance. Do this now, rather than | ||
| // when we remap instructions, because remap is called once we've | ||
| // cloned all blocks (all the clones would get the same atom | ||
| // number). |
There was a problem hiding this comment.
Ah right -- it's the fact that Something Happens (TM) during CloneBasicBlock that I think I was missing.
OCHyams
added a commit
to OCHyams/llvm-project
that referenced
this pull request
May 9, 2025
OCHyams
added a commit
to OCHyams/llvm-project
that referenced
this pull request
May 9, 2025
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.
No description provided.