[DA] Add tests for GCD MIV misses dependency due to overflow (NFC) #169926
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
Contributor
Author
This stack of pull requests is managed by Graphite. Learn more about stacking. |
This was referenced Nov 28, 2025
kasuga-fj
commented
Nov 29, 2025
Comment on lines
+161
to
+162
| %guard = icmp ne i64 %const, 0 | ||
| br i1 %guard, label %loop, label %exit |
Contributor
Author
There was a problem hiding this comment.
This is a hack to transfer the nsw flag of 3 * %m from IR to SCEV. Please let me know if there is a better way to do this.
llvmbot
added
the
llvm:analysis
kasuga-fj
changed the title
Member
|
@llvm/pr-subscribers-llvm-analysis Author: Ryotaro Kasuga (kasuga-fj) ChangesAdd two test cases where dependencies are missed due to overflows. These will be fixed by #169927 and #169928, respectively. Full diff: https://github.com/llvm/llvm-project/pull/169926.diff 1 Files Affected:
diff --git a/llvm/test/Analysis/DependenceAnalysis/gcd-miv-overflow.ll b/llvm/test/Analysis/DependenceAnalysis/gcd-miv-overflow.ll
index 9169ac323d834..618d14c75faad 100644
--- a/llvm/test/Analysis/DependenceAnalysis/gcd-miv-overflow.ll
+++ b/llvm/test/Analysis/DependenceAnalysis/gcd-miv-overflow.ll
@@ -56,6 +56,125 @@ loop:
%ec = icmp eq i64 %i.inc, 100
br i1 %ec, label %exit, label %loop
+exit:
+ ret void
+}
+
+; max_i = INT64_MAX/6 // 1537228672809129301
+; for (long long i = 0; i <= max_i; i++) {
+; A[-6*i + INT64_MAX] = 0;
+; if (i)
+; A[3*i - 2] = 1;
+; }
+;
+; FIXME: DependenceAnalysis currently detects no dependency between the two
+; stores, but it does exist. For example,
+;
+; memory access | i == 1 | i == (max_i + 1) / 2 | i == max_i
+; ---------------------|--------|----------------------|-------------------
+; A[-6*i + INT64_MAX] | | A[3*max_i - 2] | A[1]
+; A[3*i - 2] | A[1] | | A[3*max_i - 2]
+;
+;
+define void @gcdmiv_delta_ovfl(ptr %A) {
+; CHECK-ALL-LABEL: 'gcdmiv_delta_ovfl'
+; CHECK-ALL-NEXT: Src: store i8 0, ptr %idx.0, align 1 --> Dst: store i8 0, ptr %idx.0, align 1
+; CHECK-ALL-NEXT: da analyze - none!
+; CHECK-ALL-NEXT: Src: store i8 0, ptr %idx.0, align 1 --> Dst: store i8 1, ptr %idx.1, align 1
+; CHECK-ALL-NEXT: da analyze - none!
+; CHECK-ALL-NEXT: Src: store i8 1, ptr %idx.1, align 1 --> Dst: store i8 1, ptr %idx.1, align 1
+; CHECK-ALL-NEXT: da analyze - none!
+;
+; CHECK-GCD-MIV-LABEL: 'gcdmiv_delta_ovfl'
+; CHECK-GCD-MIV-NEXT: Src: store i8 0, ptr %idx.0, align 1 --> Dst: store i8 0, ptr %idx.0, align 1
+; CHECK-GCD-MIV-NEXT: da analyze - consistent output [*]!
+; CHECK-GCD-MIV-NEXT: Src: store i8 0, ptr %idx.0, align 1 --> Dst: store i8 1, ptr %idx.1, align 1
+; CHECK-GCD-MIV-NEXT: da analyze - none!
+; CHECK-GCD-MIV-NEXT: Src: store i8 1, ptr %idx.1, align 1 --> Dst: store i8 1, ptr %idx.1, align 1
+; CHECK-GCD-MIV-NEXT: da analyze - consistent output [*]!
+;
+entry:
+ br label %loop.header
+
+loop.header:
+ %i = phi i64 [ 0, %entry ], [ %i.inc, %loop.latch ]
+ %subscript.0 = phi i64 [ 9223372036854775807, %entry ], [ %subscript.0.next, %loop.latch ]
+ %subscript.1 = phi i64 [ -2, %entry ], [ %subscript.1.next, %loop.latch ]
+ %idx.0 = getelementptr inbounds i8, ptr %A, i64 %subscript.0
+ store i8 0, ptr %idx.0
+ %cond.store = icmp ne i64 %i, 0
+ br i1 %cond.store, label %if.store, label %loop.latch
+
+if.store:
+ %idx.1 = getelementptr i8, ptr %A, i64 %subscript.1
+ store i8 1, ptr %idx.1
+ br label %loop.latch
+
+loop.latch:
+ %i.inc = add nuw nsw i64 %i, 1
+ %subscript.0.next = add nsw i64 %subscript.0, -6
+ %subscript.1.next = add nsw i64 %subscript.1, 3
+ %exitcond = icmp sgt i64 %i.inc, 1537228672809129301
+ br i1 %exitcond, label %exit, label %loop.header
+
+exit:
+ ret void
+}
+
+; max_i = INT64_MAX/3 // 3074457345618258602
+; for (i = 0; i < max_i; i++) {
+; A[3*i] = 0;
+; A[-3*i - 3*m - INT64_MAX] = 1;
+; }
+;
+; FIXME: DependenceAnalysis currently detects no dependency between the two
+; stores, but it may exist. For example, consider `m = 1`. Then,
+; `-3*m - INT64_MAX` is `INT64_MAX - 1`. So `-3*i - 3*m - INT64_MAX` is
+; effectively `-3*i + (INT64_MAX - 1)`. Thus, accesses will be as follows:
+;
+; memory access | i == 1 | i == max_i - 1
+; ---------------------------|------------------|------------------
+; A[3*i] | A[3] | A[INT64_MAX - 4]
+; A[-3*i - 3*m - INT64_MAX] | A[INT64_MAX - 4] | A[3]
+;
+
+define void @gcdmiv_const_ovfl(ptr %A, i64 %m) {
+; CHECK-ALL-LABEL: 'gcdmiv_const_ovfl'
+; CHECK-ALL-NEXT: Src: store i8 0, ptr %gep.0, align 1 --> Dst: store i8 0, ptr %gep.0, align 1
+; CHECK-ALL-NEXT: da analyze - none!
+; CHECK-ALL-NEXT: Src: store i8 0, ptr %gep.0, align 1 --> Dst: store i8 1, ptr %gep.1, align 1
+; CHECK-ALL-NEXT: da analyze - none!
+; CHECK-ALL-NEXT: Src: store i8 1, ptr %gep.1, align 1 --> Dst: store i8 1, ptr %gep.1, align 1
+; CHECK-ALL-NEXT: da analyze - none!
+;
+; CHECK-GCD-MIV-LABEL: 'gcdmiv_const_ovfl'
+; CHECK-GCD-MIV-NEXT: Src: store i8 0, ptr %gep.0, align 1 --> Dst: store i8 0, ptr %gep.0, align 1
+; CHECK-GCD-MIV-NEXT: da analyze - consistent output [*]!
+; CHECK-GCD-MIV-NEXT: Src: store i8 0, ptr %gep.0, align 1 --> Dst: store i8 1, ptr %gep.1, align 1
+; CHECK-GCD-MIV-NEXT: da analyze - none!
+; CHECK-GCD-MIV-NEXT: Src: store i8 1, ptr %gep.1, align 1 --> Dst: store i8 1, ptr %gep.1, align 1
+; CHECK-GCD-MIV-NEXT: da analyze - consistent output [*]!
+;
+entry:
+ %m.3 = mul nsw i64 -3, %m
+ %const = sub i64 %m.3, 9223372036854775807
+ %guard = icmp ne i64 %const, 0
+ br i1 %guard, label %loop, label %exit
+
+loop:
+ %i = phi i64 [ 0, %entry ], [ %i.inc, %loop ]
+ %offset.0 = phi i64 [ 0, %entry ], [ %offset.0.next, %loop ]
+ %offset.1 = phi i64 [ %const, %entry ], [ %offset.1.next, %loop ]
+ %gep.0 = getelementptr inbounds i8, ptr %A, i64 %offset.0
+ %gep.1 = getelementptr inbounds i8, ptr %A, i64 %offset.1
+ store i8 0, ptr %gep.0
+ store i8 1, ptr %gep.1
+ %i.inc = add nuw nsw i64 %i, 1
+ %offset.0.next = add nsw i64 %offset.0, 3
+ %offset.1.next = add nsw i64 %offset.1, -3
+ %ec = icmp eq i64 %i.inc, 3074457345618258602
+ br i1 %ec, label %exit, label %loop
+
exit:
ret void
}
|
aahrun
pushed a commit
to aahrun/llvm-project
that referenced
this pull request
Dec 1, 2025
…lvm#169926) Add two test cases where dependencies are missed due to overflows. These will be fixed by llvm#169927 and llvm#169928, respectively.
kasuga-fj
added a commit
that referenced
this pull request
Dec 1, 2025
Add overflow check when computing `Delta` in `gcdMIVtest`. Fix one of the tests added by #169926.
kasuga-fj
added a commit
that referenced
this pull request
Dec 1, 2025
In `gcdMIVtest`, there is logic that assumes the addition(s) of `SCEVAddExpr` don't overflow without any checks. Adding overflow checks would be fine, but this part appeart to be less useful. So this patch removes it. Fix one of the tests added in #169926.
augusto2112
pushed a commit
to augusto2112/llvm-project
that referenced
this pull request
Dec 3, 2025
…lvm#169926) Add two test cases where dependencies are missed due to overflows. These will be fixed by llvm#169927 and llvm#169928, respectively.
augusto2112
pushed a commit
to augusto2112/llvm-project
that referenced
this pull request
Dec 3, 2025
Add overflow check when computing `Delta` in `gcdMIVtest`. Fix one of the tests added by llvm#169926.
augusto2112
pushed a commit
to augusto2112/llvm-project
that referenced
this pull request
Dec 3, 2025
In `gcdMIVtest`, there is logic that assumes the addition(s) of `SCEVAddExpr` don't overflow without any checks. Adding overflow checks would be fine, but this part appeart to be less useful. So this patch removes it. Fix one of the tests added in llvm#169926.
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.
Add two test cases where dependencies are missed due to overflows. These will be fixed by #169927 and #169928, respectively.