[LV] Amend check for IV increments in collectUsersInEntryBlock #108020
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-llvm-transforms Author: Hari Limaye (hazzlim) Changes…lock The check for IV increments in collectUsersInEntryBlock currently triggers for exit-block PHIs which use the IV start value, resulting in us failing to add the input value for the middle block to these PHIs. Fix this by amending the check for IV increments to only include incoming values that are instructions inside the loop. Fixes #108004 Full diff: https://github.com/llvm/llvm-project/pull/108020.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 2be3b577529258..054a170c36863c 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -8662,6 +8662,7 @@ static MapVector<PHINode *, VPValue *> collectUsersInExitBlock(
!cast<VPWidenIntOrFpInductionRecipe>(V)->getTruncInst()) ||
isa<VPWidenPointerInductionRecipe>(V) ||
(isa<Instruction>(IncomingValue) &&
+ OrigLoop->contains(cast<Instruction>(IncomingValue)) &&
any_of(IncomingValue->users(), [&Inductions](User *U) {
auto *P = dyn_cast<PHINode>(U);
return P && Inductions.contains(P);
diff --git a/llvm/test/Transforms/LoopVectorize/use-iv-start-value.ll b/llvm/test/Transforms/LoopVectorize/use-iv-start-value.ll
new file mode 100644
index 00000000000000..c7ea86533570b0
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/use-iv-start-value.ll
@@ -0,0 +1,84 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S | FileCheck %s
+
+; Check that we correctly handle the use of %start2 in the exit block, and do
+; not crash.
+
+define i64 @foo(i64 %start, i64 %end) {
+; CHECK-LABEL: define i64 @foo(
+; CHECK-SAME: i64 [[START:%.*]], i64 [[END:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[P1:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: [[P2:%.*]] = alloca [1024 x i32], align 4
+; CHECK-NEXT: call void @init_mem(ptr [[P1]], i64 1024)
+; CHECK-NEXT: call void @init_mem(ptr [[P2]], i64 1024)
+; CHECK-NEXT: [[START2:%.*]] = and i64 [[START]], 12345
+; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[END]], [[START2]]
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
+; CHECK-NEXT: [[IND_END:%.*]] = add i64 [[START2]], [[N_VEC]]
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 [[START2]], [[INDEX]]
+; CHECK-NEXT: [[IND:%.*]] = add i64 [[OFFSET_IDX]], 0
+; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i32, ptr [[P1]], i64 [[IND]]
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[ARRAYIDX1]], i32 0
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[P2]], i64 [[IND]]
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[TMP4]], i32 0
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
+; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], %[[MIDDLE_BLOCK]] ], [ [[START2]], %[[ENTRY]] ]
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[IND1:%.*]] = phi i64 [ [[IND_NEXT1:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
+; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[P1]], i64 [[IND1]]
+; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX3]], align 4
+; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[P2]], i64 [[IND1]]
+; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4
+; CHECK-NEXT: [[IND_NEXT1]] = add i64 [[IND1]], 1
+; CHECK-NEXT: [[CMP:%.*]] = icmp ne i64 [[IND_NEXT1]], [[END]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[EXIT]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: [[USE:%.*]] = phi i64 [ [[START2]], %[[FOR_BODY]] ], [ [[START2]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT: ret i64 [[USE]]
+;
+entry:
+ %p1 = alloca [1024 x i32]
+ %p2 = alloca [1024 x i32]
+ call void @init_mem(ptr %p1, i64 1024)
+ call void @init_mem(ptr %p2, i64 1024)
+ %start2 = and i64 %start, 12345
+ br label %for.body
+
+for.body:
+ %ind = phi i64 [ %ind.next, %for.body ], [ %start2, %entry ]
+ %arrayidx1 = getelementptr inbounds i32, ptr %p1, i64 %ind
+ %0 = load i32, ptr %arrayidx1, align 4
+ %arrayidx2 = getelementptr inbounds i32, ptr %p2, i64 %ind
+ %1 = load i32, ptr %arrayidx2, align 4
+ %ind.next = add i64 %ind, 1
+ %cmp = icmp ne i64 %ind.next, %end
+ br i1 %cmp, label %for.body, label %exit
+
+exit:
+ %use = phi i64 [ %start2, %for.body ]
+ ret i64 %use
+}
+
+declare void @init_mem(ptr, i64)
+
+;.
+; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
+; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
+; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
+; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
+;.
|
fhahn
reviewed
Sep 10, 2024
| ; CHECK-NEXT: ret i64 [[USE]] | ||
| ; | ||
| entry: | ||
| %p1 = alloca [1024 x i32] |
There was a problem hiding this comment.
Can p1/p2 instead be passed as function arguments, removing the need to call @init_mem?
There was a problem hiding this comment.
Good point - I've updated the test to pass p1 and p2 as function arguments.
fhahn
approved these changes
Sep 11, 2024
| br label %for.body | ||
|
|
||
| for.body: | ||
| %ind = phi i64 [ %ind.next, %for.body ], [ %start2, %entry ] |
There was a problem hiding this comment.
Suggested change
| %ind = phi i64 [ %ind.next, %for.body ], [ %start2, %entry ] | |
| %ind = phi i64 [ %start2, %entry ], [ %ind.next, %for.body ] |
nit: seems more natural to have the incoming entry for %entry first
|
(looks like the title line of the PR overflowed to the description. Might be good to shorten the title, e.g. |
The check for IV increments in collectUsersInEntryBlock currently triggers for exit-block PHIs which use the IV start value, resulting in us failing to add the input value for the middle block to these PHIs. Fix this by amending the check for IV increments to only include incoming values that are instructions inside the loop.
hazzlim
force-pushed
the
loop-vectorize-fix
branch
from
78656e4 to
65774f5
Compare
hazzlim
changed the title
Nice - done. Thank you for reviewing - I will land this once checks have passed. |
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.
The check for IV increments in collectUsersInEntryBlock currently triggers for exit-block PHIs which use the IV start value, resulting in us failing to add the input value for the middle block to these PHIs.
Fix this by amending the check for IV increments to only include incoming values that are instructions inside the loop.
Fixes #108004