[LoopInterchange] Precommit tests for scalar dependencies. NFC. #115900
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
We are miscompiling and incorrectly interchanging loops with scalar dependencies. This precommits some tests demonstrating this. This is based on the tests in https://reviews.llvm.org/D87879 by `mdchen`.
|
@llvm/pr-subscribers-llvm-transforms Author: Sjoerd Meijer (sjoerdmeijer) ChangesWe are miscompiling and incorrectly interchanging loops with scalar dependencies that are live-out and conditionally set. This precommits some tests demonstrating this. This is based on the tests in https://reviews.llvm.org/D87879 by Full diff: https://github.com/llvm/llvm-project/pull/115900.diff 1 Files Affected:
diff --git a/llvm/test/Transforms/LoopInterchange/gh54176-scalar-deps.ll b/llvm/test/Transforms/LoopInterchange/gh54176-scalar-deps.ll
new file mode 100644
index 00000000000000..b3383655668980
--- /dev/null
+++ b/llvm/test/Transforms/LoopInterchange/gh54176-scalar-deps.ll
@@ -0,0 +1,225 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=loop-interchange -S | FileCheck %s
+
+@f = dso_local local_unnamed_addr global [4 x [9 x i32]] [[9 x i32] [i32 5, i32 3, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0], [9 x i32] zeroinitializer, [9 x i32] zeroinitializer, [9 x i32] zeroinitializer], align 4
+@g = common dso_local local_unnamed_addr global i32 0, align 4
+
+; int32_t f[3][3];
+; int32_t g;
+;
+; int32_t test1(_Bool cond) {
+; for (int64_t i = 0; i < 3; i++) {
+; for (int64_t j = 0; j < 3; j++) {
+; int32_t val = f[i][j];
+; if (val == 0)
+; if (!cond)
+; g++;
+; else
+; g = 3;
+; if (!cond)
+; g++;
+; }
+; }
+; return g;
+; }
+;
+; FIXME: if there's an output dependency inside the loop and Src doesn't
+; dominate Dst, we should not interchange. Thus, this currently miscompiles.
+;
+define dso_local i32 @test1(i1 %cond) {
+; CHECK-LABEL: define dso_local i32 @test1(
+; CHECK-SAME: i1 [[COND:%.*]]) {
+; CHECK-NEXT: [[FOR_PREHEADER:.*:]]
+; CHECK-NEXT: br label %[[INNERLOOP_PREHEADER:.*]]
+; CHECK: [[OUTERLOOP_PREHEADER:.*]]:
+; CHECK-NEXT: br label %[[OUTERLOOP:.*]]
+; CHECK: [[OUTERLOOP]]:
+; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[INDVARS_IV_NEXT21_I:%.*]], %[[FOR_LATCH:.*]] ], [ 0, %[[OUTERLOOP_PREHEADER]] ]
+; CHECK-NEXT: br label %[[INNERLOOP_SPLIT:.*]]
+; CHECK: [[INNERLOOP_PREHEADER]]:
+; CHECK-NEXT: br label %[[INNERLOOP:.*]]
+; CHECK: [[INNERLOOP]]:
+; CHECK-NEXT: [[J:%.*]] = phi i64 [ [[TMP0:%.*]], %[[IF_END_SPLIT:.*]] ], [ 0, %[[INNERLOOP_PREHEADER]] ]
+; CHECK-NEXT: br label %[[OUTERLOOP_PREHEADER]]
+; CHECK: [[INNERLOOP_SPLIT]]:
+; CHECK-NEXT: [[ARRAYIDX6_I:%.*]] = getelementptr inbounds [4 x [9 x i32]], ptr @f, i64 0, i64 [[J]], i64 [[I]]
+; CHECK-NEXT: [[I1:%.*]] = load i32, ptr [[ARRAYIDX6_I]], align 4
+; CHECK-NEXT: [[TOBOOL_I:%.*]] = icmp eq i32 [[I1]], 0
+; CHECK-NEXT: br i1 [[TOBOOL_I]], label %[[LAND_END:.*]], label %[[LAND_RHS:.*]]
+; CHECK: [[LAND_RHS]]:
+; CHECK-NEXT: store i32 3, ptr @g, align 4
+; CHECK-NEXT: br label %[[LAND_END]]
+; CHECK: [[LAND_END]]:
+; CHECK-NEXT: br i1 [[COND]], label %[[IF_END:.*]], label %[[IF_THEN:.*]]
+; CHECK: [[IF_THEN]]:
+; CHECK-NEXT: [[I2:%.*]] = load i32, ptr @g, align 4
+; CHECK-NEXT: [[INC_I:%.*]] = add i32 [[I2]], 1
+; CHECK-NEXT: store i32 [[INC_I]], ptr @g, align 4
+; CHECK-NEXT: br label %[[IF_END]]
+; CHECK: [[IF_END]]:
+; CHECK-NEXT: [[J_NEXT:%.*]] = add nuw nsw i64 [[J]], 1
+; CHECK-NEXT: [[EXITCOND_I:%.*]] = icmp eq i64 [[J_NEXT]], 3
+; CHECK-NEXT: br label %[[FOR_LATCH]]
+; CHECK: [[IF_END_SPLIT]]:
+; CHECK-NEXT: [[TMP0]] = add nuw nsw i64 [[J]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[TMP0]], 3
+; CHECK-NEXT: br i1 [[TMP1]], label %[[EXIT:.*]], label %[[INNERLOOP]]
+; CHECK: [[FOR_LATCH]]:
+; CHECK-NEXT: [[INDVARS_IV_NEXT21_I]] = add nsw i64 [[I]], 1
+; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i64 [[I]], 2
+; CHECK-NEXT: br i1 [[CMP_I]], label %[[OUTERLOOP]], label %[[IF_END_SPLIT]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: [[I3:%.*]] = load i32, ptr @g, align 4
+; CHECK-NEXT: ret i32 [[I3]]
+;
+for.preheader:
+ br label %outerloop
+
+outerloop:
+ %i = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next21.i, %for.latch ]
+ br label %innerloop
+
+innerloop:
+ %j = phi i64 [ 0, %outerloop ], [ %j.next, %if.end ]
+ %arrayidx6.i = getelementptr inbounds [4 x [9 x i32]], [4 x [9 x i32]]* @f, i64 0, i64 %j, i64 %i
+ %i1 = load i32, i32* %arrayidx6.i, align 4
+ %tobool.i = icmp eq i32 %i1, 0
+ br i1 %tobool.i, label %land.end, label %land.rhs
+
+land.rhs:
+ store i32 3, i32* @g, align 4
+ br label %land.end
+
+land.end:
+ br i1 %cond, label %if.end, label %if.then
+
+if.then:
+ %i2 = load i32, i32* @g, align 4
+ %inc.i = add i32 %i2, 1
+ store i32 %inc.i, i32* @g, align 4
+ br label %if.end
+
+if.end:
+ %j.next = add nuw nsw i64 %j, 1
+ %exitcond.i = icmp eq i64 %j.next, 3
+ br i1 %exitcond.i, label %for.latch, label %innerloop
+
+for.latch:
+ %indvars.iv.next21.i = add nsw i64 %i, 1
+ %cmp.i = icmp slt i64 %i, 2
+ br i1 %cmp.i, label %outerloop, label %exit
+
+exit:
+ %i3 = load i32, i32* @g, align 4
+ ret i32 %i3
+}
+
+; int32_t f[3][3];
+; int32_t g;
+;
+; int32_t test2(_Bool cond) {
+; for (int64_t i = 0; i < 3; i++) {
+; for (int64_t j = 0; j < 3; j++) {
+; int32_t val = f[i][j];
+; g = 3;
+; if (val == 0)
+; if (!cond)
+; g++;
+; else
+; if (!cond)
+; g++;
+; }
+; }
+; return g;
+; }
+;
+; FIXME: if there's an output dependency inside the loop and Src doesn't
+; dominate Dst, we should not interchange. Thus, this currently miscompiles.
+;
+define dso_local i32 @test2(i1 %cond) {
+; CHECK-LABEL: define dso_local i32 @test2(
+; CHECK-SAME: i1 [[COND:%.*]]) {
+; CHECK-NEXT: [[FOR_PREHEADER:.*:]]
+; CHECK-NEXT: br label %[[INNERLOOP_PREHEADER:.*]]
+; CHECK: [[OUTERLOOP_PREHEADER:.*]]:
+; CHECK-NEXT: br label %[[OUTERLOOP:.*]]
+; CHECK: [[OUTERLOOP]]:
+; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[INDVARS_IV_NEXT21_I:%.*]], %[[FOR_LATCH:.*]] ], [ 0, %[[OUTERLOOP_PREHEADER]] ]
+; CHECK-NEXT: br label %[[INNERLOOP_SPLIT:.*]]
+; CHECK: [[INNERLOOP_PREHEADER]]:
+; CHECK-NEXT: br label %[[INNERLOOP:.*]]
+; CHECK: [[INNERLOOP]]:
+; CHECK-NEXT: [[J:%.*]] = phi i64 [ [[TMP0:%.*]], %[[IF_END_SPLIT:.*]] ], [ 0, %[[INNERLOOP_PREHEADER]] ]
+; CHECK-NEXT: br label %[[OUTERLOOP_PREHEADER]]
+; CHECK: [[INNERLOOP_SPLIT]]:
+; CHECK-NEXT: [[ARRAYIDX6_I:%.*]] = getelementptr inbounds [4 x [9 x i32]], ptr @f, i64 0, i64 [[J]], i64 [[I]]
+; CHECK-NEXT: [[I1:%.*]] = load i32, ptr [[ARRAYIDX6_I]], align 4
+; CHECK-NEXT: [[TOBOOL_I:%.*]] = icmp eq i32 [[I1]], 0
+; CHECK-NEXT: store i32 3, ptr @g, align 4
+; CHECK-NEXT: br i1 [[TOBOOL_I]], label %[[LAND_END:.*]], label %[[LAND_RHS:.*]]
+; CHECK: [[LAND_RHS]]:
+; CHECK-NEXT: br label %[[LAND_END]]
+; CHECK: [[LAND_END]]:
+; CHECK-NEXT: br i1 [[COND]], label %[[IF_END:.*]], label %[[IF_THEN:.*]]
+; CHECK: [[IF_THEN]]:
+; CHECK-NEXT: [[I2:%.*]] = load i32, ptr @g, align 4
+; CHECK-NEXT: [[INC_I:%.*]] = add i32 [[I2]], 1
+; CHECK-NEXT: store i32 [[INC_I]], ptr @g, align 4
+; CHECK-NEXT: br label %[[IF_END]]
+; CHECK: [[IF_END]]:
+; CHECK-NEXT: [[J_NEXT:%.*]] = add nuw nsw i64 [[J]], 1
+; CHECK-NEXT: [[EXITCOND_I:%.*]] = icmp eq i64 [[J_NEXT]], 3
+; CHECK-NEXT: br label %[[FOR_LATCH]]
+; CHECK: [[IF_END_SPLIT]]:
+; CHECK-NEXT: [[TMP0]] = add nuw nsw i64 [[J]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[TMP0]], 3
+; CHECK-NEXT: br i1 [[TMP1]], label %[[EXIT:.*]], label %[[INNERLOOP]]
+; CHECK: [[FOR_LATCH]]:
+; CHECK-NEXT: [[INDVARS_IV_NEXT21_I]] = add nsw i64 [[I]], 1
+; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i64 [[I]], 2
+; CHECK-NEXT: br i1 [[CMP_I]], label %[[OUTERLOOP]], label %[[IF_END_SPLIT]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: [[I3:%.*]] = load i32, ptr @g, align 4
+; CHECK-NEXT: ret i32 [[I3]]
+;
+for.preheader:
+ br label %outerloop
+
+outerloop:
+ %i = phi i64 [ 0, %for.preheader ], [ %indvars.iv.next21.i, %for.latch ]
+ br label %innerloop
+
+innerloop:
+ %j = phi i64 [ 0, %outerloop ], [ %j.next, %if.end ]
+ %arrayidx6.i = getelementptr inbounds [4 x [9 x i32]], [4 x [9 x i32]]* @f, i64 0, i64 %j, i64 %i
+ %i1 = load i32, i32* %arrayidx6.i, align 4
+ %tobool.i = icmp eq i32 %i1, 0
+ store i32 3, i32* @g, align 4
+ br i1 %tobool.i, label %land.end, label %land.rhs
+
+land.rhs:
+ br label %land.end
+
+land.end:
+ br i1 %cond, label %if.end, label %if.then
+
+if.then:
+ %i2 = load i32, i32* @g, align 4
+ %inc.i = add i32 %i2, 1
+ store i32 %inc.i, i32* @g, align 4
+ br label %if.end
+
+if.end:
+ %j.next = add nuw nsw i64 %j, 1
+ %exitcond.i = icmp eq i64 %j.next, 3
+ br i1 %exitcond.i, label %for.latch, label %innerloop
+
+for.latch:
+ %indvars.iv.next21.i = add nsw i64 %i, 1
+ %cmp.i = icmp slt i64 %i, 2
+ br i1 %cmp.i, label %outerloop, label %exit
+
+exit:
+ %i3 = load i32, i32* @g, align 4
+ ret i32 %i3
+}
|
sjoerdmeijer
changed the title
|
I am going ahead with pre-committing these tests which I think is fine. Next, I am going to look into addressing the functional issue (#54176), so we can discuss things further when I put something up for review. |
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.
We are miscompiling and incorrectly interchanging loops with scalar dependencies that are live-out and conditionally set. This precommits some tests demonstrating this. This is based on the tests in https://reviews.llvm.org/D87879 by
mdchen.