[Pipeline] Eliminate dead loops introduced by InstCombine #69073
Conversation
This file contains 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
dtcxzyw
requested review from
nikic,
topperc,
paulwalker-arm and
sdesmalen-arm
|
@llvm/pr-subscribers-llvm-transforms Author: Yingwei Zheng (dtcxzyw) ChangesThis PR adds a Fixes #68282. Full diff: https://github.com/llvm/llvm-project/pull/69073.diff 7 Files Affected:
diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index 600f8d43caaf216..011d0119d18ade1 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -1280,10 +1280,14 @@ void PassBuilder::addVectorPasses(OptimizationLevel Level,
// divide result.
// 2. It helps to clean up some loop-invariant code created by the loop
// unroll pass when IsFullLTO=false.
- FPM.addPass(createFunctionToLoopPassAdaptor(
- LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap,
- /*AllowSpeculation=*/true),
- /*UseMemorySSA=*/true, /*UseBlockFrequencyInfo=*/false));
+ // 3. It deletes dead loops exposed by instcombine.
+ LoopPassManager LPM;
+ LPM.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap,
+ /*AllowSpeculation=*/true));
+ LPM.addPass(LoopDeletionPass());
+ FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM),
+ /*UseMemorySSA=*/true,
+ /*UseBlockFrequencyInfo=*/false));
// Now that we've vectorized and unrolled loops, we may have more refined
// alignment information, try to re-derive it here.
diff --git a/llvm/test/Other/new-pm-defaults.ll b/llvm/test/Other/new-pm-defaults.ll
index 098db2b959a29ec..68af447faef6293 100644
--- a/llvm/test/Other/new-pm-defaults.ll
+++ b/llvm/test/Other/new-pm-defaults.ll
@@ -265,6 +265,7 @@
; CHECK-O-NEXT: Running pass: LoopSimplifyPass
; CHECK-O-NEXT: Running pass: LCSSAPass
; CHECK-O-NEXT: Running pass: LICMPass
+; CHECK-O-NEXT: Running pass: LoopDeletionPass
; CHECK-O-NEXT: Running pass: AlignmentFromAssumptionsPass
; CHECK-O-NEXT: Running pass: LoopSinkPass
; CHECK-O-NEXT: Running pass: InstSimplifyPass
diff --git a/llvm/test/Other/new-pm-lto-defaults.ll b/llvm/test/Other/new-pm-lto-defaults.ll
index d451d2897f673cd..f627a0079fd6fae 100644
--- a/llvm/test/Other/new-pm-lto-defaults.ll
+++ b/llvm/test/Other/new-pm-lto-defaults.ll
@@ -134,6 +134,7 @@
; CHECK-O23SZ-NEXT: Running pass: LoopSimplifyPass
; CHECK-O23SZ-NEXT: Running pass: LCSSAPass
; CHECK-O23SZ-NEXT: Running pass: LICMPass
+; CHECK-O23SZ-NEXT: Running pass: LoopDeletionPass
; CHECK-O23SZ-NEXT: Running pass: AlignmentFromAssumptionsPass on foo
; CHECK-EP-Peephole-NEXT: Running pass: NoOpFunctionPass on foo
; CHECK-O23SZ-NEXT: Running pass: JumpThreadingPass on foo
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
index 1103c8f984fa296..28b68dc80b1cabd 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
@@ -192,6 +192,7 @@
; CHECK-POSTLINK-O-NEXT: Running pass: LoopSimplifyPass
; CHECK-POSTLINK-O-NEXT: Running pass: LCSSAPass
; CHECK-POSTLINK-O-NEXT: Running pass: LICMPass
+; CHECK-POSTLINK-O-NEXT: Running pass: LoopDeletionPass
; CHECK-POSTLINK-O-NEXT: Running pass: AlignmentFromAssumptionsPass
; CHECK-POSTLINK-O-NEXT: Running pass: LoopSinkPass
; CHECK-POSTLINK-O-NEXT: Running pass: InstSimplifyPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
index 05d165ae399f903..b6ae0690f76797c 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
@@ -178,6 +178,7 @@
; CHECK-O-NEXT: Running pass: LoopSimplifyPass
; CHECK-O-NEXT: Running pass: LCSSAPass
; CHECK-O-NEXT: Running pass: LICMPass
+; CHECK-O-NEXT: Running pass: LoopDeletionPass
; CHECK-O-NEXT: Running pass: AlignmentFromAssumptionsPass
; CHECK-O-NEXT: Running pass: LoopSinkPass
; CHECK-O-NEXT: Running pass: InstSimplifyPass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
index 2393ba9dbde4b64..5954feccacc5f6c 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
@@ -185,6 +185,7 @@
; CHECK-O-NEXT: Running pass: LoopSimplifyPass
; CHECK-O-NEXT: Running pass: LCSSAPass
; CHECK-O-NEXT: Running pass: LICMPass
+; CHECK-O-NEXT: Running pass: LoopDeletionPass
; CHECK-O-NEXT: Running pass: AlignmentFromAssumptionsPass
; CHECK-O-NEXT: Running pass: LoopSinkPass
; CHECK-O-NEXT: Running pass: InstSimplifyPass
diff --git a/llvm/test/Transforms/PhaseOrdering/X86/pr68282.ll b/llvm/test/Transforms/PhaseOrdering/X86/pr68282.ll
new file mode 100644
index 000000000000000..6f1e3dfa7ad3436
--- /dev/null
+++ b/llvm/test/Transforms/PhaseOrdering/X86/pr68282.ll
@@ -0,0 +1,59 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
+; RUN: opt < %s -passes='default<O3>' -mcpu=generic -S | FileCheck %s
+
+; C version of test case
+; int pr68282() {
+; int outputValue = 0;
+; for (int x = 0; x < 1024 * 1024; ++x) {
+; outputValue += outputValue + 0;
+; outputValue += outputValue + 1;
+; outputValue += outputValue + 2;
+; outputValue += outputValue + 3;
+; outputValue += outputValue + 4;
+; outputValue += outputValue + 5;
+; outputValue += outputValue + 6;
+; outputValue += outputValue + 7;
+; }
+; return outputValue;
+; }
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define i32 @pr68282() {
+; CHECK-LABEL: define i32 @pr68282(
+; CHECK-SAME: ) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: ret i32 -134744073
+;
+entry:
+ br label %for.cond
+
+for.cond:
+ %outputValue.0 = phi i32 [ 0, %entry ], [ %add15, %for.body ]
+ %x.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+ %cmp = icmp slt i32 %x.0, 1048576
+ br i1 %cmp, label %for.body, label %for.cond.cleanup
+
+for.cond.cleanup:
+ ret i32 %outputValue.0
+
+for.body:
+ %add1 = add nsw i32 %outputValue.0, %outputValue.0
+ %add2 = add nsw i32 %add1, 1
+ %add3 = add nsw i32 %add1, %add2
+ %add4 = add nsw i32 %add3, 2
+ %add5 = add nsw i32 %add3, %add4
+ %add6 = add nsw i32 %add5, 3
+ %add7 = add nsw i32 %add5, %add6
+ %add8 = add nsw i32 %add7, 4
+ %add9 = add nsw i32 %add7, %add8
+ %add10 = add nsw i32 %add9, 5
+ %add11 = add nsw i32 %add9, %add10
+ %add12 = add nsw i32 %add11, 6
+ %add13 = add nsw i32 %add11, %add12
+ %add14 = add nsw i32 %add13, 7
+ %add15 = add nsw i32 %add13, %add14
+ %inc = add nsw i32 %x.0, 1
+ br label %for.cond
+}
|
|
@nikic I would like to evaluate the compile-time impact of this PR. Could you please add my fork https://github.com/dtcxzyw/llvm-project to http://llvm-compile-time-tracker.com/? |
Done, here are the results: https://llvm-compile-time-tracker.com/compare.php?from=ed2c80151d6be0e8c2480a68d301d2c8a447ef13&to=060c47fa33edcf27e647c6c11cd1fd2e589bc272&stat=instructions:u |
nikic
requested changes
Oct 14, 2023
There was a problem hiding this comment.
Architecturally, trying to exploit optimization opportunities that are only opened up by runtime unrolling is pretty much always the wrong thing to do. At that point the optimization happens mostly by accident, not by intent, and only on some subtargets.
In this case, what you want to do instead is allow calculating the exit value in the loop earlier. Possibly in SCEV/IndVars. A possibly relevant observation here is that the exit value reaches a fixed point after a small number of iterations.
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.
This PR adds a
LoopDeletionPassat the end of the vectorization pipeline. It will eliminate dead loops introduced byInstCombine.Fixes #68282.