[DFAJumpThreading] Add MaxOuterUseBlocks threshold #163428
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
|
@llvm/pr-subscribers-llvm-transforms Author: Hongyu Chen (XChy) ChangesFor every threadable path Note that the O3 statistics in llvm-test-suite are not influenced. Full diff: https://github.com/llvm/llvm-project/pull/163428.diff 1 Files Affected:
diff --git a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
index ff5f390d6fe18..11d65d5a59175 100644
--- a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
+++ b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
@@ -122,16 +122,22 @@ static cl::opt<unsigned>
cl::desc("Maximum cost accepted for the transformation"),
cl::Hidden, cl::init(50));
-extern cl::opt<bool> ProfcheckDisableMetadataFixes;
-
-} // namespace llvm
-
static cl::opt<double> MaxClonedRate(
"dfa-max-cloned-rate",
cl::desc(
"Maximum cloned instructions rate accepted for the transformation"),
cl::Hidden, cl::init(7.5));
+static cl::opt<unsigned>
+ MaxOuterUseBlocks("dfa-max-out-use-blocks",
+ cl::desc("Maximum unduplicated blocks with outer uses "
+ "accepted for the transformation"),
+ cl::Hidden, cl::init(40));
+
+extern cl::opt<bool> ProfcheckDisableMetadataFixes;
+
+} // namespace llvm
+
namespace {
class SelectInstToUnfold {
SelectInst *SI;
@@ -966,8 +972,16 @@ struct TransformDFA {
// SLPVectorizer.
// TODO: Thread the switch partially before reaching the threshold.
uint64_t NumOrigInst = 0;
- for (auto *BB : DuplicateMap.keys())
+ uint64_t NumOuterUseBlock = 0;
+ for (auto *BB : DuplicateMap.keys()) {
NumOrigInst += BB->sizeWithoutDebug();
+ // Only unduplicated blocks with single predecessor require new phi
+ // nodes.
+ for (auto *Succ : successors(BB))
+ if (!DuplicateMap.count(Succ) && Succ->getSinglePredecessor())
+ NumOuterUseBlock++;
+ }
+
if (double(NumClonedInst) / double(NumOrigInst) > MaxClonedRate) {
LLVM_DEBUG(dbgs() << "DFA Jump Threading: Not jump threading, too much "
"instructions wll be cloned\n");
@@ -978,6 +992,20 @@ struct TransformDFA {
return false;
}
+ // Too much unduplicated blocks with outer uses may cause too much
+ // insertions of phi nodes for duplicated definitions. TODO: Drop this
+ // threshold if we come up with another way to reduce the number of inserted
+ // phi nodes.
+ if (NumOuterUseBlock > MaxOuterUseBlocks) {
+ LLVM_DEBUG(dbgs() << "DFA Jump Threading: Not jump threading, too much "
+ "blocks with outer uses\n");
+ ORE->emit([&]() {
+ return OptimizationRemarkMissed(DEBUG_TYPE, "NotProfitable", Switch)
+ << "Too much blocks with outer uses.";
+ });
+ return false;
+ }
+
InstructionCost DuplicationCost = 0;
unsigned JumpTableSize = 0;
|
|
ping. |
nikic
reviewed
Oct 27, 2025
XChy
force-pushed
the
dfa-jump-thread-max-outer-uses
branch
from
8704e9a to
c23bc4b
Compare
nikic
reviewed
Oct 29, 2025
| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6 | ||
| ; RUN: opt -S -passes=dfa-jump-threading -dfa-max-out-use-blocks=5 %s | FileCheck %s | ||
|
|
||
| define void @max_outer_uses_by_switch(i32 %cond) { |
There was a problem hiding this comment.
Hm, it seems like this doesn't really demonstrate the phi explosion issue. Without the limit you get:
diff --git a/llvm/test/Transforms/DFAJumpThreading/max-outer-uses.ll b/llvm/test/Transforms/DFAJumpThreading/max-outer-uses.ll
index f059754b8221..24f06c77614c 100644
--- a/llvm/test/Transforms/DFAJumpThreading/max-outer-uses.ll
+++ b/llvm/test/Transforms/DFAJumpThreading/max-outer-uses.ll
@@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
-; RUN: opt -S -passes=dfa-jump-threading -dfa-max-out-use-blocks=5 %s | FileCheck %s
+; RUN: opt -S -passes=dfa-jump-threading -dfa-max-out-use-blocks=10 %s | FileCheck %s
define void @max_outer_uses_by_switch(i32 %cond) {
; CHECK-LABEL: define void @max_outer_uses_by_switch(
@@ -7,23 +7,36 @@ define void @max_outer_uses_by_switch(i32 %cond) {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: br label %[[SWITCH_BB:.*]]
; CHECK: [[SWITCH_BB]]:
-; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[ONE:%.*]], %[[CASE1_SUCC:.*]] ], [ 2, %[[CASE2:.*]] ]
+; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ poison, %[[CASE1_SUCC:.*]] ]
; CHECK-NEXT: switch i32 [[PHI]], label %[[DEFAULT_DEST:.*]] [
; CHECK-NEXT: i32 0, label %[[CASE1:.*]]
-; CHECK-NEXT: i32 1, label %[[CASE2]]
+; CHECK-NEXT: i32 1, label %[[CASE2:.*]]
; CHECK-NEXT: ]
+; CHECK: [[SWITCH_BB_JT2:.*]]:
+; CHECK-NEXT: [[PHI_JT2:%.*]] = phi i32 [ 2, %[[CASE2]] ]
+; CHECK-NEXT: br label %[[DEFAULT_DEST]]
+; CHECK: [[SWITCH_BB_JT1:.*]]:
+; CHECK-NEXT: [[PHI_JT1:%.*]] = phi i32 [ [[ONE_JT1:%.*]], %[[CASE1_SUCC_JT1:.*]] ]
+; CHECK-NEXT: br label %[[CASE2]]
; CHECK: [[CASE1]]:
-; CHECK-NEXT: br label %[[CASE1_SUCC]]
+; CHECK-NEXT: br label %[[CASE1_SUCC_JT1]]
; CHECK: [[CASE1_SUCC]]:
-; CHECK-NEXT: [[ONE]] = phi i32 [ 1, %[[CASE1]] ]
; CHECK-NEXT: switch i32 [[COND]], label %[[SWITCH_BB]] [
; CHECK-NEXT: i32 0, label %[[OUTER1:.*]]
; CHECK-NEXT: i32 1, label %[[OUTER2:.*]]
; CHECK-NEXT: i32 2, label %[[OUTER3:.*]]
; CHECK-NEXT: i32 3, label %[[OUTER4:.*]]
; CHECK-NEXT: ]
+; CHECK: [[CASE1_SUCC_JT1]]:
+; CHECK-NEXT: [[ONE_JT1]] = phi i32 [ 1, %[[CASE1]] ]
+; CHECK-NEXT: switch i32 [[COND]], label %[[SWITCH_BB_JT1]] [
+; CHECK-NEXT: i32 0, label %[[OUTER1]]
+; CHECK-NEXT: i32 1, label %[[OUTER2]]
+; CHECK-NEXT: i32 2, label %[[OUTER3]]
+; CHECK-NEXT: i32 3, label %[[OUTER4]]
+; CHECK-NEXT: ]
; CHECK: [[CASE2]]:
-; CHECK-NEXT: br label %[[SWITCH_BB]]
+; CHECK-NEXT: br label %[[SWITCH_BB_JT2]]
; CHECK: [[OUTER1]]:
; CHECK-NEXT: ret void
; CHECK: [[OUTER2]]:which does not introduce any new (multi-operand) phis.
There was a problem hiding this comment.
Yeah, I updated the tests now. They should show new phis now.
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/88/builds/17633 Here is the relevant piece of the build log for the reference |
aokblast
pushed a commit
to aokblast/llvm-project
that referenced
this pull request
Oct 30, 2025
For every threadable path `B1 -> B2 -> ... -> Bn`, we need to insert phi nodes into every unduplicated successor of `Bi` if there are outer uses of duplicated definitions in `B_i`. To prevent the booming of phi nodes, this patch adds a threshold for the maximum number of unduplicated successors that may contain outer uses. This threshold makes sense especially when multi-target branches like switch/indirectbr/callbr are duplicated. Note that the O3 statistics in llvm-test-suite are not influenced.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
For every threadable path
B1 -> B2 -> ... -> Bn, we need to insert phi nodes into every unduplicated successor ofBiif there are outer uses of duplicated definitions inB_i. To prevent the booming of phi nodes, this patch adds a threshold for the maximum number of unduplicated successors that may contain outer uses. This threshold makes sense especially when multi-target branches like switch/indirectbr/callbr are duplicated.Note that the O3 statistics in llvm-test-suite are not influenced.