[llvm] [DFAJumpThreading] Don't thread switch without multiple successors (PR #71060)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Nov 2 06:25:57 PDT 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
Author: XChy (XChy)
<details>
<summary>Changes</summary>
Fixes #<!-- -->56882.
Fixes #<!-- -->60254.
When switch has only one successor, it make no sense to thread it. And computing the cost of it brings div-by-zero exception. We prevent it in this patch.
---
Full diff: https://github.com/llvm/llvm-project/pull/71060.diff
2 Files Affected:
- (modified) llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp (+6)
- (added) llvm/test/Transforms/DFAJumpThreading/single_succ_switch.ll (+43)
``````````diff
diff --git a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
index f2efe60bdf886a2..66407a3b2d504a2 100644
--- a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
+++ b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp
@@ -728,6 +728,10 @@ struct TransformDFA {
CodeMetrics Metrics;
SwitchInst *Switch = SwitchPaths->getSwitchInst();
+ // Don't thread switch without multiple successors.
+ if (Switch->getNumSuccessors() <= 1)
+ return false;
+
// Note that DuplicateBlockMap is not being used as intended here. It is
// just being used to ensure (BB, State) pairs are only counted once.
DuplicateBlockMap DuplicateMap;
@@ -805,6 +809,8 @@ struct TransformDFA {
// using binary search, hence the LogBase2().
unsigned CondBranches =
APInt(32, Switch->getNumSuccessors()).ceilLogBase2();
+ assert(CondBranches > 0 &&
+ "The threaded switch must have multiple branches");
DuplicationCost = Metrics.NumInsts / CondBranches;
} else {
// Compared with jump tables, the DFA optimizer removes an indirect branch
diff --git a/llvm/test/Transforms/DFAJumpThreading/single_succ_switch.ll b/llvm/test/Transforms/DFAJumpThreading/single_succ_switch.ll
new file mode 100644
index 000000000000000..00500a7a2598d85
--- /dev/null
+++ b/llvm/test/Transforms/DFAJumpThreading/single_succ_switch.ll
@@ -0,0 +1,43 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
+; RUN: opt -S -passes=dfa-jump-threading %s | FileCheck %s
+
+define void @pr60254() {
+; CHECK-LABEL: define void @pr60254() {
+; CHECK-NEXT: entry_1:
+; CHECK-NEXT: br label [[BB_2:%.*]]
+; CHECK: bb_2:
+; CHECK-NEXT: [[PTR_I32_25_0:%.*]] = phi i32 [ 0, [[BB_4:%.*]] ], [ 0, [[BB_2]] ], [ 0, [[ENTRY_1:%.*]] ]
+; CHECK-NEXT: switch i32 [[PTR_I32_25_0]], label [[BB_2]] [
+; CHECK-NEXT: ]
+; CHECK: bb_4:
+; CHECK-NEXT: br label [[BB_2]]
+;
+entry_1:
+ br label %bb_2
+
+bb_2: ; preds = %bb_4, %bb_2, %entry_1
+ %ptr_i32_25.0 = phi i32 [ 0, %bb_4 ], [ 0, %bb_2 ], [ 0, %entry_1 ]
+ switch i32 %ptr_i32_25.0, label %bb_2 [
+ ]
+
+bb_4: ; No predecessors!
+ br label %bb_2
+}
+
+define void @pr56882() {
+; CHECK-LABEL: define void @pr56882() {
+; CHECK-NEXT: entry_1:
+; CHECK-NEXT: br label [[BB_2:%.*]]
+; CHECK: bb_2:
+; CHECK-NEXT: [[PTR_I64_16_0:%.*]] = phi i64 [ -1317805584074026212, [[ENTRY_1:%.*]] ], [ -158622699357888703, [[BB_2]] ]
+; CHECK-NEXT: switch i64 [[PTR_I64_16_0]], label [[BB_2]] [
+; CHECK-NEXT: ]
+;
+entry_1:
+ br label %bb_2
+
+bb_2: ; preds = %bb_2, %entry_1
+ %ptr_i64_16.0 = phi i64 [ -1317805584074026212, %entry_1 ], [ -158622699357888703, %bb_2 ]
+ switch i64 %ptr_i64_16.0, label %bb_2 [
+ ]
+}
``````````
</details>
https://github.com/llvm/llvm-project/pull/71060
More information about the llvm-commits
mailing list