[llvm] 0c70a26 - [SLP]Clear root node reordering only if the root node is not re-used in graph
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Mon Feb 3 06:05:31 PST 2025
Author: Alexey Bataev
Date: 2025-02-03T06:05:19-08:00
New Revision: 0c70a26f46e4efd5a29eb281ff99d2cf7f04c6f6
URL: https://github.com/llvm/llvm-project/commit/0c70a26f46e4efd5a29eb281ff99d2cf7f04c6f6
DIFF: https://github.com/llvm/llvm-project/commit/0c70a26f46e4efd5a29eb281ff99d2cf7f04c6f6.diff
LOG: [SLP]Clear root node reordering only if the root node is not re-used in graph
The reordering of the root node can be safely cleared only if the root
node is not reused, otherwise the graph might be broken
Fixes #125357
Added:
llvm/test/Transforms/SLPVectorizer/X86/root-node-reordered-reused.ll
Modified:
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 4f83b9cfa18d655..539c9227af7e3a2 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -20147,7 +20147,7 @@ class HorizontalReduction {
}
V.reorderTopToBottom();
// No need to reorder the root node at all.
- V.reorderBottomToTop(/*IgnoreReorder=*/true);
+ V.reorderBottomToTop(!V.doesRootHaveInTreeUses());
// Keep extracted other reduction values, if they are used in the
// vectorization trees.
BoUpSLP::ExtraValueToDebugLocsMap LocalExternallyUsedValues(
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/root-node-reordered-reused.ll b/llvm/test/Transforms/SLPVectorizer/X86/root-node-reordered-reused.ll
new file mode 100644
index 000000000000000..64b43ebf3e140f2
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/root-node-reordered-reused.ll
@@ -0,0 +1,106 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
+
+ at d = global i32 0, align 4
+ at c = global i32 0, align 4
+ at f = global i32 0, align 4
+ at b = global i32 0, align 4
+ at a = global i32 0, align 4
+ at e = global i32 0, align 4
+
+define i32 @test() {
+; CHECK-LABEL: define i32 @test() {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr @d, align 4
+; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[TMP0]], 0
+; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label %[[IF_END6:.*]], label %[[IF_THEN:.*]]
+; CHECK: [[IF_THEN]]:
+; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr @c, align 4
+; CHECK-NEXT: [[TMP2:%.*]] = load volatile i32, ptr @f, align 4
+; CHECK-NEXT: [[TOBOOL1_NOT:%.*]] = icmp eq i32 [[TMP2]], 0
+; CHECK-NEXT: br i1 [[TOBOOL1_NOT]], label %[[IF_END:.*]], label %[[I:.*]]
+; CHECK: [[IF_END]]:
+; CHECK-NEXT: [[TMP3:%.*]] = load volatile i32, ptr @f, align 4
+; CHECK-NEXT: [[TOBOOL3_NOT:%.*]] = icmp eq i32 [[TMP3]], 0
+; CHECK-NEXT: br i1 [[TOBOOL3_NOT]], label %[[IF_END6]], label %[[I]]
+; CHECK: [[IF_END6]]:
+; CHECK-NEXT: [[H_0:%.*]] = phi i32 [ [[TMP1]], %[[IF_END]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr @b, align 4
+; CHECK-NEXT: [[TOBOOL7_NOT:%.*]] = icmp eq i32 [[TMP4]], 0
+; CHECK-NEXT: br i1 [[TOBOOL7_NOT]], label %[[IF_END13:.*]], label %[[I]]
+; CHECK: [[LAND_RHS:.*]]:
+; CHECK-NEXT: [[TMP5:%.*]] = load volatile i32, ptr @f, align 4
+; CHECK-NEXT: br label %[[IF_END13]]
+; CHECK: [[IF_END13]]:
+; CHECK-NEXT: [[H_2:%.*]] = phi i32 [ [[H_0]], %[[IF_END6]] ], [ [[H_1:%.*]], %[[LAND_RHS]] ], [ [[H_1]], %[[I]] ]
+; CHECK-NEXT: [[J_1:%.*]] = phi i32 [ [[TMP0]], %[[IF_END6]] ], [ [[J_0:%.*]], %[[LAND_RHS]] ], [ [[J_0]], %[[I]] ]
+; CHECK-NEXT: [[K_1:%.*]] = phi i32 [ [[TMP0]], %[[IF_END6]] ], [ [[K_0:%.*]], %[[LAND_RHS]] ], [ [[K_0]], %[[I]] ]
+; CHECK-NEXT: [[L_1:%.*]] = phi i32 [ [[TMP0]], %[[IF_END6]] ], [ [[L_0:%.*]], %[[LAND_RHS]] ], [ [[L_0]], %[[I]] ]
+; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr @a, align 4
+; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr @e, align 4
+; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[TMP6]], [[TMP7]]
+; CHECK-NEXT: br i1 [[CMP_NOT]], label %[[IF_END15:.*]], label %[[I]]
+; CHECK: [[I]]:
+; CHECK-NEXT: [[J_0]] = phi i32 [ 0, %[[IF_THEN]] ], [ [[J_1]], %[[IF_END13]] ], [ 0, %[[IF_END]] ], [ [[TMP0]], %[[IF_END6]] ]
+; CHECK-NEXT: [[K_0]] = phi i32 [ 0, %[[IF_THEN]] ], [ [[K_1]], %[[IF_END13]] ], [ 0, %[[IF_END]] ], [ [[TMP0]], %[[IF_END6]] ]
+; CHECK-NEXT: [[L_0]] = phi i32 [ 0, %[[IF_THEN]] ], [ [[L_1]], %[[IF_END13]] ], [ 0, %[[IF_END]] ], [ [[TMP0]], %[[IF_END6]] ]
+; CHECK-NEXT: [[H_1]] = phi i32 [ [[TMP1]], %[[IF_THEN]] ], [ [[H_2]], %[[IF_END13]] ], [ [[TMP1]], %[[IF_END]] ], [ [[H_0]], %[[IF_END6]] ]
+; CHECK-NEXT: [[OR:%.*]] = or i32 [[J_0]], [[H_1]]
+; CHECK-NEXT: [[OR9:%.*]] = or i32 [[OR]], [[K_0]]
+; CHECK-NEXT: [[OR10:%.*]] = or i32 [[OR9]], [[L_0]]
+; CHECK-NEXT: [[TOBOOL11_NOT:%.*]] = icmp eq i32 [[OR10]], 0
+; CHECK-NEXT: br i1 [[TOBOOL11_NOT]], label %[[IF_END13]], label %[[LAND_RHS]]
+; CHECK: [[IF_END15]]:
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ %0 = load i32, ptr @d, align 4
+ %tobool.not = icmp eq i32 %0, 0
+ br i1 %tobool.not, label %if.end6, label %if.then
+
+if.then:
+ %1 = load i32, ptr @c, align 4
+ %2 = load volatile i32, ptr @f, align 4
+ %tobool1.not = icmp eq i32 %2, 0
+ br i1 %tobool1.not, label %if.end, label %i
+
+if.end:
+ %3 = load volatile i32, ptr @f, align 4
+ %tobool3.not = icmp eq i32 %3, 0
+ br i1 %tobool3.not, label %if.end6, label %i
+
+if.end6:
+ %h.0 = phi i32 [ %1, %if.end ], [ 0, %entry ]
+ %4 = load i32, ptr @b, align 4
+ %tobool7.not = icmp eq i32 %4, 0
+ br i1 %tobool7.not, label %if.end13, label %i
+
+land.rhs:
+ %5 = load volatile i32, ptr @f, align 4
+ br label %if.end13
+
+if.end13:
+ %h.2 = phi i32 [ %h.0, %if.end6 ], [ %h.1, %land.rhs ], [ %h.1, %i ]
+ %j.1 = phi i32 [ %0, %if.end6 ], [ %j.0, %land.rhs ], [ %j.0, %i ]
+ %k.1 = phi i32 [ %0, %if.end6 ], [ %k.0, %land.rhs ], [ %k.0, %i ]
+ %l.1 = phi i32 [ %0, %if.end6 ], [ %l.0, %land.rhs ], [ %l.0, %i ]
+ %6 = load i32, ptr @a, align 4
+ %7 = load i32, ptr @e, align 4
+ %cmp.not = icmp eq i32 %6, %7
+ br i1 %cmp.not, label %if.end15, label %i
+
+i:
+ %j.0 = phi i32 [ 0, %if.then ], [ %j.1, %if.end13 ], [ 0, %if.end ], [ %0, %if.end6 ]
+ %k.0 = phi i32 [ 0, %if.then ], [ %k.1, %if.end13 ], [ 0, %if.end ], [ %0, %if.end6 ]
+ %l.0 = phi i32 [ 0, %if.then ], [ %l.1, %if.end13 ], [ 0, %if.end ], [ %0, %if.end6 ]
+ %h.1 = phi i32 [ %1, %if.then ], [ %h.2, %if.end13 ], [ %1, %if.end ], [ %h.0, %if.end6 ]
+ %or = or i32 %j.0, %h.1
+ %or9 = or i32 %or, %k.0
+ %or10 = or i32 %or9, %l.0
+ %tobool11.not = icmp eq i32 %or10, 0
+ br i1 %tobool11.not, label %if.end13, label %land.rhs
+
+if.end15:
+ ret i32 0
+}
+
More information about the llvm-commits
mailing list