[llvm] c5a49fb - [LAA] Add tests with 128 bit inductions and 128 bit pointers.
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Sun Jun 29 03:31:31 PDT 2025
Author: Florian Hahn
Date: 2025-06-29T11:28:35+01:00
New Revision: c5a49fb62d4196bf26f1a4e2340accd4f242de5d
URL: https://github.com/llvm/llvm-project/commit/c5a49fb62d4196bf26f1a4e2340accd4f242de5d
DIFF: https://github.com/llvm/llvm-project/commit/c5a49fb62d4196bf26f1a4e2340accd4f242de5d.diff
LOG: [LAA] Add tests with 128 bit inductions and 128 bit pointers.
Adds extra test coverage for
https://github.com/llvm/llvm-project/pull/140048.
Added:
llvm/test/Analysis/LoopAccessAnalysis/dependences-i128-inductions.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/dependences-i128-inductions.ll b/llvm/test/Analysis/LoopAccessAnalysis/dependences-i128-inductions.ll
new file mode 100644
index 0000000000000..5108cbf07d706
--- /dev/null
+++ b/llvm/test/Analysis/LoopAccessAnalysis/dependences-i128-inductions.ll
@@ -0,0 +1,153 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -disable-output -passes='print<access-info>' %s 2>&1 | FileCheck %s
+
+target datalayout = "e-m:e-p:128:128:128"
+
+; Tests with 128 bit inductions and 128 bit pointer width and index size.
+
+
+define void @backward_i128(ptr %A, i128 %n) {
+; CHECK-LABEL: 'backward_i128'
+; CHECK-NEXT: loop:
+; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
+; CHECK-NEXT: Unsafe indirect dependence.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: IndirectUnsafe:
+; CHECK-NEXT: %l = load i32, ptr %gep.A, align 4 ->
+; CHECK-NEXT: store i32 %l, ptr %gep.A.1, align 4
+; CHECK-EMPTY:
+; CHECK-NEXT: Run-time memory checks:
+; CHECK-NEXT: Grouped accesses:
+; CHECK-EMPTY:
+; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT: SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT: Expressions re-written:
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i128 [ 0, %entry ], [ %iv.next, %loop ]
+ %gep.A = getelementptr inbounds i32, ptr %A, i128 %iv
+ %l = load i32, ptr %gep.A
+ %iv.1 = add nuw nsw i128 %iv, 1
+ %gep.A.1 = getelementptr inbounds i32, ptr %A, i128 %iv.1
+ store i32 %l, ptr %gep.A.1
+ %iv.next = add i128 %iv, 1
+ %ec = icmp eq i128 %iv.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_i128(ptr %A, i128 %n) {
+; CHECK-LABEL: 'forward_i128'
+; CHECK-NEXT: loop:
+; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
+; CHECK-NEXT: Unsafe indirect dependence.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: IndirectUnsafe:
+; CHECK-NEXT: %l = load i32, ptr %gep.A.1, align 4 ->
+; CHECK-NEXT: store i32 %l, ptr %gep.A, align 4
+; CHECK-EMPTY:
+; CHECK-NEXT: Run-time memory checks:
+; CHECK-NEXT: Grouped accesses:
+; CHECK-EMPTY:
+; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT: SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT: Expressions re-written:
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i128 [ 0, %entry ], [ %iv.next, %loop ]
+ %iv.1 = add nuw nsw i128 %iv, 1
+ %gep.A.1 = getelementptr inbounds i32, ptr %A, i128 %iv.1
+ %l = load i32, ptr %gep.A.1
+ %gep.A = getelementptr inbounds i32, ptr %A, i128 %iv
+ store i32 %l, ptr %gep.A
+ %iv.next = add i128 %iv, 1
+ %ec = icmp eq i128 %iv.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_negative_step(ptr %A, i128 %n) {
+; CHECK-LABEL: 'forward_negative_step'
+; CHECK-NEXT: loop:
+; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
+; CHECK-NEXT: Unsafe indirect dependence.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: IndirectUnsafe:
+; CHECK-NEXT: %l = load i32, ptr %gep.A, align 4 ->
+; CHECK-NEXT: store i32 %l, ptr %gep.A.1, align 4
+; CHECK-EMPTY:
+; CHECK-NEXT: Run-time memory checks:
+; CHECK-NEXT: Grouped accesses:
+; CHECK-EMPTY:
+; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT: SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT: Expressions re-written:
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i128 [ 36893488147419103232, %entry ], [ %iv.next, %loop ]
+ %gep.A = getelementptr inbounds i32, ptr %A, i128 %iv
+ %l = load i32, ptr %gep.A
+ %iv.1 = add nuw nsw i128 %iv, 1
+ %gep.A.1 = getelementptr inbounds i32, ptr %A, i128 %iv.1
+ store i32 %l, ptr %gep.A.1
+ %iv.next = add nsw i128 %iv, -1
+ %ec = icmp eq i128 %iv.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+; The induction step is i128 2^63 + 1, which does not fit into a 64 bit signed
+; integer.
+define void @forward_i128_step_63bit_plus_one(ptr %A, i128 %n) {
+; CHECK-LABEL: 'forward_i128_step_63bit_plus_one'
+; CHECK-NEXT: loop:
+; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
+; CHECK-NEXT: Unsafe indirect dependence.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: IndirectUnsafe:
+; CHECK-NEXT: %l = load i32, ptr %gep.A.1, align 4 ->
+; CHECK-NEXT: store i32 %l, ptr %gep.A, align 4
+; CHECK-EMPTY:
+; CHECK-NEXT: Run-time memory checks:
+; CHECK-NEXT: Grouped accesses:
+; CHECK-EMPTY:
+; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT: SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT: Expressions re-written:
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i128 [ 0, %entry ], [ %iv.next, %loop ]
+ %iv.1 = add nuw nsw i128 %iv, 1
+ %gep.A.1 = getelementptr inbounds i32, ptr %A, i128 %iv.1
+ %l = load i32, ptr %gep.A.1
+ %gep.A = getelementptr inbounds i32, ptr %A, i128 %iv
+ store i32 %l, ptr %gep.A
+ %iv.next = add i128 %iv, 9223372036854775809
+ %ec = icmp eq i128 %iv.next, %n
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
More information about the llvm-commits
mailing list