[llvm] 184290e - [LAA] Add tests with dependencies may preventing st-to-ld forwarding.
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Sun Dec 10 05:57:28 PST 2023
Author: Florian Hahn
Date: 2023-12-10T13:56:53Z
New Revision: 184290e579ccb909faf9a03e35266ed23cf26e23
URL: https://github.com/llvm/llvm-project/commit/184290e579ccb909faf9a03e35266ed23cf26e23
DIFF: https://github.com/llvm/llvm-project/commit/184290e579ccb909faf9a03e35266ed23cf26e23.diff
LOG: [LAA] Add tests with dependencies may preventing st-to-ld forwarding.
Add test cases with varying distances between stores and loads that may
prevent store-to-load forwarding.
Added:
llvm/test/Analysis/LoopAccessAnalysis/num-iters-for-store-load-conflict.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/num-iters-for-store-load-conflict.ll b/llvm/test/Analysis/LoopAccessAnalysis/num-iters-for-store-load-conflict.ll
new file mode 100644
index 0000000000000..d3eda21dee27e
--- /dev/null
+++ b/llvm/test/Analysis/LoopAccessAnalysis/num-iters-for-store-load-conflict.ll
@@ -0,0 +1,310 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -passes='print<access-info>' -disable-output < %s 2>&1 | FileCheck %s
+
+define void @forward_dist_7(ptr %A, ptr noalias %B) {
+; CHECK-LABEL: 'forward_dist_7'
+; 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: Forward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: ForwardButPreventsForwarding:
+; CHECK-NEXT: store i32 0, ptr %gep.2, align 4 ->
+; CHECK-NEXT: %l = load i32, ptr %gep.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 i64 [ 0, %entry ], [ %next, %loop ]
+ %gep.1 = getelementptr i32, ptr %A, i64 %iv
+ %gep.2 = getelementptr i32, ptr %gep.1, i64 7
+ store i32 0, ptr %gep.2, align 4
+ %l = load i32, ptr %gep.1
+ store i32 %l, ptr %B
+ %next = add nuw nsw i64 %iv, 1
+ %ec = icmp eq i64 %iv, 1000
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_dist_9(ptr %A, ptr noalias %B) {
+; CHECK-LABEL: 'forward_dist_9'
+; 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: Forward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: ForwardButPreventsForwarding:
+; CHECK-NEXT: store i32 0, ptr %gep.2, align 4 ->
+; CHECK-NEXT: %l = load i32, ptr %gep.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 i64 [ 0, %entry ], [ %next, %loop ]
+ %gep.1 = getelementptr i32, ptr %A, i64 %iv
+ %gep.2 = getelementptr i32, ptr %gep.1, i64 9
+ store i32 0, ptr %gep.2, align 4
+ %l = load i32, ptr %gep.1
+ store i32 %l, ptr %B
+ %next = add nuw nsw i64 %iv, 1
+ %ec = icmp eq i64 %iv, 1000
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_dist_11(ptr %A, ptr noalias %B) {
+; CHECK-LABEL: 'forward_dist_11'
+; 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: Forward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: ForwardButPreventsForwarding:
+; CHECK-NEXT: store i32 0, ptr %gep.2, align 4 ->
+; CHECK-NEXT: %l = load i32, ptr %gep.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 i64 [ 0, %entry ], [ %next, %loop ]
+ %gep.1 = getelementptr i32, ptr %A, i64 %iv
+ %gep.2 = getelementptr i32, ptr %gep.1, i64 9
+ store i32 0, ptr %gep.2, align 4
+ %l = load i32, ptr %gep.1
+ store i32 %l, ptr %B
+ %next = add nuw nsw i64 %iv, 1
+ %ec = icmp eq i64 %iv, 1000
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_dist_13(ptr %A, ptr noalias %B) {
+; CHECK-LABEL: 'forward_dist_13'
+; 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: Forward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: ForwardButPreventsForwarding:
+; CHECK-NEXT: store i32 0, ptr %gep.2, align 4 ->
+; CHECK-NEXT: %l = load i32, ptr %gep.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 i64 [ 0, %entry ], [ %next, %loop ]
+ %gep.1 = getelementptr i32, ptr %A, i64 %iv
+ %gep.2 = getelementptr i32, ptr %gep.1, i64 13
+ store i32 0, ptr %gep.2, align 4
+ %l = load i32, ptr %gep.1
+ store i32 %l, ptr %B
+ %next = add nuw nsw i64 %iv, 1
+ %ec = icmp eq i64 %iv, 1000
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_dist_15(ptr %A, ptr noalias %B) {
+; CHECK-LABEL: 'forward_dist_15'
+; 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: Forward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: ForwardButPreventsForwarding:
+; CHECK-NEXT: store i32 0, ptr %gep.2, align 4 ->
+; CHECK-NEXT: %l = load i32, ptr %gep.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 i64 [ 0, %entry ], [ %next, %loop ]
+ %gep.1 = getelementptr i32, ptr %A, i64 %iv
+ %gep.2 = getelementptr i32, ptr %gep.1, i64 13
+ store i32 0, ptr %gep.2, align 4
+ %l = load i32, ptr %gep.1
+ store i32 %l, ptr %B
+ %next = add nuw nsw i64 %iv, 1
+ %ec = icmp eq i64 %iv, 1000
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_dist_17(ptr %A, ptr noalias %B) {
+; CHECK-LABEL: 'forward_dist_17'
+; 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: Forward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: ForwardButPreventsForwarding:
+; CHECK-NEXT: store i32 0, ptr %gep.2, align 4 ->
+; CHECK-NEXT: %l = load i32, ptr %gep.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 i64 [ 0, %entry ], [ %next, %loop ]
+ %gep.1 = getelementptr i32, ptr %A, i64 %iv
+ %gep.2 = getelementptr i32, ptr %gep.1, i64 17
+ store i32 0, ptr %gep.2, align 4
+ %l = load i32, ptr %gep.1
+ store i32 %l, ptr %B
+ %next = add nuw nsw i64 %iv, 1
+ %ec = icmp eq i64 %iv, 1000
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @forward_dist_19(ptr %A, ptr noalias %B) {
+; CHECK-LABEL: 'forward_dist_19'
+; 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: Forward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: ForwardButPreventsForwarding:
+; CHECK-NEXT: store i32 0, ptr %gep.2, align 4 ->
+; CHECK-NEXT: %l = load i32, ptr %gep.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 i64 [ 0, %entry ], [ %next, %loop ]
+ %gep.1 = getelementptr i32, ptr %A, i64 %iv
+ %gep.2 = getelementptr i32, ptr %gep.1, i64 19
+ store i32 0, ptr %gep.2, align 4
+ %l = load i32, ptr %gep.1
+ store i32 %l, ptr %B
+ %next = add nuw nsw i64 %iv, 1
+ %ec = icmp eq i64 %iv, 1000
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+ at A = global [37 x [37 x double]] zeroinitializer, align 8
+
+define void @unknown_loop_bounds(i64 %x, i64 %y) {
+; CHECK-LABEL: 'unknown_loop_bounds'
+; CHECK-NEXT: inner:
+; 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: Backward loop carried data dependence that prevents store-to-load forwarding.
+; CHECK-NEXT: Dependences:
+; CHECK-NEXT: BackwardVectorizableButPreventsForwarding:
+; CHECK-NEXT: %l = load double, ptr %gep.0, align 8 ->
+; CHECK-NEXT: store double %l, ptr %gep.1, align 8
+; 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:
+; CHECK-NEXT: outer.header:
+; CHECK-NEXT: Report: loop is not the innermost loop
+; CHECK-NEXT: Dependences:
+; 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 %outer.header
+
+outer.header:
+ %outer.iv = phi i64 [ 0, %entry ], [ %outer.iv.next, %outer.latch ]
+ %outer.iv.next = add nuw nsw i64 %outer.iv, 1
+ br label %inner
+
+inner:
+ %inner.iv = phi i64 [ 0, %outer.header ], [ %inner.iv.next, %inner ]
+ %gep.0 = getelementptr inbounds [37 x [37 x double]], ptr @A, i64 0, i64 %outer.iv, i64 %inner.iv
+ %l = load double, ptr %gep.0, align 8
+ %gep.1 = getelementptr inbounds [37 x [37 x double]], ptr @A, i64 0, i64 %outer.iv.next, i64 %inner.iv
+ store double %l, ptr %gep.1, align 8
+ %inner.iv.next = add nuw nsw i64 %inner.iv, 1
+ %inner.ec = icmp eq i64 %inner.iv.next, %y
+ br i1 %inner.ec, label %outer.latch, label %inner
+
+outer.latch:
+ %outer.ec = icmp eq i64 %outer.iv.next, %x
+ br i1 %outer.ec, label %exit, label %outer.header
+
+exit:
+ ret void
+}
More information about the llvm-commits
mailing list