[llvm-branch-commits] [llvm] 9070c25 - [LAA] Require AddRecs to be in the innermost loop for diff-checks.

[Tobias Hieta via llvm-branch-commits]

Author: Florian Hahn
Date: 2022-08-30T08:08:28+02:00
New Revision: 9070c258c143d6666274da3c70971bd266cabb82

URL: https://github.com/llvm/llvm-project/commit/9070c258c143d6666274da3c70971bd266cabb82
DIFF: https://github.com/llvm/llvm-project/commit/9070c258c143d6666274da3c70971bd266cabb82.diff

LOG: [LAA] Require AddRecs to be in the innermost loop for diff-checks.

The simpler diff-checks require pointers with add-recs from the same
innermost loop, but this property wasn't check completely. Add the
missing check to ensure both addrecs are in the innermost loop.

Fixes #57315.

(cherry picked from commit 9405af1c850139f6ddd6d35ff395756fe2aef31f)

Added: 
    

Modified: 
    llvm/include/llvm/Analysis/LoopAccessAnalysis.h
    llvm/lib/Analysis/LoopAccessAnalysis.cpp
    llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/Analysis/LoopAccessAnalysis.h b/llvm/include/llvm/Analysis/LoopAccessAnalysis.h
index 8f71ce9e96c05..af8e8d22269e4 100644
--- a/llvm/include/llvm/Analysis/LoopAccessAnalysis.h
+++ b/llvm/include/llvm/Analysis/LoopAccessAnalysis.h
@@ -253,6 +253,8 @@ class MemoryDepChecker {
     return {};
   }
 
+  const Loop *getInnermostLoop() const { return InnermostLoop; }
+
 private:
   /// A wrapper around ScalarEvolution, used to add runtime SCEV checks, and
   /// applies dynamic knowledge to simplify SCEV expressions and convert them

diff  --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index aa35f253bc5f0..8311b480ab099 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -280,7 +280,8 @@ void RuntimePointerChecking::tryToCreateDiffCheck(
 
   auto *SrcAR = dyn_cast<SCEVAddRecExpr>(Src->Expr);
   auto *SinkAR = dyn_cast<SCEVAddRecExpr>(Sink->Expr);
-  if (!SrcAR || !SinkAR) {
+  if (!SrcAR || !SinkAR || SrcAR->getLoop() != DC.getInnermostLoop() ||
+      SinkAR->getLoop() != DC.getInnermostLoop()) {
     CanUseDiffCheck = false;
     return;
   }

diff  --git a/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll b/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll
index c8595aee2fe9d..bdafe9f4daaae 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll
@@ -172,18 +172,29 @@ exit:
   ret void
 }
 
-; FIXME: Full no-overlap checks are required instead of 
diff erence checks, as
+; Full no-overlap checks are required instead of 
diff erence checks, as
 ; one of the add-recs used is invariant in the inner loop.
 ; Test case for PR57315.
 define void @nested_loop_outer_iv_addrec_invariant_in_inner1(ptr %a, ptr %b, i64 %n) {
 ; CHECK-LABEL: @nested_loop_outer_iv_addrec_invariant_in_inner1(
-; CHECK:       entry:
-; CHECK-NEXT:    [[B:%.*]] = ptrtoint ptr %b to i64
-; CHECK-NEXT:    [[A:%.*]] = ptrtoint ptr %a to i64
+; CHECK:        entry:
+; CHECK-NEXT:    [[N_SHL_2:%.]] = shl i64 %n, 2
+; CHECK-NEXT:    [[B_GEP_UPPER:%.*]] = getelementptr i8, ptr %b, i64 [[N_SHL_2]]
+; CHECK-NEXT:    br label %outer
+
+; CHECK:       outer.header:
+; CHECK:         [[OUTER_IV_SHL_2:%.]] = shl i64 %outer.iv, 2
+; CHECK-NEXT:    [[A_GEP_UPPER:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_SHL_2]]
+; CHECK-NEXT:    [[OUTER_IV_4:%.]] = add i64 [[OUTER_IV_SHL_2]], 4
+; CHECK-NEXT:    [[A_GEP_UPPER_4:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_4]]
+; CHECK:         [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
+
 ; CHECK:       vector.memcheck:
-; CHECK-NEXT:    [[TMP1:%.*]] = sub i64 [[A]], [[B]]
-; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP1]], 16
-; CHECK-NEXT:    br i1 [[DIFF_CHECK]], label %scalar.ph, label %vector.ph
+; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult ptr [[A_GEP_UPPER]], [[B_GEP_UPPER]]
+; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult ptr %b, [[A_GEP_UPPER_4]]
+; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
+; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label %scalar.ph, label %vector.ph
 ;
 entry:
   br label %outer.header
@@ -216,13 +227,24 @@ exit:
 ; sink and source swapped.
 define void @nested_loop_outer_iv_addrec_invariant_in_inner2(ptr %a, ptr %b, i64 %n) {
 ; CHECK-LABEL: @nested_loop_outer_iv_addrec_invariant_in_inner2(
-; CHECK:       entry:
-; CHECK-NEXT:    [[A:%.*]] = ptrtoint ptr %a to i64
-; CHECK-NEXT:    [[B:%.*]] = ptrtoint ptr %b to i64
+; CHECK:        entry:
+; CHECK-NEXT:    [[N_SHL_2:%.]] = shl i64 %n, 2
+; CHECK-NEXT:    [[B_GEP_UPPER:%.*]] = getelementptr i8, ptr %b, i64 [[N_SHL_2]]
+; CHECK-NEXT:    br label %outer
+
+; CHECK:       outer.header:
+; CHECK:         [[OUTER_IV_SHL_2:%.]] = shl i64 %outer.iv, 2
+; CHECK-NEXT:    [[A_GEP_UPPER:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_SHL_2]]
+; CHECK-NEXT:    [[OUTER_IV_4:%.]] = add i64 [[OUTER_IV_SHL_2]], 4
+; CHECK-NEXT:    [[A_GEP_UPPER_4:%.*]] = getelementptr i8, ptr %a, i64 [[OUTER_IV_4]]
+; CHECK:         [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %scalar.ph, label %vector.memcheck
+
 ; CHECK:       vector.memcheck:
-; CHECK-NEXT:    [[TMP1:%.*]] = sub i64 [[B]], [[A]]
-; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP1]], 16
-; CHECK-NEXT:    br i1 [[DIFF_CHECK]], label %scalar.ph, label %vector.ph
+; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult ptr %b, [[A_GEP_UPPER_4]]
+; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult ptr [[A_GEP_UPPER]], [[B_GEP_UPPER]]
+; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
+; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label %scalar.ph, label %vector.ph
 ;
 entry:
   br label %outer.header