[llvm] e775efc - [LV] Apply loop guards when checking recur during hoisting RT checks.

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2024-06-04T20:37:46+01:00
New Revision: e775efcec476b3ec4eacfb14dc5bc7e86d9aa8fd

URL: https://github.com/llvm/llvm-project/commit/e775efcec476b3ec4eacfb14dc5bc7e86d9aa8fd
DIFF: https://github.com/llvm/llvm-project/commit/e775efcec476b3ec4eacfb14dc5bc7e86d9aa8fd.diff

LOG: [LV] Apply loop guards when checking recur during hoisting RT checks.

Apply loop guards when checking if the recurrence is non-negative in
cases where runtime checks are hoisted out of an inner loop.

Added: 
    

Modified: 
    llvm/lib/Transforms/Utils/LoopUtils.cpp
    llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Utils/LoopUtils.cpp b/llvm/lib/Transforms/Utils/LoopUtils.cpp
index de3eb4a4ed5be..922d4b6437c7e 100644
--- a/llvm/lib/Transforms/Utils/LoopUtils.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUtils.cpp
@@ -1760,7 +1760,8 @@ static PointerBounds expandBounds(const RuntimeCheckingPtrGroup *CG,
           Low = cast<SCEVAddRecExpr>(Low)->getStart();
           // If there is a possibility that the stride is negative then we have
           // to generate extra checks to ensure the stride is positive.
-          if (!SE.isKnownNonNegative(Recur)) {
+          if (!SE.isKnownNonNegative(
+                  SE.applyLoopGuards(Recur, HighAR->getLoop()))) {
             Stride = Recur;
             LLVM_DEBUG(dbgs() << "LAA: ... but need to check stride is "
                                  "positive: "

diff  --git a/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll b/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
index 8e6dbfc68d121..e3e98c3a7b627 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
@@ -1509,7 +1509,8 @@ outer.exit:
   ret void
 }
 
-; TODO: STRIDE_CHECK can be eliminated via loop guards.
+; The stride for the access in the inner loop is known to be non-negative via
+; loop guards.
 define void @stride_check_known_via_loop_guard(ptr %C, ptr %A, i32 %Acols) {
 ; CHECK-LABEL: define void @stride_check_known_via_loop_guard
 ; CHECK-SAME: (ptr [[C:%.*]], ptr [[A:%.*]], i32 [[ACOLS:%.*]]) {
@@ -1518,8 +1519,6 @@ define void @stride_check_known_via_loop_guard(ptr %C, ptr %A, i32 %Acols) {
 ; CHECK-NEXT:    br i1 [[PRE_C]], label [[EXIT:%.*]], label [[OUTER_HEADER_PREHEADER:%.*]]
 ; CHECK:       outer.header.preheader:
 ; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A]], i64 8
-; CHECK-NEXT:    [[TMP0:%.*]] = sext i32 [[ACOLS]] to i64
-; CHECK-NEXT:    [[TMP1:%.*]] = shl nsw i64 [[TMP0]], 3
 ; CHECK-NEXT:    [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[C]], i64 34359738368
 ; CHECK-NEXT:    br label [[OUTER_HEADER:%.*]]
 ; CHECK:       outer.header:
@@ -1533,23 +1532,21 @@ define void @stride_check_known_via_loop_guard(ptr %C, ptr %A, i32 %Acols) {
 ; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult ptr [[A]], [[SCEVGEP1]]
 ; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult ptr [[C]], [[SCEVGEP]]
 ; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
-; CHECK-NEXT:    [[STRIDE_CHECK:%.*]] = icmp slt i64 [[TMP1]], 0
-; CHECK-NEXT:    [[TMP2:%.*]] = or i1 [[FOUND_CONFLICT]], [[STRIDE_CHECK]]
-; CHECK-NEXT:    br i1 [[TMP2]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
+; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
 ; CHECK:       vector.ph:
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[INDEX]], 0
-; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds double, ptr [[C]], i32 [[TMP3]]
-; CHECK-NEXT:    [[TMP5:%.*]] = load double, ptr [[ARRAYIDX_US]], align 8, !alias.scope [[META69:![0-9]+]], !noalias [[META72:![0-9]+]]
-; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x double> poison, double [[TMP5]], i64 0
+; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds double, ptr [[C]], i32 [[TMP0]]
+; CHECK-NEXT:    [[TMP2:%.*]] = load double, ptr [[ARRAYIDX_US]], align 8, !alias.scope [[META69:![0-9]+]], !noalias [[META72:![0-9]+]]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x double> poison, double [[TMP2]], i64 0
 ; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x double> [[BROADCAST_SPLATINSERT]], <4 x double> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds double, ptr [[TMP4]], i32 0
-; CHECK-NEXT:    store <4 x double> [[BROADCAST_SPLAT]], ptr [[TMP6]], align 8, !alias.scope [[META72]]
+; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds double, ptr [[TMP1]], i32 0
+; CHECK-NEXT:    store <4 x double> [[BROADCAST_SPLAT]], ptr [[TMP3]], align 8, !alias.scope [[META72]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
-; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], 0
-; CHECK-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP74:![0-9]+]]
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 0
+; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP74:![0-9]+]]
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    br i1 true, label [[OUTER_LATCH]], label [[SCALAR_PH]]
 ; CHECK:       scalar.ph: