[PATCH] D108112: [LoopIdiom] Let LIR fold memset pointer / stride SCEV regarding loop guards

[Yueh-Ting Chen via Phabricator via llvm-commits]

eopXD updated this revision to Diff 393150.
eopXD added a comment.

Rebase to latest main.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D108112/new/

https://reviews.llvm.org/D108112

Files:
  llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
  llvm/test/Transforms/LoopIdiom/memset-runtime-debug.ll


Index: llvm/test/Transforms/LoopIdiom/memset-runtime-debug.ll
===================================================================
--- llvm/test/Transforms/LoopIdiom/memset-runtime-debug.ll
+++ llvm/test/Transforms/LoopIdiom/memset-runtime-debug.ll
@@ -19,6 +19,9 @@
 ; CHECK-NEXT: memset size is non-constant
 ; CHECK-NEXT: MemsetSizeSCEV: (4 * (sext i32 %m to i64))<nsw>
 ; CHECK-NEXT: PositiveStrideSCEV: (4 + (4 * (sext i32 %m to i64))<nsw>)<nsw>
+; CHECK-NEXT: Try to fold SCEV based on loop guard
+; CHECK-NEXT: FoldedMemsetSize: (4 * (sext i32 %m to i64))<nsw>
+; CHECK-NEXT: FoldedPositiveStride: (4 + (4 * (sext i32 %m to i64))<nsw>)<nsw>
 ; CHECK-NEXT: SCEV don't match, abort
 ; CHECK: loop-idiom Scanning: F[NonZeroAddressSpace] Countable Loop %for.cond1.preheader
 ; CHECK-NEXT: memset size is non-constant
Index: llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
===================================================================
--- llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
+++ llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
@@ -307,6 +307,24 @@
   }
 };
 
+// The Folder will fold expressions that are guarded by the loop entry.
+class SCEVFolder : public SCEVRewriteVisitor<SCEVFolder> {
+public:
+  ScalarEvolution &SE;
+  const Loop *CurLoop;
+  SCEVFolder(ScalarEvolution &SE, const Loop *CurLoop)
+      : SCEVRewriteVisitor(SE), SE(SE), CurLoop(CurLoop) {}
+
+  const SCEV *visitSignExtendExpr(const SCEVSignExtendExpr *Expr) {
+    // If expression is guarded by CurLoop to be greater or equal to zero
+    // then convert sext to zext. Otherwise return the original expression.
+    if (SE.isLoopEntryGuardedByCond(CurLoop, ICmpInst::ICMP_SGE, Expr,
+                                    SE.getZero(Expr->getType())))
+      return SE.getZeroExtendExpr(visit(Expr->getOperand()), Expr->getType());
+    return Expr;
+  }
+};
+
 } // end anonymous namespace
 
 char LoopIdiomRecognizeLegacyPass::ID = 0;
@@ -967,12 +985,22 @@
                       << "\n");
 
     if (PositiveStrideSCEV != MemsetSizeSCEV) {
-      // TODO: folding can be done to the SCEVs
-      // The folding is to fold expressions that is covered by the loop guard
-      // at loop entry. After the folding, compare again and proceed
-      // optimization if equal.
-      LLVM_DEBUG(dbgs() << "  SCEV don't match, abort\n");
-      return false;
+      // The folding is to fold an expression that is covered by the loop guard
+      // at loop entry. After the folding, compare again and proceed with
+      // optimization, if equal.
+      SCEVFolder Folder(*SE, CurLoop);
+      const SCEV *FoldedPositiveStride = Folder.visit(PositiveStrideSCEV);
+      const SCEV *FoldedMemsetSize = Folder.visit(MemsetSizeSCEV);
+
+      LLVM_DEBUG(dbgs() << "  Try to fold SCEV based on loop guard\n"
+                        << "    FoldedMemsetSize: " << *FoldedMemsetSize << "\n"
+                        << "    FoldedPositiveStride: " << *FoldedPositiveStride
+                        << "\n");
+
+      if (FoldedPositiveStride != FoldedMemsetSize) {
+        LLVM_DEBUG(dbgs() << "  SCEV don't match, abort\n");
+        return false;
+      }
     }
   }
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D108112.393150.patch
Type: text/x-patch
Size: 3167 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211209/5c9b268f/attachment.bin>