[polly] r252451 - [FIX] Use same alloca for invariant loads and the scalar users

[Johannes Doerfert via llvm-commits]

Author: jdoerfert
Date: Mon Nov  9 00:28:45 2015
New Revision: 252451

URL: http://llvm.org/viewvc/llvm-project?rev=252451&view=rev
Log:
[FIX] Use same alloca for invariant loads and the scalar users


Added:
    polly/trunk/test/Isl/CodeGen/invariant_load_scalar_escape_alloca_sharing.ll
Modified:
    polly/trunk/lib/CodeGen/IslNodeBuilder.cpp

Modified: polly/trunk/lib/CodeGen/IslNodeBuilder.cpp
URL: http://llvm.org/viewvc/llvm-project/polly/trunk/lib/CodeGen/IslNodeBuilder.cpp?rev=252451&r1=252450&r2=252451&view=diff
==============================================================================
--- polly/trunk/lib/CodeGen/IslNodeBuilder.cpp (original)
+++ polly/trunk/lib/CodeGen/IslNodeBuilder.cpp Mon Nov  9 00:28:45 2015
@@ -991,25 +991,35 @@ bool IslNodeBuilder::preloadInvariantEqu
     isl_id_free(ParamId);
   }
 
+  BasicBlock *EntryBB = &Builder.GetInsertBlock()->getParent()->getEntryBlock();
+  auto *Alloca = new AllocaInst(AccInstTy, AccInst->getName() + ".preload.s2a");
+  Alloca->insertBefore(&*EntryBB->getFirstInsertionPt());
+  Builder.CreateStore(PreloadVal, Alloca);
+
   for (auto *DerivedSAI : SAI->getDerivedSAIs()) {
     Value *BasePtr = DerivedSAI->getBasePtr();
 
-    // As the derived SAI information is quite coarse, any load from the current
-    // SAI could be the base pointer of the derived SAI, however we should only
-    // change the base pointer of the derived SAI if we actually preloaded it.
     for (const MemoryAccess *MA : MAs) {
-      if (BasePtr != MA->getBaseAddr())
-        continue;
-      BasePtr = Builder.CreateBitOrPointerCast(PreloadVal, BasePtr->getType());
-      DerivedSAI->setBasePtr(BasePtr);
+      // As the derived SAI information is quite coarse, any load from the
+      // current SAI could be the base pointer of the derived SAI, however we
+      // should only change the base pointer of the derived SAI if we actually
+      // preloaded it.
+      if (BasePtr == MA->getBaseAddr()) {
+        BasePtr =
+            Builder.CreateBitOrPointerCast(PreloadVal, BasePtr->getType());
+        DerivedSAI->setBasePtr(BasePtr);
+      }
+
+      // For scalar derived SAIs we remap the alloca used for the derived value.
+      if (BasePtr == MA->getAccessInstruction()) {
+        if (DerivedSAI->isPHI())
+          PHIOpMap[BasePtr] = Alloca;
+        else
+          ScalarMap[BasePtr] = Alloca;
+      }
     }
   }
 
-  BasicBlock *EntryBB = &Builder.GetInsertBlock()->getParent()->getEntryBlock();
-  auto *Alloca = new AllocaInst(AccInstTy, AccInst->getName() + ".preload.s2a");
-  Alloca->insertBefore(&*EntryBB->getFirstInsertionPt());
-  Builder.CreateStore(PreloadVal, Alloca);
-
   const Region &R = S.getRegion();
   for (const MemoryAccess *MA : MAs) {
 

Added: polly/trunk/test/Isl/CodeGen/invariant_load_scalar_escape_alloca_sharing.ll
URL: http://llvm.org/viewvc/llvm-project/polly/trunk/test/Isl/CodeGen/invariant_load_scalar_escape_alloca_sharing.ll?rev=252451&view=auto
==============================================================================
--- polly/trunk/test/Isl/CodeGen/invariant_load_scalar_escape_alloca_sharing.ll (added)
+++ polly/trunk/test/Isl/CodeGen/invariant_load_scalar_escape_alloca_sharing.ll Mon Nov  9 00:28:45 2015
@@ -0,0 +1,66 @@
+; RUN: opt %loadPolly -polly-codegen -S < %s | FileCheck %s
+;
+; Verify the preloaded %0 is stored and communicated in the same alloca.
+;
+; CHECK-NOT: alloca
+; CHECK:     %dec3.s2a = alloca i32
+; CHECK-NOT: alloca
+; CHECK:     %dec3.in.phiops = alloca i32
+; CHECK-NOT: alloca
+; CHECK:     %.preload.s2a = alloca i32
+; CHECK-NOT: alloca
+;
+; CHECK:       %ncol.load = load i32, i32* @ncol
+; CHECK-NEXT:  store i32 %ncol.load, i32* %.preload.s2a
+;
+; CHECK:      polly.stmt.while.body.lr.ph:
+; CHECK-NEXT:   %.preload.s2a.reload = load i32, i32* %.preload.s2a
+; CHECK-NEXT:   store i32 %.preload.s2a.reload, i32* %dec3.in.phiops
+;
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+
+ at ncol = external global i32, align 4
+
+define void @melt_data(i32* %data1, i32* %data2) {
+entry:
+  br label %entry.split
+
+entry.split:                                      ; preds = %entry
+  %0 = load i32, i32* @ncol, align 4
+  %tobool.2 = icmp eq i32 %0, 0
+  br i1 %tobool.2, label %while.end, label %while.body.lr.ph
+
+while.body.lr.ph:                                 ; preds = %entry.split
+  br label %while.body
+
+while.body:                                       ; preds = %while.body.lr.ph, %while.cond.backedge
+  %dec3.in = phi i32 [ %0, %while.body.lr.ph ], [ %dec3, %while.cond.backedge ]
+  %dec3 = add nsw i32 %dec3.in, -1
+  %idxprom = sext i32 %dec3 to i64
+  %arrayidx = getelementptr inbounds i32, i32* %data1, i64 %idxprom
+  %1 = load i32, i32* %arrayidx, align 4
+  %idxprom1 = sext i32 %dec3 to i64
+  %arrayidx2 = getelementptr inbounds i32, i32* %data2, i64 %idxprom1
+  %2 = load i32, i32* %arrayidx2, align 4
+  %cmp = icmp sgt i32 %1, %2
+  br i1 %cmp, label %if.then, label %while.cond.backedge
+
+if.then:                                          ; preds = %while.body
+  %idxprom3 = sext i32 %dec3 to i64
+  %arrayidx4 = getelementptr inbounds i32, i32* %data2, i64 %idxprom3
+  %3 = load i32, i32* %arrayidx4, align 4
+  %idxprom5 = sext i32 %dec3 to i64
+  %arrayidx6 = getelementptr inbounds i32, i32* %data1, i64 %idxprom5
+  store i32 %3, i32* %arrayidx6, align 4
+  br label %while.cond.backedge
+
+while.cond.backedge:                              ; preds = %if.then, %while.body
+  %tobool = icmp eq i32 %dec3, 0
+  br i1 %tobool, label %while.cond.while.end_crit_edge, label %while.body
+
+while.cond.while.end_crit_edge:                   ; preds = %while.cond.backedge
+  br label %while.end
+
+while.end:                                        ; preds = %while.cond.while.end_crit_edge, %entry.split
+  ret void
+}