[PATCH] D14485: Vector GEP - a fix in InstCombine

[Elena Demikhovsky via llvm-commits]

delena created this revision.
delena added a reviewer: joker-eph.
delena added a subscriber: llvm-commits.
delena set the repository for this revision to rL LLVM.

The current implementation of GEP visitor in InstCombine fails with assertion on Vector GEP with mix of scalar and vector types, like this:

getelementptr double, double* %a, <8 x i32> %i
(It fails to create a "sext" from <8 x i32> to <8 x i64>)

I fixed it and added some tests.

Repository:
  rL LLVM

http://reviews.llvm.org/D14485

Files:
  ../lib/Transforms/InstCombine/InstructionCombining.cpp
  ../test/Transforms/InstCombine/vector_gep2.ll

Index: ../test/Transforms/InstCombine/vector_gep2.ll
===================================================================
--- ../test/Transforms/InstCombine/vector_gep2.ll
+++ ../test/Transforms/InstCombine/vector_gep2.ll
@@ -9,3 +9,26 @@
 ; CHECK: getelementptr i8, <2 x i8*> %a, <2 x i64> <i64 0, i64 1>
   ret <2 x i8*> %g
 }
+
+define <8 x double*> @vgep_s_v8i64(double* %a, <8 x i64>%i) {
+; CHECK-LABEL: @vgep_s_v8i64
+; CHECK: getelementptr double, double* %a, <8 x i64> %i
+  %VectorGep = getelementptr double, double* %a, <8 x i64> %i
+  ret <8 x double*> %VectorGep
+}
+
+define <8 x double*> @vgep_s_v8i32(double* %a, <8 x i32>%i) {
+; CHECK-LABEL: @vgep_s_v8i32
+; CHECK: %1 = sext <8 x i32> %i to <8 x i64>
+; CHECK: getelementptr double, double* %a, <8 x i64> %1
+  %VectorGep = getelementptr double, double* %a, <8 x i32> %i
+  ret <8 x double*> %VectorGep
+}
+
+define <8 x i8*> @vgep_v8iPtr_i32(<8 x i8*> %a, i32 %i) {
+; CHECK-LABEL: @vgep_v8iPtr_i32
+; CHECK:  %1 = sext i32 %i to i64
+; CHECK:  %VectorGep = getelementptr i8, <8 x i8*> %a, i64 %1
+  %VectorGep = getelementptr i8, <8 x i8*> %a, i32 %i
+  ret <8 x i8*> %VectorGep
+}
Index: ../lib/Transforms/InstCombine/InstructionCombining.cpp
===================================================================
--- ../lib/Transforms/InstCombine/InstructionCombining.cpp
+++ ../lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -1355,7 +1355,8 @@
   // Eliminate unneeded casts for indices, and replace indices which displace
   // by multiples of a zero size type with zero.
   bool MadeChange = false;
-  Type *IntPtrTy = DL.getIntPtrType(GEP.getPointerOperandType());
+  Type *IntPtrTy =
+    DL.getIntPtrType(GEP.getPointerOperandType()->getScalarType());
 
   gep_type_iterator GTI = gep_type_begin(GEP);
   for (User::op_iterator I = GEP.op_begin() + 1, E = GEP.op_end(); I != E;
@@ -1365,21 +1366,25 @@
     if (!SeqTy)
       continue;
 
+    // Index type should have the same width as IntPtr
+    Type *IndexTy = (*I)->getType();
+    Type *NewIndexType = IndexTy->isVectorTy() ?
+      VectorType::get(IntPtrTy, IndexTy->getVectorNumElements()) : IntPtrTy;
+ 
     // If the element type has zero size then any index over it is equivalent
     // to an index of zero, so replace it with zero if it is not zero already.
     if (SeqTy->getElementType()->isSized() &&
         DL.getTypeAllocSize(SeqTy->getElementType()) == 0)
       if (!isa<Constant>(*I) || !cast<Constant>(*I)->isNullValue()) {
-        *I = Constant::getNullValue(IntPtrTy);
+        *I = Constant::getNullValue(NewIndexType);
         MadeChange = true;
       }
 
-    Type *IndexTy = (*I)->getType();
-    if (IndexTy != IntPtrTy) {
+    if (IndexTy != NewIndexType) {
       // If we are using a wider index than needed for this platform, shrink
       // it to what we need.  If narrower, sign-extend it to what we need.
       // This explicit cast can make subsequent optimizations more obvious.
-      *I = Builder->CreateIntCast(*I, IntPtrTy, true);
+      *I = Builder->CreateIntCast(*I, NewIndexType, true);
       MadeChange = true;
     }
   }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D14485.39650.patch
Type: text/x-patch
Size: 3113 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151108/978bd6ad/attachment.bin>