[llvm] [LAA] Improve code in findForkedSCEVs (NFC) (PR #140384)

[David Sherwood via llvm-commits]

================
@@ -999,24 +999,17 @@ static void findForkedSCEVs(
       break;
     }
 
-    // Find the pointer type we need to extend to.
-    Type *IntPtrTy = SE->getEffectiveSCEVType(
-        SE->getSCEV(GEP->getPointerOperand())->getType());
-
-    // Find the size of the type being pointed to. We only have a single
-    // index term (guarded above) so we don't need to index into arrays or
-    // structures, just get the size of the scalar value.
-    const SCEV *Size = SE->getSizeOfExpr(IntPtrTy, SourceTy);
-
-    // Scale up the offsets by the size of the type, then add to the bases.
-    const SCEV *Scaled1 = SE->getMulExpr(
-        Size, SE->getTruncateOrSignExtend(get<0>(OffsetScevs[0]), IntPtrTy));
-    const SCEV *Scaled2 = SE->getMulExpr(
-        Size, SE->getTruncateOrSignExtend(get<0>(OffsetScevs[1]), IntPtrTy));
-    ScevList.emplace_back(SE->getAddExpr(get<0>(BaseScevs[0]), Scaled1),
-                          NeedsFreeze);
-    ScevList.emplace_back(SE->getAddExpr(get<0>(BaseScevs[1]), Scaled2),
-                          NeedsFreeze);
+    Type *PtrTy = GEP->getPointerOperandType();
----------------
david-arm wrote:

Hmm, it does look odd though, and you're also relying upon the internal behaviour of SCEV to do the right thing, which may be fine I guess. It doesn't seem to be documented in ScalarEvolution.h, although I can believe the author(s) wanted to make life simpler internally by doing the conversion. However, if you try to do the equivalent of `getTruncateOrSignExtend` at the IR level (see CastInst::Create), then it looks like it will assert when trying to truncate to something that isn't an integer type.

What is your objection to using integer types here?

https://github.com/llvm/llvm-project/pull/140384