[LAA] Use DL to get element size for bound computation.

[fhahn]

Summary

Currently LAA uses getScalarSizeInBits to compute the size of an element
when computing the end bound of an access.

This does not work as expected for pointers to pointers, because
getScalarSizeInBits will return 0 for pointer types.

By using DataLayout to get the size of the element we can also correctly
handle pointer element types.

Note the changes to the existing test, which seems to also use the wrong
offset for the end.

Fixes PR47751.

[efriedma]

The code change looks right to me, but it's my first time looking at this code.

I verified this fixes the original program I was looking at.

On a related note, the amount of code we generate in runtime checks for the given testcase is completely ridiculous. But that's not a correctness issue. :)

[anemet]

LGTM too.

[fhahn]

In D88953#2317212, @efriedma wrote:

On a related note, the amount of code we generate in runtime checks for the given testcase is completely ridiculous. But that's not a correctness issue. :)

Agreed, but I think this mostly boils down to improving SCEV expansion and follow-up simplifications, as LAA mostly just hands off SCEV expression to the expander.

llvm/lib/Analysis/LoopAccessAnalysis.cpp

else {
     } else {
       // Fallback case: the step is not constant, but we can still
       // get the upper and lower bounds of the interval by using min/max
       // expressions.
       ScStart = SE->getUMinExpr(ScStart, ScEnd);
       ScEnd = SE->getUMaxExpr(AR->getStart(), ScEnd);
     }
     // Add the size of the pointed element to ScEnd.
+    auto &DL = Lp->getHeader()->getModule()->getDataLayout();
     unsigned EltSize =
-      Ptr->getType()->getPointerElementType()->getScalarSizeInBits() / 8;
+        DL.getTypeStoreSizeInBits(Ptr->getType()->getPointerElementType()) / 8;
     const SCEV *EltSizeSCEV = SE->getConstant(ScEnd->getType(), EltSize);
     ScEnd = SE->getAddExpr(ScEnd, EltSizeSCEV);
   }
 
   Pointers.emplace_back(Ptr, ScStart, ScEnd, WritePtr, DepSetId, ASId, Sc);
 }
 
 SmallVector<RuntimePointerCheck, 4>

llvm/test/Transforms/LoopVectorize/X86/pr23997.ll

 ; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP2:%.*]], 1
 ; CHECK-NEXT:    [[UMAX:%.*]] = select i1 [[TMP3]], i64 [[TMP2]], i64 1
 ; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[UMAX]], 16
 ; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
 ; CHECK:       vector.memcheck:
 ; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i64 [[TMP2]], 1
 ; CHECK-NEXT:    [[UMAX1:%.*]] = select i1 [[TMP4]], i64 [[TMP2]], i64 1
 ; CHECK-NEXT:    [[TMP5:%.*]] = shl i64 [[UMAX1]], 3
-; CHECK-NEXT:    [[TMP6:%.*]] = add i64 [[TMP5]], 8
+; CHECK-NEXT:    [[TMP6:%.*]] = add i64 [[TMP5]], 16
 ; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr i8, i8 addrspace(1)* [[TMP0]], i64 [[TMP6]]
 ; CHECK-NEXT:    [[SCEVGEP2:%.*]] = getelementptr i8, i8 addrspace(1)* [[TMP1]], i64 [[TMP6]]
 ; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult i8 addrspace(1)* [[DOT10]], [[SCEVGEP2]]
 ; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult i8 addrspace(1)* [[DOT12]], [[SCEVGEP]]
-; CHECK-NEXT:    [[MEMCHECK_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
-; CHECK-NEXT:    br i1 [[MEMCHECK_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
+; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
+; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
 ; CHECK:       vector.ph:
 ; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[UMAX]], -16
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[DOT13]], i64 [[INDEX]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP7]] to <4 x i8 addrspace(1)*> addrspace(1)*
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i8 addrspace(1)*>, <4 x i8 addrspace(1)*> addrspace(1)* [[TMP8]], align 8, !alias.scope !0
 ; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[TMP7]], i64 4
 ; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP9]] to <4 x i8 addrspace(1)*> addrspace(1)*
-; CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i8 addrspace(1)*>, <4 x i8 addrspace(1)*> addrspace(1)* [[TMP10]], align 8, !alias.scope !0
+; CHECK-NEXT:    [[WIDE_LOAD3:%.*]] = load <4 x i8 addrspace(1)*>, <4 x i8 addrspace(1)*> addrspace(1)* [[TMP10]], align 8, !alias.scope !0
 ; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[TMP7]], i64 8
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP11]] to <4 x i8 addrspace(1)*> addrspace(1)*
-; CHECK-NEXT:    [[WIDE_LOAD7:%.*]] = load <4 x i8 addrspace(1)*>, <4 x i8 addrspace(1)*> addrspace(1)* [[TMP12]], align 8, !alias.scope !0
+; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x i8 addrspace(1)*>, <4 x i8 addrspace(1)*> addrspace(1)* [[TMP12]], align 8, !alias.scope !0
 ; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[TMP7]], i64 12
 ; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP13]] to <4 x i8 addrspace(1)*> addrspace(1)*
-; CHECK-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x i8 addrspace(1)*>, <4 x i8 addrspace(1)*> addrspace(1)* [[TMP14]], align 8, !alias.scope !0
+; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x i8 addrspace(1)*>, <4 x i8 addrspace(1)*> addrspace(1)* [[TMP14]], align 8, !alias.scope !0
 ; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[DOT11]], i64 [[INDEX]]
 ; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP15]] to <4 x i8 addrspace(1)*> addrspace(1)*
 ; CHECK-NEXT:    store <4 x i8 addrspace(1)*> [[WIDE_LOAD]], <4 x i8 addrspace(1)*> addrspace(1)* [[TMP16]], align 8, !alias.scope !3, !noalias !0
 ; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[TMP15]], i64 4
 ; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP17]] to <4 x i8 addrspace(1)*> addrspace(1)*
-; CHECK-NEXT:    store <4 x i8 addrspace(1)*> [[WIDE_LOAD6]], <4 x i8 addrspace(1)*> addrspace(1)* [[TMP18]], align 8, !alias.scope !3, !noalias !0
+; CHECK-NEXT:    store <4 x i8 addrspace(1)*> [[WIDE_LOAD3]], <4 x i8 addrspace(1)*> addrspace(1)* [[TMP18]], align 8, !alias.scope !3, !noalias !0
 ; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[TMP15]], i64 8
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP19]] to <4 x i8 addrspace(1)*> addrspace(1)*
-; CHECK-NEXT:    store <4 x i8 addrspace(1)*> [[WIDE_LOAD7]], <4 x i8 addrspace(1)*> addrspace(1)* [[TMP20]], align 8, !alias.scope !3, !noalias !0
+; CHECK-NEXT:    store <4 x i8 addrspace(1)*> [[WIDE_LOAD4]], <4 x i8 addrspace(1)*> addrspace(1)* [[TMP20]], align 8, !alias.scope !3, !noalias !0
 ; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[TMP15]], i64 12
 ; CHECK-NEXT:    [[TMP22:%.*]] = bitcast i8 addrspace(1)* addrspace(1)* [[TMP21]] to <4 x i8 addrspace(1)*> addrspace(1)*
-; CHECK-NEXT:    store <4 x i8 addrspace(1)*> [[WIDE_LOAD8]], <4 x i8 addrspace(1)*> addrspace(1)* [[TMP22]], align 8, !alias.scope !3, !noalias !0
+; CHECK-NEXT:    store <4 x i8 addrspace(1)*> [[WIDE_LOAD5]], <4 x i8 addrspace(1)*> addrspace(1)* [[TMP22]], align 8, !alias.scope !3, !noalias !0
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !5
+; CHECK-NEXT:    br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP5:!llvm.loop !.*]]
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[UMAX]], [[N_VEC]]
 ; CHECK-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
 ; CHECK:       scalar.ph:
 ; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[INDVARS_IV3:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT4:%.*]], [[LOOP]] ]
 ; CHECK-NEXT:    [[DOT18:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[DOT13]], i64 [[INDVARS_IV3]]
 ; CHECK-NEXT:    [[V:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[DOT18]], align 8
 ; CHECK-NEXT:    [[DOT20:%.*]] = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* [[DOT11]], i64 [[INDVARS_IV3]]
 ; CHECK-NEXT:    store i8 addrspace(1)* [[V]], i8 addrspace(1)* addrspace(1)* [[DOT20]], align 8
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT4]] = add nuw nsw i64 [[INDVARS_IV3]], 1
 ; CHECK-NEXT:    [[DOT21:%.*]] = icmp ult i64 [[INDVARS_IV_NEXT4]], [[TMP2]]
-; CHECK-NEXT:    br i1 [[DOT21]], label [[LOOP]], label [[LOOPEXIT]], !llvm.loop !7
+; CHECK-NEXT:    br i1 [[DOT21]], label [[LOOP]], label [[LOOPEXIT]], [[LOOP7:!llvm.loop !.*]]
 ; CHECK:       loopexit:
 ; CHECK-NEXT:    ret void
 ;
 entry:
   br label %preheader
 
 preheader:                                       ; preds = %4
   %.10 = getelementptr inbounds i8, i8 addrspace(1)* %0, i64 16

llvm/test/Transforms/LoopVectorize/runtime-check-pointer-element-type.ll

 ; RUN: opt -loop-vectorize -force-vector-width=2 %s -S -debug 2>&1 | FileCheck %s
 ; RUN: opt -passes='loop-vectorize' -force-vector-width=2 %s -S -debug 2>&1 | FileCheck %s
 
 ; REQUIRES: asserts
 
 target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 
 ; FIXME
 ; Test case for PR47751. Make sure the runtime check includes a required
 ; addition of the size of the element type (a pointer) for the end bound.
 
 define void @test(i64 %arg, i32 %arg1, i8** %base) {
 ; CHECK:      LAA: Adding RT check for range:
 ; CHECK-NEXT:  Start: ((8 * (zext i32 (-1 + %arg1)<nsw> to i64))<nuw><nsw> + (8 * (1 smin %arg)) + (-8 * %arg) + %base)
-; CHECK-SAME:  End: ((8 * (zext i32 (-1 + %arg1)<nsw> to i64))<nuw><nsw> + %base)
+; CHECK-SAME:  End: (8 + (8 * (zext i32 (-1 + %arg1)<nsw> to i64))<nuw><nsw> + %base)
 ; CHECK-NEXT: LAA: Adding RT check for range:
 ; CHECK-NEXT:  Start: ((8 * (1 smin %arg)) + %base)
-; CHECK-SAME:  End: ((8 * %arg) + %base)<nsw>
+; CHECK-SAME:  End: (8 + (8 * %arg) + %base)
 
 ; CHECK: vector.body
 
 entry:
   br label %loop
 
 loop:
   %iv.1 = phi i64 [ %arg, %entry ], [ %iv.1.next, %loop ]