[InstCombine] combineLoadToOperationType(): don't fold int<->ptr cast into load

[lebedev.ri]

Summary

And another step towards transforms not introducing inttoptr and/or
ptrtoint casts that weren't there already.

As we've been establishing (see D88788/D88789), if there is a int<->ptr cast,
it basically must stay as-is, we can't do much with it.

I've looked, and the most source of new such casts being introduces,
as far as i can tell, is this transform, which, ironically,
tries to reduce count of casts..

On vanilla llvm test-suite + RawSpeed, @ -O3, this results in
-33.58% less IntToPtrs (19014 -> 12629)
and +76.20% more PtrToInts (18589 -> 32753),
which is an increase of +20.69% in total.

However just on RawSpeed, where i know there are basically
none IntToPtr in the original source code,
this results in -99.27% less IntToPtrs (2724 -> 20)
and +82.92% more PtrToInts (4513 -> 8255).
which is again an increase of 14.34% in total.

To me this does seem like the step in the right direction,
we end up with strictly less IntToPtr, but strictly more PtrToInt,
which seems like a reasonable trade-off.

I'm not presently sure what preparatory fixes are needed before this though.

(Eventually, CastInst::isNoopCast()/CastInst::isEliminableCastPair
should be taught about this, yes)

[lebedev.ri]

Rebased, NFC.

[nlopes]

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

563

What about fixing isNoopCast() directly?
Is the impact too bad/large?

[lebedev.ri]

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

563

I don't recall, actually. That is the final step, yes,
but i think we can get most of the way there with small incremental changes.

[lebedev.ri]

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

563

(what i mean is, if "let's just do it" is the review feedback, then i can of course do it all at once..)

[nikic]

LGTM

Looking through other uses of isNoopCast(), I don't think it makes sense to push this change into it, as many other usages do need it to work with ptrtoint/inttoptr (some of them using it specifically for them). The comment above the function indicates that "no-op" is to be understood as "generates no code" here. Possibly it could do with a rename.

[nikic]

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

560

You might want to be a tad more explicit here, e.g. add "because they have different provenance" or so.

[lebedev.ri]

In D88979#2323935, @nikic wrote:

LGTM

@nlopes does this look good to you?

Looking through other uses of isNoopCast(), I don't think it makes sense to push this change into it, as many other usages do need it to work with ptrtoint/inttoptr (some of them using it specifically for them). The comment above the function indicates that "no-op" is to be understood as "generates no code" here. Possibly it could do with a rename.

I think i don't agree with you there.
I agree with @nlopes, the end goal will be to basically disallow fusing of inttoptr/ptrtoint into loads,
disallow dropping inttoptr-of-ptrtoint/ptrtoint-of-inttoptr, etc.
And all that eventually boils down to updating CastInst::isNoopCast()/CastInst::isEliminableCastPair().

[nlopes]

In D88979#2323940, @lebedev.ri wrote:

In D88979#2323935, @nikic wrote:

LGTM

@nlopes does this look good to you?

Looking through other uses of isNoopCast(), I don't think it makes sense to push this change into it, as many other usages do need it to work with ptrtoint/inttoptr (some of them using it specifically for them). The comment above the function indicates that "no-op" is to be understood as "generates no code" here. Possibly it could do with a rename.

I think i don't agree with you there.
I agree with @nlopes, the end goal will be to basically disallow fusing of inttoptr/ptrtoint into loads,
disallow dropping inttoptr-of-ptrtoint/ptrtoint-of-inttoptr, etc.
And all that eventually boils down to updating CastInst::isNoopCast()/CastInst::isEliminableCastPair().

I'm ok with this change, and I agree it's a good step forwards. Thanks!
My only concern was that it seems this code will need further changes going forward (maybe even revert this change?) and I wouldn't want us to forget to revisit this code if/when needed.

[lebedev.ri]

@nlopes @nikic thank you for the review!

In D88979#2323948, @nlopes wrote:

In D88979#2323940, @lebedev.ri wrote:

In D88979#2323935, @nikic wrote:

LGTM

@nlopes does this look good to you?

Looking through other uses of isNoopCast(), I don't think it makes sense to push this change into it, as many other usages do need it to work with ptrtoint/inttoptr (some of them using it specifically for them). The comment above the function indicates that "no-op" is to be understood as "generates no code" here. Possibly it could do with a rename.

I think i don't agree with you there.
I agree with @nlopes, the end goal will be to basically disallow fusing of inttoptr/ptrtoint into loads,
disallow dropping inttoptr-of-ptrtoint/ptrtoint-of-inttoptr, etc.
And all that eventually boils down to updating CastInst::isNoopCast()/CastInst::isEliminableCastPair().

I'm ok with this change, and I agree it's a good step forwards. Thanks!
My only concern was that it seems this code will need further changes going forward (maybe even revert this change?) and I wouldn't want us to forget to revisit this code if/when needed.

Once those methods are fixed, sure, the uses will need dead code cleanup.

[nikic]

In D88979#2323940, @lebedev.ri wrote:

In D88979#2323935, @nikic wrote:

Looking through other uses of isNoopCast(), I don't think it makes sense to push this change into it, as many other usages do need it to work with ptrtoint/inttoptr (some of them using it specifically for them). The comment above the function indicates that "no-op" is to be understood as "generates no code" here. Possibly it could do with a rename.

I think i don't agree with you there.
I agree with @nlopes, the end goal will be to basically disallow fusing of inttoptr/ptrtoint into loads,
disallow dropping inttoptr-of-ptrtoint/ptrtoint-of-inttoptr, etc.
And all that eventually boils down to updating CastInst::isNoopCast()/CastInst::isEliminableCastPair().

I agree with the general goal -- my point here is that changing isNoopCast() may not be the way to achieve that, because at least some of the current usages do need to include ptr/int casts, and can include them safely (not all usages result in type punning).

In fact, we already have a way to write "isNoopCast() without ptrtoint or inttoptr": isa<BitCast>. This didn't quite click before.... You might want to replace CI->isNoopCast(DL) && LI.getType()->isPtrOrPtrVectorTy() == CI->getDestTy()->isPtrOrPtrVectorTy()) here with just isa<BitCast>(CI).

[aqjune]

+1 for this patch!

clang/test/CodeGen/arm64_32-vaarg.c

 
 
 typedef struct {
   long long a;
 } OneLongLong;
 
 // Minimum slot size is 4 bytes, so address needs rounding up to multiple of 8.
 long long test_longlong(OneLongLong input, va_list *mylist) {
   // CHECK-LABEL: define i64 @test_longlong(i64 %input
-// CHECK: [[STARTPTR:%.*]] = bitcast i8** %mylist to i32*
-// CHECK: [[START:%.*]] = load i32, i32* [[STARTPTR]]
+  // CHECK: [[STARTPTR:%.*]] = load i8*, i8** %mylist
+  // CHECK: [[START:%.*]] = ptrtoint i8* [[STARTPTR]] to i32
 
   // CHECK: [[ALIGN_TMP:%.*]] = add i32 [[START]], 7
   // CHECK: [[ALIGNED:%.*]] = and i32 [[ALIGN_TMP]], -8
   // CHECK: [[ALIGNED_ADDR:%.*]] = inttoptr i32 [[ALIGNED]] to i8*
   // CHECK: [[NEXT:%.*]] = getelementptr inbounds i8, i8* [[ALIGNED_ADDR]], i32 8
   // CHECK: store i8* [[NEXT]], i8** %mylist
 
   // CHECK: [[ADDR_STRUCT:%.*]] = inttoptr i32 [[ALIGNED]] to %struct.OneLongLong*
   // CHECK: [[ADDR_I64:%.*]] = getelementptr inbounds %struct.OneLongLong, %struct.OneLongLong* [[ADDR_STRUCT]], i32 0, i32 0
   // CHECK: [[RES:%.*]] = load i64, i64* [[ADDR_I64]]
   // CHECK: ret i64 [[RES]]
 
   return va_arg(*mylist, OneLongLong).a;
 }
 
 
 typedef struct {
   float arr[4];
 } HFA;

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

static Instruction *combineLoadToOperationType(InstCombinerImpl &IC,
 
   // swifterror values can't be bitcasted.
   if (LI.getPointerOperand()->isSwiftError())
     return nullptr;
 
   const DataLayout &DL = IC.getDataLayout();
 
   // Fold away bit casts of the loaded value by loading the desired type.
-  // We can do this for BitCastInsts as well as casts from and to pointer types,
-  // as long as those are noops (i.e., the source or dest type have the same
-  // bitwidth as the target's pointers).
+  // Note that we should not do this for pointer<->integer casts!
   if (LI.hasOneUse())
     if (auto* CI = dyn_cast<CastInst>(LI.user_back()))
-      if (CI->isNoopCast(DL))
+      if (CI->isNoopCast(DL) && LI.getType()->isPtrOrPtrVectorTy() ==
+                                    CI->getDestTy()->isPtrOrPtrVectorTy())
         if (!LI.isAtomic() || isSupportedAtomicType(CI->getDestTy())) {
           LoadInst *NewLoad = IC.combineLoadToNewType(LI, CI->getDestTy());
           CI->replaceAllUsesWith(NewLoad);
           IC.eraseInstFromFunction(*CI);
           return &LI;
         }
 
   // FIXME: We should also canonicalize loads of vectors when their elements are

llvm/test/Transforms/InstCombine/PR30597.ll

entry-block:
   ret i1 %switchtmp
 
 }
 
 ; Function Attrs: readonly uwtable
 define i64* @function(i64* noalias nocapture readonly dereferenceable(8)) {
 ; CHECK-LABEL: @function(
 ; CHECK-NEXT:  entry-block:
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[TMP0:%.*]] to i64**
-; CHECK-NEXT:    [[LOADED1:%.*]] = load i64*, i64** [[TMP1]], align 8, !nonnull !0
-; CHECK-NEXT:    ret i64* [[LOADED1]]
+; CHECK-NEXT:    [[LOADED:%.*]] = load i64, i64* [[TMP0:%.*]], align 8, [[RNG0:!range !.*]]
+; CHECK-NEXT:    [[INTTOPTR:%.*]] = inttoptr i64 [[LOADED]] to i64*
+; CHECK-NEXT:    ret i64* [[INTTOPTR]]
 ;
 entry-block:
   %loaded = load i64, i64* %0, align 8, !range !1
   %inttoptr = inttoptr i64 %loaded to i64*
   ret i64* %inttoptr
 }
 
 
 !0 = !{}
 !1 = !{i64 1, i64 140737488355327}

llvm/test/Transforms/InstCombine/intptr1.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt < %s  -instcombine  -S | FileCheck %s
 
 define void @test1(float* %a, float* readnone %a_end, i64* %b.i64) {
 ; CHECK-LABEL: @test1(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult float* [[A:%.*]], [[A_END:%.*]]
 ; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
 ; CHECK:       for.body.preheader:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i64* [[B_I64:%.*]] to float**
-; CHECK-NEXT:    [[B1:%.*]] = load float*, float** [[TMP0]], align 8
+; CHECK-NEXT:    [[B:%.*]] = load i64, i64* [[B_I64:%.*]], align 8
+; CHECK-NEXT:    [[B_PTR:%.*]] = inttoptr i64 [[B]] to float*
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[A_ADDR_03:%.*]] = phi float* [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[A]], [[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT:    [[B_ADDR_02_PTR:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B1]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[B_ADDR_02_PTR:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B_PTR]], [[FOR_BODY_PREHEADER]] ]
 ; CHECK-NEXT:    [[TMP1:%.*]] = load float, float* [[B_ADDR_02_PTR]], align 4
 ; CHECK-NEXT:    [[MUL_I:%.*]] = fmul float [[TMP1]], 4.200000e+01
 ; CHECK-NEXT:    store float [[MUL_I]], float* [[A_ADDR_03]], align 4
 ; CHECK-NEXT:    [[ADD]] = getelementptr inbounds float, float* [[B_ADDR_02_PTR]], i64 1
 ; CHECK-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds float, float* [[A_ADDR_03]], i64 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ult float* [[INCDEC_PTR]], [[A_END]]
 ; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END]]
 ; CHECK:       for.end:

 
 
 define void @test2(float* %a, float* readnone %a_end, float** %b.float) {
 ; CHECK-LABEL: @test2(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult float* [[A:%.*]], [[A_END:%.*]]
 ; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
 ; CHECK:       for.body.preheader:
-; CHECK-NEXT:    [[B1:%.*]] = load float*, float** [[B_FLOAT:%.*]], align 8
+; CHECK-NEXT:    [[B_I64:%.*]] = bitcast float** [[B_FLOAT:%.*]] to i64*
+; CHECK-NEXT:    [[B:%.*]] = load i64, i64* [[B_I64]], align 8
+; CHECK-NEXT:    [[B_PTR:%.*]] = inttoptr i64 [[B]] to float*
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[A_ADDR_03:%.*]] = phi float* [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[A]], [[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT:    [[B_ADDR_02_PTR:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B1]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[B_ADDR_02_PTR:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B_PTR]], [[FOR_BODY_PREHEADER]] ]
 ; CHECK-NEXT:    [[TMP1:%.*]] = load float, float* [[B_ADDR_02_PTR]], align 4
 ; CHECK-NEXT:    [[MUL_I:%.*]] = fmul float [[TMP1]], 4.200000e+01
 ; CHECK-NEXT:    store float [[MUL_I]], float* [[A_ADDR_03]], align 4
 ; CHECK-NEXT:    [[ADD]] = getelementptr inbounds float, float* [[B_ADDR_02_PTR]], i64 1
 ; CHECK-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds float, float* [[A_ADDR_03]], i64 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ult float* [[INCDEC_PTR]], [[A_END]]
 ; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END]]
 ; CHECK:       for.end:

 
 
 define void @test3(float* %a, float* readnone %a_end, i8** %b.i8p) {
 ; CHECK-LABEL: @test3(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult float* [[A:%.*]], [[A_END:%.*]]
 ; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
 ; CHECK:       for.body.preheader:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8** [[B_I8P:%.*]] to float**
-; CHECK-NEXT:    [[B1:%.*]] = load float*, float** [[TMP0]], align 8
+; CHECK-NEXT:    [[B_I64:%.*]] = bitcast i8** [[B_I8P:%.*]] to i64*
+; CHECK-NEXT:    [[B:%.*]] = load i64, i64* [[B_I64]], align 8
+; CHECK-NEXT:    [[B_PTR:%.*]] = inttoptr i64 [[B]] to float*
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[A_ADDR_03:%.*]] = phi float* [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[A]], [[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT:    [[B_ADDR_02_PTR:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B1]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[B_ADDR_02_PTR:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B_PTR]], [[FOR_BODY_PREHEADER]] ]
 ; CHECK-NEXT:    [[TMP1:%.*]] = load float, float* [[B_ADDR_02_PTR]], align 4
 ; CHECK-NEXT:    [[MUL_I:%.*]] = fmul float [[TMP1]], 4.200000e+01
 ; CHECK-NEXT:    store float [[MUL_I]], float* [[A_ADDR_03]], align 4
 ; CHECK-NEXT:    [[ADD]] = getelementptr inbounds float, float* [[B_ADDR_02_PTR]], i64 1
 ; CHECK-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds float, float* [[A_ADDR_03]], i64 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ult float* [[INCDEC_PTR]], [[A_END]]
 ; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END]]
 ; CHECK:       for.end:

 
 
 define void @test4(float* %a, float* readnone %a_end, float** %b.float) {
 ; CHECK-LABEL: @test4(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult float* [[A:%.*]], [[A_END:%.*]]
 ; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
 ; CHECK:       for.body.preheader:
-; CHECK-NEXT:    [[B_F12:%.*]] = load float*, float** [[B_FLOAT:%.*]], align 8
+; CHECK-NEXT:    [[B_F:%.*]] = load float*, float** [[B_FLOAT:%.*]], align 8
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[A_ADDR_03:%.*]] = phi float* [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[A]], [[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT:    [[B_ADDR_02_PTR:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B_F12]], [[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT:    [[TMP1:%.*]] = load float, float* [[B_ADDR_02_PTR]], align 4
+; CHECK-NEXT:    [[B_ADDR_02_IN:%.*]] = phi float* [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[B_F]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = load float, float* [[B_ADDR_02_IN]], align 4
 ; CHECK-NEXT:    [[MUL_I:%.*]] = fmul float [[TMP1]], 4.200000e+01
 ; CHECK-NEXT:    store float [[MUL_I]], float* [[A_ADDR_03]], align 4
-; CHECK-NEXT:    [[ADD]] = getelementptr inbounds float, float* [[B_ADDR_02_PTR]], i64 1
+; CHECK-NEXT:    [[ADD]] = getelementptr inbounds float, float* [[B_ADDR_02_IN]], i64 1
 ; CHECK-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds float, float* [[A_ADDR_03]], i64 1
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ult float* [[INCDEC_PTR]], [[A_END]]
 ; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END]]
 ; CHECK:       for.end:
 ; CHECK-NEXT:    ret void
 ;
 entry:
   %cmp1 = icmp ult float* %a, %a_end

llvm/test/Transforms/InstCombine/load-bitcast32.ll

entry:
   %c = inttoptr i64 %b to i64*
 
   ret i64* %c
 }
 
 define i32* @test2(i8* %x) {
 ; CHECK-LABEL: @test2(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i32**
-; CHECK-NEXT:    [[B1:%.*]] = load i32*, i32** [[TMP0]], align 4
-; CHECK-NEXT:    ret i32* [[B1]]
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i32*
+; CHECK-NEXT:    [[B:%.*]] = load i32, i32* [[A]], align 4
+; CHECK-NEXT:    [[C:%.*]] = inttoptr i32 [[B]] to i32*
+; CHECK-NEXT:    ret i32* [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i32*
   %b = load i32, i32* %a
   %c = inttoptr i32 %b to i32*
 
   ret i32* %c
 }
 
 define i64* @test3(i8* %x) {
 ; CHECK-LABEL: @test3(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i64**
-; CHECK-NEXT:    [[B1:%.*]] = load i64*, i64** [[TMP0]], align 4
-; CHECK-NEXT:    ret i64* [[B1]]
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i32*
+; CHECK-NEXT:    [[B:%.*]] = load i32, i32* [[A]], align 4
+; CHECK-NEXT:    [[C:%.*]] = inttoptr i32 [[B]] to i64*
+; CHECK-NEXT:    ret i64* [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i32*
   %b = load i32, i32* %a
   %c = inttoptr i32 %b to i64*
 
   ret i64* %c
 }
 
 define i64 @test4(i8* %x) {
 ; CHECK-LABEL: @test4(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i32*
-; CHECK-NEXT:    [[B1:%.*]] = load i32, i32* [[TMP0]], align 4
-; CHECK-NEXT:    [[C:%.*]] = zext i32 [[B1]] to i64
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i64**
+; CHECK-NEXT:    [[B:%.*]] = load i64*, i64** [[A]], align 4
+; CHECK-NEXT:    [[TMP0:%.*]] = ptrtoint i64* [[B]] to i32
+; CHECK-NEXT:    [[C:%.*]] = zext i32 [[TMP0]] to i64
 ; CHECK-NEXT:    ret i64 [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i64**
   %b = load i64*, i64** %a
   %c = ptrtoint i64* %b to i64
 
   ret i64 %c
 }
 
 define i32 @test5(i8* %x) {
 ; CHECK-LABEL: @test5(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i32*
-; CHECK-NEXT:    [[B1:%.*]] = load i32, i32* [[TMP0]], align 4
-; CHECK-NEXT:    ret i32 [[B1]]
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i32**
+; CHECK-NEXT:    [[B:%.*]] = load i32*, i32** [[A]], align 4
+; CHECK-NEXT:    [[C:%.*]] = ptrtoint i32* [[B]] to i32
+; CHECK-NEXT:    ret i32 [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i32**
   %b = load i32*, i32** %a
   %c = ptrtoint i32* %b to i32
 
   ret i32 %c
 }
 
 define i64 @test6(i8* %x) {
 ; CHECK-LABEL: @test6(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i32*
-; CHECK-NEXT:    [[B1:%.*]] = load i32, i32* [[TMP0]], align 4
-; CHECK-NEXT:    [[C:%.*]] = zext i32 [[B1]] to i64
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i32**
+; CHECK-NEXT:    [[B:%.*]] = load i32*, i32** [[A]], align 4
+; CHECK-NEXT:    [[TMP0:%.*]] = ptrtoint i32* [[B]] to i32
+; CHECK-NEXT:    [[C:%.*]] = zext i32 [[TMP0]] to i64
 ; CHECK-NEXT:    ret i64 [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i32**
   %b = load i32*, i32** %a
   %c = ptrtoint i32* %b to i64
 
   ret i64 %c
 }
 

llvm/test/Transforms/InstCombine/load-bitcast64.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -instcombine -S < %s | FileCheck %s
 
 target datalayout = "p:64:64:64-i64:32:32"
 
 
 define i64* @test1(i8* %x) {
 ; CHECK-LABEL: @test1(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i64**
-; CHECK-NEXT:    [[B1:%.*]] = load i64*, i64** [[TMP0]], align 4
-; CHECK-NEXT:    ret i64* [[B1]]
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i64*
+; CHECK-NEXT:    [[B:%.*]] = load i64, i64* [[A]], align 4
+; CHECK-NEXT:    [[C:%.*]] = inttoptr i64 [[B]] to i64*
+; CHECK-NEXT:    ret i64* [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i64*
   %b = load i64, i64* %a
   %c = inttoptr i64 %b to i64*
 
   ret i64* %c
 }
entry:
   %c = inttoptr i32 %b to i64*
 
   ret i64* %c
 }
 
 define i64 @test4(i8* %x) {
 ; CHECK-LABEL: @test4(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i64*
-; CHECK-NEXT:    [[B1:%.*]] = load i64, i64* [[TMP0]], align 8
-; CHECK-NEXT:    ret i64 [[B1]]
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i64**
+; CHECK-NEXT:    [[B:%.*]] = load i64*, i64** [[A]], align 8
+; CHECK-NEXT:    [[C:%.*]] = ptrtoint i64* [[B]] to i64
+; CHECK-NEXT:    ret i64 [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i64**
   %b = load i64*, i64** %a
   %c = ptrtoint i64* %b to i64
 
   ret i64 %c
 }
 
 define i32 @test5(i8* %x) {
 ; CHECK-LABEL: @test5(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i64*
-; CHECK-NEXT:    [[B1:%.*]] = load i64, i64* [[TMP0]], align 8
-; CHECK-NEXT:    [[C:%.*]] = trunc i64 [[B1]] to i32
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i32**
+; CHECK-NEXT:    [[B:%.*]] = load i32*, i32** [[A]], align 8
+; CHECK-NEXT:    [[TMP0:%.*]] = ptrtoint i32* [[B]] to i64
+; CHECK-NEXT:    [[C:%.*]] = trunc i64 [[TMP0]] to i32
 ; CHECK-NEXT:    ret i32 [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i32**
   %b = load i32*, i32** %a
   %c = ptrtoint i32* %b to i32
 
   ret i32 %c
 }
 
 define i64 @test6(i8* %x) {
 ; CHECK-LABEL: @test6(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[X:%.*]] to i64*
-; CHECK-NEXT:    [[B1:%.*]] = load i64, i64* [[TMP0]], align 8
-; CHECK-NEXT:    ret i64 [[B1]]
+; CHECK-NEXT:    [[A:%.*]] = bitcast i8* [[X:%.*]] to i32**
+; CHECK-NEXT:    [[B:%.*]] = load i32*, i32** [[A]], align 8
+; CHECK-NEXT:    [[C:%.*]] = ptrtoint i32* [[B]] to i64
+; CHECK-NEXT:    ret i64 [[C]]
 ;
 entry:
   %a = bitcast i8* %x to i32**
   %b = load i32*, i32** %a
   %c = ptrtoint i32* %b to i64
 
   ret i64 %c
 }
 

llvm/test/Transforms/InstCombine/memset_chk-1.ll

 ; CHECK-LABEL: @test_rauw(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CALL49:%.*]] = call i64 @strlen(i8* nonnull dereferenceable(1) [[A:%.*]])
 ; CHECK-NEXT:    [[ADD180:%.*]] = add i64 [[CALL49]], 1
 ; CHECK-NEXT:    [[YO107:%.*]] = call i64 @llvm.objectsize.i64.p0i8(i8* [[B:%.*]], i1 false, i1 false, i1 false)
 ; CHECK-NEXT:    [[CALL50:%.*]] = call i8* @__memmove_chk(i8* [[B]], i8* [[A]], i64 [[ADD180]], i64 [[YO107]])
 ; CHECK-NEXT:    [[STRLEN:%.*]] = call i64 @strlen(i8* nonnull dereferenceable(1) [[B]])
 ; CHECK-NEXT:    [[STRCHR1:%.*]] = getelementptr i8, i8* [[B]], i64 [[STRLEN]]
-; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8** [[C:%.*]] to i64*
-; CHECK-NEXT:    [[D2:%.*]] = load i64, i64* [[TMP0]], align 8
+; CHECK-NEXT:    [[D:%.*]] = load i8*, i8** [[C:%.*]], align 8
+; CHECK-NEXT:    [[SUB182:%.*]] = ptrtoint i8* [[D]] to i64
 ; CHECK-NEXT:    [[SUB183:%.*]] = ptrtoint i8* [[B]] to i64
-; CHECK-NEXT:    [[SUB184:%.*]] = sub i64 [[D2]], [[SUB183]]
+; CHECK-NEXT:    [[SUB184:%.*]] = sub i64 [[SUB182]], [[SUB183]]
 ; CHECK-NEXT:    [[ADD52_I_I:%.*]] = add nsw i64 [[SUB184]], 1
 ; CHECK-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 1 [[STRCHR1]], i8 0, i64 [[ADD52_I_I]], i1 false)
 ; CHECK-NEXT:    ret i32 4
 ;
 entry:
   %call49 = call i64 @strlen(i8* %a)
   %add180 = add i64 %call49, 1
   %yo107 = call i64 @llvm.objectsize.i64.p0i8(i8* %b, i1 false, i1 false, i1 false)