Skip to content

Don't create invalid bitcasts. #38424

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 18, 2020
Merged

Don't create invalid bitcasts. #38424

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 7 additions & 2 deletions src/codegen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1214,8 +1214,13 @@ static Value *emit_inttoptr(jl_codectx_t &ctx, Value *v, Type *ty)
{
// Almost all of our inttoptr are generated due to representing `Ptr` with `T_size`
// in LLVM and most of these integers are generated from `ptrtoint` in the first place.
if (auto I = dyn_cast<PtrToIntInst>(v))
return ctx.builder.CreateBitCast(I->getOperand(0), ty);
if (auto I = dyn_cast<PtrToIntInst>(v)) {
auto ptr = I->getOperand(0);
if (ty->getPointerAddressSpace() == ptr->getType()->getPointerAddressSpace())
return ctx.builder.CreateBitCast(ptr, ty);
else if (ty->getPointerElementType() == ptr->getType()->getPointerElementType())
return ctx.builder.CreateAddrSpaceCast(ptr, ty);
Copy link
Member

@vchuravy vchuravy Nov 13, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add a lit test?

Also what happens when As != As && eType !== eType?

Copy link
Contributor

@yuyichao yuyichao Nov 13, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is just an optimization so falling through is fine.
I assume this is for non-default codegen with normal non default address space?

Copy link
Member Author

@maleadt maleadt Nov 16, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Default codegen, but with LLVMPtr:

(gdb) call jl_(lam)
map(typeof(CUDA.cudaconvert), Tuple{CUDA.CuArray{Float32, 2}, CUDA.CuArray{Float32, 2}, CUDA.CuArray{Float32, 2}}) from map(Any, Tuple{Any, Any, Any})

(gdb) call jl_(src)
Core.CodeInfo(code=Array{Any, (44,)}[
  Expr(:call, Base.getfield, Core.Argument(n=3), 1, true),
  Expr(:call, Base.getfield, SSAValue(1), :(:dims)),
  Expr(:call, Base.getfield, SSAValue(1), :(:baseptr)),
  Base.bitcast,
  Expr(:call, SSAValue(4), CUDA.CuPtr{Float32}, SSAValue(3)),
  Expr(:call, Base.getfield, SSAValue(1), :(:offset)),
  Base.add_ptr,
  Base.bitcast,
  Expr(:call, SSAValue(8), CUDA.UInt, SSAValue(5)),
  Expr(:call, Base.bitcast, UInt64, SSAValue(6)),
  Expr(:call, SSAValue(7), SSAValue(9), SSAValue(10)),
  Expr(:call, #<intrinsic #0 bitcast>, CUDA.CuPtr{Float32}, SSAValue(11)),
  Expr(:call, Base.bitcast, Core.LLVMPtr{Float32, 1}, SSAValue(12)),
  ...

I couldn't find an easy reproducer though.

}
return ctx.builder.CreateIntToPtr(v, ty);
}

Expand Down