Make size_of, align_of, and element_offset functions return u64 instead of uint in trans::machine #11759
Conversation
| @@ -341,7 +341,7 @@ pub fn trans_native_call<'a>( | |||
| let llalign = cmp::min(llforeign_align, llrust_align); | |||
| debug!("llrust_size={:?}", llrust_size); | |||
| base::call_memcpy(bcx, llretptr_i8, llscratch_i8, | |||
| C_uint(ccx, llrust_size), llalign as u32); | |||
| C_u64(llrust_size), llalign as u32); | |||
There was a problem hiding this comment.
This looks a little suspicious, I don't think that we want to change a target's call to memcpy, we only want to change the compiler's representation of offsets and sizes to u64 to accommodate when a 32-bit platform cross-compiles to a 64-bit platform.
|
cc @jld, I have some comments, but I'm unfamiliar with this code enough to know whether this is all of the places which need to get changed. |
|
Ah so C_* represents values in the code the compiler produces (target)? That makes sense. Though, how could I safely create a C_uint for a 64-bit platform when compiled on a 32-bit platform? Wouldn't C_uint(cx, someU64 as uint) downcast to u32 if cross-compiled on a 32-bit platform, for a 64-bit platform? |
|
A 32-bit platform cannot create any structs or have any offsets larger than 32-bits, so downcasting should always be fine. |
|
OK I've updated it so that there shouldn't be anymore target code using C_u64's. |
|
This looks good to me! Could you rebase the commits into one commit? |
…ad of uint in trans::machine (per rust-lang#5172)
|
Yup should be good now. |
For #5172