[llvm-commits] [llvm] r171866 - in /llvm/trunk: lib/CodeGen/AsmPrinter/AsmPrinter.cpp test/CodeGen/ARM/fp128.ll test/CodeGen/PowerPC/fp128.ll

[Duncan Sands]

Hi Tim,

On 09/01/13 14:00, Tim Northover wrote:
>> while this seems correct and the approach is in keeping with the rest of the
>> function, it would be nicer to get rid of all the cases.
>
> I've attached a patch which should implement this. It definitely seems
> like an improvement to me. I'm including the final function inline as
> well since it's one of those diffs that's very difficult to read. Ok
> to commit?

thanks for doing this, it looks great to me.  Do the existing testcases give
complete code coverage (i.e. test the output for every floating point type
on both big and little endian platforms)?  Some small nits:

>    // Now iterate through the APInt chunks, emitting them in endian-correct
>    // order, possibly with a smaller chunk at beginning/end (e.g. for x87 80-bit
>    // floats)

Missing full stop at end of comment.

>    } else {
>      unsigned Chunk;
>      for (Chunk = 0; Chunk < NumBytes / sizeof(uint64_t); ++Chunk)

Can you get NumBytes / sizeof(uint64_t) from API.getNumWords() instead?

Ciao, Duncan.