On Mon, 7 Dec 2020, Liu Hao wrote:
在 2020/12/7 上午5:30, Martin Storsjö 写道:
My thought here was that previously by including malloc.h, you got a working
_mm_malloc and _mm_free
(defined by pointing at other functions), so there could be code relying on
getting these functions
visible by just relying on malloc.h. (The fact that MSVC doesn't have it
shouldn't really matter
afaik, as our headers don't support being built with MSVC in general afaik?)
I have an impression that some part of our CRT could be built with MSVC
because there are some `#if` checks in CRT code. Probably they are just
legacies.
Yeah, there's some stray ifdef in some places, but overall I believe that
use case isn't maintained or really working (except for maybe very small
subsets/headers/source files).
The Intel Intrinsics Guide [1] says those functions are of SSE
technology and
suggests `#include <xmmintrin.h>` instead.
I guess that'd work, but that ends up declaring a whole lot more than what the
header did earlier.
// Martin
As long as both GCC and Clang have <mm_malloc.h>, these patches look good.
Thanks, will push them soon, then!
// Martin
_______________________________________________
Mingw-w64-public mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
