Nick Terrell

> GAS also supports Intel syntax, maybe it's better to switch to Intel syntax in the long run. In the medium to long term we will look into switching to...

@gvanem `HUF_ENABLE_ASM_X86_64_BMI2` definitely isn't enabled on x86 32-bit builds. I've added tests for Windows + CMake builds to our CI, and don't see any failures. I've further restricted the assembly...

@gvanem I mean regarding this failure: ``` LINK : fatal error LNK1104: cannot open file 'CMakeFiles\libzstd_shared.dir\F_\MingW32\src\Compression\zstd\lib\decompress\huf_decompress_amd64.S.obj' ``` I haven't been able to reproduce the failure, but I've done more to...

> I strongly think that the assembly file should be moved to intrinsics. What do you mean by intrinsics? Like compiler builtins e.g. bextr? > You get real cross-platform support...

`--rsyncable` is hardwired to a 32-byte hash, which could likely be increased to a 64-byte hash without any issues. I quickly tried to change it out, and saw a 2x...

We can't do that because `ZDICT_trainFromBuffer` is part of our ABI stable interface. I'd recommend trying the `ZDICT_optimizeTrainFromBuffer_cover` or the `ZDICT_optimizeTrainFromBuffer_fastCover` function, depending on if you care about speed of...

Is this PR ready for review now?

> I suspect it's another case of msan test, but one where zlibwrapper is compiled with msan, but the linked libzstd is not, thus failing the msan test. That is...

Oh, thats a valgrind check, not MSAN.

I think a sane approach would be to change the order that the cwksp stores elements. Instead of: ``` [ ... workspace ... ] [objects][tables ... ->] free space [