John Platts
John Platts
The issue is that the `__x86_64__` and `__aarch64__` macros are both defined, and only one of the two macros should be defined. The Highway C++ files have to be compiled...
@7sharp9 I did successfully build Google Highway for macOS for both x86_64 and ARM64 using a GitHub Workflow, and the results of that workflow run can be found at https://github.com/johnplatts/jep_google_highway/actions/runs/13345004943/job/37274387258....
Another use case for the ShiftRightAndDemoteTo, ShiftRightAndReorderDemote2To, and ShiftRightAndOrderedDemote2To ops are to further optimize F32 to BF16 DemoteTo/ReorderDemote2To/OrderedDemote2To on NEON/LSX/LASX.
Here is how GF2P8MulInverse can be implemented on targets other than HWY_SCALAR: ``` template HWY_API V GF2P8MulInverse(V v) { using D = DFromV; const D d; #if HWY_TARGET == HWY_RVV...
Here is how GaloisAffine can be implemented: ``` template HWY_API V GaloisAffine(V v) { constexpr uint8_t kMask0 = static_cast(((kMatrix >> 56) & 0x01) | ((kMatrix >> 47) & 0x02) |...
@jan-wassenberg and @anauta00 I am able to get Google Highway to compile successfully with Visual C++ 2022 (with both cl 19.43.34810 and clang-cl 19.1.1) with the fix in pull request...
One of the issues is that the Hypot implementation in math-inl.h requires that `MaxExponentField() / 4 + MantissaBits() > MaxExponentField() / 2` which is true for F32 and F64 but...
> Thanks for reporting! I've reviewed the EMU128 implementation of this test and it looks like a compiler bug. Can you help report it on https://gcc.gnu.org/bugzilla/ so they can fix...
@malaterre I fixed these compiler errors in pull request #2812, and there were some changes to Z14/Z15 compiler intrinsics in GCC 15 and Clang 20.
I reported this GCC 15 compiler bug at https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122692 with a reduced test case.