Vladislav Shchapov
Vladislav Shchapov
zlib source code from macOS 26.0: https://github.com/apple-oss-distributions/zlib/tree/zlib-100
> There is no performance impact as indirect calls are infrequent. zlib-ng uses indirection calls (function pointers) to do optimized functions. I think performance testing is needed.
@PERFACCT-JS Could you provide the full build log? Or compiler flags for `adler32_avx2.c`
@PERFACCT-JS Try replacing (on current zlib-ng develop branch): ``` CMAKE_C_COMPILER_ID MATCHES "GNU" OR CMAKE_C_COMPILER_ID MATCHES "Clang" ``` with: ``` CMAKE_C_COMPILER_ID MATCHES "GNU" OR CMAKE_C_COMPILER_ID MATCHES "Clang" OR CMAKE_C_COMPILER_ID MATCHES "NVHPC"...
@Dead2 You're right. I remembered that nvc was option-compatible with gcc, but I was mistaken :(( NVC requires new logic for checking compiler support for instruction sets based on the...
NVC supports the parameter: ``` -mcpu=[] ``` https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-ref-guide/index.html#mcpu Apparently, it can be used to enable/disable additional instruction sets in addition to the base one.
Are there any plans to use `functable`?
`loongarch64` supports `BITREV` instruction. After merging this PR, I can add usage of `BITREV` instruction.