dreamcmi/ruapu
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Detect x86 AMX(AMX-FP16 AMX-BF16 AMX-INT8 AMX-TILE) (#79)

Browse Source
This commit is contained in:
張小凡 2024-03-06 20:14:44 +08:00 committed by GitHub
parent b113072379
commit 35b28cd404
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 31 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
README.md
View File

@ -456,6 +456,10 @@ avx5124vnniw = 0
avxvnni = 0 avxvnni = 0
avxvnniint8 = 0 avxvnniint8 = 0
avxifma = 0 avxifma = 0
amxfp16 = 0
amxbf16 = 0
amxint8 = 0
amxtile = 0
``` ```
</details> </details>
@ -494,6 +498,10 @@ avx5124vnniw = 0
avxvnni = 0 avxvnni = 0
avxvnniint8 = 0 avxvnniint8 = 0
avxifma = 0 avxifma = 0
amxfp16 = 0
amxbf16 = 0
amxint8 = 0
amxtile = 0
``` ```
</details> </details>
@ -553,6 +561,10 @@ avx5124vnniw = 0
avxvnni = 0 avxvnni = 0
avxvnniint8 = 0 avxvnniint8 = 0
avxifma = 0 avxifma = 0
amxfp16 = 0
amxbf16 = 0
amxint8 = 0
amxtile = 0
``` ```
</details> </details>
@ -592,6 +604,10 @@ avx5124vnniw = 0
avxvnni = 0 avxvnni = 0
avxvnniint8 = 0 avxvnniint8 = 0
avxifma = 0 avxifma = 0
amxfp16 = 0
amxbf16 = 0
amxint8 = 0
amxtile = 0
``` ```
</details> </details>
@ -616,7 +632,7 @@ _`fma4` on zen1, ISA in hypervisor, etc._
|CPU|ISA| |CPU|ISA|
|:---:|---| |:---:|---|
|x86|`mmx` `sse` `sse2` `sse3` `ssse3` `sse41` `sse42` `sse4a` `xop` `avx` `f16c` `fma` `fma4` `avx2` `avx512f` `avx512bw` `avx512cd` `avx512dq` `avx512vl` `avx512vnni` `avx512bf16` `avx512ifma` `avx512vbmi` `avx512vbmi2` `avx512fp16` `avx512er` `avx5124fmaps` `avx5124vnniw` `avxvnni` `avxvnniint8` `avxifma`| |x86|`mmx` `sse` `sse2` `sse3` `ssse3` `sse41` `sse42` `sse4a` `xop` `avx` `f16c` `fma` `fma4` `avx2` `avx512f` `avx512bw` `avx512cd` `avx512dq` `avx512vl` `avx512vnni` `avx512bf16` `avx512ifma` `avx512vbmi` `avx512vbmi2` `avx512fp16` `avx512er` `avx5124fmaps` `avx5124vnniw` `avxvnni` `avxvnniint8` `avxifma` `amxfp16` `amxbf16` `amxint8` `amxtile`|
|arm|`half` `edsp` `neon` `vfpv4` `idiv`| |arm|`half` `edsp` `neon` `vfpv4` `idiv`|
|aarch64|`neon` `vfpv4` `cpuid` `asimdrdm` `asimdhp` `asimddp` `asimdfhm` `bf16` `i8mm` `mte` `sve` `sve2` `svebf16` `svei8mm` `svef32mm` `pmull` `crc32` `aes` `sha1` `sha2` `sha3` `sha512` `sm3` `sm4` `amx`| |aarch64|`neon` `vfpv4` `cpuid` `asimdrdm` `asimdhp` `asimddp` `asimdfhm` `bf16` `i8mm` `mte` `sve` `sve2` `svebf16` `svei8mm` `svef32mm` `pmull` `crc32` `aes` `sha1` `sha2` `sha3` `sha512` `sm3` `sm4` `amx`|
|mips|`msa`| |mips|`msa`|
@ -671,6 +687,7 @@ ruapu determines whether the CPU supports certain instruction sets by trying to
* [@dreamcmi](https://github.com/dreamcmi) &emsp;_Detect more risc-v ISA_ * [@dreamcmi](https://github.com/dreamcmi) &emsp;_Detect more risc-v ISA_
* [@cocoa-xu](https://github.com/cocoa-xu) &emsp;_Add FreeBSD support, python support_ * [@cocoa-xu](https://github.com/cocoa-xu) &emsp;_Add FreeBSD support, python support_
* [@YuzukiTsuru](https://github.com/YuzukiTsuru) &emsp;_Add OpenRISC support_ * [@YuzukiTsuru](https://github.com/YuzukiTsuru) &emsp;_Add OpenRISC support_
* [@whyb](https://github.com/whyb) &emsp;_Detect x86 AMX_
## License ## License
MIT License MIT License

5
main.c
View File

@ -47,6 +47,11 @@ int main()
PRINT_ISA_SUPPORT(avxvnni) PRINT_ISA_SUPPORT(avxvnni)
PRINT_ISA_SUPPORT(avxvnniint8) PRINT_ISA_SUPPORT(avxvnniint8)
PRINT_ISA_SUPPORT(avxifma) PRINT_ISA_SUPPORT(avxifma)
PRINT_ISA_SUPPORT(amxfp16)
PRINT_ISA_SUPPORT(amxbf16)
PRINT_ISA_SUPPORT(amxint8)
PRINT_ISA_SUPPORT(amxtile)
#elif __aarch64__ || defined(_M_ARM64) #elif __aarch64__ || defined(_M_ARM64)
PRINT_ISA_SUPPORT(neon) PRINT_ISA_SUPPORT(neon)

8
ruapu.h
View File

@ -216,6 +216,10 @@ RUAPU_INSTCODE(avxvnni, 0xc4, 0xe2, 0x7d, 0x52, 0xc0) // vpdpwssd ymm0,ymm0,ymm0
RUAPU_INSTCODE(avxvnniint8, 0xc4, 0xe2, 0x7f, 0x50, 0xc0) // vpdpbssd ymm0,ymm0,ymm0 RUAPU_INSTCODE(avxvnniint8, 0xc4, 0xe2, 0x7f, 0x50, 0xc0) // vpdpbssd ymm0,ymm0,ymm0
// TODO:avxvnniint16, vpdpwusd xmm,xmm,xmm // TODO:avxvnniint16, vpdpwusd xmm,xmm,xmm
RUAPU_INSTCODE(avxifma, 0xc4, 0xe2, 0xfd, 0xb4, 0xc0) // vpmadd52luq ymm0,ymm0,ymm0 RUAPU_INSTCODE(avxifma, 0xc4, 0xe2, 0xfd, 0xb4, 0xc0) // vpmadd52luq ymm0,ymm0,ymm0
RUAPU_INSTCODE(amxfp16, 0xc4, 0xe2, 0x7b, 0x5c, 0xd1) // tdpfp16ps %tmm0, %tmm1, %tmm2
RUAPU_INSTCODE(amxbf16, 0xc4, 0xe2, 0x7a, 0x5c, 0xd1) // tdpbf16ps %tmm0, %tmm1, %tmm2
RUAPU_INSTCODE(amxint8, 0xc4, 0xe2, 0x7b, 0x5e, 0xd1) // tdpbssd %tmm0, %tmm1, %tmm2
RUAPU_INSTCODE(amxtile, 0xc4, 0xe2, 0x7a, 0x49, 0xc0) // tilezero %tmm0
#elif __aarch64__ || defined(_M_ARM64) #elif __aarch64__ || defined(_M_ARM64)
RUAPU_INSTCODE(neon, 0x4e20d400) // fadd v0.4s,v0.4s,v0.4s RUAPU_INSTCODE(neon, 0x4e20d400) // fadd v0.4s,v0.4s,v0.4s
@ -335,6 +339,10 @@ RUAPU_ISAENTRY(avxvnni)
RUAPU_ISAENTRY(avxvnniint8) RUAPU_ISAENTRY(avxvnniint8)
// TODO:avxvnniint16 // TODO:avxvnniint16
RUAPU_ISAENTRY(avxifma) RUAPU_ISAENTRY(avxifma)
RUAPU_ISAENTRY(amxfp16)
RUAPU_ISAENTRY(amxbf16)
RUAPU_ISAENTRY(amxint8)
RUAPU_ISAENTRY(amxtile)
#elif __aarch64__ || defined(_M_ARM64) #elif __aarch64__ || defined(_M_ARM64)
RUAPU_ISAENTRY(neon) RUAPU_ISAENTRY(neon)