diff options
| -rw-r--r-- | src/cmake/cpu.cmake | 10 | ||||
| -rw-r--r-- | src/vppinfra/cpu.h | 23 |
2 files changed, 31 insertions, 2 deletions
diff --git a/src/cmake/cpu.cmake b/src/cmake/cpu.cmake index 031a9bcfe93..b54155d4bb3 100644 --- a/src/cmake/cpu.cmake +++ b/src/cmake/cpu.cmake @@ -136,6 +136,11 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*") OFF ) + add_vpp_march_variant(znver3 + FLAGS -march=znver3 -mtune=znver3 -mprefer-vector-width=256 + OFF + ) + if (GNU_ASSEMBLER_AVX512_BUG) message(WARNING "AVX-512 multiarch variant(s) disabled due to GNU Assembler bug") else() @@ -151,6 +156,11 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "amd64.*|x86_64.*|AMD64.*") FLAGS -march=sapphirerapids -mtune=sapphirerapids -mprefer-vector-width=512 OFF ) + + add_vpp_march_variant(znver4 + FLAGS -march=znver4 -mtune=znver4 -mprefer-vector-width=512 + OFF + ) endif() elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64.*|AARCH64.*)") set(VPP_DEFAULT_MARCH_FLAGS -march=armv8-a+crc) diff --git a/src/vppinfra/cpu.h b/src/vppinfra/cpu.h index 29992bb11eb..017ecb1ca0e 100644 --- a/src/vppinfra/cpu.h +++ b/src/vppinfra/cpu.h @@ -27,7 +27,9 @@ _ (skx, "Intel Skylake (server) / Cascade Lake") \ _ (icl, "Intel Ice Lake") \ _ (adl, "Intel Alder Lake") \ - _ (spr, "Intel Sapphire Rapids") + _ (spr, "Intel Sapphire Rapids") \ + _ (znver3, "AMD Milan") \ + _ (znver4, "AMD Genoa") #elif defined(__aarch64__) #define foreach_march_variant \ _ (octeontx2, "Marvell Octeon TX2") \ @@ -147,7 +149,8 @@ _CLIB_MARCH_FN_REGISTRATION(fn) _ (movdir64b, 7, ecx, 28) \ _ (enqcmd, 7, ecx, 29) \ _ (avx512_fp16, 7, edx, 23) \ - _ (invariant_tsc, 0x80000007, edx, 8) + _ (invariant_tsc, 0x80000007, edx, 8) \ + _ (monitorx, 0x80000001, ecx, 29) #define foreach_aarch64_flags \ _ (fp, 0) \ @@ -300,6 +303,22 @@ clib_cpu_march_priority_hsw () return -1; } +static inline int +clib_cpu_march_priority_znver4 () +{ + if (clib_cpu_supports_avx512_bitalg () && clib_cpu_supports_monitorx ()) + return 250; + return -1; +} + +static inline int +clib_cpu_march_priority_znver3 () +{ + if (clib_cpu_supports_avx2 () && clib_cpu_supports_monitorx ()) + return 70; + return -1; +} + #define X86_CPU_ARCH_PERF_FUNC 0xA static inline int |