[CI/Build] CPU release supports both of AVX2 and AVX512 (#35466)

Signed-off-by: jiang1.li <jiang1.li@intel.com>
Co-authored-by: jiang1.li <jiang1.li@intel.com>
This commit is contained in:
Ma Jian
2026-02-28 04:35:21 +00:00
committed by GitHub
co-authored by jiang1.li
parent 2562e0271e
commit 90805ff464
6 changed files with 160 additions and 146 deletions
+1 -3
View File
@@ -178,9 +178,7 @@ def mla_decode_kvcache_cpu(
block_tables: torch.Tensor,
seq_lens: torch.Tensor,
) -> None:
torch.ops._C_cpu.mla_decode_kvcache(
out, query, kv_cache, scale, block_tables, seq_lens
)
torch.ops._C.mla_decode_kvcache(out, query, kv_cache, scale, block_tables, seq_lens)
# merge attn states ops
+24
View File
@@ -483,3 +483,27 @@ class CpuPlatform(Platform):
@classmethod
def support_hybrid_kv_cache(cls) -> bool:
return True
@classmethod
def import_kernels(cls) -> None:
if Platform.get_cpu_architecture() in (CpuArchEnum.X86,):
if torch._C._cpu._is_avx512_supported():
try:
import vllm._C # noqa: F401
except ImportError as e:
logger.warning("Failed to import from vllm._C: %r", e)
else:
# Note: The lib name is _C_AVX2, but the module name is _C.
# This will cause a exception "dynamic module does define
# module export function". But the library is imported
# successfully. So ignore the exception for now, until we find
# a solution.
try:
import vllm._C_AVX2 # noqa: F401
except ImportError as e:
logger.warning("Failed to import from vllm._C_AVX2: %r", e)
else:
try:
import vllm._C # noqa: F401
except ImportError as e:
logger.warning("Failed to import from vllm._C: %r", e)
+1 -1
View File
@@ -85,7 +85,7 @@ class CPUWorker(Worker):
self.local_omp_cpuid = omp_cpuids_list[self.rank]
if self.local_omp_cpuid != "nobind":
ret = torch.ops._C_utils.init_cpu_threads_env(self.local_omp_cpuid)
ret = torch.ops._C.init_cpu_threads_env(self.local_omp_cpuid)
if ret:
logger.info(ret)