A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 22:35 UTC
[Hardware][CPU] Build fix for ARM without BF16 (#21848)
Signed-off-by: Eric Curtin <ecurtin@redhat.com>
Eric Curtin committed
Jul 30, 2025 at 14:22 UTC
b876860c6214d03279e79e0babb7eb4e3e286cbd
1 file changed
+2
csrc/cpu/quant.cpp
+2
@@ -16,12 +16,14 @@ struct KernelVecType<float> {
16
using cvt_vec_type = vec_op::FP32Vec16;
17
};
18
19
+#if !defined(__aarch64__) || defined(ARM_BF16_SUPPORT)
20
template <>
21
struct KernelVecType<c10::BFloat16> {
22
using load_vec_type = vec_op::BF16Vec16;
23
using azp_adj_load_vec_type = vec_op::INT32Vec16;
24
using cvt_vec_type = vec_op::FP32Vec16;
25
};
26
+#endif
27
28
template <>
29
struct KernelVecType<c10::Half> {