c68ddc11ce
### What this PR does / why we need it? Add a new fusion ops to custom_op, which can cobime the torch.bmm() and transpsose to achieve better peformance. This ops is used in mla_v1 to replace the bmm and transpose ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? - vLLM version: v0.11.2 --------- Signed-off-by: hust17yixuan <303660421@qq.com>
16 lines
301 B
C++
16 lines
301 B
C++
#ifndef KERNEL_MATH_UTILS_H
|
|
#define KERNEL_MATH_UTILS_H
|
|
#include <cstdint>
|
|
|
|
namespace device_utils {
|
|
|
|
template <typename T, T roundVal>
|
|
__aicore__ __force_inline__ T RoundUp(const T &val)
|
|
{
|
|
return (val + roundVal - 1) / roundVal * roundVal;
|
|
}
|
|
|
|
}; // namespace device_utils
|
|
|
|
#endif
|