Nvfp4 A16 Blackwell Gemv Jit Runner

class Nvfp4A16BlackwellGemvJitRunner

Public Functions

void load(Nvfp4A16BlackwellGemvJitKernel const &kernel)

void launch(

void const *activation,

uint8_t const *qweights,

uint8_t const *blockScales,

float const *globalScale,

void *output,

void *workspace,

size_t workspaceSize,

int32_t m,

int32_t splitK,

cudaStream_t stream

) const

inline bool isLoaded() const noexcept

inline Nvfp4A16BlackwellGemvJitKey const &getKey() const noexcept

bool trt_edgellm::isNvfp4A16BlackwellGemvJitSupported(

Nvfp4A16BlackwellGemvJitKey const &key,

int32_t const m,

int32_t const splitK

) noexcept

size_t trt_edgellm::getNvfp4A16BlackwellGemvJitWorkspaceSize(

Nvfp4A16BlackwellGemvJitKey const &key,

int32_t const m,

int32_t const splitK

) noexcept