Nvfp4 A16 Blackwell Gemv Jit Compiler

struct Nvfp4A16BlackwellGemvJitKey

Public Functions

inline auto asTuple() const noexcept

inline bool operator==(

Nvfp4A16BlackwellGemvJitKey const &other

) const noexcept

Public Members

int32_t sm = {nvfp4_a16_blackwell::kTargetSm}

uint32_t layout = {kNVFP4_A16_BLACKWELL_GEMV_LAYOUT_ABI}

int32_t n = {}

int32_t k = {}

Nvfp4A16BlackwellGemvDataType dataType = {Nvfp4A16BlackwellGemvDataType::kHALF}

uint32_t sourceAbi = {kNVFP4_A16_BLACKWELL_GEMV_SOURCE_ABI}

struct Nvfp4A16BlackwellGemvJitDigest

Public Functions

inline bool operator==(

Nvfp4A16BlackwellGemvJitDigest const &other

) const noexcept

Public Members

uint64_t lo = {}

uint64_t hi = {}

struct Nvfp4A16BlackwellGemvJitKernel

Public Members

Nvfp4A16BlackwellGemvJitKey key

Nvfp4A16BlackwellGemvJitDigest digest

std::vector<uint8_t> cubin

bool trt_edgellm::canCompileNvfp4A16BlackwellGemvJitKernel(

Nvfp4A16BlackwellGemvJitKey const &key

) noexcept

Nvfp4A16BlackwellGemvJitDigest trt_edgellm::computeNvfp4A16BlackwellGemvJitDigest(

Nvfp4A16BlackwellGemvJitKey const &key,

void const *const cubinData,

size_t const cubinSize

)

Nvfp4A16BlackwellGemvJitKernel trt_edgellm::compileNvfp4A16BlackwellGemvJitKernel(

Nvfp4A16BlackwellGemvJitKey const &key

)

std::vector<uint8_t> trt_edgellm::serializeNvfp4A16BlackwellGemvJitKernel(

Nvfp4A16BlackwellGemvJitKernel const &kernel

)

Nvfp4A16BlackwellGemvJitKernel trt_edgellm::deserializeNvfp4A16BlackwellGemvJitKernel(

void const *const data,

size_t const size

)