Ple Scatter

cudaError_t trt_edgellm::kernel::launchPleScatter(

void const *rows,

int32_t const *slots,

void *pool,

int64_t rowCount,

int64_t rowBytes,

cudaStream_t stream

) noexcept

Scatter packed rows into reserved pool slots. Caller guarantees distinct, valid destination slots.

cudaError_t trt_edgellm::kernel::launchPleScatterConvert(

void const *rows,

int32_t const *slots,

void *pool,

int64_t rowCount,

int64_t rowWidth,

trt_edgellm::DataType sourceType,

float scale,

cudaStream_t stream

) noexcept

Convert selected source rows to FP16, then scale and scatter into distinct reserved slots.

cudaError_t trt_edgellm::kernel::launchPleConvert(

void const *rows,

void *output,

int64_t rowCount,

int64_t rowWidth,

trt_edgellm::DataType sourceType,

float scale,

cudaStream_t stream

) noexcept

Convert contiguous source rows to FP16, then scale into a contiguous destination.