33 std::string predictor =
"Lorenzo1DPredictor";
34 std::vector<std::string> transforms;
35 std::string coder =
"AdaptiveBitpackCoder";
36 int elems_per_lane = 1;
62 const uint8_t* pred_params,
size_t params_bytes,
63 uint8_t* d_out,
MemoryPool* pool, fz::stream_t stream,
64 std::string* execution_path =
nullptr);
81 const WarpFusionSpec& spec,
const uint8_t* d_archive,
size_t archive_bytes,
82 size_t n_elems,
size_t n_out,
float ebx2,
83 const uint8_t* pred_params,
size_t params_bytes,
84 float* d_out,
MemoryPool* pool, fz::stream_t stream);
size_t launchNvrtcWarpInverseFused(const WarpFusionSpec &spec, const uint8_t *d_archive, size_t archive_bytes, size_t n_elems, size_t n_out, float ebx2, const uint8_t *pred_params, size_t params_bytes, float *d_out, MemoryPool *pool, fz::stream_t stream)
size_t launchNvrtcWarpFused(const WarpFusionSpec &spec, const float *d_in, size_t n_ab, const uint8_t *pred_params, size_t params_bytes, uint8_t *d_out, MemoryPool *pool, fz::stream_t stream, std::string *execution_path=nullptr)
std::string generateWarpFusionSource(const WarpFusionSpec &spec)
The CUDA source the codegen emits for spec (exposed for tests/inspection).
Definition nvrtc_warp_fusion.h:32
std::string predictor_ti
Definition nvrtc_warp_fusion.h:43
Backend-neutral GPU type aliases.