33 std::string predictor =
"Lorenzo1DPredictor";
34 std::vector<std::string> transforms;
35 std::string coder =
"AdaptiveBitpackCoder";
36 int elems_per_lane = 1;
54 const uint8_t* pred_params,
size_t params_bytes,
55 uint8_t* 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 generateWarpFusionSource(const WarpFusionSpec &spec)
The CUDA source the codegen emits for spec (exposed for tests/inspection).
Definition nvrtc_warp_fusion.h:32
Backend-neutral GPU type aliases.