|
FZGPUModules 2.0
GPU-accelerated modular compression pipelines
|
Everything a fused runner needs to compress one group in place. More...
#include <fusion_registry.h>
Public Attributes | |
| const std::vector< Stage * > * | stages |
| group stages, producer→consumer | |
| const void * | d_input |
| the group's input buffer | |
| size_t | input_bytes |
| bytes of d_input | |
| void * | d_output |
| the group's MAIN output buffer (tail port 0, >= worst case) | |
| MemoryPool * | pool |
| pool for the runner's scratch | |
| fz::stream_t | stream |
| std::vector< FusedSideOutput > * | side_outputs = nullptr |
Everything a fused runner needs to compress one group in place.
| fz::stream_t fz::FusedRunContext::stream |
stream (runner may synchronise)
| std::vector<FusedSideOutput>* fz::FusedRunContext::side_outputs = nullptr |
Escaping side outputs of group members (e.g. outlier lists), in member/port order. nullptr or empty for single-output pipelines. The runner writes each d_ptr and fills each size; a runner that produces none leaves them alone.