|
FZGPUModules 2.0
GPU-accelerated modular compression pipelines
|
Minimal context a fused runner hands a stage so it can establish the forward-computed state its OWN inverse will later read. More...
#include <fusion.h>
Public Attributes | |
| const void * | d_input = nullptr |
| device input buffer | |
| size_t | input_bytes = 0 |
| its size in bytes | |
| MemoryPool * | pool = nullptr |
| scratch pool | |
| fz::stream_t | stream = nullptr |
| stream to prime on | |
Minimal context a fused runner hands a stage so it can establish the forward-computed state its OWN inverse will later read.
A fused pipeline reuses the same stage objects for decompress but bypasses their forward execute(), so any state normally computed there (e.g. a quantizer's resolved error bound from a value-range scan) must be primed explicitly, or the inverse reconstructs with defaults. See Stage::primeFusedForwardState.