onnx2versal
Loading...
Searching...
No Matches
QgemmChunkNGraph< QGEMM, CONCAT, NCHUNK, TT, TTPARAM, M, K, N > Class Template Reference

Multiinstance graph for MxK times KxN that stores weights and biases Requires KxN_RND weight, NCHUNK%8=0, N%4=0 Chunks KxN weights by N dimension into NCHUNK chunks. Each instance has max size = 16384 and 4096 bytes respectively. Places maximum of 3x3 tiles, 8 conv tiles surrounding concat tile (max AIE DMA input=8)

#include <graph_qgemm.h>

Inherits adf::graph.

Public Member Functions

 QgemmChunkNGraph (std::vector< TTPARAM > weights, std::vector< int32_t > bias, float x_scale, float w_scale, float y_scale, TT x_zero, TTPARAM w_zero, TT y_zero)
 

Public Attributes

adf::kernel k [CHUNK_COUNT]
 
ConcatStreamGraph< CONCAT, TT, CHUNK_COUNT, M, NCHUNK, N > concat_g
 
adf::port< input > pin [1]
 
adf::port< output > pout [1]
 

Static Public Attributes

static const int CHUNK_COUNT = (N + NCHUNK - 1) / NCHUNK
 

The documentation for this class was generated from the following file: