vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat: cuda implementation for ggml_conv_transpose_1d (ggml/854)

Browse source 
* conv transpose 1d passing test for 1d input and kernel

* working for different input and output channel counts, added test for variable stride

* initial draft appears to work with stride other than 1

* working with all old and new conv1d  tests

* added a test for large tensors

* removed use cuda hardcoding

* restored test-conv-transpose.c

* removed unused arugments, and fixed bug where test failure would cause subsequent tests to fail

* fixed accumulator bug

* added test to test-backend-ops

* fixed mistake

* addressed review

* fixed includes

* removed blank lines

* style and warning fixes

* return failure when test fails

* fix supports_op

---------

Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
John Balis 2024-07-02 11:09:52 -05:00 • committed by Georgi Gerganov
parent 470939d483
commit fde13b3bb9
4 changed files with 146 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml/src/ggml-cuda/conv-transpose-1d.cuh Normal file
View file

@ -0,0 +1,5 @@
#include "common.cuh"
#define CUDA_CONV_TRANPOSE_1D_BLOCK_SIZE 256
void ggml_cuda_op_conv_transpose_1d(ggml_backend_cuda_context & ctx, ggml_tensor * dst);