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

CUDA: fixed row rounding for 0 tensor splits (#4594)

Browse source
This commit is contained in:
Johannes Gäßler 2023-12-23 09:16:33 +01:00 committed by GitHub
parent 7082d24cec
commit e0a4002273
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
ggml-cuda.cu
View file

@ -7937,15 +7937,19 @@ static void ggml_cuda_op_mul_mat(
if (id != 0) {
row_low[id] = ne01*g_tensor_split[id];
if (row_low[id] < ne01) {
row_low[id] -= row_low[id] % rounding;
}
}
if (id != g_device_count - 1) {
row_high[id] = ne01*g_tensor_split[id + 1];
if (row_high[id] < ne01) {
row_high[id] -= row_high[id] % rounding;
}
}
}
}
for (int64_t id = 0; id < g_device_count; ++id) {
if ((!split && id != g_main_device) || row_low[id] == row_high[id]) {