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

Add further missing barrier

Browse source
This commit is contained in:
0cc4m 2023-08-06 05:25:33 +02:00
parent 44bbc85aaf
commit ccd2592782
1 changed files with 1 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

1
ggml-vulkan.cpp
View file

@ -2346,6 +2346,7 @@ static void ggml_vk_mul_f32(const ggml_tensor * src0, const ggml_tensor * src1,
ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_X), ggml_vk_subbuffer(d_Y) }, vk_device.compute_queue, vk::AccessFlagBits::eTransferWrite, vk::AccessFlagBits::eShaderRead, true); ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_X), ggml_vk_subbuffer(d_Y) }, vk_device.compute_queue, vk::AccessFlagBits::eTransferWrite, vk::AccessFlagBits::eShaderRead, true);
ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_D) }, vk_device.compute_queue, vk::AccessFlagBits::eTransferRead, vk::AccessFlagBits::eShaderWrite, false); ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_D) }, vk_device.compute_queue, vk::AccessFlagBits::eTransferRead, vk::AccessFlagBits::eShaderWrite, false);
ggml_vk_dispatch_pipeline(s, vk_pipeline_mul_f32, { { d_X, 0, buf_sz }, { d_Y, 0, (uint32_t) d_Y.size }, { d_D, 0, buf_sz } }, sizeof(int) * pc.size(), pc.data(), { (uint32_t)ne00, (uint32_t)ne01, 1}); ggml_vk_dispatch_pipeline(s, vk_pipeline_mul_f32, { { d_X, 0, buf_sz }, { d_Y, 0, (uint32_t) d_Y.size }, { d_D, 0, buf_sz } }, sizeof(int) * pc.size(), pc.data(), { (uint32_t)ne00, (uint32_t)ne01, 1});
ggml_vk_sync_buffers(s.buffer, { ggml_vk_subbuffer(d_D) }, vk_device.compute_queue, vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eTransferRead, true);
// copy dst to host // copy dst to host
float * d = (float *) ((char *) dst->data + i02*nb2 + i03*nb3); float * d = (float *) ((char *) dst->data + i02*nb2 + i03*nb3);