mozempk/ggml-org-llama-cpp-mirror
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-10-11 07:20:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

opencl: fix get_tensor for q5_K adreno gemm_nonshuffle kernel (#29555)

Browse Source
This commit is contained in:
lhez
2026-09-29 20:44:42 +03:00
committed by GitHub
parent 5c200e0c8d
commit 6dbbac4429
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-opencl/ggml-opencl.cpp
+5 -1
View File
@@ -12454,9 +12454,13 @@ static void ggml_backend_opencl_buffer_get_tensor(ggml_backend_buffer_t buffer,
buf_trans_d.allocate(backend_ctx->context, size_d);
buf_trans_dm.allocate(backend_ctx->context, size_dm);
// bin kernel transposes s but src kernel does not
cl_mem buf_s = extra->s;
if (use_q5_k_bin_kernels(backend_ctx, tensor)) {
transpose_2d_as_32b(backend_ctx, extra->q, buf_trans_q.buffer, size_q, M, K/8);
transpose_2d_as_8b (backend_ctx, extra->s, buf_trans_s.buffer, size_s, M, K/256*12, true, true);
buf_s = buf_trans_s.buffer;
} else {
transpose_2d_as_16b(backend_ctx, extra->q, buf_trans_q.buffer, size_q, M, K/4);
}
@@ -12467,7 +12471,7 @@ static void ggml_backend_opencl_buffer_get_tensor(ggml_backend_buffer_t buffer,
cl_kernel kernel = backend_ctx->kernel_restore_block_q5_K_noshuffle;
CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), &buf_trans_q.buffer));
CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), &buf_trans_qh.buffer));
CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), &buf_trans_s.buffer));
CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), &buf_s));
CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), &buf_trans_d.buffer));
CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), &buf_trans_dm.buffer));
CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), &data_device));