Function ggml_cl_transform_tensor

ggml-opencl.cpp:1872–1899

}

void ggml_cl_transform_tensor(void * data, ggml_tensor * tensor) {
    const int64_t ne0 = tensor->ne[0];
    const int64_t ne1 = tensor->ne[1];
    const int64_t ne2 = tensor->ne[2];
    const int64_t ne3 = tensor->ne[3];

    const ggml_type type = tensor->type;
    const size_t s_sz = ggml_type_size(type) * (size_t) (ne0 * ne1 / ggml_blck_size(type));
    const size_t q_sz = s_sz * (size_t) (ne2 * ne3);

    size_t q_size;
    cl_mem dst = ggml_cl_pool_malloc(q_sz, &q_size);

    tensor->data = data;
    // copy tensor to device
    size_t offset = 0;
    for (int64_t i3 = 0; i3 < ne3; i3++) {
        for (int64_t i2 = 0; i2 < ne2; i2++) {
            CL_CHECK(ggml_cl_h2d_tensor_2d(queue, dst, offset, tensor, i3, i2, NULL));
            offset += s_sz;
        }
    }

    CL_CHECK(clFinish(queue));

    tensor->extra = dst;
    GGML_ASSERT(tensor->backend == GGML_BACKEND_GPU);
}