Function ggml_compute_forward_mul_mat

ggml.c:9598–9807

#endif

static void ggml_compute_forward_mul_mat(
        const struct ggml_compute_params * params,
        const struct ggml_tensor * src0,
        const struct ggml_tensor * src1,
              struct ggml_tensor * dst) {
    int64_t t0 = ggml_perf_time_us();
    UNUSED(t0);

    GGML_TENSOR_BINARY_OP_LOCALS

    const int ith = params->ith;
    const int nth = params->nth;

    const enum ggml_type type = src0->type;

    const bool src1_cont = ggml_is_contiguous(src1);

    ggml_vec_dot_t    const vec_dot               = type_traits[type].vec_dot;
    enum ggml_type    const vec_dot_type          = type_traits[type].vec_dot_type;
    ggml_from_float_t const from_float_to_vec_dot = type_traits[vec_dot_type].from_float;

    GGML_ASSERT(ne0 == ne01);
    GGML_ASSERT(ne1 == ne11);
    GGML_ASSERT(ne2 == ne12);
    GGML_ASSERT(ne3 == ne13);

    // we don't support permuted src0 or src1
    GGML_ASSERT(nb00 == ggml_type_size(type));
    GGML_ASSERT(nb10 == ggml_type_size(src1->type));

    // dst cannot be transposed or permuted
    GGML_ASSERT(nb0 == sizeof(float));
    GGML_ASSERT(nb0 <= nb1);
    GGML_ASSERT(nb1 <= nb2);
    GGML_ASSERT(nb2 <= nb3);

    // broadcast factors
    const int64_t r2 = ne12/ne02;
    const int64_t r3 = ne13/ne03;

    // nb01 >= nb00 - src0 is not transposed
    //   compute by src0 rows

#if defined(GGML_USE_CLBLAST)
    if (ggml_cl_can_mul_mat(src0, src1, dst)) {
        if (params->ith == 0 && params->type == GGML_TASK_COMPUTE) {
            ggml_cl_mul_mat(src0, src1, dst, params->wdata, params->wsize);
        }
        return;
    }
#endif

#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
    if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
        if (params->ith != 0) {
            return;
        }