Function f32_kern

dnn/src/fallback/matrix_mul/algos.cpp:427–443

/* ===================== F32 algo ===================== */
namespace {
void f32_kern(const MatrixMulImpl::KernParam& kern_param) {
    MIDOUT_BEGIN(megdnn_fb_gi_f32_4x12, midout_iv("f32_kern"_hash)) {
        auto M = kern_param.M, N = kern_param.N, K = kern_param.K;
        auto trA = kern_param.trA, trB = kern_param.trB;
        auto LDA = kern_param.LDA, LDB = kern_param.LDB, LDC = kern_param.LDC;
        auto A_type = kern_param.A_type, B_type = kern_param.B_type,
             C_type = kern_param.C_type;
        const auto Aptr = kern_param.A<float>(), Bptr = kern_param.B<float>();
        auto Cptr = kern_param.C<float>();

        matmul::fallback::gi_sgemm_4x12 strategy(M, N, K, A_type, B_type, C_type);
        megdnn::matmul::GemmInterleaved<matmul::fallback::gi_sgemm_4x12>(
                M, N, K, trA, trB, strategy)
                .execute(Aptr, LDA, Bptr, LDB, Cptr, LDC, kern_param.workspace_ptr);
    }
    MIDOUT_END();
}

}  // anonymous namespace