Custom plugin "pack schema"/partitioning error with naive GEMM ref_kernels #866

void naive_gemm
     (
       const obj_t*  alpha,
       const obj_t*  a,
       const obj_t*  b,
       const obj_t*  beta,
       const obj_t*  c,
       const cntx_t* cntx,
       const rntm_t* rntm
     )
{
    bli_init_once();

    if ( bli_error_checking_is_enabled() )
        bli_gemm_check( alpha, a, b, beta, c, cntx );

    if ( bli_l3_return_early_if_trivial( alpha, a, b, beta, c ) == BLIS_SUCCESS )
        return;

    num_t dt = bli_obj_dt( c );
    ind_t im = BLIS_NAT;

    obj_t a_local;
    obj_t b_local;
    obj_t c_local;
    bli_obj_alias_submatrix(  a, &a_local );
    bli_obj_alias_submatrix(  b, &b_local );
    bli_obj_alias_submatrix(  c, &c_local );

    if ( cntx == NULL ) cntx = bli_gks_query_cntx();
    gemm_cntl_t cntl;
    bli_gemm_cntl_init
    (
      im,
      BLIS_GEMM,
      alpha,
      &a_local,
      &b_local,
      beta,
      &c_local,
      cntx,
      &cntl
    );

    bli_gemm_cntl_finalize
    (
      BLIS_GEMM,
      &a_local,
      &b_local,
      &c_local,
      &cntl
    );

    bli_l3_thread_decorator
    ( &a_local, &b_local, &c_local, cntx, ( cntl_t* )&cntl, rntm );
}

#include <blis.h>

int main(int argc, char** argv) {
	num_t dt_r, dt_c;
	num_t dt_s, dt_d;
	num_t dt_a, dt_b;
	dim_t m, n, k;
	inc_t rs, cs;

	obj_t a, b, D, c_s, c;

	// Create some matrix operands to work with.
	m = 4; n = 5; rs = 0; cs = 0;
	bli_obj_create( BLIS_DOUBLE, m, m, 1, m, &c );
	bli_obj_create( BLIS_DOUBLE, m, n, 1, m, &a );
	bli_obj_create( BLIS_DOUBLE, n, m, 1, n, &b );

	const obj_t* alpha = &BLIS_ONE;
	const obj_t* beta  = &BLIS_ONE;

	bli_randm( &a );
	bli_randm( &b );
	bli_setm( &BLIS_ZERO, &c );

	naive_gemm(alpha, &a, &b, beta, &c, NULL, NULL);

	// Free the objects.
	bli_obj_free( &a );
	bli_obj_free( &b );
	bli_obj_free( &c );
	return 0;
}