strsm.cu

Nividia提供的CUDA的BLAS库源码

/*
 * Copyright 1993-2008 NVIDIA Corporation.  All rights reserved.
 *
 * NOTICE TO USER:   
 *
 * This source code is subject to NVIDIA ownership rights under U.S. and
 * international Copyright laws.  
 *
 * This software and the information contained herein is being provided 
 * under the terms and conditions of a Source Code License Agreement.     
 *
 * NVIDIA MAKES NO REPRESENTATION ABOUT THE SUITABILITY OF THIS SOURCE
 * CODE FOR ANY PURPOSE.  IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR 
 * IMPLIED WARRANTY OF ANY KIND.  NVIDIA DISCLAIMS ALL WARRANTIES WITH
 * REGARD TO THIS SOURCE CODE, INCLUDING ALL IMPLIED WARRANTIES OF
 * MERCHANTABILITY, NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR PURPOSE.
 * IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL,
 * OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS
 * OF USE, DATA OR PROFITS,  WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
 * OR OTHER TORTIOUS ACTION,  ARISING OUT OF OR IN CONNECTION WITH THE USE
 * OR PERFORMANCE OF THIS SOURCE CODE.  
 *
 * U.S. Government End Users.   This source code is a "commercial item" as 
 * that term is defined at  48 C.F.R. 2.101 (OCT 1995), consisting  of
 * "commercial computer  software"  and "commercial computer software 
 * documentation" as such terms are  used in 48 C.F.R. 12.212 (SEPT 1995)
 * and is provided to the U.S. Government only as a commercial end item.
 * Consistent with 48 C.F.R.12.212 and 48 C.F.R. 227.7202-1 through
 * 227.7202-4 (JUNE 1995), all U.S. Government End Users acquire the 
 * source code with only those rights set forth herein.
 */

/* This file contains the implementation of the BLAS-3 function strsm */

#include <stdlib.h>
#include <assert.h>
#include <string.h>
#include <stdio.h>
#include <limits.h>
#include <ctype.h>
#include <math.h>
#include "cublas.h"   /* CUBLAS public header file  */
#include "cublasP.h"  /* CUBLAS private header file */

__global__ void strsm_l_up_tr_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);

__global__ void strsm_l_up_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_sw_nu(struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_sw_nu(struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_sw_nu(struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_up_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_l_lo_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_alpha0_sw_nu(struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_up_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_l_lo_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);



__global__ void strsm_r_up_tr_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_alpha0_hw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_alpha0_sw (struct cublasStrsmParams parms);

__global__ void strsm_r_up_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_alpha0_hw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_up_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void strsm_r_lo_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_up_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_tr_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);
__global__ void fast_strsm_r_lo_nt_main_fulltile_alpha0_sw_nu (struct cublasStrsmParams parms);


typedef void (*pf) (struct cublasStrsmParams parms);

static pf strsm_l_hw[64] = {
    strsm_l_lo_tr_main_hw,
    strsm_l_lo_nt_main_hw,
    strsm_l_up_tr_main_hw,
    strsm_l_up_nt_main_hw,
    strsm_l_lo_tr_main_fulltile_hw,
    strsm_l_lo_nt_main_fulltile_hw,
    strsm_l_up_tr_main_fulltile_hw,
    strsm_l_up_nt_main_fulltile_hw,
    strsm_l_lo_tr_main_alpha0_hw,
    strsm_l_lo_nt_main_alpha0_hw,
    strsm_l_up_tr_main_alpha0_hw,
    strsm_l_up_nt_main_alpha0_hw,
    strsm_l_lo_tr_main_fulltile_alpha0_hw,
    strsm_l_lo_nt_main_fulltile_alpha0_hw,
    strsm_l_up_tr_main_fulltile_alpha0_hw,
    strsm_l_up_nt_main_fulltile_alpha0_hw,
    fast_strsm_l_lo_tr_main_hw,
    fast_strsm_l_lo_nt_main_hw,
    fast_strsm_l_up_tr_main_hw,
    fast_strsm_l_up_nt_main_hw,
    fast_strsm_l_lo_tr_main_fulltile_hw,
    fast_strsm_l_lo_nt_main_fulltile_hw,
    fast_strsm_l_up_tr_main_fulltile_hw,
    fast_strsm_l_up_nt_main_fulltile_hw,
    fast_strsm_l_lo_tr_main_alpha0_hw,
    fast_strsm_l_lo_nt_main_alpha0_hw,
    fast_strsm_l_up_tr_main_alpha0_hw,
    fast_strsm_l_up_nt_main_alpha0_hw,
    fast_strsm_l_lo_tr_main_fulltile_alpha0_hw,
    fast_strsm_l_lo_nt_main_fulltile_alpha0_hw,
    fast_strsm_l_up_tr_main_fulltile_alpha0_hw,
    fast_strsm_l_up_nt_main_fulltile_alpha0_hw,
    strsm_l_lo_tr_main_hw_nu,
    strsm_l_lo_nt_main_hw_nu,
    strsm_l_up_tr_main_hw_nu,
    strsm_l_up_nt_main_hw_nu,
    strsm_l_lo_tr_main_fulltile_hw_nu,
    strsm_l_lo_nt_main_fulltile_hw_nu,
    strsm_l_up_tr_main_fulltile_hw_nu,
    strsm_l_up_nt_main_fulltile_hw_nu,
    strsm_l_lo_tr_main_alpha0_hw_nu,
    strsm_l_lo_nt_main_alpha0_hw_nu,
    strsm_l_up_tr_main_alpha0_hw_nu,
    strsm_l_up_nt_main_alpha0_hw_nu,
    strsm_l_lo_tr_main_fulltile_alpha0_hw_nu,
    strsm_l_lo_nt_main_fulltile_alpha0_hw_nu,
    strsm_l_up_tr_main_fulltile_alpha0_hw_nu,
    strsm_l_up_nt_main_fulltile_alpha0_hw_nu,
    fast_strsm_l_lo_tr_main_hw_nu,
    fast_strsm_l_lo_nt_main_hw_nu,
    fast_strsm_l_up_tr_main_hw_nu,
    fast_strsm_l_up_nt_main_hw_nu,
    fast_strsm_l_lo_tr_main_fulltile_hw_nu,
    fast_strsm_l_lo_nt_main_fulltile_hw_nu,
    fast_strsm_l_up_tr_main_fulltile_hw_nu,
    fast_strsm_l_up_nt_main_fulltile_hw_nu,
    fast_strsm_l_lo_tr_main_alpha0_hw_nu,