⭐ 欢迎来到虫虫下载站! | 📦 资源下载 📁 资源专辑 ℹ️ 关于我们
⭐ 虫虫下载站

📄 strmm.cu

📁 Nividia提供的CUDA的BLAS库源码
💻 CU
📖 第 1 页 / 共 5 页
字号:
/* * 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 strmm */#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 */#define BLK_LOG             (5)#define BLK                 (1 << BLK_LOG)#define JINC                (BLK * CUBLAS_STRMM_CTAS)  // used by strmm_l#define IINC                (BLK * CUBLAS_STRMM_CTAS)  // used by strmm_r#if ((CUBLAS_STRMM_THREAD_COUNT<BLK))#error block dimension must be >= threadcount#endif#if ((CUBLAS_STRMM_THREAD_COUNT%BLK)!=0)#error threadcount and block dimensions do not divide evenly#endif#define A_NBR_COLS          (CUBLAS_STRMM_THREAD_COUNT/BLK)#define B_NBR_COLS          (CUBLAS_STRMM_THREAD_COUNT/BLK)#if (((BLK*BLK)%CUBLAS_STRMM_THREAD_COUNT)!=0)#error blocksize of A and B not evenly divided by threadcount!#endif#define A_ELEMS_PER_THREAD  ((BLK * BLK) / CUBLAS_STRMM_THREAD_COUNT)#define B_ELEMS_PER_THREAD  ((BLK * BLK) / CUBLAS_STRMM_THREAD_COUNT)__global__ void strmm_r_lo_tr_main_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_alpha0_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_alpha0_fulltile_sw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_alpha0_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_lo_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_r_up_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_lo_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void strmm_l_up_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_lo_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_r_up_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_tr_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_lo_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);__global__ void fast_strmm_l_up_nt_main_unit_alpha0_fulltile_hw (struct cublasStrmmParams parms);typedef void (*pf) (struct cublasStrmmParams parms);static pf strmm_sw[128] = {    strmm_r_lo_tr_main_sw,    strmm_r_up_tr_main_sw,    strmm_r_lo_nt_main_sw,    strmm_r_up_nt_main_sw,    strmm_l_lo_tr_main_sw,    strmm_l_up_tr_main_sw,    strmm_l_lo_nt_main_sw,    strmm_l_up_nt_main_sw,    strmm_r_lo_tr_main_unit_sw,    strmm_r_up_tr_main_unit_sw,    strmm_r_lo_nt_main_unit_sw,    strmm_r_up_nt_main_unit_sw,    strmm_l_lo_tr_main_unit_sw,    strmm_l_up_tr_main_unit_sw,    strmm_l_lo_nt_main_unit_sw,    strmm_l_up_nt_main_unit_sw,    strmm_r_lo_tr_main_alpha0_sw,    strmm_r_up_tr_main_alpha0_sw,    strmm_r_lo_nt_main_alpha0_sw,    strmm_r_up_nt_main_alpha0_sw,    strmm_l_lo_tr_main_alpha0_sw,    strmm_l_up_tr_main_alpha0_sw,    strmm_l_lo_nt_main_alpha0_sw,    strmm_l_up_nt_main_alpha0_sw,    strmm_r_lo_tr_main_unit_alpha0_sw,    strmm_r_up_tr_main_unit_alpha0_sw,    strmm_r_lo_nt_main_unit_alpha0_sw,    strmm_r_up_nt_main_unit_alpha0_sw,    strmm_l_lo_tr_main_unit_alpha0_sw,    strmm_l_up_tr_main_unit_alpha0_sw,    strmm_l_lo_nt_main_unit_alpha0_sw,    strmm_l_up_nt_main_unit_alpha0_sw,    fast_strmm_r_lo_tr_main_sw,    fast_strmm_r_up_tr_main_sw,    fast_strmm_r_lo_nt_main_sw,    fast_strmm_r_up_nt_main_sw,    fast_strmm_l_lo_tr_main_sw,    fast_strmm_l_up_tr_main_sw,

⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -