📄 gemm_beta.s
字号:
/*********************************************************************//* *//* Optimized BLAS libraries *//* By Kazushige Goto <kgoto@tacc.utexas.edu> *//* *//* Copyright (c) The University of Texas, 2005. All rights reserved. *//* UNIVERSITY EXPRESSLY DISCLAIMS ANY AND ALL WARRANTIES CONCERNING *//* THIS SOFTWARE AND DOCUMENTATION, INCLUDING ANY WARRANTIES OF *//* MERCHANTABILITY, FITNESS FOR ANY PARTICULAR PURPOSE, *//* NON-INFRINGEMENT AND WARRANTIES OF PERFORMANCE, AND ANY WARRANTY *//* THAT MIGHT OTHERWISE ARISE FROM COURSE OF DEALING OR USAGE OF *//* TRADE. NO WARRANTY IS EITHER EXPRESS OR IMPLIED WITH RESPECT TO *//* THE USE OF THE SOFTWARE OR DOCUMENTATION. *//* Under no circumstances shall University be liable for incidental, *//* special, indirect, direct or consequential damages or loss of *//* profits, interruption of business, or related expenses which may *//* arise from use of Software or Documentation, including but not *//* limited to those resulting from defects in Software and/or *//* Documentation, or loss or inaccuracy of data of any kind. *//*********************************************************************/#define ASSEMBLER#include "common.h"#define M r3#define N r4#define C r10#define LDC r11#define J r5#define PRE r6#define CO1 r7#define ALPHA f31#define STACKSIZE 32 PROLOGUE PROFCODE addi SP, SP, -STACKSIZE li r0, 0 stfd f14, 0(SP) stfd f15, 8(SP) stfd f31, 16(SP) stw r0, 24(SP)#ifdef linux#ifndef __64BIT__ lwz LDC, 8 + STACKSIZE(SP)#else ld C, 112 + STACKSIZE(SP) ld LDC, 120 + STACKSIZE(SP)#endif#endif#if defined(_AIX) || defined(__APPLE__)#ifdef __64BIT__ ld C, 112 + STACKSIZE(SP) ld LDC, 120 + STACKSIZE(SP)#else#ifdef DOUBLE lwz C, 60 + STACKSIZE(SP) lwz LDC, 64 + STACKSIZE(SP)#else lwz C, 56 + STACKSIZE(SP) lwz LDC, 60 + STACKSIZE(SP)#endif#endif#endif slwi LDC, LDC, BASE_SHIFT fmr ALPHA, f1 lfs f0, 24(SP) cmpwi cr0, M, 0 ble- LL(999) cmpwi cr0, N, 0 ble- LL(999) mr J, N fcmpu cr7, f1, f0 bne cr7, LL(20) .align 4LL(10): mr CO1, C add C, C, LDC addi PRE, 0, 32 * SIZE srawi. r0, M, 4 mtspr CTR, r0 ble LL(15) .align 4LL(12): STFD f0, 0 * SIZE(CO1) STFD f0, 1 * SIZE(CO1) STFD f0, 2 * SIZE(CO1) STFD f0, 3 * SIZE(CO1) STFD f0, 4 * SIZE(CO1) STFD f0, 5 * SIZE(CO1) STFD f0, 6 * SIZE(CO1) STFD f0, 7 * SIZE(CO1) STFD f0, 8 * SIZE(CO1) STFD f0, 9 * SIZE(CO1) STFD f0, 10 * SIZE(CO1) STFD f0, 11 * SIZE(CO1) STFD f0, 12 * SIZE(CO1) STFD f0, 13 * SIZE(CO1) STFD f0, 14 * SIZE(CO1) STFD f0, 15 * SIZE(CO1) dcbst PRE, CO1 addi CO1, CO1, 16 * SIZE bdnz LL(12) .align 4 LL(15): andi. r0, M, 15 mtspr CTR, r0 beq LL(19) .align 4LL(16): STFD f0, 0 * SIZE(CO1) addi CO1, CO1, 1 * SIZE bdnz LL(16) .align 4LL(19): addic. J, J, -1 bgt LL(10) b LL(999) .align 4LL(20): mr CO1, C add C, C, LDC addi PRE, 0, 16 * SIZE srawi. r0, M, 4 mtspr CTR, r0 ble LL(25) .align 4LL(22): LFD f0, 0 * SIZE(CO1) LFD f1, 1 * SIZE(CO1) LFD f2, 2 * SIZE(CO1) LFD f3, 3 * SIZE(CO1) LFD f4, 4 * SIZE(CO1) LFD f5, 5 * SIZE(CO1) LFD f6, 6 * SIZE(CO1) LFD f7, 7 * SIZE(CO1) LFD f8, 8 * SIZE(CO1) LFD f9, 9 * SIZE(CO1) LFD f10, 10 * SIZE(CO1) LFD f11, 11 * SIZE(CO1) LFD f12, 12 * SIZE(CO1) LFD f13, 13 * SIZE(CO1) LFD f14, 14 * SIZE(CO1) LFD f15, 15 * SIZE(CO1) FMUL f0, ALPHA, f0 FMUL f1, ALPHA, f1 FMUL f2, ALPHA, f2 FMUL f3, ALPHA, f3 FMUL f4, ALPHA, f4 FMUL f5, ALPHA, f5 FMUL f6, ALPHA, f6 FMUL f7, ALPHA, f7 FMUL f8, ALPHA, f8 FMUL f9, ALPHA, f9 FMUL f10, ALPHA, f10 FMUL f11, ALPHA, f11 FMUL f12, ALPHA, f12 FMUL f13, ALPHA, f13 FMUL f14, ALPHA, f14 FMUL f15, ALPHA, f15 STFD f0, 0 * SIZE(CO1) STFD f1, 1 * SIZE(CO1) STFD f2, 2 * SIZE(CO1) STFD f3, 3 * SIZE(CO1) STFD f4, 4 * SIZE(CO1) STFD f5, 5 * SIZE(CO1) STFD f6, 6 * SIZE(CO1) STFD f7, 7 * SIZE(CO1) STFD f8, 8 * SIZE(CO1) STFD f9, 9 * SIZE(CO1) STFD f10, 10 * SIZE(CO1) STFD f11, 11 * SIZE(CO1) STFD f12, 12 * SIZE(CO1) STFD f13, 13 * SIZE(CO1) STFD f14, 14 * SIZE(CO1) STFD f15, 15 * SIZE(CO1) addi CO1, CO1, 16 * SIZE dcbtst PRE, CO1 bdnz LL(22) .align 4 LL(25): andi. r0, M, 15 mtspr CTR, r0 ble LL(29) .align 4LL(26): LFD f0, 0 * SIZE(CO1) FMUL f0, f0, ALPHA STFD f0, 0 * SIZE(CO1) addi CO1, CO1, 1 * SIZE bdnz LL(26) .align 4LL(29): addic. J, J, -1 bgt LL(20) .align 4LL(999): li r3, 0 lfd f14, 0(SP) lfd f15, 8(SP) lfd f31, 16(SP) addi SP, SP, STACKSIZE blr EPILOGUE
⌨️ 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 + -