📄 zgemm_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_R f30#define ALPHA_I f31#define STACKSIZE 32 PROLOGUE PROFCODE addi SP, SP, -STACKSIZE li r0, 0 stfd f30, 0(SP) stfd f31, 8(SP) stw r0, 16(SP)#ifdef linux#ifndef __64BIT__ lwz LDC, 8 + STACKSIZE(SP)#else ld C, 120 + STACKSIZE(SP) ld LDC, 128 + STACKSIZE(SP)#endif#endif#if defined(_AIX) || defined(__APPLE__)#ifdef __64BIT__ ld C, 120 + STACKSIZE(SP) ld LDC, 128 + STACKSIZE(SP)#else#ifdef DOUBLE lwz C, 68 + STACKSIZE(SP) lwz LDC, 72 + STACKSIZE(SP)#else lwz C, 60 + STACKSIZE(SP) lwz LDC, 64 + STACKSIZE(SP)#endif#endif#endif slwi LDC, LDC, ZBASE_SHIFT lfs f0, 16(SP) fmr ALPHA_R, f1 fmr ALPHA_I, f2 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) fcmpu cr7, f2, f0 bne cr7, LL(20) .align 4LL(10): mr CO1, C add C, C, LDC addi PRE, 0, 32 * SIZE srawi. r0, M, 3 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, 7 mtspr CTR, r0 beq LL(19) .align 4LL(16): STFD f0, 0 * SIZE(CO1) STFD f0, 1 * SIZE(CO1) addi CO1, CO1, 2 * 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, 2 mtspr CTR, r0 ble LL(25) .align 4LL(22): LFD f3, 0 * SIZE(CO1) LFD f4, 1 * SIZE(CO1) LFD f5, 2 * SIZE(CO1) LFD f6, 3 * SIZE(CO1) LFD f7, 4 * SIZE(CO1) LFD f8, 5 * SIZE(CO1) LFD f9, 6 * SIZE(CO1) LFD f10, 7 * SIZE(CO1) FMUL f0, ALPHA_I, f4 FMUL f4, ALPHA_R, f4 FMUL f11, ALPHA_I, f6 FMUL f6, ALPHA_R, f6 FMUL f12, ALPHA_I, f8 FMUL f8, ALPHA_R, f8 FMUL f13, ALPHA_I, f10 FMUL f10, ALPHA_R, f10 FMADD f4, ALPHA_I, f3, f4 FMSUB f3, ALPHA_R, f3, f0 FMADD f6, ALPHA_I, f5, f6 FMSUB f5, ALPHA_R, f5, f11 FMADD f8, ALPHA_I, f7, f8 FMSUB f7, ALPHA_R, f7, f12 FMADD f10, ALPHA_I, f9, f10 FMSUB f9, ALPHA_R, f9, f13 STFD f3, 0 * SIZE(CO1) STFD f4, 1 * SIZE(CO1) STFD f5, 2 * SIZE(CO1) STFD f6, 3 * SIZE(CO1) STFD f7, 4 * SIZE(CO1) STFD f8, 5 * SIZE(CO1) STFD f9, 6 * SIZE(CO1) STFD f10, 7 * SIZE(CO1) addi CO1, CO1, 8 * SIZE dcbtst PRE, CO1 bdnz LL(22) .align 4 LL(25): andi. r0, M, 3 mtspr CTR, r0 ble LL(29) .align 4LL(26): LFD f0, 0 * SIZE(CO1) LFD f1, 1 * SIZE(CO1) FMUL f5, ALPHA_I, f1 FMUL f1, ALPHA_R, f1 FMADD f1, ALPHA_I, f0, f1 FMSUB f0, ALPHA_R, f0, f5 STFD f0, 0 * SIZE(CO1) STFD f1, 1 * SIZE(CO1) addi CO1, CO1, 2 * SIZE bdnz LL(26) .align 4LL(29): addic. J, J, -1 bgt LL(20) .align 4LL(999): li r3, 0 lfd f30, 0(SP) lfd f31, 8(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 + -