📄 zgemm_tcopy_hummer_2.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 A r5#define LDA r6#define B r7#define AO1 r8#define AO2 r9#define J r10#define B1 r11#define B2 r28#define M4 r29#define INC r30#define INC2 r31 #define c01 f0#define c02 f1#define c03 f2#define c04 f3#define c05 f4#define c06 f5#define c07 f6#define c08 f7 PROLOGUE PROFCODE stwu r31, -4(SP) stwu r30, -4(SP) stwu r29, -4(SP) stwu r28, -4(SP) slwi LDA, LDA, ZBASE_SHIFT slwi M4, M, 1 + ZBASE_SHIFT li r9, -2 and B2, N, r9 mullw B2, B2, M slwi B2, B2, ZBASE_SHIFT add B2, B2, B cmpwi cr0, M, 0 ble- LL(99) cmpwi cr0, N, 0 ble- LL(99) subi B2, B2, 2 * SIZE subi M4, M4, 6 * SIZE li INC, 1 * SIZE li INC2, 2 * SIZE andi. r0, A, 2 * SIZE - 1 bne LL(100) subi A, A, 2 * SIZE srawi. J, M, 1 ble LL(20) .align 4LL(10): mr AO1, A add AO2, A, LDA add A, AO2, LDA sub B1, B, M4 addi B, B, 8 * SIZE srawi. r0, N, 1 mtspr CTR, r0 ble LL(15) .align 4LL(12): LFPDUX c01, AO1, INC2 LFPDUX c02, AO1, INC2 LFPDUX c03, AO2, INC2 LFPDUX c04, AO2, INC2 STFPDUX c01, B1, M4 STFPDUX c02, B1, INC2 STFPDUX c03, B1, INC2 STFPDUX c04, B1, INC2 bdnz LL(12) .align 4 LL(15): andi. r0, N, 1 ble LL(19) LFPDUX c01, AO1, INC2 LFPDUX c02, AO2, INC2 STFPDUX c01, B2, INC2 STFPDUX c02, B2, INC2 .align 4LL(19): addic. J, J, -1 bgt LL(10) .align 4LL(20): andi. J, M, 1 addi M4, M4, 4 * SIZE ble LL(99) mr AO1, A sub B1, B, M4 srawi. r0, N, 1 mtspr CTR, r0 ble LL(23) .align 4LL(22): LFPDUX c01, AO1, INC2 LFPDUX c02, AO1, INC2 STFPDUX c01, B1, M4 STFPDUX c02, B1, INC2 bdnz LL(22) .align 4 LL(23): andi. r0, N, 1 ble LL(99) LFPDUX c01, AO1, INC2 STFPDUX c01, B2, INC2 .align 4LL(99): addi SP, SP, -4 lwzu r28, 4(SP) lwzu r29, 4(SP) lwzu r30, 4(SP) lwzu r31, 4(SP) addi SP, SP, 4 blr .align 4LL(100): subi A, A, SIZE srawi. J, M, 1 ble LL(120) .align 4LL(110): mr AO1, A add AO2, A, LDA add A, AO2, LDA sub B1, B, M4 addi B, B, 8 * SIZE srawi. r0, N, 1 mtspr CTR, r0 ble LL(115) .align 4LL(112): LFDUX c01, AO1, INC LFDUX c02, AO1, INC LFDUX c03, AO1, INC LFDUX c04, AO1, INC LFDUX c05, AO2, INC fsmfp c01, c02 LFDUX c06, AO2, INC fsmfp c03, c04 LFDUX c07, AO2, INC fsmfp c05, c06 LFDUX c08, AO2, INC fsmfp c07, c08 STFPDUX c01, B1, M4 STFPDUX c03, B1, INC2 STFPDUX c05, B1, INC2 STFPDUX c07, B1, INC2 bdnz LL(112) .align 4 LL(115): andi. r0, N, 1 ble LL(119) LFDUX c01, AO1, INC LFDUX c02, AO1, INC LFDUX c03, AO2, INC LFDUX c04, AO2, INC fsmfp c01, c02 fsmfp c03, c04 STFPDUX c01, B2, INC2 STFPDUX c03, B2, INC2 .align 4LL(119): addic. J, J, -1 bgt LL(110) .align 4LL(120): andi. J, M, 1 addi M4, M4, 4 * SIZE ble LL(999) mr AO1, A sub B1, B, M4 srawi. r0, N, 1 mtspr CTR, r0 ble LL(123) .align 4LL(122): LFDUX c01, AO1, INC LFDUX c02, AO1, INC LFDUX c03, AO1, INC LFDUX c04, AO1, INC fsmfp c01, c02 fsmfp c03, c04 STFPDUX c01, B1, M4 STFPDUX c03, B1, INC2 bdnz LL(122) .align 4 LL(123): andi. r0, N, 1 ble LL(999) LFDUX c01, AO1, INC LFDUX c02, AO1, INC fsmfp c01, c02 STFPDUX c01, B2, INC2 .align 4LL(999): addi SP, SP, -4 lwzu r28, 4(SP) lwzu r29, 4(SP) lwzu r30, 4(SP) lwzu r31, 4(SP) addi SP, SP, 4 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 + -