⭐ 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

📄 gemm_tcopy_hummer_4.s

📁 Optimized GotoBLAS libraries
💻 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 AO3	r10#define AO4	r11#define J	r25#define B1	r26#define B2	r27#define B3	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)	stwu	r27,  -4(SP)	stwu	r26,  -4(SP)	stwu	r25,  -4(SP)	slwi	LDA, LDA, BASE_SHIFT	slwi	M4, M, 2 + BASE_SHIFT	li	r8, -4	li	r9, -2	and	B2, N, r8	and	B3, N, r9	mullw	B2, B2, M	mullw	B3, B3, M	slwi	B2, B2, BASE_SHIFT	slwi	B3, B3, BASE_SHIFT	add	B2, B2, B	add	B3, B3, B	cmpwi	cr0, M, 0	ble-	.L99	cmpwi	cr0, N, 0	ble-	.L99	subi	B2, B2, 2 * SIZE	subi	B3, B3, 2 * SIZE	subi	M4, M4, 14 * SIZE	li	INC,  1 * SIZE	li	INC2, 2 * SIZE	andi.	r0, A,   2 * SIZE - 1	bne	.L100	andi.	r0, LDA, 2 * SIZE - 1	bne	.L100		subi	A, A, 2 * SIZE	srawi.	J,  M,  2	ble	.L20	.align 4.L10:	mr	AO1, A	add	AO2, A,   LDA	add	AO3, AO2, LDA	add	AO4, AO3, LDA	add	A,   AO4, LDA	sub	B1, B, M4	addi	B, B, 16 * SIZE	srawi.	r0,  N,  2	mtspr	CTR, r0	ble	.L15	.align 4.L12:	LFPDUX	c01,   AO1, INC2	LFPDUX	c02,   AO1, INC2	LFPDUX	c03,   AO2, INC2	LFPDUX	c04,   AO2, INC2 	LFPDUX	c05,   AO3, INC2	LFPDUX	c06,   AO3, INC2	LFPDUX	c07,   AO4, INC2	LFPDUX	c08,   AO4, INC2	STFPDUX	c01,   B1, M4	STFPDUX	c02,   B1, INC2	STFPDUX	c03,   B1, INC2	STFPDUX	c04,   B1, INC2	STFPDUX	c05,   B1, INC2	STFPDUX	c06,   B1, INC2	STFPDUX	c07,   B1, INC2	STFPDUX	c08,   B1, INC2	bdnz	.L12	.align 4	.L15:	andi.	r0,  N,  3	ble	.L19	andi.	r0,  N,  2	ble	.L17	LFPDUX	c01,   AO1, INC2	LFPDUX	c03,   AO2, INC2	LFPDUX	c05,   AO3, INC2	LFPDUX	c07,   AO4, INC2	STFPDUX	c01,   B2, INC2	STFPDUX	c03,   B2, INC2	STFPDUX	c05,   B2, INC2	STFPDUX	c07,   B2, INC2	.align 4.L17:	andi.	r0,  N,  1	ble	.L19	LFDUX	c01,   AO1, INC2	LFDUX	c02,   AO2, INC2	LFDUX	c03,   AO3, INC2	LFDUX	c04,   AO4, INC2	fsmfp	c01, c02	fsmfp	c03, c04	STFPDUX	c01,   B3, INC2	STFPDUX	c03,   B3, INC2	.align 4.L19:	addic.	J, J, -1	bgt	.L10	.align 4.L20:	andi.	J,  M,  2	addi	M4, M4, 8 * SIZE	ble	.L30	mr	AO1, A	add	AO2, A,   LDA	add	A,   AO2, LDA	sub	B1, B, M4	addi	B, B, 8 * SIZE	srawi.	r0,  N,  2	mtspr	CTR, r0	ble	.L23	.align 4.L22:	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	.L22	.align 4	.L23:	andi.	r0,  N,  2	ble	.L24	LFPDUX	c01,   AO1, INC2	LFPDUX	c03,   AO2, INC2	STFPDUX	c01,   B2, INC2	STFPDUX	c03,   B2, INC2	.align 4.L24:	andi.	r0,  N,  1	ble	.L30	LFDUX	c01,   AO1, INC2	LFDUX	c02,   AO2, INC2	fsmfp	c01, c02	STFPDUX	c01,   B3, INC2	.align 4.L30:	andi.	J,  M,  1	addi	M4, M4, 4 * SIZE	ble	.L99	mr	AO1, A	sub	B1, B, M4	srawi.	r0,  N,  2	mtspr	CTR, r0	ble	.L33	.align 4.L32:	LFPDUX	c01,   AO1, INC2	LFPDUX	c02,   AO1, INC2	STFPDUX	c01,   B1, M4	STFPDUX	c02,   B1, INC2	bdnz	.L32	.align 4	.L33:	andi.	r0,  N,  2	ble	.L34	LFPDUX	c01,   AO1, INC2	STFPDUX	c01,   B2, INC2	.align 4.L34:	andi.	r0,  N,  1	ble	.L99	LFDX	c01,   AO1, INC2	STFDX	c01,   B3, INC2	.align 4.L99:	addi	SP, SP, -4	lwzu	r25,   4(SP)	lwzu	r26,   4(SP)	lwzu	r27,   4(SP)	lwzu	r28,   4(SP)	lwzu	r29,   4(SP)	lwzu	r30,   4(SP)	lwzu	r31,   4(SP)	addi	SP, SP, 4	blr.L100:	subi	A, A, SIZE	srawi.	J,  M,  2	ble	.L120	.align 4.L110:	mr	AO1, A	add	AO2, A,   LDA	add	AO3, AO2, LDA	add	AO4, AO3, LDA	add	A,   AO4, LDA	sub	B1, B, M4	addi	B, B, 16 * SIZE	srawi.	r0,  N,  2	mtspr	CTR, r0	ble	.L115	.align 4.L112:	LFDUX	c01,   AO1, INC	LFDUX	c03,   AO2, INC 	LFDUX	c05,   AO3, INC	LFDUX	c07,   AO4, INC	LFSDUX	c01,   AO1, INC	LFSDUX	c03,   AO2, INC	LFSDUX	c05,   AO3, INC	LFSDUX	c07,   AO4, INC	LFDUX	c02,   AO1, INC	LFDUX	c04,   AO2, INC	LFDUX	c06,   AO3, INC	LFDUX	c08,   AO4, INC	LFSDUX	c02,   AO1, INC	LFSDUX	c04,   AO2, INC	LFSDUX	c06,   AO3, INC	LFSDUX	c08,   AO4, INC	STFPDUX	c01,   B1, M4	STFPDUX	c02,   B1, INC2	STFPDUX	c03,   B1, INC2	STFPDUX	c04,   B1, INC2	STFPDUX	c05,   B1, INC2	STFPDUX	c06,   B1, INC2	STFPDUX	c07,   B1, INC2	STFPDUX	c08,   B1, INC2	bdnz	.L112	.align 4	.L115:	andi.	r0,  N,  3	ble	.L119	andi.	r0,  N,  2	ble	.L117	LFDUX	c01,   AO1, INC	LFDUX	c03,   AO2, INC	LFDUX	c05,   AO3, INC	LFDUX	c07,   AO4, INC	LFSDUX	c01,   AO1, INC	LFSDUX	c03,   AO2, INC	LFSDUX	c05,   AO3, INC	LFSDUX	c07,   AO4, INC	STFPDUX	c01,   B2, INC2	STFPDUX	c03,   B2, INC2	STFPDUX	c05,   B2, INC2	STFPDUX	c07,   B2, INC2	.align 4.L117:	andi.	r0,  N,  1	ble	.L119	LFDUX	c01,   AO1, INC	LFDUX	c02,   AO2, INC	LFDUX	c03,   AO3, INC	LFDUX	c04,   AO4, INC	fsmfp	c01, c02	fsmfp	c03, c04	STFPDUX	c01,   B3, INC2	STFPDUX	c03,   B3, INC2	.align 4.L119:	addic.	J, J, -1	bgt	.L110	.align 4.L120:	andi.	J,  M,  2	addi	M4, M4, 8 * SIZE	ble	.L130	mr	AO1, A	add	AO2, A,   LDA	add	A,   AO2, LDA	sub	B1, B, M4	addi	B, B, 8 * SIZE	srawi.	r0,  N,  2	mtspr	CTR, r0	ble	.L123	.align 4.L122:	LFDUX	c01,   AO1, INC	LFDUX	c03,   AO2, INC	LFSDUX	c01,   AO1, INC	LFSDUX	c03,   AO2, INC	LFDUX	c02,   AO1, INC	LFDUX	c04,   AO2, INC	LFSDUX	c02,   AO1, INC	LFSDUX	c04,   AO2, INC	STFPDUX	c01,   B1, M4	STFPDUX	c02,   B1, INC2	STFPDUX	c03,   B1, INC2	STFPDUX	c04,   B1, INC2	bdnz	.L122	.align 4	.L123:	andi.	r0,  N,  2	ble	.L124	LFDUX	c01,   AO1, INC	LFDUX	c03,   AO2, INC	LFSDUX	c01,   AO1, INC	LFSDUX	c03,   AO2, INC	STFPDUX	c01,   B2, INC2	STFPDUX	c03,   B2, INC2	.align 4.L124:	andi.	r0,  N,  1	ble	.L130	LFDUX	c01,   AO1, INC	LFDUX	c02,   AO2, INC	fsmfp	c01, c02	STFPDUX	c01,   B3, INC2	.align 4.L130:	andi.	J,  M,  1	addi	M4, M4, 4 * SIZE	ble	.L999	mr	AO1, A	sub	B1, B, M4	srawi.	r0,  N,  2	mtspr	CTR, r0	ble	.L133	.align 4.L132:	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	.L132	.align 4	.L133:	andi.	r0,  N,  2	ble	.L134	LFDUX	c01,   AO1, INC	LFDUX	c02,   AO1, INC	fsmfp	c01, c02	STFPDUX	c01,   B2, INC2	.align 4.L134:	andi.	r0,  N,  1	ble	.L999	LFDX	c01,   AO1, INC	STFDX	c01,   B3, INC2	.align 4.L999:	addi	SP, SP, -4	lwzu	r25,   4(SP)	lwzu	r26,   4(SP)	lwzu	r27,   4(SP)	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 + -