iamax_sse2.s

Optimized GotoBLAS libraries

/*********************************************************************/
/*                                                                   */
/*             Optimized BLAS libraries                              */
/*                     By Kazushige Goto <[e-mail]>     */
/*                                                                   */
/* 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 STACK	16
#define ARGS	 0
	
#define STACK_M		 4 + STACK + ARGS(%esp)
#define STACK_X		 8 + STACK + ARGS(%esp)
#define STACK_INCX	12 + STACK + ARGS(%esp)

#define RET	%eax
#define	M	%ebx
#define X	%ecx
#define INCX	%edx
#define I	%esi
#define MM	%ebp
#define XX	%edi
#define TEMP	%ebx
	
#ifdef MIN
#define maxpd	minpd
#define maxsd	minsd
#endif

	PROLOGUE

	pushl	%ebp
	pushl	%edi
	pushl	%esi
	pushl	%ebx

	PROFCODE

	movl	STACK_M, M
	movl	STACK_X, X
	movl	STACK_INCX, INCX

#ifdef F_INTERFACE
	movl	(M), M
	movl	(INCX), INCX
#endif

	pxor	%xmm0, %xmm0
#ifdef ABS
	pxor	%xmm7, %xmm7
#endif
	xor	RET, RET
	testl	M, M
	jle	.L999
	leal	(, INCX, SIZE), INCX
	testl	INCX, INCX
	jle	.L999

	movl	M, MM
	movl	X, XX

#ifdef ABS
	cmpeqpd	%xmm7, %xmm7
	psrlq	$1, %xmm7
#endif

	movsd	(XX), %xmm0
	addl	INCX, XX
	decl	MM
#ifdef ABS
	andpd	 %xmm7, %xmm0
#endif
	unpcklpd %xmm0, %xmm0
	movapd	 %xmm0, %xmm1
	movapd	 %xmm0, %xmm2
	movapd	 %xmm0, %xmm3
	cmpl	$SIZE, INCX
	jne	.L80

/* Analigned Check */
	cmpl	$7, MM
	jle	.L50

	testl	$7, XX
	jne	.L50		# Purely Unaligned Mode

	testl	$15, XX		# Checking for 128bit align
	je	.L05

	movsd	0 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	unpcklpd %xmm4, %xmm4
	maxpd	 %xmm4, %xmm3
	decl	MM
	addl	$SIZE, XX
	ALIGN_3

.L05:
	movl	MM,  I
	sarl	$4, I
	jle	.L15
	ALIGN_4
	
#define PREFETCHSIZE 256

.L11:
	prefetcht0	PREFETCHSIZE * SIZE(XX)

	movapd	0 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movapd	2 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1

	movapd	4 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm2

	movapd	6 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm3

	prefetcht0	(PREFETCHSIZE + 8) * SIZE(XX)

	movapd	8 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movapd	10 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1

	movapd	12 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm2

	movapd	14 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm3

	addl	$16 * SIZE, XX
	decl	I
	jg	.L11
	ALIGN_4

.L15:
	andl	$15,  MM
	jle	.L20

	testl	$8, MM
	je	.L16

	movapd	0 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movapd	2 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1

	movapd	4 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm2

	movapd	6 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm3
	addl	$8 * SIZE, XX
	ALIGN_3

.L16:
	testl	$4, MM
	je	.L17

	movapd	0 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movapd	2 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1
	addl	$4 * SIZE, XX
	ALIGN_3	

.L17:
	testl	$2, MM
	je	.L18

	movapd	0 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm2
	addl	$2 * SIZE, XX
	
.L18:
	testl	$1, MM
	je	.L20

	movsd	0 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	unpcklpd  %xmm4, %xmm4
	maxpd	%xmm4, %xmm3
	ALIGN_3

/* Finding Index */
.L20:
	movl	X, XX
	movl	M, MM

	maxpd	 %xmm1, %xmm0
	maxpd	 %xmm3, %xmm2
	maxpd	 %xmm2, %xmm0
	movapd	 %xmm0, %xmm1
	unpckhpd %xmm0, %xmm0
	maxsd	 %xmm1, %xmm0
	unpcklpd %xmm0, %xmm0

	testl	$15, XX		# Checking for 128bit align
	je	.L21

	movsd	0 * SIZE(XX), %xmm1
#ifdef ABS
	andpd	 %xmm7, %xmm1
#endif
	incl	RET
	comisd	%xmm0, %xmm1
	je	.L999
	addl	$SIZE, XX
	decl	MM
	ALIGN_3

.L21:
	movl	MM,  I
	sarl	$3, I
	jle	.L25
	ALIGN_4
	
.L22:
	prefetcht0	PREFETCHSIZE * SIZE(XX)

	movapd	0 * SIZE(XX), %xmm1
#ifdef ABS
	andpd	 %xmm7, %xmm1
#endif
	cmpeqpd	%xmm0, %xmm1

	movapd	2 * SIZE(XX), %xmm2
#ifdef ABS
	andpd	 %xmm7, %xmm2
#endif
	cmpeqpd	%xmm0, %xmm2

	movapd	4 * SIZE(XX), %xmm3
#ifdef ABS
	andpd	 %xmm7, %xmm3
#endif
	cmpeqpd	%xmm0, %xmm3

	movapd	6 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	cmpeqpd	%xmm0, %xmm4

	orpd	%xmm2, %xmm1
	orpd	%xmm4, %xmm3
	orpd	%xmm3, %xmm1
	movmskpd %xmm1, TEMP
	testl	 $3, TEMP
	jne	 .L23

	addl	$8 * SIZE, XX
	addl	$8, RET
	decl	I
	jg	.L22
	jmp	.L25
	ALIGN_4

.L23:
	movsd	0 * SIZE(XX), %xmm1
	movsd	1 * SIZE(XX), %xmm2
	movsd	2 * SIZE(XX), %xmm3
	movsd	3 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm1
	andpd	 %xmm7, %xmm2
	andpd	 %xmm7, %xmm3
	andpd	 %xmm7, %xmm4
#endif

	incl	RET
	comisd	%xmm0, %xmm1
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm2
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm3
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm4
	je	.L999

	movsd	4 * SIZE(XX), %xmm1
	movsd	5 * SIZE(XX), %xmm2
	movsd	6 * SIZE(XX), %xmm3

#ifdef ABS
	andpd	 %xmm7, %xmm1
	andpd	 %xmm7, %xmm2
	andpd	 %xmm7, %xmm3
#endif

	incl	RET
	comisd	%xmm0, %xmm1
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm2
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm3
	je	.L999
	incl	RET
	jmp	.L999
	ALIGN_3

.L25:
	testl	$4, MM
	je	.L27

	movsd	0 * SIZE(XX), %xmm1
	movsd	1 * SIZE(XX), %xmm2
	movsd	2 * SIZE(XX), %xmm3
	movsd	3 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm1
	andpd	 %xmm7, %xmm2
	andpd	 %xmm7, %xmm3
	andpd	 %xmm7, %xmm4
#endif
	addl	$4 * SIZE, XX
	incl	RET
	comisd	%xmm0, %xmm1
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm2
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm3
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm4
	je	.L999
	ALIGN_3	

.L27:
	testl	$2, MM
	je	.L28

	movsd	0 * SIZE(XX), %xmm1
	movsd	1 * SIZE(XX), %xmm2
#ifdef ABS
	andpd	 %xmm7, %xmm1
	andpd	 %xmm7, %xmm2
#endif
	addl	$2 * SIZE, XX
	incl	RET
	comisd	%xmm0, %xmm1
	je	.L999
	incl	RET
	comisd	%xmm0, %xmm2
	je	.L999
	ALIGN_3
	
.L28:
	incl	RET
	jmp	.L999
	ALIGN_3

.L50:
/* Unaligned Mode */
	movl	MM,  I
	sarl	$4, I
	jle	.L55
	ALIGN_4

.L51:
	prefetcht0	PREFETCHSIZE * SIZE(XX)

	movsd	0 * SIZE(XX), %xmm4
	movhpd	1 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movsd	2 * SIZE(XX), %xmm4
	movhpd	3 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1

	movsd	4 * SIZE(XX), %xmm4
	movhpd	5 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm2

	movsd	6 * SIZE(XX), %xmm4
	movhpd	7 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm3

	prefetcht0	(PREFETCHSIZE + 8) * SIZE(XX)

	movsd	8 * SIZE(XX), %xmm4
	movhpd	9 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movsd	10 * SIZE(XX), %xmm4
	movhpd	11 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1

	movsd	12 * SIZE(XX), %xmm4
	movhpd	13 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm2

	movsd	14 * SIZE(XX), %xmm4
	movhpd	15 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm3

	addl	$16 * SIZE, XX
	decl	I
	jg	.L51
	ALIGN_4

.L55:
	andl	$15,  MM
	jle	.L60

	testl	$8, MM
	je	.L56

	movsd	0 * SIZE(XX), %xmm4
	movhpd	1 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movsd	2 * SIZE(XX), %xmm4
	movhpd	3 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1

	movsd	4 * SIZE(XX), %xmm4
	movhpd	5 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm2

	movsd	6 * SIZE(XX), %xmm4
	movhpd	7 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm3

	addl	$8 * SIZE, XX
	ALIGN_3

.L56:
	testl	$4, MM
	je	.L57

	movsd	0 * SIZE(XX), %xmm4
	movhpd	1 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm0

	movsd	2 * SIZE(XX), %xmm4
	movhpd	3 * SIZE(XX), %xmm4
#ifdef ABS
	andpd	 %xmm7, %xmm4
#endif
	maxpd	%xmm4, %xmm1
	addl	$4 * SIZE, XX