zscal.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	 8
	
#define STACK_N		 4 + STACK(%esp)
#ifdef XDOUBLE
#define ALPHA_R		16 + STACK(%esp)
#define ALPHA_I		32 + STACK(%esp)
#define STACK_X		48 + STACK(%esp)
#define STACK_INCX	52 + STACK(%esp)
#elif defined(DOUBLE)
#define ALPHA_R		16 + STACK(%esp)
#define ALPHA_I		24 + STACK(%esp)
#define STACK_X		32 + STACK(%esp)
#define STACK_INCX	36 + STACK(%esp)
#else
#define ALPHA_R		16 + STACK(%esp)
#define ALPHA_I		20 + STACK(%esp)
#define STACK_X		24 + STACK(%esp)
#define STACK_INCX	28 + STACK(%esp)
#endif

#define N		%esi
#define X		%edx
#define INCX		%ebx

#define I		%ecx

	PROLOGUE

	pushl	%esi
	pushl	%ebx

	PROFCODE

#ifdef F_INTERFACE_GFORT
	EMMS
#endif

	movl	STACK_N,    N
	movl	STACK_X,    X
	movl	STACK_INCX, INCX

	sall	$ZBASE_SHIFT, INCX

	FLD	ALPHA_R
	FLD	ALPHA_I

	testl	N, N
	jle	.L999

	fld	%st(1)
	fabs
	fld	%st(1)
	fabs
	faddp	%st, %st(1)

	fldz
	fcomip	%st(1), %st
	ffreep	%st
	jne	.L30

	EMMS

	pxor	%mm0, %mm0

	cmpl	$2 * SIZE, INCX
	jne	.L20

	movl	N,  I
	sarl	$2, I
	jle	.L15
	ALIGN_4

.L12:
#ifdef XDOUBLE
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
	movq	%mm0,  32(X)
	movq	%mm0,  40(X)
	movq	%mm0,  48(X)
	movq	%mm0,  56(X)
	movq	%mm0,  64(X)
	movq	%mm0,  72(X)
	movq	%mm0,  80(X)
	movq	%mm0,  88(X)
	movq	%mm0,  96(X)
	movq	%mm0, 104(X)
	movq	%mm0, 112(X)
	movq	%mm0, 120(X)
#elif defined(DOUBLE)
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
	movq	%mm0,  32(X)
	movq	%mm0,  40(X)
	movq	%mm0,  48(X)
	movq	%mm0,  56(X)
#else
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
#endif

	addl	$8 * SIZE, X
	decl	I
	jg	.L12
	ALIGN_3

.L15:
	movl	N,  I
	andl	$3, I
	jle	.L18
	ALIGN_2

.L16:
#ifdef XDOUBLE
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
#elif defined(DOUBLE)
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
#else
	movq	%mm0,   0(X)
#endif

	addl	$2 * SIZE, X
	decl	I
	jg	.L16

.L18:
	EMMS

	xorl	%eax, %eax
	popl	%ebx
	popl	%esi
	ret
	ALIGN_2

.L20:
	movl	N,  I
	sarl	$2, I
	jle	.L25
	ALIGN_3

.L22:
#ifdef XDOUBLE
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
	addl	INCX, X

#elif defined(DOUBLE)
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	addl	INCX, X
#else
	movq	%mm0,   0(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	addl	INCX, X

	movq	%mm0,   0(X)
	addl	INCX, X
#endif

	decl	I
	jg	.L22
	ALIGN_3

.L25:
	movl	N,  I
	andl	$3, I
	jle	.L28
	ALIGN_3

.L26:
#ifdef XDOUBLE
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	movq	%mm0,  16(X)
	movq	%mm0,  24(X)
	addl	INCX, X
#elif defined(DOUBLE)
	movq	%mm0,   0(X)
	movq	%mm0,   8(X)
	addl	INCX, X
#else
	movq	%mm0,   0(X)
	addl	INCX, X
#endif

	decl	I
	jg	.L26

.L28:
	EMMS

	xorl	%eax, %eax
	popl	%ebx
	popl	%esi
	ret
	ALIGN_3

.L30:
	movl	N, I
	ALIGN_2

.L32:
	FLD	0 * SIZE(X)
	fmul	%st(1),%st
	FLD	1 * SIZE(X)
	fmul	%st(3),%st
	faddp	%st,%st(1)

	FLD	0 * SIZE(X)
	fmul	%st(3),%st
	FLD	1 * SIZE(X)
	fmul	%st(3),%st
	fsubrp	%st,%st(1)

	FST	0 * SIZE(X)
	FST	1 * SIZE(X)
	addl	INCX, X
	decl	I
	jg	.L32
	ALIGN_2

.L999:
	ffreep	%st
	ffreep	%st

	xorl	%eax,%eax
	popl	%ebx
	popl	%esi
	ret

	EPILOGUE