zgemm_beta.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	12
#define ARGS	 0
	
#define M	 4 + STACK + ARGS(%esp)
#define N	 8 + STACK + ARGS(%esp)
#ifdef DOUBLE
#define BETA_R	16 + STACK + ARGS(%esp)
#define BETA_I	24 + STACK + ARGS(%esp)
#define C	48 + STACK + ARGS(%esp)
#define LDC	52 + STACK + ARGS(%esp)
#else
#define BETA_R	16 + STACK + ARGS(%esp)
#define BETA_I	20 + STACK + ARGS(%esp)
#define C	40 + STACK + ARGS(%esp)
#define LDC	44 + STACK + ARGS(%esp)
#endif

	PROLOGUE

	pushl	%ebp
	pushl	%edi
	pushl	%esi

	PROFCODE

	movl	M,   %ebp
	movl	N,   %ecx
	movl	LDC, %edx
	movl	C,   %edi

	FLD	BETA_R
	FLD	BETA_I

	testl	%ebp, %ebp		# if n <= 0 goto End
	jle	.L83
	testl	%ecx, %ecx		# if m <= 0 goto End
	jle	.L83

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

	sall	$ZBASE_SHIFT,  %edx

	ftst
	fnstsw	%ax
	andb	$68, %ah
	ffreep	%st(0)
	je	.L71
	ALIGN_2

.L53:
	movl	%edi, %esi		# c_offset1 = c_offset
	addl	%edx, %edi		# c_offset += ldc

	movl	%ebp, %eax
	sarl	$2,   %eax
	jle	.L56
	ALIGN_2

.L57:
#if defined(HAS_PREFETCH) && defined(PENTIUM3)
	prefetchnta	16 * SIZE(%esi)
	prefetchnta	24 * SIZE(%esi)
#endif

	FSTU	0 * SIZE(%esi)		# c_offset1
	FSTU	1 * SIZE(%esi)
	FSTU	2 * SIZE(%esi)
	FSTU	3 * SIZE(%esi)
	FSTU	4 * SIZE(%esi)
	FSTU	5 * SIZE(%esi)
	FSTU	6 * SIZE(%esi)
	FSTU	7 * SIZE(%esi)
	addl	$8 * SIZE, %esi		# c_offset1 += 8
	decl	%eax			# i--
	jg	.L57
	ALIGN_2

.L56:
	movl	%ebp, %eax
	andl	$3,   %eax
	jle	.L62
	ALIGN_2

.L63:
	FSTU	0 * SIZE(%esi)
	FSTU	1 * SIZE(%esi)
	addl	$2 * SIZE,%esi
	decl	%eax
	jg	.L63
	ALIGN_2

.L62:
	decl	%ecx			# j --
	jg	.L53
	jmp	.L83
	ALIGN_3

.L71:
	movl	%edi, %esi
	addl	%edx, %edi		# c_offset += ldc


	movl	%ebp, %eax
	sarl	$1,   %eax
	jle	.L84
	ALIGN_3

.L85:
#if defined(HAS_PREFETCH) && defined(PENTIUM3)
	prefetchnta	16 * SIZE(%esi)
#endif
	fld	%st(0)
	FMUL	0 * SIZE(%esi)
	fld	%st(2)
	FMUL	1 * SIZE(%esi)
	faddp	%st,%st(1)
	fld	%st(2)
	FMUL	0 * SIZE(%esi)
	fld	%st(2)
	FMUL	1 * SIZE(%esi)
	fsubrp	%st,%st(1)

	FST	0 * SIZE(%esi)
	FST	1 * SIZE(%esi)

	fld	%st(0)
	FMUL	2 * SIZE(%esi)
	fld	%st(2)
	FMUL	3 * SIZE(%esi)
	faddp	%st,%st(1)
	fld	%st(2)
	FMUL	2 * SIZE(%esi)
	fld	%st(2)
	FMUL	3 * SIZE(%esi)
	fsubrp	%st,%st(1)

	FST	2 * SIZE(%esi)
	FST	3 * SIZE(%esi)
	addl	$4 * SIZE, %esi

	decl	%eax
	jg	.L85
	ALIGN_3

.L84:
	movl	%ebp, %eax
	andl	$1,   %eax
	jle	.L74
	ALIGN_3

.L75:
#if defined(HAS_PREFETCH) && defined(PENTIUM3)
	prefetchnta	16 * SIZE(%esi)
#endif

	fld	%st(0)
	FMUL	0 * SIZE(%esi)
	fld	%st(2)
	FMUL	1 * SIZE(%esi)
	faddp	%st,%st(1)
	fld	%st(2)
	FMUL	0 * SIZE(%esi)
	fld	%st(2)
	FMUL	1 * SIZE(%esi)
	fsubrp	%st,%st(1)

	FST	0 * SIZE(%esi)
	FST	1 * SIZE(%esi)
	ALIGN_2

.L74:
	decl	%ecx
	jg	.L71
	ALIGN_2

.L83:
	ffreep	%st
	ffreep	%st

	popl	%esi
	popl	%edi
	popl	%ebp
	ret

	EPILOGUE