qgemv_n.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 SP	r12

#define M	r32
#define N	r33
#ifndef XDOUBLE
#define A	r36
#define LDA	r37
#define X	r38
#define INCX	r39
#define Y	r34
#define INCY	r35
#else
#define A	r38
#define LDA	r39
#define X	r34
#define INCX	r35
#define Y	r36
#define INCY	r37
#endif

#define BUFFER	r11

#define I	r14
#define J	r15
#define AO1	r16
#define AO2	r17
#define AO3	r18
#define AO4	r19
#define AO5	r20
#define AO6	r21
#define AO7	r22
#define AO8	r23
#define YLD1	r24
#define YLD2	r25
#define YST1	r26
#define YST2	r27
#define II	r28
#define YY	r29

#define ARLC	r30
#define PR	r31
	
#define LDA7M8	r8
#define PREA	r9
#define PREB	r10

#define ALPHA1	f8
#define ALPHA2	f9
#define ALPHA3	f10
#define ALPHA4	f11
#define ALPHA5	f12
#define ALPHA6	f13
#define ALPHA7	f14
#define ALPHA8	f15

#define RPREFETCHSIZE	( 8 * 1 +  6)
#define WPREFETCHSIZE	( 8 * 1 +  6)

#define RPREFETCH	lfetch.nt1
#define WPREFETCH	lfetch.excl.nt1

#define ALPHA	f6

	PROLOGUE
	.prologue
	PROFCODE
	{ .mmi
	mov	ARLC  = ar.lc
	}
	;;
	mov	PR = pr
	adds	r14 = 16, SP
	adds	r15 = 24, SP
	adds	r16 = 32, SP
	.body
	;;	

#ifdef XDOUBLE
	ld8	X      = [r14], 16
	ld8	INCX   = [r15], 16
	;;
#endif
	ld8	Y      = [r14], 16
	ld8	INCY   = [r15], 16
	;;
	ld8	BUFFER = [r14]
	;;

	mov	ALPHA = f8
	cmp.ge	p7, p0 = 0, M
	cmp.ge	p6, p0 = 0, N
	;;
	shladd	INCX = INCX, BASE_SHIFT, r0
	shladd	LDA  = LDA,  BASE_SHIFT, r0
	shladd	INCY = INCY, BASE_SHIFT, r0
	;;
	(p7) br.cond.dpnt .L999
	(p6) br.cond.dpnt .L999
	;;
	sub	I = A, Y
	mov	YY = Y
	;;
	cmp.eq	p10, p0 = SIZE, INCY
	(p10) br.cond.dptk .L10
	;;
	shr	J = M, 3
	mov	YY = BUFFER
	;;
	(p8) adds  YY = SIZE, BUFFER
	;;
	mov	ar.lc = J
	mov	YST1 = YY
	adds	YST2 = 4 * SIZE, YY
	;;
.L02:
	STFD	[YST1] = f0, 1 * SIZE
	STFD	[YST2] = f0, 1 * SIZE
	;;
	STFD	[YST1] = f0, 1 * SIZE
	STFD	[YST2] = f0, 1 * SIZE
	;;
	STFD	[YST1] = f0, 1 * SIZE
	STFD	[YST2] = f0, 1 * SIZE
	;;
	STFD	[YST1] = f0, 5 * SIZE
	STFD	[YST2] = f0, 5 * SIZE
	br.cloop.sptk.few .L02
	;;

.L10:
	shr	J   = N, 3
	;;
	cmp.eq	p6, p0 = r0, J
	(p6) br.cond.dpnt .L20
	;;
	.align 16

.L11:
	shladd	LDA7M8 = LDA, 3, r0
	;;
	sub	LDA7M8 = LDA, LDA7M8	
	;;
	adds	LDA7M8 = 8 * SIZE, LDA7M8
	;;	
	mov	YLD1 = YY
	mov	YST1 = YY
	adds	YLD2 = 1 * SIZE, YY
	adds	YST2 = 1 * SIZE, YY
	;;
	LDFD	ALPHA1 = [X], INCX
	;;
	LDFD	ALPHA2 = [X], INCX
	;;
	LDFD	ALPHA3 = [X], INCX
	;;
	LDFD	ALPHA4 = [X], INCX
	;;
	LDFD	ALPHA5 = [X], INCX
	;;
	LDFD	ALPHA6 = [X], INCX
	;;
	LDFD	ALPHA7 = [X], INCX
	;;
	LDFD	ALPHA8 = [X], INCX
	;;
	FMPY	ALPHA1 = ALPHA, ALPHA1
	FMPY	ALPHA2 = ALPHA, ALPHA2
	FMPY	ALPHA3 = ALPHA, ALPHA3
	FMPY	ALPHA4 = ALPHA, ALPHA4
	FMPY	ALPHA5 = ALPHA, ALPHA5
	FMPY	ALPHA6 = ALPHA, ALPHA6
	;;
	mov	AO1 = A
	adds	AO2 = 1 * SIZE, A
	adds	AO3 = 2 * SIZE, A
	adds	AO4 = 3 * SIZE, A
	adds	AO5 = 4 * SIZE, A
	adds	AO6 = 5 * SIZE, A
	adds	AO7 = 6 * SIZE, A
	adds	AO8 = 7 * SIZE, A
	shladd	A   = LDA, 3, A
	;;
	shr	I = M, 3
	mov	pr.rot= 0
	;;
	cmp.eq	p16, p0 = r0, r0
	;;
	adds	I = -1, I
	adds J = -1, J
	;;
	adds	PREB = (WPREFETCHSIZE) * SIZE, YY
	;;
	cmp.lt p7, p8 = r0, J
	tbit.nz	p13, p11 = M, 2
	mov	ar.ec= 2
	;;
	FMPY	ALPHA7 = ALPHA, ALPHA7
	;;
	{ .mfi
	and	II = 7, M
	FMPY	ALPHA8 = ALPHA, ALPHA8
	mov	ar.lc = I
	}
	{ .mib
	cmp.eq	p6, p0 = -1, I
	tbit.nz	p14, p12 = M, 1
	(p6) br.cond.dpnt .L15
	}
	;;
	.align 16

.L12:
	{ .mmf
	(p17) LDFD	f93  = [AO5], LDA7M8
	(p17) LDFD	f94  = [AO6], LDA7M8
	(p17) FMA	f101 = ALPHA1,  f33, f101
	}
	{ .mmf
	(p17) LDFD	f95  = [AO7], LDA7M8
	(p17) LDFD	f96  = [AO8], LDA7M8
	(p17) FMA	f104 = ALPHA1,  f34, f104
	}
	;;
	{ .mmf
	(p16) LDFD	f32  = [AO1]
	(p16) LDFD	f33  = [AO2], LDA
	(p17) FMA	f107 = ALPHA1,  f35, f107
	}
	{ .mmf
	(p16) LDFD	f34  = [AO3], LDA
	(p16) LDFD	f35  = [AO4], LDA
	(p17) FMA	f110 = ALPHA1,  f36, f110
	}
	;;
	{ .mmf
	(p16) LDFD	f100  = [YLD1], 2 * SIZE
	(p16) LDFD	f103  = [YLD2], 2 * SIZE
	(p17) FMA	f113 = ALPHA1,  f37, f113
	}
	{ .mmf
	(p16) adds	PREA = (RPREFETCHSIZE) * SIZE, AO1
	(p16) add	AO1 = AO1, LDA
	(p17) FMA	f116 = ALPHA1,  f38, f116
	}
	;;
	{ .mmf
	(p18) STFD	[YST1] = f102, 2 * SIZE
	(p18) STFD	[YST2] = f105, 2 * SIZE
	(p17) FMA	f119 = ALPHA1,  f39, f119
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f122 = ALPHA1,  f40, f122
	}
	;;
	{ .mmf
	(p16) LDFD	f36  = [AO5], LDA
	(p16) LDFD	f37  = [AO6], LDA
	(p17) FMA	f101 = ALPHA2,  f41, f101
	}
	{ .mmf
	(p16) LDFD	f38  = [AO7], LDA
	(p16) LDFD	f39  = [AO8], LDA
	(p17) FMA	f104 = ALPHA2,  f42, f104
	}
	;;
	{ .mmf
	(p16) LDFD	f40  = [AO1], LDA
	(p16) LDFD	f41  = [AO2], LDA
	(p17) FMA	f107 = ALPHA2,  f43, f107
	}
	{ .mmf
	(p16) LDFD	f42  = [AO3], LDA
	(p16) LDFD	f43  = [AO4], LDA
	(p17) FMA	f110 = ALPHA2,  f44, f110
	}
	;;
	{ .mmf
	(p16) LDFD	f106  = [YLD1], 2 * SIZE
	(p16) LDFD	f109  = [YLD2], 2 * SIZE
	(p17) FMA	f113 = ALPHA2,  f45, f113
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f116 = ALPHA2,  f46, f116
	}
	;;
	{ .mmf
	(p18) STFD	[YST1] = f108, 2 * SIZE
	(p18) STFD	[YST2] = f111, 2 * SIZE
	(p17) FMA	f119 = ALPHA2,  f47, f119
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f122 = ALPHA2,  f48, f122
	}
	;;
	{ .mmf
	(p16) LDFD	f44  = [AO5], LDA
	(p16) LDFD	f45  = [AO6], LDA
	(p17) FMA	f101 = ALPHA3,  f49, f101
	}
	{ .mmf
	(p16) LDFD	f46  = [AO7], LDA
	(p16) LDFD	f47  = [AO8], LDA
	(p17) FMA	f104 = ALPHA3,  f50, f104
	}
	;;
	{ .mmf
	(p16) LDFD	f48  = [AO1], LDA
	(p16) LDFD	f49  = [AO2], LDA
	(p17) FMA	f107 = ALPHA3,  f51, f107
	}
	{ .mmf
	(p16) LDFD	f50  = [AO3], LDA
	(p16) LDFD	f51  = [AO4], LDA
	(p17) FMA	f110 = ALPHA3,  f52, f110
	}
	;;
	{ .mmf
	(p16) LDFD	f112 = [YLD1], 2 * SIZE
	(p16) LDFD	f115 = [YLD2], 2 * SIZE
	(p17) FMA	f113 = ALPHA3,  f53, f113
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f116 = ALPHA3,  f54, f116
	}
	;;
	{ .mmf
	(p18) STFD	[YST1] = f114, 2 * SIZE
	(p18) STFD	[YST2] = f117, 2 * SIZE
	(p17) FMA	f119 = ALPHA3,  f55, f119
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f122 = ALPHA3,  f56, f122
	}
	;;
	{ .mmf
	(p16) LDFD	f52  = [AO5], LDA
	(p16) LDFD	f53  = [AO6], LDA
	(p17) FMA	f101 = ALPHA4,  f57, f101
	}
	{ .mmf
	(p16) LDFD	f54  = [AO7], LDA
	(p16) LDFD	f55  = [AO8], LDA
	(p17) FMA	f104 = ALPHA4,  f58, f104
	}
	;;
	{ .mmf
	(p16) LDFD	f56  = [AO1], LDA
	(p16) LDFD	f57  = [AO2], LDA
	(p17) FMA	f107 = ALPHA4,  f59, f107
	}
	{ .mmf
	(p16) LDFD	f58  = [AO3], LDA
	(p16) LDFD	f59  = [AO4], LDA
	(p17) FMA	f110 = ALPHA4,  f60, f110
	}
	;;
	{ .mmf
	(p16) LDFD	f118 = [YLD1], 2 * SIZE
	(p16) LDFD	f121 = [YLD2], 2 * SIZE
	(p17) FMA	f113 = ALPHA4,  f61, f113
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f116 = ALPHA4,  f62, f116
	}
	;;
	{ .mmf
	(p18) STFD	[YST1] = f120, 2 * SIZE
	(p18) STFD	[YST2] = f123, 2 * SIZE
	(p17) FMA	f119 = ALPHA4,  f63, f119
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f122 = ALPHA4,  f64, f122
	}
	;;
	{ .mmf
	(p16) LDFD	f60  = [AO5], LDA
	(p16) LDFD	f61  = [AO6], LDA
	(p17) FMA	f101 = ALPHA5,  f65, f101
	}
	{ .mmf
	(p16) LDFD	f62  = [AO7], LDA
	(p16) LDFD	f63  = [AO8], LDA
	(p17) FMA	f104 = ALPHA5,  f66, f104
	}
	;;
	{ .mmf
	(p16) LDFD	f64  = [AO1], LDA
	(p16) LDFD	f65  = [AO2], LDA
	(p17) FMA	f107 = ALPHA5,  f67, f107
	}
	{ .mmf
	(p16) LDFD	f66  = [AO3], LDA
	(p16) LDFD	f67  = [AO4], LDA
	(p17) FMA	f110 = ALPHA5,  f68, f110
	}
	;;
	{ .mmf
	(p16) WPREFETCH [PREB], 8 * SIZE
	nop   __LINE__
	(p17) FMA	f113 = ALPHA5,  f69, f113
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f116 = ALPHA5,  f70, f116
	}
	;;
	{ .mmf
	(p16) RPREFETCH [PREA]
	nop   __LINE__
	(p17) FMA	f119 = ALPHA5,  f71, f119
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f122 = ALPHA5,  f72, f122
	}
	;;
	{ .mmf
	(p16) LDFD	f68  = [AO5], LDA
	(p16) LDFD	f69  = [AO6], LDA
	(p17) FMA	f101 = ALPHA6,  f73, f101
	}
	{ .mmf
	(p16) LDFD	f70  = [AO7], LDA
	(p16) LDFD	f71  = [AO8], LDA
	(p17) FMA	f104 = ALPHA6,  f74, f104
	}
	;;
	{ .mmf
	(p16) LDFD	f72  = [AO1], LDA
	(p16) LDFD	f73  = [AO2], LDA
	(p17) FMA	f107 = ALPHA6,  f75, f107
	}
	{ .mmf
	(p16) LDFD	f74  = [AO3], LDA
	(p16) LDFD	f75  = [AO4], LDA
	(p17) FMA	f110 = ALPHA6,  f76, f110
	}
	;;
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f113 = ALPHA6,  f77, f113
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f116 = ALPHA6,  f78, f116
	}
	;;
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f119 = ALPHA6,  f79, f119
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f122 = ALPHA6,  f80, f122
	}
	;;
	{ .mmf
	(p16) LDFD	f76  = [AO5], LDA
	(p16) LDFD	f77  = [AO6], LDA
	(p17) FMA	f101 = ALPHA7,  f81, f101
	}
	{ .mmf
	(p16) LDFD	f78  = [AO7], LDA
	(p16) LDFD	f79  = [AO8], LDA
	(p17) FMA	f104 = ALPHA7,  f82, f104
	}
	;;
	{ .mmf
	(p16) LDFD	f80  = [AO1], LDA
	(p16) LDFD	f81  = [AO2], LDA
	(p17) FMA	f107 = ALPHA7,  f83, f107
	}
	{ .mmf
	(p16) LDFD	f82  = [AO3], LDA
	(p16) LDFD	f83  = [AO4], LDA
	(p17) FMA	f110 = ALPHA7,  f84, f110
	}
	;;
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f113 = ALPHA7,  f85, f113
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f116 = ALPHA7,  f86, f116
	}
	;;
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f119 = ALPHA7,  f87, f119
	}
	{ .mmf
	nop   __LINE__
	nop   __LINE__
	(p17) FMA	f122 = ALPHA7,  f88, f122
	}
	;; 
	{ .mmf
	(p16) LDFD	f84  = [AO5], LDA
	(p16) LDFD	f85  = [AO6], LDA
	(p17) FMA	f101 = ALPHA8,  f89, f101
	}
	{ .mmf
	(p16) LDFD	f86  = [AO7], LDA
	(p16) LDFD	f87  = [AO8], LDA
	(p17) FMA	f104 = ALPHA8,  f90, f104
	}