head	1.1;
branch	1.1.1;
access;
symbols
	netbsd-11-0-RC4:1.1.1.2
	netbsd-11-0-RC3:1.1.1.2
	netbsd-11-0-RC2:1.1.1.2
	netbsd-11-0-RC1:1.1.1.2
	perseant-exfatfs-base-20250801:1.1.1.2
	netbsd-11:1.1.1.2.0.18
	netbsd-11-base:1.1.1.2
	netbsd-10-1-RELEASE:1.1.1.2
	perseant-exfatfs-base-20240630:1.1.1.2
	perseant-exfatfs:1.1.1.2.0.16
	perseant-exfatfs-base:1.1.1.2
	netbsd-8-3-RELEASE:1.1.1.1
	netbsd-9-4-RELEASE:1.1.1.2
	netbsd-10-0-RELEASE:1.1.1.2
	netbsd-10-0-RC6:1.1.1.2
	netbsd-10-0-RC5:1.1.1.2
	netbsd-10-0-RC4:1.1.1.2
	netbsd-10-0-RC3:1.1.1.2
	netbsd-10-0-RC2:1.1.1.2
	netbsd-10-0-RC1:1.1.1.2
	netbsd-10:1.1.1.2.0.14
	netbsd-10-base:1.1.1.2
	netbsd-9-3-RELEASE:1.1.1.2
	gmp-6-2-1:1.1.1.2
	cjep_sun2x-base1:1.1.1.2
	cjep_sun2x:1.1.1.2.0.12
	cjep_sun2x-base:1.1.1.2
	cjep_staticlib_x-base1:1.1.1.2
	netbsd-9-2-RELEASE:1.1.1.2
	cjep_staticlib_x:1.1.1.2.0.10
	cjep_staticlib_x-base:1.1.1.2
	netbsd-9-1-RELEASE:1.1.1.2
	gmp-6-2-0:1.1.1.2
	phil-wifi-20200421:1.1.1.2
	phil-wifi-20200411:1.1.1.2
	is-mlppp:1.1.1.2.0.8
	is-mlppp-base:1.1.1.2
	phil-wifi-20200406:1.1.1.2
	netbsd-8-2-RELEASE:1.1.1.1
	netbsd-9-0-RELEASE:1.1.1.2
	netbsd-9-0-RC2:1.1.1.2
	netbsd-9-0-RC1:1.1.1.2
	phil-wifi-20191119:1.1.1.2
	netbsd-9:1.1.1.2.0.6
	netbsd-9-base:1.1.1.2
	phil-wifi-20190609:1.1.1.2
	netbsd-8-1-RELEASE:1.1.1.1
	netbsd-8-1-RC1:1.1.1.1
	pgoyette-compat-merge-20190127:1.1.1.2
	pgoyette-compat-20190127:1.1.1.2
	pgoyette-compat-20190118:1.1.1.2
	pgoyette-compat-1226:1.1.1.2
	pgoyette-compat-1126:1.1.1.2
	pgoyette-compat-1020:1.1.1.2
	pgoyette-compat-0930:1.1.1.2
	pgoyette-compat-0906:1.1.1.2
	netbsd-7-2-RELEASE:1.1.1.1
	pgoyette-compat-0728:1.1.1.2
	netbsd-8-0-RELEASE:1.1.1.1
	phil-wifi:1.1.1.2.0.4
	phil-wifi-base:1.1.1.2
	pgoyette-compat-0625:1.1.1.2
	netbsd-8-0-RC2:1.1.1.1
	pgoyette-compat-0521:1.1.1.2
	pgoyette-compat-0502:1.1.1.2
	pgoyette-compat-0422:1.1.1.2
	netbsd-8-0-RC1:1.1.1.1
	pgoyette-compat-0415:1.1.1.2
	pgoyette-compat-0407:1.1.1.2
	pgoyette-compat-0330:1.1.1.2
	pgoyette-compat-0322:1.1.1.2
	pgoyette-compat-0315:1.1.1.2
	netbsd-7-1-2-RELEASE:1.1.1.1
	pgoyette-compat:1.1.1.2.0.2
	pgoyette-compat-base:1.1.1.2
	netbsd-7-1-1-RELEASE:1.1.1.1
	matt-nb8-mediatek:1.1.1.1.0.26
	matt-nb8-mediatek-base:1.1.1.1
	gmp-6-1-2:1.1.1.2
	perseant-stdc-iso10646:1.1.1.1.0.24
	perseant-stdc-iso10646-base:1.1.1.1
	netbsd-8:1.1.1.1.0.22
	netbsd-8-base:1.1.1.1
	prg-localcount2-base3:1.1.1.1
	prg-localcount2-base2:1.1.1.1
	prg-localcount2-base1:1.1.1.1
	prg-localcount2:1.1.1.1.0.20
	prg-localcount2-base:1.1.1.1
	pgoyette-localcount-20170426:1.1.1.1
	bouyer-socketcan-base1:1.1.1.1
	pgoyette-localcount-20170320:1.1.1.1
	netbsd-7-1:1.1.1.1.0.18
	netbsd-7-1-RELEASE:1.1.1.1
	netbsd-7-1-RC2:1.1.1.1
	netbsd-7-nhusb-base-20170116:1.1.1.1
	bouyer-socketcan:1.1.1.1.0.16
	bouyer-socketcan-base:1.1.1.1
	pgoyette-localcount-20170107:1.1.1.1
	netbsd-7-1-RC1:1.1.1.1
	pgoyette-localcount-20161104:1.1.1.1
	netbsd-7-0-2-RELEASE:1.1.1.1
	localcount-20160914:1.1.1.1
	netbsd-7-nhusb:1.1.1.1.0.14
	netbsd-7-nhusb-base:1.1.1.1
	pgoyette-localcount-20160806:1.1.1.1
	pgoyette-localcount-20160726:1.1.1.1
	pgoyette-localcount:1.1.1.1.0.12
	pgoyette-localcount-base:1.1.1.1
	netbsd-7-0-1-RELEASE:1.1.1.1
	netbsd-7-0:1.1.1.1.0.10
	netbsd-7-0-RELEASE:1.1.1.1
	netbsd-7-0-RC3:1.1.1.1
	netbsd-7-0-RC2:1.1.1.1
	netbsd-7-0-RC1:1.1.1.1
	tls-maxphys-base:1.1.1.1
	tls-maxphys:1.1.1.1.0.8
	netbsd-7:1.1.1.1.0.6
	netbsd-7-base:1.1.1.1
	yamt-pagecache:1.1.1.1.0.4
	yamt-pagecache-base9:1.1.1.1
	tls-earlyentropy:1.1.1.1.0.2
	tls-earlyentropy-base:1.1.1.1
	riastradh-xf86-video-intel-2-7-1-pre-2-21-15:1.1.1.1
	riastradh-drm2-base3:1.1.1.1
	gmp-5-1-3:1.1.1.1
	gmp:1.1.1;
locks; strict;
comment	@;; @;


1.1
date	2013.11.29.07.49.49;	author mrg;	state Exp;
branches
	1.1.1.1;
next	;
commitid	L2Av4PuGmdoL39fx;

1.1.1.1
date	2013.11.29.07.49.49;	author mrg;	state Exp;
branches
	1.1.1.1.4.1
	1.1.1.1.8.1;
next	1.1.1.2;
commitid	L2Av4PuGmdoL39fx;

1.1.1.2
date	2017.08.22.09.40.49;	author mrg;	state Exp;
branches;
next	;
commitid	W5kmAIk8hwVpSb4A;

1.1.1.1.4.1
date	2013.11.29.07.49.49;	author yamt;	state dead;
branches;
next	1.1.1.1.4.2;
commitid	nx2BSsHy0NPeAxBx;

1.1.1.1.4.2
date	2014.05.22.14.09.04;	author yamt;	state Exp;
branches;
next	;
commitid	nx2BSsHy0NPeAxBx;

1.1.1.1.8.1
date	2013.11.29.07.49.49;	author tls;	state dead;
branches;
next	1.1.1.1.8.2;
commitid	jTnpym9Qu0o4R1Nx;

1.1.1.1.8.2
date	2014.08.19.23.59.53;	author tls;	state Exp;
branches;
next	;
commitid	jTnpym9Qu0o4R1Nx;


desc
@@


1.1
log
@Initial revision
@
text
@dnl  SPARC v9 64-bit mpn_addmul_2 -- Multiply an n limb number with 2-limb
dnl  number and add the result to a n limb vector.

dnl  Copyright 2002, 2003 Free Software Foundation, Inc.

dnl  This file is part of the GNU MP Library.

dnl  The GNU MP Library is free software; you can redistribute it and/or modify
dnl  it under the terms of the GNU Lesser General Public License as published
dnl  by the Free Software Foundation; either version 3 of the License, or (at
dnl  your option) any later version.

dnl  The GNU MP Library is distributed in the hope that it will be useful, but
dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
dnl  License for more details.

dnl  You should have received a copy of the GNU Lesser General Public License
dnl  along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.

include(`../config.m4')

C                  cycles/limb
C UltraSPARC 1&2:      9
C UltraSPARC 3:       10

C Algorithm: We use 16 floating-point multiplies per limb product, with the
C 2-limb v operand split into eight 16-bit pieces, and the n-limb u operand
C split into 32-bit pieces.  We sum four 48-bit partial products using
C floating-point add, then convert the resulting four 50-bit quantities and
C transfer them to the integer unit.

C Possible optimizations:
C   1. Align the stack area where we transfer the four 50-bit product-sums
C      to a 32-byte boundary.  That would minimize the cache collision.
C      (UltraSPARC-1/2 use a direct-mapped cache.)  (Perhaps even better would
C      be to align the area to map to the area immediately before up?)
C   2. Perform two of the fp->int conversions with integer instructions.  We
C      can get almost ten free IEU slots, if we clean up bookkeeping and the
C      silly carry-limb code.
C   3. For an mpn_addmul_1 based on this, we need to fix the silly carry-limb
C      code.

C OSP (Overlapping software pipeline) version of mpn_mul_basecase:
C Operand swap will require 8 LDDA and 8 FXTOD, which will mean 8 cycles.
C FI	= 20
C L	=  9 x un * vn
C WDFI	= 10 x vn / 2
C WD	= 4

C Instruction classification (as per UltraSPARC functional units).
C Assuming silly carry code is fixed.  Includes bookkeeping.
C
C               mpn_addmul_X     mpn_mul_X
C                1       2       1       2
C               ==========      ==========
C      FM        8      16       8      16
C      FA       10      18      10      18
C     MEM       12      12      10      10
C  ISHIFT        6       6       6       6
C IADDLOG       11      11      10      10
C  BRANCH        1       1       1       1
C
C TOTAL IEU     17      17      16      16
C TOTAL         48      64      45      61
C
C IEU cycles     8.5     8.5     8       8
C MEM cycles    12      12      10      10
C ISSUE cycles  12      16      11.25   15.25
C FPU cycles    10      18      10      18
C cycles/loop   12      18      12      18
C cycles/limb   12       9      12       9


C INPUT PARAMETERS
C rp[n + 1]	i0
C up[n]		i1
C n		i2
C vp[2]		i3


ASM_START()
	REGISTER(%g2,#scratch)
	REGISTER(%g3,#scratch)

C Combine registers:
C u00_hi= u32_hi
C u00_lo= u32_lo
C a000  = out000
C a016  = out016
C Free: f52 f54


define(`p000', `%f8')  define(`p016',`%f10')
define(`p032',`%f12')  define(`p048',`%f14')
define(`p064',`%f16')  define(`p080',`%f18')
define(`p096a',`%f20') define(`p112a',`%f22')
define(`p096b',`%f56') define(`p112b',`%f58')

define(`out000',`%f0') define(`out016',`%f6')

define(`v000',`%f24')  define(`v016',`%f26')
define(`v032',`%f28')  define(`v048',`%f30')
define(`v064',`%f44')  define(`v080',`%f46')
define(`v096',`%f48')  define(`v112',`%f50')

define(`u00',`%f32')   define(`u32', `%f34')

define(`a000',`%f36')  define(`a016',`%f38')
define(`a032',`%f40')  define(`a048',`%f42')
define(`a064',`%f60')  define(`a080',`%f62')

define(`u00_hi',`%f2') define(`u32_hi',`%f4')
define(`u00_lo',`%f3') define(`u32_lo',`%f5')

define(`cy',`%g1')
define(`rlimb',`%g3')
define(`i00',`%l0')    define(`i16',`%l1')
define(`r00',`%l2')    define(`r32',`%l3')
define(`xffffffff',`%l7')
define(`xffff',`%o0')


PROLOGUE(mpn_addmul_2)

C Initialization.  (1) Split v operand into eight 16-bit chunks and store them
C as IEEE double in fp registers.  (2) Clear upper 32 bits of fp register pairs
C f2 and f4.  (3) Store masks in registers aliased to `xffff' and `xffffffff'.
C This code could be better scheduled.

	save	%sp, -256, %sp

ifdef(`HAVE_VIS',
`	mov	-1, %g4
	wr	%g0, 0xD2, %asi
	srlx	%g4, 32, xffffffff	C store mask in register `xffffffff'
	ldda	[%i3+6] %asi, v000
	ldda	[%i3+4] %asi, v016
	ldda	[%i3+2] %asi, v032
	ldda	[%i3+0] %asi, v048
	fxtod	v000, v000
	ldda	[%i3+14] %asi, v064
	fxtod	v016, v016
	ldda	[%i3+12] %asi, v080
	fxtod	v032, v032
	ldda	[%i3+10] %asi, v096
	fxtod	v048, v048
	ldda	[%i3+8] %asi, v112
	fxtod	v064, v064
	fxtod	v080, v080
	fxtod	v096, v096
	fxtod	v112, v112
	fzero	u00_hi
	fzero	u32_hi
',
`	mov	-1, %g4
	ldx	[%i3+0], %l0		C vp[0]
	srlx	%g4, 48, xffff		C store mask in register `xffff'
	ldx	[%i3+8], %l1		C vp[1]

	and	%l0, xffff, %g2
	stx	%g2, [%sp+2223+0]
	srlx	%l0, 16, %g3
	and	%g3, xffff, %g3
	stx	%g3, [%sp+2223+8]
	srlx	%l0, 32, %g2
	and	%g2, xffff, %g2
	stx	%g2, [%sp+2223+16]
	srlx	%l0, 48, %g3
	stx	%g3, [%sp+2223+24]
	and	%l1, xffff, %g2
	stx	%g2, [%sp+2223+32]
	srlx	%l1, 16, %g3
	and	%g3, xffff, %g3
	stx	%g3, [%sp+2223+40]
	srlx	%l1, 32, %g2
	and	%g2, xffff, %g2
	stx	%g2, [%sp+2223+48]
	srlx	%l1, 48, %g3
	stx	%g3, [%sp+2223+56]

	srlx	%g4, 32, xffffffff	C store mask in register `xffffffff'

	ldd	[%sp+2223+0], v000
	ldd	[%sp+2223+8], v016
	ldd	[%sp+2223+16], v032
	ldd	[%sp+2223+24], v048
	fxtod	v000, v000
	ldd	[%sp+2223+32], v064
	fxtod	v016, v016
	ldd	[%sp+2223+40], v080
	fxtod	v032, v032
	ldd	[%sp+2223+48], v096
	fxtod	v048, v048
	ldd	[%sp+2223+56], v112
	fxtod	v064, v064
	ld	[%sp+2223+0], u00_hi	C zero u00_hi
	fxtod	v080, v080
	ld	[%sp+2223+0], u32_hi	C zero u32_hi
	fxtod	v096, v096
	fxtod	v112, v112
')
C Initialization done.
	mov	0, %g2
	mov	0, rlimb
	mov	0, %g4
	add	%i0, -8, %i0		C BOOKKEEPING

C Start software pipeline.

	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	fxtod	u00_hi, u00
C mid
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	fmuld	u00, v000, a000
	fmuld	u00, v016, a016
	fmuld	u00, v032, a032
	fmuld	u00, v048, a048
	add	%i2, -1, %i2		C BOOKKEEPING
	fmuld	u00, v064, p064
	add	%i1, 8, %i1		C BOOKKEEPING
	fxtod	u32_hi, u32
	fmuld	u00, v080, p080
	fmuld	u00, v096, p096a
	brnz,pt	%i2, .L_2_or_more
	 fmuld	u00, v112, p112a

.L1:	fdtox	a000, out000
	fmuld	u32, v000, p000
	fdtox	a016, out016
	fmuld	u32, v016, p016
	fmovd	p064, a064
	fmuld	u32, v032, p032
	fmovd	p080, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	fdtox	a000, out000
	fdtox	a016, out016
	faddd	p064, p096a, a064
	faddd	p080, p112a, a080
	std	out000, [%sp+2223+0]
	b	.L_wd2
	 std	out016, [%sp+2223+8]

.L_2_or_more:
	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	fdtox	a000, out000
	fmuld	u32, v000, p000
	fdtox	a016, out016
	fmuld	u32, v016, p016
	fmovd	p064, a064
	fmuld	u32, v032, p032
	fmovd	p080, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	fdtox	a000, out000
	fmuld	u00, v000, p000
	fdtox	a016, out016
	fmuld	u00, v016, p016
	faddd	p064, p096a, a064
	fmuld	u00, v032, p032
	faddd	p080, p112a, a080
	fmuld	u00, v048, p048
	add	%i2, -1, %i2		C BOOKKEEPING
	std	out000, [%sp+2223+0]
	faddd	p000, a032, a000
	fmuld	u00, v064, p064
	add	%i1, 8, %i1		C BOOKKEEPING
	std	out016, [%sp+2223+8]
	fxtod	u32_hi, u32
	faddd	p016, a048, a016
	fmuld	u00, v080, p080
	faddd	p032, a064, a032
	fmuld	u00, v096, p096a
	faddd	p048, a080, a048
	brnz,pt	%i2, .L_3_or_more
	 fmuld	u00, v112, p112a

	b	.Lend
	 nop

C  64      32       0
C   .       .       .
C   .       |__rXXX_|	32
C   .      |___cy___|	34
C   .  |_______i00__|	50
C  |_______i16__|   .	50


C BEGIN MAIN LOOP
	.align	16
.L_3_or_more:
.Loop:	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u32, v000, p000
C
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u32, v016, p016
C
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	faddd	p064, p096b, a064
	fmuld	u32, v032, p032
C
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	faddd	p080, p112b, a080
	fmuld	u32, v048, p048
C
	nop
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
C
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
C
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
C
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
C
	stw	%l5, [%i0+4]
	nop
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C midloop
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u00, v000, p000
C
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u00, v016, p016
C
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	faddd	p064, p096a, a064
	fmuld	u00, v032, p032
C
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	faddd	p080, p112a, a080
	fmuld	u00, v048, p048
C
	add	%i2, -1, %i2		C BOOKKEEPING
	std	out000, [%sp+2223+0]
	faddd	p000, a032, a000
	fmuld	u00, v064, p064
C
	add	i00, r32, rlimb
	add	%i1, 8, %i1		C BOOKKEEPING
	std	out016, [%sp+2223+8]
	fxtod	u32_hi, u32
C
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u00, v080, p080
C
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u00, v096, p096a
C
	stw	%l5, [%i0+0]
	faddd	p048, a080, a048
	brnz,pt	%i2, .Loop
	 fmuld	u00, v112, p112a
C END MAIN LOOP

C WIND-DOWN PHASE 1
.Lend:	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u32, v000, p000
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u32, v016, p016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	faddd	p064, p096b, a064
	fmuld	u32, v032, p032
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	faddd	p080, p112b, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	stw	%l5, [%i0+4]
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	faddd	p064, p096a, a064
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	faddd	p080, p112a, a080
	std	out000, [%sp+2223+0]
	add	i00, r32, rlimb
	std	out016, [%sp+2223+8]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

C WIND-DOWN PHASE 2
.L_wd2:	and	%g2, xffffffff, %g2
	fdtox	a032, out000
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a048, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	std	out000, [%sp+2223+16]
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+4]
C mid
	and	%g2, xffffffff, %g2
	fdtox	a064, out000
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a080, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	std	out000, [%sp+2223+0]
	add	i00, r32, rlimb
	std	out016, [%sp+2223+8]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

C WIND-DOWN PHASE 3
.L_wd3:	and	%g2, xffffffff, %g2
	fdtox	p096b, out000
	add	%g2, rlimb, %l5
	fdtox	p112b, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], rlimb
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	std	out000, [%sp+2223+16]
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+4]
C mid
	and	%g2, xffffffff, %g2
	add	%g2, rlimb, %l5
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], rlimb
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

	and	%g2, xffffffff, %g2
	add	%g2, rlimb, %l5
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16

	sllx	i16, 16, %g2
	add	i00, cy, cy
	return	%i7+8
	add	%g2, cy, %o0
EPILOGUE(mpn_addmul_2)
@


1.1.1.1
log
@initial import GMP 5.1.3 sources.  changes include:

fixes for:
- mpn_sbpi1_div_qr_sec and mpn_sbpi1_div_r_sec
- mpz_powm_ui
- AMD family 11h
- mpz_powm_sec and mpn_powm_sec
- ASSERT() fixes
- gcd, gcdext, and invert function fixes
- some PPC division operations
@
text
@@


1.1.1.2
log
@initial import of GMP 6.1.2.  main changes from 5.1.3 below.

notes:
 - support for thumb-less ARM chips was in our port of 5.1.3, but a
   similar method has been provided upstream now
 - someone should look at the AVX failure reports, and fix them

Changes between GMP version 6.1.0 and 6.1.1

  FEATURES
  * Work around faulty cpuid on some recent Intel chips (this allows GMP to run
    on Skylake Pentiums).
  * Support thumb-less ARM chips.

Changes between GMP version 6.0.* and 6.1.0

  BUGS FIXED
  * The public function mpn_com is now correctly declared in gmp.h.
  * Healed possible failures of mpn_sec_sqr for non-cryptographic sizes for
    some obsolete CPUs.
  * Various problems related to precision for mpf have been fixed.
  * Fixed ABI incompatible stack alignment in calls from assembly code.
  * Fixed PIC bug in popcount affecting Intel processors using the 32-bit ABI.
  SPEEDUPS
  * Speedup for Intel Broadwell and Skylake through assembly code making use of
    new ADX instructions.
  * Square root is now faster when the remainder is not needed. Also the speed
    to compute the k-th root improved, for small sizes.
  FEATURES
  * New C++ functions gcd and lcm for mpz_class.
  * New public mpn functions mpn_divexact_1, mpn_zero_p, and mpn_cnd_swap.
  * New public mpq_cmp_z function, to efficiently compare rationals with
    integers.
  * Support for more 32-bit arm processors.
  * Support for AVX-less modern x86 CPUs. (Such support might be missing either
    because the CPU vendor chose to disable AVX, or because the running kernel
    lacks AVX context switch support.)
  * Support for NetBSD under Xen; we switch off AVX unconditionally under
    NetBSD since a bug in NetBSD makes AVX fail under Xen.
  MISC
  * Tuned values for FFT multiplications are provided for larger number on
    many platforms.

Changes between GMP version 5.1.* and 6.0.0
  BUGS FIXED
  * The function mpz_invert now considers any number invertible in Z/1Z.
  * The mpn multiply code now handles operands of more than 2^31 limbs
    correctly.  (Note however that the mpz code is limited to 2^32 bits on
    32-bit hosts and 2^37 bits on 64-bit hosts.)
  SPEEDUPS
  * Plain division of large operands is faster and more monotonous in operand
    size.
  * Major speedup for ARM, in particular ARM Cortex-A15, thanks to improved
    assembly.
  * Speedup for Intel Sandy Bridge, Ivy Bridge, Haswell, thanks to rewritten
    and vastly expanded assembly support.  Speedup also for the older Core 2
    and Nehalem.
  * Faster mixed arithmetic between mpq_class and double.
  FEATURES
  * Support for new Intel and AMD CPUs.
  * New public functions mpn_sec_mul and mpn_sec_sqr, implementing side-channel
    silent multiplication and squaring.
  * New public functions mpn_sec_div_qr and mpn_sec_div_r, implementing
    side-channel silent division.
  * New public functions mpn_cnd_add_n and mpn_cnd_sub_n.  Side-channel silent
    conditional addition and subtraction.
  * New public function mpn_sec_powm, implementing side-channel silent modexp.
  * New public function mpn_sec_invert, implementing side-channel silent
    modular inversion.
  * Better support for applications which use the mpz_t type, but nevertheless
    need to call some of the lower-level mpn functions.  See the documentation
    for mpz_limbs_read and related functions.
@
text
@d7 1
a7 1
dnl
d9 4
a12 14
dnl  it under the terms of either:
dnl
dnl    * the GNU Lesser General Public License as published by the Free
dnl      Software Foundation; either version 3 of the License, or (at your
dnl      option) any later version.
dnl
dnl  or
dnl
dnl    * the GNU General Public License as published by the Free Software
dnl      Foundation; either version 2 of the License, or (at your option) any
dnl      later version.
dnl
dnl  or both in parallel, as here.
dnl
d15 5
a19 6
dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
dnl  for more details.
dnl
dnl  You should have received copies of the GNU General Public License and the
dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
dnl  see https://www.gnu.org/licenses/.
@


1.1.1.1.8.1
log
@file addmul_2.asm was added on branch tls-maxphys on 2014-08-19 23:59:53 +0000
@
text
@d1 540
@


1.1.1.1.8.2
log
@Rebase to HEAD as of a few days ago.
@
text
@a0 540
dnl  SPARC v9 64-bit mpn_addmul_2 -- Multiply an n limb number with 2-limb
dnl  number and add the result to a n limb vector.

dnl  Copyright 2002, 2003 Free Software Foundation, Inc.

dnl  This file is part of the GNU MP Library.

dnl  The GNU MP Library is free software; you can redistribute it and/or modify
dnl  it under the terms of the GNU Lesser General Public License as published
dnl  by the Free Software Foundation; either version 3 of the License, or (at
dnl  your option) any later version.

dnl  The GNU MP Library is distributed in the hope that it will be useful, but
dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
dnl  License for more details.

dnl  You should have received a copy of the GNU Lesser General Public License
dnl  along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.

include(`../config.m4')

C                  cycles/limb
C UltraSPARC 1&2:      9
C UltraSPARC 3:       10

C Algorithm: We use 16 floating-point multiplies per limb product, with the
C 2-limb v operand split into eight 16-bit pieces, and the n-limb u operand
C split into 32-bit pieces.  We sum four 48-bit partial products using
C floating-point add, then convert the resulting four 50-bit quantities and
C transfer them to the integer unit.

C Possible optimizations:
C   1. Align the stack area where we transfer the four 50-bit product-sums
C      to a 32-byte boundary.  That would minimize the cache collision.
C      (UltraSPARC-1/2 use a direct-mapped cache.)  (Perhaps even better would
C      be to align the area to map to the area immediately before up?)
C   2. Perform two of the fp->int conversions with integer instructions.  We
C      can get almost ten free IEU slots, if we clean up bookkeeping and the
C      silly carry-limb code.
C   3. For an mpn_addmul_1 based on this, we need to fix the silly carry-limb
C      code.

C OSP (Overlapping software pipeline) version of mpn_mul_basecase:
C Operand swap will require 8 LDDA and 8 FXTOD, which will mean 8 cycles.
C FI	= 20
C L	=  9 x un * vn
C WDFI	= 10 x vn / 2
C WD	= 4

C Instruction classification (as per UltraSPARC functional units).
C Assuming silly carry code is fixed.  Includes bookkeeping.
C
C               mpn_addmul_X     mpn_mul_X
C                1       2       1       2
C               ==========      ==========
C      FM        8      16       8      16
C      FA       10      18      10      18
C     MEM       12      12      10      10
C  ISHIFT        6       6       6       6
C IADDLOG       11      11      10      10
C  BRANCH        1       1       1       1
C
C TOTAL IEU     17      17      16      16
C TOTAL         48      64      45      61
C
C IEU cycles     8.5     8.5     8       8
C MEM cycles    12      12      10      10
C ISSUE cycles  12      16      11.25   15.25
C FPU cycles    10      18      10      18
C cycles/loop   12      18      12      18
C cycles/limb   12       9      12       9


C INPUT PARAMETERS
C rp[n + 1]	i0
C up[n]		i1
C n		i2
C vp[2]		i3


ASM_START()
	REGISTER(%g2,#scratch)
	REGISTER(%g3,#scratch)

C Combine registers:
C u00_hi= u32_hi
C u00_lo= u32_lo
C a000  = out000
C a016  = out016
C Free: f52 f54


define(`p000', `%f8')  define(`p016',`%f10')
define(`p032',`%f12')  define(`p048',`%f14')
define(`p064',`%f16')  define(`p080',`%f18')
define(`p096a',`%f20') define(`p112a',`%f22')
define(`p096b',`%f56') define(`p112b',`%f58')

define(`out000',`%f0') define(`out016',`%f6')

define(`v000',`%f24')  define(`v016',`%f26')
define(`v032',`%f28')  define(`v048',`%f30')
define(`v064',`%f44')  define(`v080',`%f46')
define(`v096',`%f48')  define(`v112',`%f50')

define(`u00',`%f32')   define(`u32', `%f34')

define(`a000',`%f36')  define(`a016',`%f38')
define(`a032',`%f40')  define(`a048',`%f42')
define(`a064',`%f60')  define(`a080',`%f62')

define(`u00_hi',`%f2') define(`u32_hi',`%f4')
define(`u00_lo',`%f3') define(`u32_lo',`%f5')

define(`cy',`%g1')
define(`rlimb',`%g3')
define(`i00',`%l0')    define(`i16',`%l1')
define(`r00',`%l2')    define(`r32',`%l3')
define(`xffffffff',`%l7')
define(`xffff',`%o0')


PROLOGUE(mpn_addmul_2)

C Initialization.  (1) Split v operand into eight 16-bit chunks and store them
C as IEEE double in fp registers.  (2) Clear upper 32 bits of fp register pairs
C f2 and f4.  (3) Store masks in registers aliased to `xffff' and `xffffffff'.
C This code could be better scheduled.

	save	%sp, -256, %sp

ifdef(`HAVE_VIS',
`	mov	-1, %g4
	wr	%g0, 0xD2, %asi
	srlx	%g4, 32, xffffffff	C store mask in register `xffffffff'
	ldda	[%i3+6] %asi, v000
	ldda	[%i3+4] %asi, v016
	ldda	[%i3+2] %asi, v032
	ldda	[%i3+0] %asi, v048
	fxtod	v000, v000
	ldda	[%i3+14] %asi, v064
	fxtod	v016, v016
	ldda	[%i3+12] %asi, v080
	fxtod	v032, v032
	ldda	[%i3+10] %asi, v096
	fxtod	v048, v048
	ldda	[%i3+8] %asi, v112
	fxtod	v064, v064
	fxtod	v080, v080
	fxtod	v096, v096
	fxtod	v112, v112
	fzero	u00_hi
	fzero	u32_hi
',
`	mov	-1, %g4
	ldx	[%i3+0], %l0		C vp[0]
	srlx	%g4, 48, xffff		C store mask in register `xffff'
	ldx	[%i3+8], %l1		C vp[1]

	and	%l0, xffff, %g2
	stx	%g2, [%sp+2223+0]
	srlx	%l0, 16, %g3
	and	%g3, xffff, %g3
	stx	%g3, [%sp+2223+8]
	srlx	%l0, 32, %g2
	and	%g2, xffff, %g2
	stx	%g2, [%sp+2223+16]
	srlx	%l0, 48, %g3
	stx	%g3, [%sp+2223+24]
	and	%l1, xffff, %g2
	stx	%g2, [%sp+2223+32]
	srlx	%l1, 16, %g3
	and	%g3, xffff, %g3
	stx	%g3, [%sp+2223+40]
	srlx	%l1, 32, %g2
	and	%g2, xffff, %g2
	stx	%g2, [%sp+2223+48]
	srlx	%l1, 48, %g3
	stx	%g3, [%sp+2223+56]

	srlx	%g4, 32, xffffffff	C store mask in register `xffffffff'

	ldd	[%sp+2223+0], v000
	ldd	[%sp+2223+8], v016
	ldd	[%sp+2223+16], v032
	ldd	[%sp+2223+24], v048
	fxtod	v000, v000
	ldd	[%sp+2223+32], v064
	fxtod	v016, v016
	ldd	[%sp+2223+40], v080
	fxtod	v032, v032
	ldd	[%sp+2223+48], v096
	fxtod	v048, v048
	ldd	[%sp+2223+56], v112
	fxtod	v064, v064
	ld	[%sp+2223+0], u00_hi	C zero u00_hi
	fxtod	v080, v080
	ld	[%sp+2223+0], u32_hi	C zero u32_hi
	fxtod	v096, v096
	fxtod	v112, v112
')
C Initialization done.
	mov	0, %g2
	mov	0, rlimb
	mov	0, %g4
	add	%i0, -8, %i0		C BOOKKEEPING

C Start software pipeline.

	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	fxtod	u00_hi, u00
C mid
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	fmuld	u00, v000, a000
	fmuld	u00, v016, a016
	fmuld	u00, v032, a032
	fmuld	u00, v048, a048
	add	%i2, -1, %i2		C BOOKKEEPING
	fmuld	u00, v064, p064
	add	%i1, 8, %i1		C BOOKKEEPING
	fxtod	u32_hi, u32
	fmuld	u00, v080, p080
	fmuld	u00, v096, p096a
	brnz,pt	%i2, .L_2_or_more
	 fmuld	u00, v112, p112a

.L1:	fdtox	a000, out000
	fmuld	u32, v000, p000
	fdtox	a016, out016
	fmuld	u32, v016, p016
	fmovd	p064, a064
	fmuld	u32, v032, p032
	fmovd	p080, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	fdtox	a000, out000
	fdtox	a016, out016
	faddd	p064, p096a, a064
	faddd	p080, p112a, a080
	std	out000, [%sp+2223+0]
	b	.L_wd2
	 std	out016, [%sp+2223+8]

.L_2_or_more:
	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	fdtox	a000, out000
	fmuld	u32, v000, p000
	fdtox	a016, out016
	fmuld	u32, v016, p016
	fmovd	p064, a064
	fmuld	u32, v032, p032
	fmovd	p080, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	fdtox	a000, out000
	fmuld	u00, v000, p000
	fdtox	a016, out016
	fmuld	u00, v016, p016
	faddd	p064, p096a, a064
	fmuld	u00, v032, p032
	faddd	p080, p112a, a080
	fmuld	u00, v048, p048
	add	%i2, -1, %i2		C BOOKKEEPING
	std	out000, [%sp+2223+0]
	faddd	p000, a032, a000
	fmuld	u00, v064, p064
	add	%i1, 8, %i1		C BOOKKEEPING
	std	out016, [%sp+2223+8]
	fxtod	u32_hi, u32
	faddd	p016, a048, a016
	fmuld	u00, v080, p080
	faddd	p032, a064, a032
	fmuld	u00, v096, p096a
	faddd	p048, a080, a048
	brnz,pt	%i2, .L_3_or_more
	 fmuld	u00, v112, p112a

	b	.Lend
	 nop

C  64      32       0
C   .       .       .
C   .       |__rXXX_|	32
C   .      |___cy___|	34
C   .  |_______i00__|	50
C  |_______i16__|   .	50


C BEGIN MAIN LOOP
	.align	16
.L_3_or_more:
.Loop:	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u32, v000, p000
C
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u32, v016, p016
C
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	faddd	p064, p096b, a064
	fmuld	u32, v032, p032
C
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	faddd	p080, p112b, a080
	fmuld	u32, v048, p048
C
	nop
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
C
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
C
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
C
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
C
	stw	%l5, [%i0+4]
	nop
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C midloop
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u00, v000, p000
C
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u00, v016, p016
C
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	faddd	p064, p096a, a064
	fmuld	u00, v032, p032
C
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	faddd	p080, p112a, a080
	fmuld	u00, v048, p048
C
	add	%i2, -1, %i2		C BOOKKEEPING
	std	out000, [%sp+2223+0]
	faddd	p000, a032, a000
	fmuld	u00, v064, p064
C
	add	i00, r32, rlimb
	add	%i1, 8, %i1		C BOOKKEEPING
	std	out016, [%sp+2223+8]
	fxtod	u32_hi, u32
C
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u00, v080, p080
C
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u00, v096, p096a
C
	stw	%l5, [%i0+0]
	faddd	p048, a080, a048
	brnz,pt	%i2, .Loop
	 fmuld	u00, v112, p112a
C END MAIN LOOP

C WIND-DOWN PHASE 1
.Lend:	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u32, v000, p000
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u32, v016, p016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	faddd	p064, p096b, a064
	fmuld	u32, v032, p032
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	faddd	p080, p112b, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	stw	%l5, [%i0+4]
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	faddd	p064, p096a, a064
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	faddd	p080, p112a, a080
	std	out000, [%sp+2223+0]
	add	i00, r32, rlimb
	std	out016, [%sp+2223+8]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

C WIND-DOWN PHASE 2
.L_wd2:	and	%g2, xffffffff, %g2
	fdtox	a032, out000
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a048, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	std	out000, [%sp+2223+16]
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+4]
C mid
	and	%g2, xffffffff, %g2
	fdtox	a064, out000
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a080, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	std	out000, [%sp+2223+0]
	add	i00, r32, rlimb
	std	out016, [%sp+2223+8]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

C WIND-DOWN PHASE 3
.L_wd3:	and	%g2, xffffffff, %g2
	fdtox	p096b, out000
	add	%g2, rlimb, %l5
	fdtox	p112b, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], rlimb
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	std	out000, [%sp+2223+16]
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+4]
C mid
	and	%g2, xffffffff, %g2
	add	%g2, rlimb, %l5
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], rlimb
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

	and	%g2, xffffffff, %g2
	add	%g2, rlimb, %l5
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16

	sllx	i16, 16, %g2
	add	i00, cy, cy
	return	%i7+8
	add	%g2, cy, %o0
EPILOGUE(mpn_addmul_2)
@


1.1.1.1.4.1
log
@file addmul_2.asm was added on branch yamt-pagecache on 2014-05-22 14:09:04 +0000
@
text
@d1 540
@


1.1.1.1.4.2
log
@sync with head.

for a reference, the tree before this commit was tagged
as yamt-pagecache-tag8.

this commit was splitted into small chunks to avoid
a limitation of cvs.  ("Protocol error: too many arguments")
@
text
@a0 540
dnl  SPARC v9 64-bit mpn_addmul_2 -- Multiply an n limb number with 2-limb
dnl  number and add the result to a n limb vector.

dnl  Copyright 2002, 2003 Free Software Foundation, Inc.

dnl  This file is part of the GNU MP Library.

dnl  The GNU MP Library is free software; you can redistribute it and/or modify
dnl  it under the terms of the GNU Lesser General Public License as published
dnl  by the Free Software Foundation; either version 3 of the License, or (at
dnl  your option) any later version.

dnl  The GNU MP Library is distributed in the hope that it will be useful, but
dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
dnl  License for more details.

dnl  You should have received a copy of the GNU Lesser General Public License
dnl  along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.

include(`../config.m4')

C                  cycles/limb
C UltraSPARC 1&2:      9
C UltraSPARC 3:       10

C Algorithm: We use 16 floating-point multiplies per limb product, with the
C 2-limb v operand split into eight 16-bit pieces, and the n-limb u operand
C split into 32-bit pieces.  We sum four 48-bit partial products using
C floating-point add, then convert the resulting four 50-bit quantities and
C transfer them to the integer unit.

C Possible optimizations:
C   1. Align the stack area where we transfer the four 50-bit product-sums
C      to a 32-byte boundary.  That would minimize the cache collision.
C      (UltraSPARC-1/2 use a direct-mapped cache.)  (Perhaps even better would
C      be to align the area to map to the area immediately before up?)
C   2. Perform two of the fp->int conversions with integer instructions.  We
C      can get almost ten free IEU slots, if we clean up bookkeeping and the
C      silly carry-limb code.
C   3. For an mpn_addmul_1 based on this, we need to fix the silly carry-limb
C      code.

C OSP (Overlapping software pipeline) version of mpn_mul_basecase:
C Operand swap will require 8 LDDA and 8 FXTOD, which will mean 8 cycles.
C FI	= 20
C L	=  9 x un * vn
C WDFI	= 10 x vn / 2
C WD	= 4

C Instruction classification (as per UltraSPARC functional units).
C Assuming silly carry code is fixed.  Includes bookkeeping.
C
C               mpn_addmul_X     mpn_mul_X
C                1       2       1       2
C               ==========      ==========
C      FM        8      16       8      16
C      FA       10      18      10      18
C     MEM       12      12      10      10
C  ISHIFT        6       6       6       6
C IADDLOG       11      11      10      10
C  BRANCH        1       1       1       1
C
C TOTAL IEU     17      17      16      16
C TOTAL         48      64      45      61
C
C IEU cycles     8.5     8.5     8       8
C MEM cycles    12      12      10      10
C ISSUE cycles  12      16      11.25   15.25
C FPU cycles    10      18      10      18
C cycles/loop   12      18      12      18
C cycles/limb   12       9      12       9


C INPUT PARAMETERS
C rp[n + 1]	i0
C up[n]		i1
C n		i2
C vp[2]		i3


ASM_START()
	REGISTER(%g2,#scratch)
	REGISTER(%g3,#scratch)

C Combine registers:
C u00_hi= u32_hi
C u00_lo= u32_lo
C a000  = out000
C a016  = out016
C Free: f52 f54


define(`p000', `%f8')  define(`p016',`%f10')
define(`p032',`%f12')  define(`p048',`%f14')
define(`p064',`%f16')  define(`p080',`%f18')
define(`p096a',`%f20') define(`p112a',`%f22')
define(`p096b',`%f56') define(`p112b',`%f58')

define(`out000',`%f0') define(`out016',`%f6')

define(`v000',`%f24')  define(`v016',`%f26')
define(`v032',`%f28')  define(`v048',`%f30')
define(`v064',`%f44')  define(`v080',`%f46')
define(`v096',`%f48')  define(`v112',`%f50')

define(`u00',`%f32')   define(`u32', `%f34')

define(`a000',`%f36')  define(`a016',`%f38')
define(`a032',`%f40')  define(`a048',`%f42')
define(`a064',`%f60')  define(`a080',`%f62')

define(`u00_hi',`%f2') define(`u32_hi',`%f4')
define(`u00_lo',`%f3') define(`u32_lo',`%f5')

define(`cy',`%g1')
define(`rlimb',`%g3')
define(`i00',`%l0')    define(`i16',`%l1')
define(`r00',`%l2')    define(`r32',`%l3')
define(`xffffffff',`%l7')
define(`xffff',`%o0')


PROLOGUE(mpn_addmul_2)

C Initialization.  (1) Split v operand into eight 16-bit chunks and store them
C as IEEE double in fp registers.  (2) Clear upper 32 bits of fp register pairs
C f2 and f4.  (3) Store masks in registers aliased to `xffff' and `xffffffff'.
C This code could be better scheduled.

	save	%sp, -256, %sp

ifdef(`HAVE_VIS',
`	mov	-1, %g4
	wr	%g0, 0xD2, %asi
	srlx	%g4, 32, xffffffff	C store mask in register `xffffffff'
	ldda	[%i3+6] %asi, v000
	ldda	[%i3+4] %asi, v016
	ldda	[%i3+2] %asi, v032
	ldda	[%i3+0] %asi, v048
	fxtod	v000, v000
	ldda	[%i3+14] %asi, v064
	fxtod	v016, v016
	ldda	[%i3+12] %asi, v080
	fxtod	v032, v032
	ldda	[%i3+10] %asi, v096
	fxtod	v048, v048
	ldda	[%i3+8] %asi, v112
	fxtod	v064, v064
	fxtod	v080, v080
	fxtod	v096, v096
	fxtod	v112, v112
	fzero	u00_hi
	fzero	u32_hi
',
`	mov	-1, %g4
	ldx	[%i3+0], %l0		C vp[0]
	srlx	%g4, 48, xffff		C store mask in register `xffff'
	ldx	[%i3+8], %l1		C vp[1]

	and	%l0, xffff, %g2
	stx	%g2, [%sp+2223+0]
	srlx	%l0, 16, %g3
	and	%g3, xffff, %g3
	stx	%g3, [%sp+2223+8]
	srlx	%l0, 32, %g2
	and	%g2, xffff, %g2
	stx	%g2, [%sp+2223+16]
	srlx	%l0, 48, %g3
	stx	%g3, [%sp+2223+24]
	and	%l1, xffff, %g2
	stx	%g2, [%sp+2223+32]
	srlx	%l1, 16, %g3
	and	%g3, xffff, %g3
	stx	%g3, [%sp+2223+40]
	srlx	%l1, 32, %g2
	and	%g2, xffff, %g2
	stx	%g2, [%sp+2223+48]
	srlx	%l1, 48, %g3
	stx	%g3, [%sp+2223+56]

	srlx	%g4, 32, xffffffff	C store mask in register `xffffffff'

	ldd	[%sp+2223+0], v000
	ldd	[%sp+2223+8], v016
	ldd	[%sp+2223+16], v032
	ldd	[%sp+2223+24], v048
	fxtod	v000, v000
	ldd	[%sp+2223+32], v064
	fxtod	v016, v016
	ldd	[%sp+2223+40], v080
	fxtod	v032, v032
	ldd	[%sp+2223+48], v096
	fxtod	v048, v048
	ldd	[%sp+2223+56], v112
	fxtod	v064, v064
	ld	[%sp+2223+0], u00_hi	C zero u00_hi
	fxtod	v080, v080
	ld	[%sp+2223+0], u32_hi	C zero u32_hi
	fxtod	v096, v096
	fxtod	v112, v112
')
C Initialization done.
	mov	0, %g2
	mov	0, rlimb
	mov	0, %g4
	add	%i0, -8, %i0		C BOOKKEEPING

C Start software pipeline.

	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	fxtod	u00_hi, u00
C mid
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	fmuld	u00, v000, a000
	fmuld	u00, v016, a016
	fmuld	u00, v032, a032
	fmuld	u00, v048, a048
	add	%i2, -1, %i2		C BOOKKEEPING
	fmuld	u00, v064, p064
	add	%i1, 8, %i1		C BOOKKEEPING
	fxtod	u32_hi, u32
	fmuld	u00, v080, p080
	fmuld	u00, v096, p096a
	brnz,pt	%i2, .L_2_or_more
	 fmuld	u00, v112, p112a

.L1:	fdtox	a000, out000
	fmuld	u32, v000, p000
	fdtox	a016, out016
	fmuld	u32, v016, p016
	fmovd	p064, a064
	fmuld	u32, v032, p032
	fmovd	p080, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	fdtox	a000, out000
	fdtox	a016, out016
	faddd	p064, p096a, a064
	faddd	p080, p112a, a080
	std	out000, [%sp+2223+0]
	b	.L_wd2
	 std	out016, [%sp+2223+8]

.L_2_or_more:
	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	fdtox	a000, out000
	fmuld	u32, v000, p000
	fdtox	a016, out016
	fmuld	u32, v016, p016
	fmovd	p064, a064
	fmuld	u32, v032, p032
	fmovd	p080, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	fdtox	a000, out000
	fmuld	u00, v000, p000
	fdtox	a016, out016
	fmuld	u00, v016, p016
	faddd	p064, p096a, a064
	fmuld	u00, v032, p032
	faddd	p080, p112a, a080
	fmuld	u00, v048, p048
	add	%i2, -1, %i2		C BOOKKEEPING
	std	out000, [%sp+2223+0]
	faddd	p000, a032, a000
	fmuld	u00, v064, p064
	add	%i1, 8, %i1		C BOOKKEEPING
	std	out016, [%sp+2223+8]
	fxtod	u32_hi, u32
	faddd	p016, a048, a016
	fmuld	u00, v080, p080
	faddd	p032, a064, a032
	fmuld	u00, v096, p096a
	faddd	p048, a080, a048
	brnz,pt	%i2, .L_3_or_more
	 fmuld	u00, v112, p112a

	b	.Lend
	 nop

C  64      32       0
C   .       .       .
C   .       |__rXXX_|	32
C   .      |___cy___|	34
C   .  |_______i00__|	50
C  |_______i16__|   .	50


C BEGIN MAIN LOOP
	.align	16
.L_3_or_more:
.Loop:	ld	[%i1+4], u00_lo		C read low 32 bits of up[i]
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u32, v000, p000
C
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u32, v016, p016
C
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	faddd	p064, p096b, a064
	fmuld	u32, v032, p032
C
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	faddd	p080, p112b, a080
	fmuld	u32, v048, p048
C
	nop
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
C
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	fxtod	u00_hi, u00
C
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
C
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
C
	stw	%l5, [%i0+4]
	nop
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C midloop
	ld	[%i1+0], u32_lo		C read high 32 bits of up[i]
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u00, v000, p000
C
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u00, v016, p016
C
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	faddd	p064, p096a, a064
	fmuld	u00, v032, p032
C
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	faddd	p080, p112a, a080
	fmuld	u00, v048, p048
C
	add	%i2, -1, %i2		C BOOKKEEPING
	std	out000, [%sp+2223+0]
	faddd	p000, a032, a000
	fmuld	u00, v064, p064
C
	add	i00, r32, rlimb
	add	%i1, 8, %i1		C BOOKKEEPING
	std	out016, [%sp+2223+8]
	fxtod	u32_hi, u32
C
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u00, v080, p080
C
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u00, v096, p096a
C
	stw	%l5, [%i0+0]
	faddd	p048, a080, a048
	brnz,pt	%i2, .Loop
	 fmuld	u00, v112, p112a
C END MAIN LOOP

C WIND-DOWN PHASE 1
.Lend:	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	fmuld	u32, v000, p000
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	fmuld	u32, v016, p016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	faddd	p064, p096b, a064
	fmuld	u32, v032, p032
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	faddd	p080, p112b, a080
	fmuld	u32, v048, p048
	std	out000, [%sp+2223+16]
	faddd	p000, a032, a000
	fmuld	u32, v064, p064
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	faddd	p016, a048, a016
	fmuld	u32, v080, p080
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	faddd	p032, a064, a032
	fmuld	u32, v096, p096b
	stw	%l5, [%i0+4]
	faddd	p048, a080, a048
	fmuld	u32, v112, p112b
C mid
	and	%g2, xffffffff, %g2
	fdtox	a000, out000
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a016, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	faddd	p064, p096a, a064
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	faddd	p080, p112a, a080
	std	out000, [%sp+2223+0]
	add	i00, r32, rlimb
	std	out016, [%sp+2223+8]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

C WIND-DOWN PHASE 2
.L_wd2:	and	%g2, xffffffff, %g2
	fdtox	a032, out000
	lduw	[%i0+4+8], r00		C read low 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a048, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	std	out000, [%sp+2223+16]
	add	i00, r00, rlimb
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+4]
C mid
	and	%g2, xffffffff, %g2
	fdtox	a064, out000
	lduw	[%i0+0], r32		C read high 32 bits of rp[i]
	add	%g2, rlimb, %l5
	fdtox	a080, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	std	out000, [%sp+2223+0]
	add	i00, r32, rlimb
	std	out016, [%sp+2223+8]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

C WIND-DOWN PHASE 3
.L_wd3:	and	%g2, xffffffff, %g2
	fdtox	p096b, out000
	add	%g2, rlimb, %l5
	fdtox	p112b, out016
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], rlimb
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16
	std	out000, [%sp+2223+16]
	add	%i0, 8, %i0		C BOOKKEEPING
	std	out016, [%sp+2223+24]
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+4]
C mid
	and	%g2, xffffffff, %g2
	add	%g2, rlimb, %l5
	srlx	%l5, 32, cy
	ldx	[%sp+2223+0], rlimb
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+8], i16
	sllx	i16, 16, %g2
	add	cy, rlimb, rlimb
	srlx	i16, 16, %g4
	add	%g2, rlimb, %l5
	stw	%l5, [%i0+0]

	and	%g2, xffffffff, %g2
	add	%g2, rlimb, %l5
	srlx	%l5, 32, cy
	ldx	[%sp+2223+16], i00
	add	%g4, cy, cy		C new cy
	ldx	[%sp+2223+24], i16

	sllx	i16, 16, %g2
	add	i00, cy, cy
	return	%i7+8
	add	%g2, cy, %o0
EPILOGUE(mpn_addmul_2)
@