xref: /netbsd-src/external/lgpl3/gmp/dist/tune/Makefile.am (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1## Process this file with automake to generate Makefile.in
2
3# Copyright 2000-2003, 2005-2011 Free Software Foundation, Inc.
4#
5#  This file is part of the GNU MP Library.
6#
7#  The GNU MP Library is free software; you can redistribute it and/or modify
8#  it under the terms of either:
9#
10#    * the GNU Lesser General Public License as published by the Free
11#      Software Foundation; either version 3 of the License, or (at your
12#      option) any later version.
13#
14#  or
15#
16#    * the GNU General Public License as published by the Free Software
17#      Foundation; either version 2 of the License, or (at your option) any
18#      later version.
19#
20#  or both in parallel, as here.
21#
22#  The GNU MP Library is distributed in the hope that it will be useful, but
23#  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
24#  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
25#  for more details.
26#
27#  You should have received copies of the GNU General Public License and the
28#  GNU Lesser General Public License along with the GNU MP Library.  If not,
29#  see https://www.gnu.org/licenses/.
30
31
32AM_CPPFLAGS = -I$(top_srcdir) -I$(top_srcdir)/tests
33
34EXTRA_DIST = alpha.asm pentium.asm sparcv9.asm hppa.asm hppa2.asm hppa2w.asm \
35  ia64.asm powerpc.asm powerpc64.asm x86_64.asm many.pl
36noinst_HEADERS = speed.h
37
38# Prefer -static on the speed and tune programs, since that can avoid
39# overheads of shared library linkages on some systems.  Libtool tends to
40# botch -static if configured with --disable-static, perhaps reasonably
41# enough.  In any event under --disable-static the only choice is a dynamic
42# link so there's no point in -static.
43#
44if ENABLE_STATIC
45STATIC = -static
46else
47STATIC =
48endif
49
50
51EXTRA_LTLIBRARIES = libspeed.la
52
53libspeed_la_SOURCES =							\
54  common.c divrem1div.c divrem1inv.c divrem2div.c divrem2inv.c		\
55  div_qr_1n_pi1_1.c div_qr_1n_pi1_2.c div_qr_1_tune.c			\
56  freq.c								\
57  gcdext_single.c gcdext_double.c gcdextod.c gcdextos.c			\
58  hgcd_lehmer.c hgcd_appr_lehmer.c hgcd_reduce_1.c hgcd_reduce_2.c	\
59  jacbase1.c jacbase2.c jacbase3.c jacbase4.c				\
60  mod_1_div.c mod_1_inv.c mod_1_1-1.c mod_1_1-2.c modlinv.c		\
61  noop.c powm_mod.c powm_redc.c pre_divrem_1.c				\
62  set_strb.c set_strs.c set_strp.c time.c
63
64libspeed_la_DEPENDENCIES = $(SPEED_CYCLECOUNTER_OBJ) \
65  $(top_builddir)/tests/libtests.la $(top_builddir)/libgmp.la
66libspeed_la_LIBADD = $(libspeed_la_DEPENDENCIES) $(LIBM)
67libspeed_la_LDFLAGS = $(STATIC)
68
69$(top_builddir)/tests/libtests.la:
70	cd $(top_builddir)/tests; $(MAKE) $(AM_MAKEFLAGS) libtests.la
71
72
73# The library code is faster static than shared on some systems, so do
74# tuning and measuring with static, since users who care about maximizing
75# speed will be using that.  speed-dynamic exists to show the difference.
76#
77# On Solaris 8, gcc 2.95.2 -static is somehow broken (it creates executables
78# that immediately seg fault), so -all-static is not used.  The only thing
79# -all-static does is make libc static linked as well as libgmp, and that
80# makes a difference only when measuring malloc and friends in the speed
81# program.  This can always be forced with "make speed_LDFLAGS=-all-static
82# ..." if desired, see tune/README.
83
84EXTRA_PROGRAMS = speed speed-dynamic speed-ext tuneup tune-gcd-p
85
86DEPENDENCIES = libspeed.la
87LDADD = $(DEPENDENCIES) $(TUNE_LIBS)
88
89speed_SOURCES = speed.c
90speed_LDFLAGS = $(STATIC)
91
92speed_dynamic_SOURCES = speed.c
93
94speed_ext_SOURCES = speed-ext.c
95speed_ext_LDFLAGS = $(STATIC)
96
97tuneup_SOURCES = tuneup.c
98nodist_tuneup_SOURCES = sqr_basecase.c fac_ui.c $(TUNE_MPN_SRCS)
99tuneup_DEPENDENCIES = $(TUNE_SQR_OBJ) libspeed.la
100tuneup_LDADD = $(tuneup_DEPENDENCIES) $(TUNE_LIBS)
101tuneup_LDFLAGS = $(STATIC)
102
103tune_gcd_p_SOURCES = tune-gcd-p.c
104tune_gcd_p_DEPENDENCIES = ../mpn/gcd.c
105tune_gcd_p_LDFLAGS = $(STATIC)
106
107
108tune:
109	$(MAKE) $(AM_MAKEFLAGS) tuneup$(EXEEXT)
110	./tuneup
111
112allprogs: $(EXTRA_PROGRAMS)
113
114# $(MANY_CLEAN) and $(MANY_DISTCLEAN) are hooks for many.pl
115CLEANFILES = $(EXTRA_PROGRAMS) $(EXTRA_LTLIBRARIES) \
116	$(TUNE_MPN_SRCS) fac_ui.c sqr_asm.asm \
117	stg.gnuplot stg.data \
118	mtg.gnuplot mtg.data \
119	fibg.gnuplot fibg.data \
120	graph.gnuplot graph.data \
121	$(MANY_CLEAN)
122DISTCLEANFILES = sqr_basecase.c  $(MANY_DISTCLEAN)
123
124
125# Generating these little files at build time seems better than including
126# them in the distribution, since the list can be changed more easily.
127#
128# mpn/generic/tdiv_qr.c uses mpn_divrem_1 and mpn_divrem_2, but only for 1
129# and 2 limb divisors, which are never used during tuning, so it doesn't
130# matter whether it picks up a tuned or untuned version of those.
131#
132# divrem_1 and mod_1 are recompiled renamed to "_tune" to avoid a linking
133# problem.  If a native divrem_1 provides an mpn_divrem_1c entrypoint then
134# common.c will want that, but the generic divrem_1 doesn't provide it,
135# likewise for mod_1.  The simplest way around this is to have the tune
136# build versions renamed suitably.
137#
138# FIXME: Would like say mul_n.c to depend on $(top_builddir)/mul_n.c so the
139# recompiled object will be rebuilt if that file changes.
140
141TUNE_MPN_SRCS = $(TUNE_MPN_SRCS_BASIC) divrem_1.c mod_1.c
142TUNE_MPN_SRCS_BASIC = div_qr_2.c bdiv_q.c bdiv_qr.c			\
143  dcpi1_div_qr.c dcpi1_divappr_q.c dcpi1_bdiv_qr.c dcpi1_bdiv_q.c	\
144  invertappr.c invert.c binvert.c divrem_2.c gcd.c gcdext.c		\
145  get_str.c set_str.c matrix22_mul.c					\
146  hgcd.c hgcd_appr.c hgcd_reduce.c					\
147  mul_n.c sqr.c sec_powm.c						\
148  mullo_n.c mul_fft.c mul.c tdiv_qr.c mulmod_bnm1.c sqrmod_bnm1.c	\
149  mulmid.c mulmid_n.c toom42_mulmid.c sqrlo.c sqrlo_basecase.c		\
150  nussbaumer_mul.c toom6h_mul.c toom8h_mul.c toom6_sqr.c toom8_sqr.c	\
151  toom22_mul.c toom2_sqr.c toom33_mul.c toom3_sqr.c toom44_mul.c toom4_sqr.c
152
153$(TUNE_MPN_SRCS_BASIC):
154	for i in $(TUNE_MPN_SRCS_BASIC); do \
155	  echo "#define TUNE_PROGRAM_BUILD 1" >$$i; \
156	  echo "#include \"mpn/generic/$$i\"" >>$$i; \
157	done
158
159divrem_1.c:
160	echo "#define TUNE_PROGRAM_BUILD 1"                >divrem_1.c
161	echo "#define __gmpn_divrem_1  mpn_divrem_1_tune" >>divrem_1.c
162	echo "#include \"mpn/generic/divrem_1.c\""        >>divrem_1.c
163
164mod_1.c:
165	echo "#define TUNE_PROGRAM_BUILD 1"          >mod_1.c
166	echo "#define __gmpn_mod_1  mpn_mod_1_tune" >>mod_1.c
167	echo "#include \"mpn/generic/mod_1.c\""     >>mod_1.c
168
169sqr_asm.asm: $(top_builddir)/mpn/sqr_basecase.asm
170	echo 'define(SQR_TOOM2_THRESHOLD_OVERRIDE,SQR_TOOM2_THRESHOLD_MAX)' >sqr_asm.asm
171	echo 'include(../mpn/sqr_basecase.asm)' >>sqr_asm.asm
172
173# FIXME: Should it depend on $(top_builddir)/fac_ui.h too?
174fac_ui.c: $(top_builddir)/mpz/fac_ui.c
175	echo "#define TUNE_PROGRAM_BUILD 1"          >fac_ui.c
176	echo "#define __gmpz_fac_ui mpz_fac_ui_tune" >>fac_ui.c
177	echo "#define __gmpz_oddfac_1 mpz_oddfac_1_tune" >>fac_ui.c
178	echo "#include \"mpz/oddfac_1.c\""           >>fac_ui.c
179	echo "#include \"mpz/fac_ui.c\""             >>fac_ui.c
180
181include ../mpn/Makeasm.am
182