Tune fac_ui thresholds.

2 files changed, 41 insertions, 4 deletions

diff --git a/tune/Makefile.am b/tune/Makefile.am
index 646a1f4af..c0e0bbbab 100644
--- a/tune/Makefile.am
+++ b/tune/Makefile.am
@@ -1,6 +1,7 @@
 ## Process this file with automake to generate Makefile.in
 
-# Copyright 2000, 2001, 2002, 2003, 2005 Free Software Foundation, Inc.
+# Copyright 2000, 2001, 2002, 2003, 2005, 2006, 2007, 2008, 2009,
+# 2010, 2011 Free Software Foundation, Inc.
 #
 # This file is part of the GNU MP Library.
 #
@@ -83,7 +84,7 @@ speed_ext_SOURCES = speed-ext.c
 speed_ext_LDFLAGS = $(STATIC)
 
 tuneup_SOURCES = tuneup.c
-nodist_tuneup_SOURCES = sqr_basecase.c $(TUNE_MPN_SRCS)
+nodist_tuneup_SOURCES = sqr_basecase.c fac_ui.c $(TUNE_MPN_SRCS)
 tuneup_DEPENDENCIES = $(TUNE_SQR_OBJ) libspeed.la
 tuneup_LDADD = $(tuneup_DEPENDENCIES) $(TUNE_LIBS)
 tuneup_LDFLAGS = $(STATIC)
@@ -101,7 +102,7 @@ allprogs: $(EXTRA_PROGRAMS)
 
 # $(MANY_CLEAN) and $(MANY_DISTCLEAN) are hooks for many.pl
 CLEANFILES = $(EXTRA_PROGRAMS) $(EXTRA_LTLIBRARIES) \
-	$(TUNE_MPN_SRCS) sqr_asm.asm \
+	$(TUNE_MPN_SRCS) fac_ui.c sqr_asm.asm \
 	stg.gnuplot stg.data \
 	mtg.gnuplot mtg.data \
 	fibg.gnuplot fibg.data \
@@ -158,5 +159,10 @@ sqr_asm.asm: $(top_builddir)/mpn/sqr_basecase.asm
 	echo 'define(SQR_TOOM2_THRESHOLD_OVERRIDE,SQR_TOOM2_THRESHOLD_MAX)' >sqr_asm.asm
 	echo 'include(../mpn/sqr_basecase.asm)' >>sqr_asm.asm
 
+# FIXME: Should it depend on $(top_builddir)/mpz/fac_ui.h too?
+fac_ui.c: $(top_builddir)/mpz/fac_ui.c
+	echo "#define TUNE_PROGRAM_BUILD 1"          >fac_ui.c
+	echo "#define __gmpz_fac_ui mpz_fac_ui_tune" >>fac_ui.c
+	echo "#include \"mpz/fac_ui.c\""             >>fac_ui.c
 
 include ../mpn/Makeasm.am
diff --git a/tune/tuneup.c b/tune/tuneup.c
index bc7e8cc3d..ce84e8fa7 100644
--- a/tune/tuneup.c
+++ b/tune/tuneup.c
@@ -214,6 +214,8 @@ mp_size_t  get_str_dc_threshold         = MP_SIZE_T_MAX;
 mp_size_t  get_str_precompute_threshold = MP_SIZE_T_MAX;
 mp_size_t  set_str_dc_threshold         = MP_SIZE_T_MAX;
 mp_size_t  set_str_precompute_threshold = MP_SIZE_T_MAX;
+mp_size_t  fac_odd_threshold            = MP_SIZE_T_MAX;
+mp_size_t  fac_dsc_threshold            = FAC_DSC_THRESHOLD_LIMIT;
 
 mp_size_t  fft_modf_sqr_threshold = MP_SIZE_T_MAX;
 mp_size_t  fft_modf_mul_threshold = MP_SIZE_T_MAX;
@@ -366,12 +368,15 @@ analyze_dat (int final)
 }
 
 
-/* Measuring for recompiled mpn/generic/divrem_1.c and mpn/generic/mod_1.c */
+/* Measuring for recompiled mpn/generic/divrem_1.c, mpn/generic/mod_1.c
+ * and mpz/fac_ui.c */
 
 mp_limb_t mpn_divrem_1_tune
   __GMP_PROTO ((mp_ptr, mp_size_t, mp_srcptr, mp_size_t, mp_limb_t));
 mp_limb_t mpn_mod_1_tune
    __GMP_PROTO ((mp_srcptr, mp_size_t, mp_limb_t));
+void mpz_fac_ui_tune
+   __GMP_PROTO ((mpz_ptr, unsigned long));
 
 double
 speed_mpn_mod_1_tune (struct speed_params *s)
@@ -383,6 +388,11 @@ speed_mpn_divrem_1_tune (struct speed_params *s)
 {
   SPEED_ROUTINE_MPN_DIVREM_1 (mpn_divrem_1_tune);
 }
+double
+speed_mpz_fac_ui_tune (struct speed_params *s)
+{
+  SPEED_ROUTINE_MPZ_FAC_UI (mpz_fac_ui_tune);
+}
 
 
 double
@@ -2612,6 +2622,24 @@ tune_fft_sqr (void)
 }
 
 void
+tune_fac_ui (void)
+{
+  static struct param_t  param;
+
+  param.function = speed_mpz_fac_ui_tune;
+
+  param.name = "FAC_ODD_THRESHOLD";
+  param.min_size = 3;
+  param.min_is_always = 0;
+  one (&fac_odd_threshold, &param);
+
+  param.name = "FAC_DSC_THRESHOLD";
+  param.min_size = MAX (32, fac_odd_threshold);
+  param.max_size = FAC_DSC_THRESHOLD_LIMIT;
+  one (&fac_dsc_threshold, &param);
+}
+
+void
 all (void)
 {
   time_t  start_time, end_time;
@@ -2747,6 +2775,9 @@ all (void)
   tune_set_str ();
   printf("\n");
 
+  tune_fac_ui ();
+  printf("\n");
+
   time (&end_time);
   printf ("/* Tuneup completed successfully, took %ld seconds */\n",
           (long) (end_time - start_time));