[speed-ext.c for tune/speed-ext
peter@cordes.ca**20080325024627
 Your darcs repo should be gmp-.../tune instead of tests/devel.
 Hardlinks should do the trick.
 Use the patch for Makefile.in.
 See the suggested command line in speed-ext.c
] {
addfile ./Makefile.in.diff
hunk ./Makefile.in.diff 1
+--- Makefile.in~	2008-03-23 01:20:18.576446999 -0300
++++ Makefile.in	2008-03-23 01:52:46.846474317 -0300
+@@ -106,7 +106,7 @@
+ speed_dynamic_OBJECTS = $(am_speed_dynamic_OBJECTS)
+ speed_dynamic_LDADD = $(LDADD)
+ speed_dynamic_DEPENDENCIES = $(am__DEPENDENCIES_3)
+-am_speed_ext_OBJECTS = speed-ext$U.$(OBJEXT)
++am_speed_ext_OBJECTS = speed-ext$U.$(OBJEXT) rshift.o
+ speed_ext_OBJECTS = $(am_speed_ext_OBJECTS)
+ speed_ext_LDADD = $(LDADD)
+ speed_ext_DEPENDENCIES = $(am__DEPENDENCIES_3)
addfile ./speed-ext.c
hunk ./speed-ext.c 1
+/* Speed-measuring for rshift functions
+
+Copyright 2008 Peter Cordes <peter@cordes.ca>.  License=GPL version >= 3.  Share and enjoy.
+
+This file has the same name as the example speed-ext.c
+so I don't have to change the Makefile(.in) so much to build it
+
+If you have more than 1 CPU core, you can dedicate core #2 to the benchmark like so:
+make speed-ext; timeout -15 5 taskset 2 burnK7; time sudo nice --7 taskset 2 chrt -r 1 timeout -15 20 ./speed-ext -p 1000000 -c -s 1-30 rshift.8
+
+burnK7 generates load to shift your CPU to full speed before the benchmark starts.
+timeout sends a signal after the configured time.  This is important to prevent a runaway realtime process, in case your system is handling kbd interrupts with that core or something.  (it's cpu #1 on my system).
+taskset pins the task to one CPU (also the load generator, so the right core clocks up.)
+chrt is like nice but for realtime scheduling
+
+You should be able to get very reproducible cycle-accurate results this way.
+
+*/
+
+#define SPEED_EXTRA_PROTOS                                              \
+double speed_rshift_sse2_aligned (struct speed_params *s); \
+double speed_rshift_sse2 (struct speed_params *s); \
+double speed_rshift_core2 (struct speed_params *s);
+
+#define SPEED_EXTRA_ROUTINES            \
+  { "rshift",  speed_rshift_core2, FLAG_R  }, \
+  { "rshift_sse2",   speed_rshift_sse2, FLAG_R   }, \
+  { "rshift_sse2_aligned",  speed_rshift_sse2_aligned, FLAG_R  },
+
+
+#include "speed.c"
+
+mp_limb_t mpn_rshift_sse2_aligned(mp_limb_t *rp, const mp_limb_t *sp, mp_size_t n, unsigned int count );
+mp_limb_t mpn_rshift_sse2(mp_limb_t *rp, const mp_limb_t *sp, mp_size_t n, unsigned int count );
+mp_limb_t mpn_rshift_core2(mp_limb_t *rp, const mp_limb_t *sp, mp_size_t n, unsigned int count );
+
+double speed_rshift_sse2_aligned (struct speed_params *s){
+  SPEED_ROUTINE_MPN_UNARY_1 (mpn_rshift_sse2_aligned); }
+double speed_rshift_sse2 (struct speed_params *s){
+  SPEED_ROUTINE_MPN_UNARY_1 (mpn_rshift_sse2); }
+double speed_rshift_core2 (struct speed_params *s){
+  SPEED_ROUTINE_MPN_UNARY_1 (mpn_rshift_core2); }
}