speed enhancement

[Paul Zimmermann]

       Hi,

I suggest to add mpn_sqrt and mpn_root to the "speed" program, to measure
the mpn_sqrtrem and mpn_rootrem functions with NULL 2nd argument. Patch below.

I realized that mpn_rootrem(2) with NULL 2nd argument is faster than
mpn_sqrtrem with NULL 2nd argument!

Maybe mpn_sqrtrem can be improved in that case, or simply call mpn_rootrem(2)?

tarte% ./speed -s 1000 mpn_sqrt mpn_sqrtrem mpn_root.2 mpn_rootrem.2
overhead 0.000000002 secs, precision 10000 units of 3.53e-10 secs, CPU freq 2833.00 MHz
             mpn_sqrt   mpn_sqrtrem    mpn_root.2 mpn_rootrem.2
1000      0.000154164   0.000154746  #0.000145061   0.000225465

Paul Zimmermann

--- common.c.orig	2010-09-28 10:08:16.000000000 +0200
+++ common.c	2010-09-28 10:10:36.000000000 +0200
@@ -1495,11 +1495,22 @@
 }
 
 double
+speed_mpn_sqrt (struct speed_params *s)
+{
+  SPEED_ROUTINE_MPN_SQRT (mpn_sqrtrem);
+}
+
+double
 speed_mpn_rootrem (struct speed_params *s)
 {
   SPEED_ROUTINE_MPN_ROOTREM (mpn_rootrem);
 }
 
+double
+speed_mpn_root (struct speed_params *s)
+{
+  SPEED_ROUTINE_MPN_ROOT (mpn_rootrem);
+}
 
 double
 speed_mpz_fac_ui (struct speed_params *s)
--- speed.c.orig	2010-09-28 10:12:38.000000000 +0200
+++ speed.c	2010-09-28 10:13:01.000000000 +0200
@@ -347,7 +347,9 @@
   { "mpn_set_str_basecase", speed_mpn_bc_set_str,  FLAG_R_OPTIONAL },
 
   { "mpn_sqrtrem",       speed_mpn_sqrtrem          },
+  { "mpn_sqrt",          speed_mpn_sqrt             },
   { "mpn_rootrem",       speed_mpn_rootrem, FLAG_R  },
+  { "mpn_root",          speed_mpn_root,    FLAG_R  },
 
   { "mpn_fib2_ui",       speed_mpn_fib2_ui,    FLAG_NODATA },
   { "mpz_fib_ui",        speed_mpz_fib_ui,     FLAG_NODATA },
--- speed.h.orig	2010-09-28 10:07:03.000000000 +0200
+++ speed.h	2010-09-28 10:11:10.000000000 +0200
@@ -277,7 +277,9 @@
 double speed_mpn_sqr_diagonal __GMP_PROTO ((struct speed_params *s));
 double speed_mpn_sqr __GMP_PROTO ((struct speed_params *s));
 double speed_mpn_sqrtrem __GMP_PROTO ((struct speed_params *s));
+double speed_mpn_sqrt __GMP_PROTO ((struct speed_params *s));
 double speed_mpn_rootrem __GMP_PROTO ((struct speed_params *s));
+double speed_mpn_root __GMP_PROTO ((struct speed_params *s));
 double speed_mpn_sub_n __GMP_PROTO ((struct speed_params *s));
 double speed_mpn_sublsh1_n __GMP_PROTO ((struct speed_params *s));
 double speed_mpn_sublsh2_n __GMP_PROTO ((struct speed_params *s));
@@ -2751,6 +2753,33 @@
     return t;								\
   }
 
+#define SPEED_ROUTINE_MPN_SQRT(function)				\
+  {									\
+    mp_ptr    wp;							\
+    unsigned  i;							\
+    double    t;							\
+    TMP_DECL;								\
+									\
+    SPEED_RESTRICT_COND (s->size >= 1);					\
+									\
+    TMP_MARK;								\
+    SPEED_TMP_ALLOC_LIMBS (wp,	s->size, s->align_wp);			\
+									\
+    speed_operand_src (s, s->xp, s->size);				\
+    speed_operand_dst (s, wp, s->size);					\
+    speed_cache_fill (s);						\
+									\
+    speed_starttime ();							\
+    i = s->reps;							\
+    do									\
+      function (wp, NULL, s->xp, s->size);				\
+    while (--i != 0);							\
+    t = speed_endtime ();						\
+									\
+    TMP_FREE;								\
+    return t;								\
+  }
+
 #define SPEED_ROUTINE_MPN_ROOTREM(function)				\
   {									\
     mp_ptr    wp, wp2;							\
@@ -2781,6 +2810,34 @@
   }
 
 
+#define SPEED_ROUTINE_MPN_ROOT(function)				\
+  {									\
+    mp_ptr    wp;							\
+    unsigned  i;							\
+    double    t;							\
+    TMP_DECL;								\
+									\
+    SPEED_RESTRICT_COND (s->size >= 1);					\
+									\
+    TMP_MARK;								\
+    SPEED_TMP_ALLOC_LIMBS (wp,	s->size, s->align_wp);			\
+									\
+    speed_operand_src (s, s->xp, s->size);				\
+    speed_operand_dst (s, wp, s->size);					\
+    speed_cache_fill (s);						\
+									\
+    speed_starttime ();							\
+    i = s->reps;							\
+    do									\
+      function (wp, NULL, s->xp, s->size, s->r);                        \
+    while (--i != 0);							\
+    t = speed_endtime ();						\
+									\
+    TMP_FREE;								\
+    return t;								\
+  }
+
+
 /* s->size controls the number of limbs in the input, s->r is the base, or
    decimal by default. */
 #define SPEED_ROUTINE_MPN_GET_STR(function)				\