From 74469d71e6792fccdfb17f2913aaabafb7094a89 Mon Sep 17 00:00:00 2001
From: Niels M?ller <nisse@lysator.liu.se>
Date: Wed, 4 Sep 2019 20:09:20 +0200
Subject: Speed support for mpn_hgcd2.

* tune/speed.h (SPEED_ROUTINE_MPN_HGCD2): New macro.
* tune/common.c (speed_mpn_hgcd2): New function.
* tune/speed.c (routine): Add mpn_hgcd2.
---
 tune/common.c |  6 ++++++
 tune/speed.c  |  1 +
 tune/speed.h  | 35 +++++++++++++++++++++++++++++++++++
 3 files changed, 42 insertions(+)

(limited to 'tune')

diff --git a/tune/common.c b/tune/common.c
index 0b3f491b7..a918a67a4 100644
--- a/tune/common.c
+++ b/tune/common.c
@@ -1633,6 +1633,12 @@ speed_mpn_matrix22_mul (struct speed_params *s)
   return t;
 }
 
+double
+speed_mpn_hgcd2 (struct speed_params *s)
+{
+  SPEED_ROUTINE_MPN_HGCD2 (mpn_hgcd2);
+}
+
 double
 speed_mpn_hgcd (struct speed_params *s)
 {
diff --git a/tune/speed.c b/tune/speed.c
index a013a24b7..4220d6be5 100644
--- a/tune/speed.c
+++ b/tune/speed.c
@@ -285,6 +285,7 @@ const struct routine_t {
 
   { "mpn_matrix22_mul",  speed_mpn_matrix22_mul     },
 
+  { "mpn_hgcd2",         speed_mpn_hgcd2, FLAG_NODATA },
   { "mpn_hgcd",          speed_mpn_hgcd             },
   { "mpn_hgcd_lehmer",   speed_mpn_hgcd_lehmer      },
   { "mpn_hgcd_appr",     speed_mpn_hgcd_appr        },
diff --git a/tune/speed.h b/tune/speed.h
index 25b154203..044bc9c2b 100644
--- a/tune/speed.h
+++ b/tune/speed.h
@@ -214,6 +214,7 @@ double speed_mpn_div_qr_2n (struct speed_params *);
 double speed_mpn_div_qr_2u (struct speed_params *);
 double speed_mpn_fib2_ui (struct speed_params *);
 double speed_mpn_matrix22_mul (struct speed_params *);
+double speed_mpn_hgcd2 (struct speed_params *);
 double speed_mpn_hgcd (struct speed_params *);
 double speed_mpn_hgcd_lehmer (struct speed_params *);
 double speed_mpn_hgcd_appr (struct speed_params *);
@@ -2843,6 +2844,40 @@ int speed_routine_count_zeros_setup (struct speed_params *, mp_ptr, int, int);
      },									\
      function (px[j-1], py[j-1], 0))
 
+#define SPEED_ROUTINE_MPN_HGCD2(function)				\
+  {									\
+    unsigned   i, j;							\
+    struct hgcd_matrix1 m = {{{0,0},{0,0}}};				\
+    double     t;							\
+									\
+    speed_operand_src (s, s->xp_block, SPEED_BLOCK_SIZE);		\
+    speed_operand_src (s, s->yp_block, SPEED_BLOCK_SIZE);		\
+    speed_cache_fill (s);						\
+									\
+    speed_starttime ();							\
+    i = s->reps;							\
+    mp_limb_t chain = 0;						\
+    do									\
+      {									\
+	for (j = 0; j < SPEED_BLOCK_SIZE; j+= 2)			\
+	  {								\
+	    /* randomized but successively dependent */			\
+	    function (s->xp_block[j] | GMP_NUMB_HIGHBIT,		\
+		      s->xp_block[j+1] + chain,				\
+		      s->yp_block[j] | GMP_NUMB_HIGHBIT,		\
+		      s->yp_block[j+1], &m);				\
+	    chain += m.u[0][0];						\
+	  }								\
+      }									\
+    while (--i != 0);							\
+    t = speed_endtime ();						\
+									\
+    /* make sure the compiler won't optimize away chain */		\
+    noop_1 (chain);							\
+									\
+    s->time_divisor = SPEED_BLOCK_SIZE / 2;				\
+    return t;								\
+  }
 
 #define SPEED_ROUTINE_MPN_HGCD_CALL(func, itchfunc)			\
   {									\
-- 
cgit v1.2.1