Add support, testing, and timing of mpn_mul_3 and mpn_mul_4.

* gmp-impl.h (mpn_mul_3, mpn_mul_4): Declare.
* tests/tests.h (refmpn_mul_3, refmpn_mul_4): Declare.
* tests/try.c (param_init): Set things up for mpn_mul_3 and mpn_mul_4.
(choice_array): Likewise.
(call): Likewise.
* mpn/Makefile.am (nodist_EXTRA_libmpn_la_SOURCES):
Add mul_3.c and mul_4.
* mpn/asm-defs.m4: Define mul_3 and mul_4.
* tests/refmpn.c (refmpn_mul_N): New function.
(refmpn_mul_2): Remove old definition, call refmpn_mul_N.
(refmpn_mul_3, refmpn_mul_4): New functions.
* tune/common.c (speed_mpn_mul_3, speed_mpn_mul_4): New functions.
* tune/speed.h (speed_mpn_mul_3, speed_mpn_mul_4): Declare.
* tune/speed.c (routine): New entries for mpn_mul_2 and mpn_mul_3.

3 files changed, 55 insertions, 11 deletions

diff --git a/tests/devel/try.c b/tests/devel/try.c
index 772190bb2..b998cb5b8 100644
--- a/tests/devel/try.c
+++ b/tests/devel/try.c
@@ -561,6 +561,8 @@ validate_sqrtrem (void)
 #define TYPE_MUL_1C            8
 
 #define TYPE_MUL_2             9
+#define TYPE_MUL_3             92
+#define TYPE_MUL_4             93
 
 #define TYPE_ADDMUL_1         10
 #define TYPE_ADDMUL_1C        11
@@ -727,6 +729,16 @@ param_init (void)
   p->overlap = OVERLAP_NOT_SRC2;
   REFERENCE (refmpn_mul_2);
 
+  p = &param[TYPE_MUL_3];
+  COPY (TYPE_MUL_2);
+  p->msize = 3;
+  REFERENCE (refmpn_mul_3);
+
+  p = &param[TYPE_MUL_4];
+  COPY (TYPE_MUL_2);
+  p->msize = 4;
+  REFERENCE (refmpn_mul_4);
+
 
   p = &param[TYPE_ADDMUL_1];
   p->retval = 1;
@@ -1439,6 +1451,12 @@ const struct choice_t choice_array[] = {
 #if HAVE_NATIVE_mpn_mul_2
   { TRY(mpn_mul_2),      TYPE_MUL_2, 2 },
 #endif
+#if HAVE_NATIVE_mpn_mul_3
+  { TRY(mpn_mul_3),      TYPE_MUL_3, 3 },
+#endif
+#if HAVE_NATIVE_mpn_mul_4
+  { TRY(mpn_mul_4),      TYPE_MUL_4, 4 },
+#endif
 
   { TRY(mpn_rshift),     TYPE_RSHIFT },
   { TRY(mpn_lshift),     TYPE_LSHIFT },
@@ -1958,10 +1976,12 @@ call (struct each_t *e, tryfun_t function)
     break;
 
   case TYPE_MUL_2:
+  case TYPE_MUL_3:
+  case TYPE_MUL_4:
     if (size == 1)
       abort ();
     e->retval = CALLING_CONVENTIONS (function)
-      (e->d[0].p, e->s[0].p, size, e->s[1].p);
+      (e->d[0].p, e->s[0].p, size, multiplier_N);
     break;
 
   case TYPE_ADDMUL_2:
diff --git a/tests/refmpn.c b/tests/refmpn.c
index bcb887c38..b2ceaa481 100644
--- a/tests/refmpn.c
+++ b/tests/refmpn.c
@@ -1,8 +1,8 @@
 /* Reference mpn functions, designed to be simple, portable and independent
    of the normal gmp code.  Speed isn't a consideration.
 
-Copyright 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free
-Software Foundation, Inc.
+Copyright 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006,
+2007, 2008 Free Software Foundation, Inc.
 
 This file is part of the GNU MP Library.
 
@@ -790,15 +790,17 @@ refmpn_mul_1 (mp_ptr rp, mp_srcptr sp, mp_size_t size, mp_limb_t multiplier)
 
 
 mp_limb_t
-refmpn_mul_2 (mp_ptr dst, mp_srcptr src, mp_size_t size, mp_srcptr mult)
+refmpn_mul_N (mp_ptr dst, mp_srcptr src, mp_size_t size,
+	      mp_srcptr mult, mp_size_t msize)
 {
   mp_ptr     src_copy;
-  mp_limb_t  c;
+  mp_limb_t  ret;
+  mp_size_t  i;
 
   ASSERT (refmpn_overlap_fullonly_p (dst, src, size));
-  ASSERT (! refmpn_overlap_p (dst, size+1, mult, (mp_size_t) 2));
-  ASSERT (size >= 1);
-  ASSERT_MPN (mult, 2);
+  ASSERT (! refmpn_overlap_p (dst, size+msize-1, mult, msize));
+  ASSERT (size >= msize);
+  ASSERT_MPN (mult, msize);
 
   /* in case dst==src */
   src_copy = refmpn_malloc_limbs (size);
@@ -806,11 +808,29 @@ refmpn_mul_2 (mp_ptr dst, mp_srcptr src, mp_size_t size, mp_srcptr mult)
   src = src_copy;
 
   dst[size] = refmpn_mul_1 (dst, src, size, mult[0]);
-  c = refmpn_addmul_1 (dst+1, src, size, mult[1]);
+  for (i = 1; i < msize-1; i++)
+    dst[size+i] = refmpn_addmul_1 (dst+i, src, size, mult[i]);
+  ret = refmpn_addmul_1 (dst+i, src, size, mult[i]);
+
   free (src_copy);
-  return c;
+  return ret;
 }
 
+mp_limb_t
+refmpn_mul_2 (mp_ptr rp, mp_srcptr sp, mp_size_t size, mp_srcptr mult)
+{
+  return refmpn_mul_N (rp, sp, size, mult, (mp_size_t) 2);
+}
+mp_limb_t
+refmpn_mul_3 (mp_ptr rp, mp_srcptr sp, mp_size_t size, mp_srcptr mult)
+{
+  return refmpn_mul_N (rp, sp, size, mult, (mp_size_t) 3);
+}
+mp_limb_t
+refmpn_mul_4 (mp_ptr rp, mp_srcptr sp, mp_size_t size, mp_srcptr mult)
+{
+  return refmpn_mul_N (rp, sp, size, mult, (mp_size_t) 4);
+}
 
 #define AORSMUL_1C(operation_n)                                 \
   {                                                             \
diff --git a/tests/tests.h b/tests/tests.h
index 657ad2b74..c45667c11 100644
--- a/tests/tests.h
+++ b/tests/tests.h
@@ -294,7 +294,11 @@ mp_limb_t refmpn_mul_1 __GMP_PROTO ((mp_ptr wp, mp_srcptr xp, mp_size_t size,
 mp_limb_t refmpn_mul_1c __GMP_PROTO ((mp_ptr wp, mp_srcptr xp, mp_size_t size,
                                  mp_limb_t multiplier, mp_limb_t carry));
 mp_limb_t refmpn_mul_2 __GMP_PROTO ((mp_ptr dst, mp_srcptr src, mp_size_t size,
-                                mp_srcptr mult));
+				     mp_srcptr mult));
+mp_limb_t refmpn_mul_3 __GMP_PROTO ((mp_ptr dst, mp_srcptr src, mp_size_t size,
+				     mp_srcptr mult));
+mp_limb_t refmpn_mul_4 __GMP_PROTO ((mp_ptr dst, mp_srcptr src, mp_size_t size,
+				     mp_srcptr mult));
 
 void refmpn_mul_basecase __GMP_PROTO ((mp_ptr prodp,
                                   mp_srcptr up, mp_size_t usize,