Nailify. Revamp non-nail code.

1 files changed, 60 insertions, 33 deletions

diff --git a/mpn/generic/addmul_1.c b/mpn/generic/addmul_1.c
index d5c18957b..f33f50318 100644
--- a/mpn/generic/addmul_1.c
+++ b/mpn/generic/addmul_1.c
@@ -1,8 +1,7 @@
-/* mpn_addmul_1, mpn_submul_1 -- multiply the S1_SIZE long limb vector
-   pointed to by S1_PTR by S2_LIMB, add/subtract the S1_SIZE least
-   significant limbs of the product to/from the limb vector pointed to by
-   RES_PTR.  Return the most significant limb of the product, adjusted for
-   carry-out from the addition.
+/* mpn_addmul_1 -- multiply the N long limb vector pointed to by UP by VL,
+   add the N least significant limbs of the product to the limb vector
+   pointed to by RP.  Return the most significant limb of the product,
+   adjusted for carry-out from the addition.
 
 Copyright 1992, 1993, 1994, 1996, 2000, 2002 Free Software Foundation, Inc.
 
@@ -28,42 +27,70 @@ MA 02111-1307, USA. */
 #include "longlong.h"
 
 
+#if GMP_NAIL_BITS == 0
+
 mp_limb_t
-mpn_addmul_1 (register mp_ptr res_ptr,
-	      register mp_srcptr s1_ptr,
-	      mp_size_t s1_size,
-	      register mp_limb_t s2_limb)
+mpn_addmul_1 (mp_ptr rp, mp_srcptr up, mp_size_t n, mp_limb_t vl)
 {
-  register mp_limb_t cy_limb;
-  register mp_size_t j;
-  register mp_limb_t prod_high, prod_low;
-  register mp_limb_t x;
-
-  ASSERT (s1_size >= 1);
-  ASSERT (MPN_SAME_OR_SEPARATE_P (res_ptr, s1_ptr, s1_size));
-
-  /* The loop counter and index J goes from -SIZE to -1.  This way
-     the loop becomes faster.  */
-  j = -s1_size;
+  mp_limb_t ul, cl, hpl, lpl, rl;
 
-  /* Offset the base pointers to compensate for the negative indices.  */
-  res_ptr -= j;
-  s1_ptr -= j;
+  ASSERT (n >= 1);
+  ASSERT (MPN_SAME_OR_SEPARATE_P (rp, up, n));
 
-  cy_limb = 0;
+  cl = 0;
   do
     {
-      umul_ppmm (prod_high, prod_low, s1_ptr[j], s2_limb);
+      ul = *up++;
+      umul_ppmm (hpl, lpl, ul, vl);
 
-      prod_low += cy_limb;
-      cy_limb = (prod_low < cy_limb) + prod_high;
+      lpl += cl;
+      cl = (lpl < cl) + hpl;
 
-      x = res_ptr[j];
-      prod_low = x + prod_low;
-      cy_limb += prod_low < x;;
-      res_ptr[j] = prod_low;
+      rl = *rp;
+      lpl = rl + lpl;
+      cl += lpl < rl;
+      *rp++ = lpl;
     }
-  while (++j != 0);
+  while (--n != 0);
 
-  return cy_limb;
+  return cl;
 }
+
+#endif
+
+#if GMP_NAIL_BITS == 1
+you lose
+#endif
+
+#if GMP_NAIL_BITS >= 2
+
+mp_limb_t
+mpn_addmul_1 (mp_ptr rp, mp_srcptr up, mp_size_t n, mp_limb_t vl)
+{
+  mp_limb_t shifted_vl, ul, rl, lpl, hpl, prev_hpl, xw, cl, xl;
+
+  ASSERT (vl >> GMP_NAIL_BITS == 0);
+  ASSERT (n >= 1);
+  ASSERT (MPN_SAME_OR_SEPARATE_P (rp, up, n));
+
+  shifted_vl = vl << GMP_NAIL_BITS;
+  cl = 0;
+  prev_hpl = 0;
+  do
+    {
+      ul = *up++;
+      rl = *rp;
+      umul_ppmm (hpl, lpl, ul, shifted_vl);
+      lpl >>= GMP_NAIL_BITS;
+      xw = prev_hpl + lpl + rl + cl;
+      cl = xw >> GMP_NUMB_BITS;
+      xl = xw & GMP_NUMB_MASK;
+      *rp++ = xl;
+      prev_hpl = hpl;
+    }
+  while (--n != 0);
+
+  return prev_hpl + cl;
+}
+
+#endif