X-Git-Url: https://git.distorted.org.uk/~mdw/catacomb/blobdiff_plain/850dc2722b8ca17302d89c483b63693fa2fe9ef5..8f2287ef5c05d496fcb9b012629af007fe56f897:/math/mpx.c

diff --git a/math/mpx.c b/math/mpx.c
index 18baf2f2..741cd9dc 100644
--- a/math/mpx.c
+++ b/math/mpx.c
@@ -545,15 +545,21 @@ MPX_SHIFTOP(lsr, {
   size_t nr = MPW_BITS - nb;
   mpw w;
 
-  av += nw;
-  w = av < avl ? *av++ : 0;
-  while (av < avl) {
-    mpw t;
-    if (dv >= dvl) goto done;
-    t = *av++;
-    *dv++ = MPW((w >> nb) | (t << nr));
-    w = t;
+  if (nw >= avl - av)
+    w = 0;
+  else {
+    av += nw;
+    w = *av++;
+
+    while (av < avl) {
+      mpw t;
+      if (dv >= dvl) goto done;
+      t = *av++;
+      *dv++ = MPW((w >> nb) | (t << nr));
+      w = t;
+    }
   }
+
   if (dv < dvl) {
     *dv++ = MPW(w >> nb);
     MPX_ZERO(dv, dvl);
@@ -812,7 +818,7 @@ void mpx_usub(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
 
 void mpx_usubn(mpw *dv, mpw *dvl, mpw n) { MPX_USUBN(dv, dvl, n); }
 
-/* --- @mpx_uaddnlsl@ --- *
+/* --- @mpx_usubnlsl@ --- *
  *
  * Arguments:	@mpw *dv, *dvl@ = destination and first argument vector
  *		@mpw a@ = second argument
@@ -917,19 +923,25 @@ static void simple_umul(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl,
 
 #if CPUFAM_X86
   MAYBE_UMUL4(x86_sse2)
+  MAYBE_UMUL4(x86_avx)
 #endif
 
 #if CPUFAM_AMD64
   MAYBE_UMUL4(amd64_sse2)
+  MAYBE_UMUL4(amd64_avx)
 #endif
 
 static mpx_umul__functype *pick_umul(void)
 {
 #if CPUFAM_X86
+  DISPATCH_PICK_COND(mpx_umul, maybe_umul4_x86_avx,
+		     cpu_feature_p(CPUFEAT_X86_AVX));
   DISPATCH_PICK_COND(mpx_umul, maybe_umul4_x86_sse2,
 		     cpu_feature_p(CPUFEAT_X86_SSE2));
 #endif
 #if CPUFAM_AMD64
+  DISPATCH_PICK_COND(mpx_umul, maybe_umul4_amd64_avx,
+		     cpu_feature_p(CPUFEAT_X86_AVX));
   DISPATCH_PICK_COND(mpx_umul, maybe_umul4_amd64_sse2,
 		     cpu_feature_p(CPUFEAT_X86_SSE2));
 #endif