bug#22689: [PATCH] Add logcount

[Mark Oteiza]

On 30/09/17 at 04:59pm, Eli Zaretskii wrote:
> I only see that symbol in the GCC sources, which I don't think are
> relevant here.
> 
> If you drop the __POPCNT__ part, does the code still work for you?

Yes.

> > > > +#else  /* HAVE_BUILTIN_POPCOUNTLL */
> > > > +  if (XINT (value) <= UINT_MAX)
> > > > +    XSETINT (res, bitcount32 (XUINT (value)));
> > > > +  else if (XINT (value) <= ULONG_MAX)
> > > > +    XSETINT (res, bitcount64 (XUINT (value)));
> > > 
> > > The comparisons against Uxxx_MAX seem to assume that VALUE is
> > > unsigned, but that's not guaranteed, right?
> > 
> > True.  Should I instead be doing
> > 
> >   XINT (value) <= xxx_MAX &&
> >   XINT (value) >= xxx_MIN
> > 
> > or might there be a better check?  For negative values popcount
> > typically counts ones of the two's complement
> 
> I'd just assign to an unsigned temporary, IIUC the semantics.
> 
> Btw, on Windows, a long is a 32-bit type, so I think we need
> to check against ULONG_LONG_MAX as well.

I tried implementing your comments (and added a test).  I don't know
what to do in the #else for ULONG_LONG_MAX.

diff --git a/src/data.c b/src/data.c
index e070be6c20..8b3866151d 100644
--- a/src/data.c
+++ b/src/data.c
@@ -3069,6 +3069,66 @@ usage: (logxor &rest INTS-OR-MARKERS)  */)
   return arith_driver (Alogxor, nargs, args);
 }
 
+#if GNUC_PREREQ (4, 1, 0)
+#define HAVE_BUILTIN_POPCOUNTLL
+#endif
+
+#ifndef HAVE_BUILTIN_POPCOUNTLL
+static uint32_t
+logcount32 (uint32_t b)
+{
+  b -= (b >> 1) & 0x55555555;
+  b = (b & 0x33333333) + ((b >> 2) & 0x33333333);
+  b = (b + (b >> 4)) & 0x0f0f0f0f;
+  return (b * 0x01010101) >> 24;
+}  
+
+static uint64_t
+logcount64 (uint64_t b)
+{
+  b -= (b >> 1) & 0x5555555555555555ULL;
+  b = (b & 0x3333333333333333ULL) + ((b >> 2) & 0x3333333333333333ULL);
+  b = (b + (b >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
+  return (b * 0x0101010101010101ULL) >> 56;
+}
+#endif /* HAVE_BUILTIN_POPCOUNTLL */
+
+DEFUN ("logcount", Flogcount, Slogcount, 1, 1, 0,
+       doc: /* Return population count of VALUE.
+If VALUE is negative, the count is of its two's complement representation.  */)
+  (register Lisp_Object value)
+{
+  Lisp_Object res;
+  EMACS_UINT v;
+
+  CHECK_NUMBER (value);
+
+  v = XUINT (value);
+#ifdef HAVE_BUILTIN_POPCOUNTLL
+  if (v <= UINT_MAX)
+    XSETINT (res, __builtin_popcount (v));
+  else if (v <= ULONG_MAX)
+    XSETINT (res, __builtin_popcountl (v));
+  else if (v <= ULONG_LONG_MAX)
+    XSETINT (res, __builtin_popcountll (v));
+#else  /* HAVE_BUILTIN_POPCOUNTLL */
+  if (v <= UINT_MAX)
+    XSETINT (res, logcount32 (v));
+  else if (v <= ULONG_MAX)
+    XSETINT (res, logcount64 (v));
+#endif /* HAVE_BUILTIN_POPCOUNTLL */
+  else
+    {
+      unsigned int count;
+      for (count = 0; v; count++)
+        {
+          v &= v - 1;
+        }
+      XSETINT (res, v);
+    }
+  return res;
+}
+
 static Lisp_Object
 ash_lsh_impl (Lisp_Object value, Lisp_Object count, bool lsh)
 {
@@ -3856,6 +3916,7 @@ syms_of_data (void)
   defsubr (&Slogand);
   defsubr (&Slogior);
   defsubr (&Slogxor);
+  defsubr (&Slogcount);
   defsubr (&Slsh);
   defsubr (&Sash);
   defsubr (&Sadd1);
diff --git a/test/src/data-tests.el b/test/src/data-tests.el
index 8de8c145d4..d1154cc5c4 100644
--- a/test/src/data-tests.el
+++ b/test/src/data-tests.el
@@ -107,6 +107,19 @@
   (should (isnan (min 1.0 0.0e+NaN)))
   (should (isnan (min 1.0 0.0e+NaN 1.1))))
 
+(defun data-tests-popcnt (byte)
+  "Calculate the Hamming weight of BYTE."
+  (setq byte (- byte (logand (lsh byte -1) #x55555555)))
+  (setq byte (+ (logand byte #x33333333) (logand (lsh byte -2) #x33333333)))
+  (lsh (* (logand (+ byte (lsh byte -4)) #x0f0f0f0f) #x01010101) -24))
+
+(ert-deftest data-tests-logcount ()
+  (should (cl-loop for n in (number-sequence 0 255)
+                   always (= (logcount n) (data-tests-popcnt n))))
+  ;; https://oeis.org/A000120
+  (should (= 11 (logcount 9727)))
+  (should (= 8 (logcount 9999))))
+
 ;; Bool vector tests.  Compactly represent bool vectors as hex
 ;; strings.