From f32cf51e1f84ddcfc34be1ab9fe096e5b3f61631 Mon Sep 17 00:00:00 2001
From: Karl Williamson <khw@cpan.org>
Date: Tue, 20 Jul 2021 10:41:39 -0600
Subject: [PATCH] Use ffs(), ffsl() for lsb_pos, if available

ffs() is a POSIX function that serves to find the lsb of a word.  It
should be faster than anything we hand-roll, but slower than any
available intrinsics that the previous commits use; so this commit looks
for it only if they aren't available.

Some platforms have an equivalent ffsl() for longs, which we use if the
word is long-sized.
---
 inline.h | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/inline.h b/inline.h
index be5ce02895be..9cc54c865e6c 100644
--- a/inline.h
+++ b/inline.h
@@ -731,6 +731,21 @@ Perl_is_utf8_invariant_string_loc(const U8* const s, STRLEN len, const U8 ** ep)
 #  endif
 #endif
 
+#if U32SIZE == INTSIZE && defined(HAS_FFS)
+#  define PERL_FFS32 ffs
+#elif U32SIZE == LONGSIZE && defined(HAS_FFSL)
+#  define PERL_FFS32 ffsl
+#endif
+#ifdef HAS_QUAD
+#  if U64SIZE == INTSIZE && defined(HAS_FFS)
+#    define PERL_FFS64 ffs
+#  elif U64SIZE == LONGSIZE && defined(HAS_FFSL)
+#    define PERL_FFS64 ffsl
+#  elif U64SIZE == LONGLONGSIZE && defined(HAS_FFSLL)
+#    define PERL_FFS64 ffsll
+#  endif
+#endif
+
 #if defined(_MSC_VER) && _MSC_VER >= 1400
 #  include <intrin.h>
 #  pragma intrinsic(_BitScanForward)
@@ -771,6 +786,10 @@ Perl_lsbit_pos64(U64 word)
         return (unsigned)index;
     }
 
+#  elif defined(PERL_FFS64)
+
+    return PERL_FFS64(word) - 1;    /* ffs() returns bit position indexed from 1 */
+
 #  else
 
     /* Here, we didn't find a fast method for finding the lsb.  Fall back to
@@ -823,6 +842,10 @@ Perl_lsbit_pos32(U32 word)
         return (unsigned)index;
     }
 
+#elif defined(PERL_FFS32)
+
+    return PERL_FFS32(word) - 1;
+
 #else
 
     return single_1bit_pos32(word & (~word + 1));