[patch] abseil-cpp: Use __builtin_ctzg and __builtin_clzg in the implementations

file_934661.txt (text/plain), 1.97 KB, created by Florian Albrechtskirchinger on 2024-07-01 12:29:39 UTC

From c1e1b47d989978cde8c5a2a219df425b785a0c47 Mon Sep 17 00:00:00 2001
From: Derek Mauro <[e-mail]>
Date: Fri, 3 May 2024 09:14:22 -0700
Subject: [PATCH] Use __builtin_ctzg and __builtin_clzg in the implementations
 of CountTrailingZeroesNonzero16 and CountLeadingZeroes16 when they are
 available.

GCC 14 and Clang 19 adds these new builtins. The g-suffix is for
"generic". The s-suffix on __builtin_ctzs and __builtin_clzs is for
"short". GCC never implemented the short versions and #1664 reports
GCC 14 (pre-release) gives an error here, although this may be a
pre-release bug.

Fixes #1664

PiperOrigin-RevId: 630408249
Change-Id: I4aedcc82b85430f50d025f8eb1cab089c6fcd1bc
---
 absl/numeric/internal/bits.h | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/absl/numeric/internal/bits.h b/absl/numeric/internal/bits.h
index bfef06bce1f..0917464d6ad 100644
--- a/absl/numeric/internal/bits.h
+++ b/absl/numeric/internal/bits.h
@@ -167,7 +167,9 @@ CountLeadingZeroes32(uint32_t x) {
 
 ABSL_ATTRIBUTE_ALWAYS_INLINE ABSL_INTERNAL_CONSTEXPR_CLZ inline int
 CountLeadingZeroes16(uint16_t x) {
-#if ABSL_HAVE_BUILTIN(__builtin_clzs)
+#if ABSL_HAVE_BUILTIN(__builtin_clzg)
+  return x == 0 ? 16 : __builtin_clzg(x);
+#elif ABSL_HAVE_BUILTIN(__builtin_clzs)
   static_assert(sizeof(unsigned short) == sizeof(x),  // NOLINT(runtime/int)
                 "__builtin_clzs does not take 16-bit arg");
   return x == 0 ? 16 : __builtin_clzs(x);
@@ -303,7 +305,9 @@ CountTrailingZeroesNonzero64(uint64_t x) {
 
 ABSL_ATTRIBUTE_ALWAYS_INLINE ABSL_INTERNAL_CONSTEXPR_CTZ inline int
 CountTrailingZeroesNonzero16(uint16_t x) {
-#if ABSL_HAVE_BUILTIN(__builtin_ctzs)
+#if ABSL_HAVE_BUILTIN(__builtin_ctzg)
+  return __builtin_ctzg(x);
+#elif ABSL_HAVE_BUILTIN(__builtin_ctzs)
   static_assert(sizeof(unsigned short) == sizeof(x),  // NOLINT(runtime/int)
                 "__builtin_ctzs does not take 16-bit arg");
   return __builtin_ctzs(x);