Use __attribute__((target(...))) for SSE4.2 CRC-32C support.

Presently, we check for compiler support for the required
intrinsics both with and without the -msse4.2 compiler flag, and
then depending on the results of those checks, we pick which files
to compile with which flags.  This is tedious and complicated, and
it results in unsustainable coding patterns such as separate files
for each portion of code that may need to be built with different
compiler flags.

This commit makes use of the newly-added support for
__attribute__((target(...))) in the SSE4.2 CRC-32C code.  This
simplifies both the configure-time checks and the build scripts,
and it allows us to place the functions that use the intrinsics in
files that we otherwise do not want to build with special CPU
instructions (although this commit refrains from doing so).  This
is also preparatory work for a proposed follow-up commit that will
further optimize the CRC-32C code with AVX-512 instructions.

While at it, this commit modifies meson's checks for SSE4.2 CRC
support to be the same as autoconf's.  meson was choosing whether
to use a runtime check based purely on whether -msse4.2 is
required, while autoconf has long checked for the __SSE4_2__
preprocessor symbol to decide.  meson's previous approach seems to
work just fine, but this change avoids needing to build multiple
test programs and to keep track of whether to actually use
pg_attribute_target().

Ideally we'd use __attribute__((target(...))) for ARMv8 CRC
support, too, but there's little point in doing so because until
clang 16, using the ARM intrinsics still requires special compiler
flags.  Perhaps we can re-evaluate this decision after some time
has passed.

Author: Raghuveer Devulapalli
Discussion: https://postgr.es/m/PH8PR11MB8286BE735A463468415D46B5FB5C2%40PH8PR11MB8286.namprd11.prod.outlook.com

1 files changed, 15 insertions, 7 deletions

diff --git a/meson.build b/meson.build
index ff3848b1d85..6bc3bb51dc0 100644
--- a/meson.build
+++ b/meson.build
@@ -2211,14 +2211,19 @@ endif
 # If we are targeting a processor that has Intel SSE 4.2 instructions, we can
 # use the special CRC instructions for calculating CRC-32C. If we're not
 # targeting such a processor, but we can nevertheless produce code that uses
-# the SSE intrinsics, perhaps with some extra CFLAGS, compile both
-# implementations and select which one to use at runtime, depending on whether
-# SSE 4.2 is supported by the processor we're running on.
+# the SSE intrinsics, compile both implementations and select which one to use
+# at runtime, depending on whether SSE 4.2 is supported by the processor we're
+# running on.
 #
 # Similarly, if we are targeting an ARM processor that has the CRC
 # instructions that are part of the ARMv8 CRC Extension, use them. And if
 # we're not targeting such a processor, but can nevertheless produce code that
 # uses the CRC instructions, compile both, and select at runtime.
+#
+# Note that we do not use __attribute__((target("..."))) for the ARM CRC
+# instructions because until clang 16, using the ARM intrinsics still requires
+# special -march flags.  Perhaps we can re-evaluate this decision after some
+# time has passed.
 ###############################################################
 
 have_optimized_crc = false
@@ -2234,6 +2239,9 @@ if host_cpu == 'x86' or host_cpu == 'x86_64'
     prog = '''
 #include <nmmintrin.h>
 
+#if defined(__has_attribute) && __has_attribute (target)
+__attribute__((target("sse4.2")))
+#endif
 int main(void)
 {
     unsigned int crc = 0;
@@ -2244,16 +2252,16 @@ int main(void)
 }
 '''
 
-    if cc.links(prog, name: '_mm_crc32_u8 and _mm_crc32_u32 without -msse4.2',
+    if not cc.links(prog, name: '_mm_crc32_u8 and _mm_crc32_u32',
           args: test_c_args)
+      # Do not use Intel SSE 4.2
+    elif (cc.get_define('__SSE4_2__') != '')
       # Use Intel SSE 4.2 unconditionally.
       cdata.set('USE_SSE42_CRC32C', 1)
       have_optimized_crc = true
-    elif cc.links(prog, name: '_mm_crc32_u8 and _mm_crc32_u32 with -msse4.2',
-          args: test_c_args + ['-msse4.2'])
+    else
       # Use Intel SSE 4.2, with runtime check. The CPUID instruction is needed for
       # the runtime check.
-      cflags_crc += '-msse4.2'
       cdata.set('USE_SSE42_CRC32C', false)
       cdata.set('USE_SSE42_CRC32C_WITH_RUNTIME_CHECK', 1)
       have_optimized_crc = true