On Wed, 11 Aug 2021, Sebastian Huber wrote:
> 64-bit atomic operations. Allow targets to override the default type
> size with the new TARGET_GCOV_TYPE_SIZE target macro.
Hook, not macro.
> diff --git a/gcc/c-family/c-cppbuiltin.c b/gcc/c-family/c-cppbuiltin.c
> index f79f939bd10f..e85b60c79f49 100644
> --- a/gcc/c-family/c-cppbuiltin.c
> +++ b/gcc/c-family/c-cppbuiltin.c
> @@ -1450,6 +1450,8 @@ c_cpp_builtins (cpp_reader *pfile)
> /* For libgcov. */
> builtin_define_with_int_value ("__LIBGCC_VTABLE_USES_DESCRIPTORS__",
> TARGET_VTABLE_USES_DESCRIPTORS);
> + builtin_define_with_int_value ("__LIBGCC_GCOV_TYPE_SIZE",
> + TARGET_GCOV_TYPE_SIZE);
The TARGET_* macros used to initialize targetm may only be defined to
their final values in the architecture-specific .c file that actually
defines targetm. All other files should access the hook via targetm, not
the TARGET_* macros. (TARGET_VTABLE_USES_DESCRIPTORS in the diff context
is a target macro, not a hook at all.)
> +DEFHOOKPOD
> +(gcov_type_size,
> + "The gcov type size in bits. This type is used for example for counters\n\
> +incremented by profiling and code-coverage events. The default value is
> 64,\n\
> +if the type size of long long is greater than 32, otherwise the default\n\
> +value is 32. A 64-bit type is recommended to avoid overflows of the\n\
> +counters. If the @option{-fprofile-update=atomic} is used, then the\n\
> +counters are incremented using atomic operations. Targets not supporting\n\
> +64-bit atomic operations may override the default value and request a
> 32-bit\n\
> +type.",
> + HOST_WIDE_INT, (LONG_LONG_TYPE_SIZE > 32 ? 64 : 32))
LONG_LONG_TYPE_SIZE may depend on command-line options passed to the
compiler (it does for AVR). The hook thus needs to be a function
returning the desired size.
--
Joseph S. Myers
jos...@codesourcery.com
