* [PATCH] libstdc++: Stop defining _GLIBCXX_ASSERTIONS in floating_to_chars.cc
@ 2022-04-14 19:47 Patrick Palka
2022-04-19 10:42 ` Jonathan Wakely
0 siblings, 1 reply; 2+ messages in thread
From: Patrick Palka @ 2022-04-14 19:47 UTC (permalink / raw)
To: gcc-patches; +Cc: libstdc++, Patrick Palka
Assertions were originally enabled in the compiled-in floating-point
std::to_chars implementation to help shake out any bugs, but they
apparently impose a significant performance penalty, in particular for
the hex formatting which is around 25% slower with assertions enabled.
This seems too high of a cost for unconditionally enabling them.
The newly added calls to __builtin_unreachable work around the compiler
no longer knowing that the set of valid values of 'fmt' is limited (which
was previously upheld by an assert).
Tested on x86_64-pc-linux-gnu, does this look OK for trunk?
libstdc++-v3/ChangeLog:
* src/c++17/floating_to_chars.cc: Don't define
_GLIBCXX_ASSERTIONS.
(__floating_to_chars_shortest): Add __builtin_unreachable calls to
squelch false-positive -Wmaybe-uninitialized and -Wreturn-type
warnings.
(__floating_to_chars_precision): Likewise.
---
libstdc++-v3/src/c++17/floating_to_chars.cc | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/libstdc++-v3/src/c++17/floating_to_chars.cc b/libstdc++-v3/src/c++17/floating_to_chars.cc
index 66bd457cbe2..4599d68a39c 100644
--- a/libstdc++-v3/src/c++17/floating_to_chars.cc
+++ b/libstdc++-v3/src/c++17/floating_to_chars.cc
@@ -22,9 +22,6 @@
// see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
// <http://www.gnu.org/licenses/>.
-// Activate __glibcxx_assert within this file to shake out any bugs.
-#define _GLIBCXX_ASSERTIONS 1
-
#include <charconv>
#include <bit>
@@ -1114,6 +1111,7 @@ template<typename T>
}
__glibcxx_assert(false);
+ __builtin_unreachable();
}
template<typename T>
@@ -1202,6 +1200,8 @@ template<typename T>
effective_precision = min(precision, max_eff_scientific_precision);
output_specifier = "%.*Lg";
}
+ else
+ __builtin_unreachable();
const int excess_precision = (fmt != chars_format::general
? precision - effective_precision : 0);
@@ -1234,6 +1234,8 @@ template<typename T>
output_length_upper_bound = sign + strlen("0");
output_length_upper_bound += sizeof(radix) + effective_precision;
}
+ else
+ __builtin_unreachable();
// Do the sprintf into the local buffer.
char buffer[output_length_upper_bound+1];
@@ -1570,6 +1572,7 @@ template<typename T>
}
__glibcxx_assert(false);
+ __builtin_unreachable();
}
// Define the overloads for float.
--
2.36.0.rc2.10.g1ac7422e39
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] libstdc++: Stop defining _GLIBCXX_ASSERTIONS in floating_to_chars.cc
2022-04-14 19:47 [PATCH] libstdc++: Stop defining _GLIBCXX_ASSERTIONS in floating_to_chars.cc Patrick Palka
@ 2022-04-19 10:42 ` Jonathan Wakely
0 siblings, 0 replies; 2+ messages in thread
From: Jonathan Wakely @ 2022-04-19 10:42 UTC (permalink / raw)
To: Patrick Palka; +Cc: gcc Patches, libstdc++
On Thu, 14 Apr 2022 at 20:48, Patrick Palka via Libstdc++
<libstdc++@gcc.gnu.org> wrote:
>
> Assertions were originally enabled in the compiled-in floating-point
> std::to_chars implementation to help shake out any bugs, but they
> apparently impose a significant performance penalty, in particular for
> the hex formatting which is around 25% slower with assertions enabled.
> This seems too high of a cost for unconditionally enabling them.
>
> The newly added calls to __builtin_unreachable work around the compiler
> no longer knowing that the set of valid values of 'fmt' is limited (which
> was previously upheld by an assert).
>
> Tested on x86_64-pc-linux-gnu, does this look OK for trunk?
OK, thanks.
>
> libstdc++-v3/ChangeLog:
>
> * src/c++17/floating_to_chars.cc: Don't define
> _GLIBCXX_ASSERTIONS.
> (__floating_to_chars_shortest): Add __builtin_unreachable calls to
> squelch false-positive -Wmaybe-uninitialized and -Wreturn-type
> warnings.
> (__floating_to_chars_precision): Likewise.
> ---
> libstdc++-v3/src/c++17/floating_to_chars.cc | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/libstdc++-v3/src/c++17/floating_to_chars.cc b/libstdc++-v3/src/c++17/floating_to_chars.cc
> index 66bd457cbe2..4599d68a39c 100644
> --- a/libstdc++-v3/src/c++17/floating_to_chars.cc
> +++ b/libstdc++-v3/src/c++17/floating_to_chars.cc
> @@ -22,9 +22,6 @@
> // see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
> // <http://www.gnu.org/licenses/>.
>
> -// Activate __glibcxx_assert within this file to shake out any bugs.
> -#define _GLIBCXX_ASSERTIONS 1
> -
> #include <charconv>
>
> #include <bit>
> @@ -1114,6 +1111,7 @@ template<typename T>
> }
>
> __glibcxx_assert(false);
> + __builtin_unreachable();
> }
>
> template<typename T>
> @@ -1202,6 +1200,8 @@ template<typename T>
> effective_precision = min(precision, max_eff_scientific_precision);
> output_specifier = "%.*Lg";
> }
> + else
> + __builtin_unreachable();
> const int excess_precision = (fmt != chars_format::general
> ? precision - effective_precision : 0);
>
> @@ -1234,6 +1234,8 @@ template<typename T>
> output_length_upper_bound = sign + strlen("0");
> output_length_upper_bound += sizeof(radix) + effective_precision;
> }
> + else
> + __builtin_unreachable();
>
> // Do the sprintf into the local buffer.
> char buffer[output_length_upper_bound+1];
> @@ -1570,6 +1572,7 @@ template<typename T>
> }
>
> __glibcxx_assert(false);
> + __builtin_unreachable();
> }
>
> // Define the overloads for float.
> --
> 2.36.0.rc2.10.g1ac7422e39
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2022-04-19 10:42 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-04-14 19:47 [PATCH] libstdc++: Stop defining _GLIBCXX_ASSERTIONS in floating_to_chars.cc Patrick Palka
2022-04-19 10:42 ` Jonathan Wakely
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).