depends: modify FastFixedDtoa optimisation flags

This fixes a non-determinism issue in the asm produced for
this function when cross-compiling on x86_64 and aarch64 for
the arm-linux-gnueabihf HOST.

Related to #21194.
This commit is contained in:
fanquake committed 2022-07-19 12:12:26 +01:00
1 parent 47dad42833
commit c32fa85909
2 files changed
+22

No files matched your search

@@ -0,0 +1,20 @@
Modify the optimisation flags for FastFixedDtoa.
This fixes a non-determinism issue in the asm produced for
this function when cross-compiling on x86_64 and aarch64 for
the arm-linux-gnueabihf HOST.
--- a/qtbase/src/3rdparty/double-conversion/fixed-dtoa.h
+++ b/qtbase/src/3rdparty/double-conversion/fixed-dtoa.h
@@ -48,9 +48,12 @@ namespace double_conversion {
//
// This method only works for some parameters. If it can't handle the input it
// returns false. The output is null-terminated when the function succeeds.
+#pragma GCC push_options
+#pragma GCC optimize ("-O1")
bool FastFixedDtoa(double v, int fractional_count,
Vector<char> buffer, int* length, int* decimal_point);
+#pragma GCC pop_options
} // namespace double_conversion
#endif // DOUBLE_CONVERSION_FIXED_DTOA_H_