Package: gegl Version: 1:0.4.72-2 Severity: important Tags: patch Forwarded: https://gitlab.gnome.org/GNOME/gegl/-/issues/480
Dear Maintainer, gegl 1:0.4.72-2 fails to build from source on ppc64el when built with GCC 15 due to format_sensing test failure: ERROR: ../tests/simple/format-sensing.c:166:test_format_sensing: assertion failed: (babl_format_has_alpha (format)) FAIL: tests/simple/format-sensing 1/1 FAIL 0.05s (exit status: 133 or signal 5 SIGTRAP) GCC 15 on ppc64el miscompiles the union-based bitwise sign-clearing helper (int_fabsf) in operations/common/opacity.c, gegl/gegl-math.h, and gegl/buffer/gegl-sampler-cubic.c. On ppc64el, the mask load is corrupted, causing int_fabsf(value - 1.0f) <= EPSILON to evaluate to true for value = 0.5f. This bypasses the format conversion in gegl:opacity and breaks format_sensing. Replacing the union type-punning hacks with __builtin_fabsf / __builtin_fabs fixes the issue and compiles directly to the hardware 'fabs' instruction on PowerPC. Upstream issue: https://gitlab.gnome.org/GNOME/gegl/-/issues/480 Ubuntu tracking bug: https://bugs.launchpad.net/ubuntu/+source/gegl/+bug/2168951 Please find attached the patch. Regards, Viktor Pashaiev
Description: Use compiler builtins for absolute float/double in gegl-math and opacity Replace union bitwise float sign-bit clearing hacks with __builtin_fabsf and __builtin_fabs. The union type-punning hack is miscompiled by GCC 15 on ppc64el, resulting in format_sensing test failures in gegl:opacity. Author: Viktor Pashaiev <[email protected]> Bug-Ubuntu: https://bugs.launchpad.net/ubuntu/+source/gegl/+bug/2168951 Forwarded: https://gitlab.gnome.org/GNOME/gegl/-/issues/480 Last-Update: 2026-10-12 --- a/gegl/gegl-math.h +++ b/gegl/gegl-math.h @@ -25,22 +25,12 @@ static inline float gegl_fabsf (float x) { - union { - float f; - guint32 i; - } u = {x}; - u.i &= 0x7fffffff; - return u.f; + return __builtin_fabsf (x); } static inline double gegl_fabs (double x) { - union { - double d; - guint64 i; - } u = {x}; - u.i &= 0x7fffffffffffffff; - return u.d; + return __builtin_fabs (x); } static inline float gegl_floorf (float x) --- a/operations/common/opacity.c +++ b/operations/common/opacity.c @@ -56,9 +56,7 @@ static inline gfloat int_fabsf (const gfloat x) { - union {gfloat f; guint32 i;} u = {x}; - u.i &= 0x7fffffff; - return u.f; + return __builtin_fabsf (x); } static void --- a/gegl/buffer/gegl-sampler-cubic.c +++ b/gegl/buffer/gegl-sampler-cubic.c @@ -283,9 +283,7 @@ static inline gfloat int_fabsf (const gfloat x) { - union {gfloat f; guint32 i;} u = {x}; - u.i &= 0x7fffffff; - return u.f; + return __builtin_fabsf (x); } static inline gfloat
