2.2.4 failed test gdimagegrayscale/basic

[remicollet]

Only on i386 (x86_64 is OK)

[remicollet]

Diff:

Out:

[cmb69]

Might be a rounding issue. I'll try to have a closer look at this.

[remicollet]

Perhaps we could consider the result as acceptable....

[cmb69]

Perhaps we could consider the result as acceptable....

Indeed, but we'd need some tool for similarity comparison. See issue #216. I'm not completely against adding pdiff, but we may consider alternatives (SSIM comes to mind).

[remicollet]

Notice: also fails on ppc64le, ppc64 and aarch64

[remicollet]

Other arch fixed by b7b66ea (don't know why this wasn't applied in 2.2...), so only i686 affected

[cmb69]

@remicollet Have you tried with the custom CFLAGS mentioned in #242 (comment)?

[fweimer]

The test also fails for me on x86-64, with GCC 8.2 and -O2 -march=haswell -ftree-vectorize. Similar output difference. gdimagecopyresampled/bug00201 is also affected.

[J0WI]

s390x is also affected.

[kaniini]

mips64 is also affected

[pierrejoye]

Works on most other 32bit builds, including windows VC.

mingw fails, could be solve using other FP flags.

As mentionned in the related tickets, we won't change the stable code for 32bit cygwin or mingw.