KTX roundng

[andy-thomason]

Thanks for an excellent tool. I am using it in the cubemap examples for my Vulkan helper library, Vookoo.

I have seen a potential bug in the KTX write code when rounding is following
GL_UNPACK_ALIGNMENT

The image size does not seem to take account of the change in row pitch due to rounding and reports a smaller value that required. This may just be a misunderstanding of the KTX format, so please ignore
this if it is an incorrect assumption.

[dariomanesku]

Have a look at imageSaveKtx():

cmft/src/cmft/image.cpp

Line 5231 in a8c246a

bool imageSaveKtx(const char* _fileName, const Image& _image)

Unpack alignment is taken into account when writing data:

cmft/src/cmft/image.cpp

Line 5293 in a8c246a

const uint32_t pitchRounding = (KTX_UNPACK_ALIGNMENT-1)-((pitch + KTX_UNPACK_ALIGNMENT-1)&(KTX_UNPACK_ALIGNMENT-1));

but faceSize is not affected by that:

cmft/src/cmft/image.cpp

Line 5298 in a8c246a

write = fwrite(&faceSize, sizeof(uint32_t), 1, fp);

Try recalculating faceSize with rounding and see if that solves the problem.

[andy-thomason]

Thanks, Dario. I would suggest changing const uint32_t pitch = width * bytesPerPixel; const uint32_t faceSize = pitch * height; For const uint32_t pitch = width * bytesPerPixel; const uint32_t roundedPitch = (width * bytesPerPixel + KTX_UNPACK_ALIGNMENT-1) & ~(KTX_UNPACK_ALIGNMENT-1); const uint32_t faceSize = roundedPitch * height; As you suggest. I'll try this the next time I am at home with my Linux machine.

…

On Sun, Apr 9, 2017 at 11:35 PM, Dario Manesku ***@***.***> wrote: Have a look at imageSaveKtx(): https://github.com/dariomanesku/cmft/blob/a8c246a4b59670092f34209bf8247f 79337526ed/src/cmft/image.cpp#L5231 Unpack alignment is taken into account when writing data: https://github.com/dariomanesku/cmft/blob/a8c246a4b59670092f34209bf8247f 79337526ed/src/cmft/image.cpp#L5293 but faceSize is not affected by that: https://github.com/ dariomanesku/cmft/blob/a8c246a4b59670092f34209bf8247f 79337526ed/src/cmft/image.cpp#L5298 Try recalculating faceSize with rounding and see if that solves the problem. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#29 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABe_XFNIF99qfFThWAkdIQ2CIBvrhgacks5ruV0VgaJpZM4M1h9z> .