It works fine on Windows with MSVC and on x86 Linux with gcc. However, if built with gcc on a RHEL Power8 server (where it uses VMX, I guess?) color channels get messed up, the output image is mostly gray with some diagonal mosaic stripes, sort of what happens if a stride value is wrong. Am I using the resize function correctly?
(I will attach the image, if necessary.)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
As I can see you use Simd Library correctly. Especialy if you have correct result at x86. So I suppouse that there is an error in VMX version of function ResizeBilinear. I'm sorry but can't check this fact, and fix the error because I haven't got access to Power PC at the moment.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
It seems to work correctly if the system is big endian (my previous attempt was on a little endian system). Probably some of those clever binary tricks need to be different for different ending.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
View and moderate all "English Common Forum" comments posted by this user
Mark all as spam, and block user from posting to "Discussion"
I wanted to compare performance of bilinear resizing with OpenCV, so I wrote this function:
It works fine on Windows with MSVC and on x86 Linux with gcc. However, if built with gcc on a RHEL Power8 server (where it uses VMX, I guess?) color channels get messed up, the output image is mostly gray with some diagonal mosaic stripes, sort of what happens if a stride value is wrong. Am I using the resize function correctly?
(I will attach the image, if necessary.)
As I can see you use Simd Library correctly. Especialy if you have correct result at x86. So I suppouse that there is an error in VMX version of function ResizeBilinear. I'm sorry but can't check this fact, and fix the error because I haven't got access to Power PC at the moment.
View and moderate all "English Common Forum" comments posted by this user
Mark all as spam, and block user from posting to "Discussion"
It seems to work correctly if the system is big endian (my previous attempt was on a little endian system). Probably some of those clever binary tricks need to be different for different ending.
Yes. It is similar to truth. When I was porting the algorithms to PowerPC platform, I was supposing that it was big endian.