fig1024 / op_rbf Goto Github PK

when i use this image for testing , program crashed!!, here is the image. is this a problem about the size?

if (thread_index + 1 == RBF_MAX_THREADS)
height_segment = height - thread_index * height_segment;

should be changed to:

if (thread_index + 1 == RBF_MAX_THREADS)
height_segment = height - thread_index * height_segment - 1;

Please note that I removed the variable for number of threads in the class for my implementation.

Use author's （using optimized SSE2 with optional multithreading, pipelined 2 stages）Method, the left edge of the image after the calculation has filtering, why is it?

Hi, Fig1024, your optimization is awesome, it means a lot to me. thanks!
However, i found some bugs there, and below is some of my solutions

First, according to the comment statements from https://github.com/nothings/stb.git. Maybe we should add
#define STB_IMAGE_IMPLEMENTATION
#define STB_IMAGE_WRITE_IMPLEMENTATION
at the begining of RecursiveBilateralFilter.cpp. In some situation, it couldnot pass the compile if we don't.

Second, it crash in RBFilter_SSE2.cpp,when i run your code. After debug, I found it's because out_pix4 is not defined before used in function verticalFilter
pixB = _mm_shuffle_epi8(pixB, mask_pack);

out_pix4 = _mm_srli_si128(out_pix4, 4); // shift
out_pix4 = _mm_or_si128(out_pix4, pixB);
I try to change the codes to be
if (i == 0) //
{
out_pix4 = _mm_shuffle_epi8(pixB, mask_pack);
}
else
{
pixB = _mm_shuffle_epi8(pixB, mask_pack);

out_pix4 = _mm_srli_si128(out_pix4, 4); // shift 
out_pix4 = _mm_or_si128(out_pix4, pixB);

}
It seem to work. But i'm not sure it this change is right.

Can you check here,please?
Thank you.

Original Recursive Bilateral Filter implementation
Image: lw_cross.jpg, size: 1280 x 720, time ms: 70.2
Image: Thefarmhouse.jpg, size: 1440 x 1080, time ms: 119.3
Image: testGirl.jpg, size: 448 x 626, time ms: 23.6

Optimized SSE2 single threaded, single stage (non-pipelined)
Image: lw_cross.jpg, size: 1280 x 720, time ms: 86.6
Image: Thefarmhouse.jpg, size: 1440 x 1080, time ms: 142.0
Image: testGirl.jpg, size: 448 x 626, time ms: 25.6

Optimized SSE2 2x multithreading, single stage (non-pipelined)
Image: lw_cross.jpg, size: 1280 x 720, time ms: 44.3
Image: Thefarmhouse.jpg, size: 1440 x 1080, time ms: 78.4
Image: testGirl.jpg, size: 448 x 626, time ms: 13.4

Optimized SSE2 4x multithreading, single stage (non-pipelined)
Image: lw_cross.jpg, size: 1280 x 720, time ms: 24.9
Image: Thefarmhouse.jpg, size: 1440 x 1080, time ms: 41.5
Image: testGirl.jpg, size: 448 x 626, time ms: 7.5

Optimized SSE2 4x2 thread pipelined 2 stages
Image: lw_cross.jpg, size: 1280 x 720, time ms: 18.6
Image: Thefarmhouse.jpg, size: 1440 x 1080, time ms: 32.0
Image: testGirl.jpg, size: 448 x 626, time ms: 5.9
Finish

您好，我的cpu是i7-8700 [email protected]的，但是这个复现结果和您的结果差距在十多倍，请问会是什么原因呢？谢谢