GraphicsMagick 的 OpenCL 开发记录（三十八）

<2022-05-18 Wed>

AccelerateScaleImage()和AccelerateResizeImage()的性能测试

迭代100次，缩小图片50%，如下：

[ysouyno@arch gm-ocl]$ MAGICK_OCL_DEVICE=true gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -scale 50x50% ~/temp/out.jpg
Results: 8 threads 100 iter 4.92s user 5.146311s total 19.431 iter/s 20.325 iter/cpu
[ysouyno@arch gm-ocl]$ gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -scale 50x50% ~/temp/out.jpg
Results: 8 threads 100 iter 5.88s user 5.887496s total 16.985 iter/s 17.007 iter/cpu

迭代100次，放大图片200%，如下：

[ysouyno@arch gm-ocl]$ MAGICK_OCL_DEVICE=true gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -scale 200x200% ~/temp/out.jpg
Results: 8 threads 100 iter 17.73s user 15.998019s total 6.251 iter/s 5.640 iter/cpu
[ysouyno@arch gm-ocl]$ MAGICK_OCL_DEVICE=true gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -scale 200x200% ~/temp/out.jpg
Results: 8 threads 100 iter 17.61s user 15.812017s total 6.324 iter/s 5.679 iter/cpu
[ysouyno@arch gm-ocl]$ gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -scale 200x200% ~/temp/out.jpg
Results: 8 threads 100 iter 23.15s user 23.203446s total 4.310 iter/s 4.320 iter/cpu
[ysouyno@arch gm-ocl]$ gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -scale 200x200% ~/temp/out.jpg
Results: 8 threads 100 iter 23.57s user 23.621860s total 4.233 iter/s 4.243 iter/cpu

主要看total前面的值，是运行总时间。同样的方法再对比一下AccelerateResizeImage()。

迭代100次，缩小图片50%，如下：

[ysouyno@arch gm-ocl]$ MAGICK_OCL_DEVICE=true gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -resize 50x50% ~/temp/out.jpg
Results: 8 threads 100 iter 11.28s user 8.047808s total 12.426 iter/s 8.865 iter/cpu
[ysouyno@arch gm-ocl]$ gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -resize 50x50% ~/temp/out.jpg
Results: 8 threads 100 iter 44.43s user 6.364194s total 15.713 iter/s 2.251 iter/cpu

迭代100次，放大图片200%，如下：

[ysouyno@arch gm-ocl]$ MAGICK_OCL_DEVICE=true gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -resize 200x200% ~/temp/out.jpg
Results: 8 threads 100 iter 24.67s user 18.713505s total 5.344 iter/s 4.054 iter/cpu
[ysouyno@arch gm-ocl]$ gm benchmark -iterations 100 convert ~/temp/bg1a.jpg -resize 200x200% ~/temp/out.jpg
Results: 8 threads 100 iter 160.27s user 26.635967s total 3.754 iter/s 0.624 iter/cpu

见commit：some trivial changes。