Created attachment 119196 [details] source code. Look at the lines 98 and 99 in attachment. Line 98 produces incorrect cosine result, line 99 produces correct - and it works this way only in beignet. 98(incorrect): sum += x[n] * cos((float)(M_PI * (n + .5) * k / N)); 99 (correct): sum += x[n] * cos((float)(M_PI) * (float)((n + .5) * k) / (float)N); (n, k, N are integers) Seems like beignet first tries to evaluate 'k / N' as integer (or makes smth similar mistake). To see it in action compile attachment with gcc dcttest.c -O3 -ftree-loop-vectorize -ftree-loop-distribution -ftree-parallelize-loops=4 -floop-nest-optimize -floop-interchange -floop-strip-mine -floop-block -march=native -lOpenCL -lm -o dcttest And run as ./dcttest platform device (in my case './dcttest 0 0')
With beignet master(ec92544c387c5c2e3c7f6acfccddd353e7974e13), get correct result, please try again. ./dcttest 0 0 0.070000 seconds to calculate host values Succeeded to create a device group! Device: 0 Name: Intel(R) HD Graphics IvyBridge GT2 Vendor: Intel Available: Yes Compute Units: 16 Clock Frequency: 1000 mHz Global Memory: 2048 mb Max Allocateable Memory: 1536 mb Local Memory: 65536 kb Succeeded to create a compute context! Succeeded to create a command commands! Succeeded to create compute program! Succeeded to create program executable! Succeeded to create compute kernel! Computed '8132/8192' correct values! Computed '8132/8192' correct values! Computed '8132/8192' correct values! Computed '8132/8192' correct values! Computed '8132/8192' correct values! Computed '8132/8192' correct values! Computed '8132/8192' correct values! Computed '8132/8192' correct values! Computed '8132/8192' correct values! 0.019000 average seconds to calculate gpu values kernel is: void dct_ii(int N, float *x, float *X) { float sum = 0.; for (int n = 0; n < N; ++n) { sum += x[n]; } X[0] = sum; for (uint k = 1; k < N; ++k) { sum = 0.; for (int n = 0; n < N; ++n) { sum += x[n] * cos((float)(M_PI * (n + .5) * k / N)); //sum += x[n] * cos((float)(M_PI) * (float)((n + .5) * k) / (float)N); } X[k] = sum; } }
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/beignet/beignet/issues/9.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.