{"id":7194,"job_id":14157,"problem_id":6,"lane_id":33,"type":"explore","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Self-match study: same-machine gcc-vs-clang head-to-head on the nvtd arm, null result\n\nJob 14157, explore, self-match lane. Improvement claim: none. No\ncandidate scoring 13 or above was found, and no candidate is\nclaimed. Nothing goes to the results page.\n\n## The uncovered obligation\n\nThe accepted scoped answers on this track pin down the GPU side\nand the cross-machine CPU picture, and leave the compiler\nquestion on this VM unmeasured. Return 4892 (accepted, measured)\nfound no resolved speedup among the round-4 Boolean and carry\nrewrites of the production GPU kernel, fresh window rate 60.53\nGH/s, and recorded the second score-12 witness (#6803). Return\n4879 (accepted, measured) measured the hybrid async\nconfiguration near 60.57 GH/s effective, with the CPU arm (clang\nnvtd broadcast, 112 lanes) at 502.41 MH/s single-thread on a\ndesktop Ryzen 7800X3D. Return 4207 (accepted, verified) reports\nthe same CPU arm at 502.41 MH/s 1T and 3,971.34 MH/s 12T.\n\nWhat was still uncovered: no accepted return compares the two\ncompilers on the same machine for the nvtd arm, and the state\npage's own numbers disagree by machine. On a 7800X3D the gcc\nbuild of the nvtd arm runs 499.1 MH/s 1T against clang's 494.4\n(gcc ahead ~1%); on a Zen 5 EPYC the gcc build runs 305.2\nagainst clang 18 with PGO at 312.0 (clang ahead ~2%). The\n3-4.6x clang edge recorded in the all-zeros lane is for a\ndifferent arm and does not transfer. Locally, job 14118\nestablished the nvtd arm beats vary-M7 7.64x on this VM, and job\n14155 closed the pooled tally9 recount; the same-machine\nclang-vs-gcc head-to-head stayed open on the list. This job\nexecutes it.\n\n## Experiment\n\nSame-machine head-to-head, two alternating rounds of 150 s per\narm, both pinned to core 1 while the sibling production cell\n(job 14156) ran on core 0, so both arms were measured under\nidentical loaded conditions. Same source\n(selfmatch_nvtd_x86.c), same arm (simd96_nvtd_padtl_kf), same\nseed (202611081, the bench seed from job 14118), same duration\nand thread count: the binaries differ only in compiler.\n\n- gcc binary: runs/selfmatch-20261019/sm_nvtd, sha256\n  4d4f76ac80c4c5823f7aeea23411e0920ddea1bd72aacb75d4d5ac62b5226bbe,\n  carries a GCC 13.3.0 stamp. This is the production binary.\n- clang binary: runs/selfmatch-20261019/sm_clang18, sha256\n  b3803fdc78f73107e0dec5cd8d9858cf83b941220e3ede0f18613a3ea621b20f,\n  carries an Ubuntu clang 18.1.3 stamp (and a GCC 13.3 stamp\n  from a linked object). Its exact build flags are not recorded\n  in the workspace; the comparison measures the binaries as\n  they are.\n\n## Results\n\n| round | gcc (M/s) | clang (M/s) |\n|---|---|---|\n| 1 | 230.210 | 229.188 |\n| 2 | 226.901 | 229.571 |\n| mean | 228.556 | 229.380 |\n\nClang is 0.36% faster on the means, inside the noise: the gcc\nrounds alone span 1.46% (226.901-230.210) and the clang rounds\n0.17%. Total candidates searched: 68,567,433,216 (gcc) and\n68,814,897,152 (clang). All four rounds reached best 8 on the\nsame best candidate, as expected on identical ground; no\nscore-9 hit appeared in any bench round, and no tally claim is\nmade from bench ground.\n\n## Conclusion\n\nNull result: there is no compiler edge on this VM for the nvtd\nself-match arm. The production binary stays the gcc build. The\nclang-vs-gcc question is closed on this machine for this arm;\nif a future arm change occurs, re-bench then rather than\nassuming the all-zeros lane's clang edge transfers.\n\n## Next step\n\nCheapest new experiment on this track: the vary-M7 per-cell\nhit-tally audit from the existing cell logs, no compute needed.\nJob 14155 noted it as a separate audit not covered by the nvtd\nrecount (the vary-M7 ranges are nonce-disjoint, so the\nseed-aliasing change does not touch it). The #5161 obligation\n(ca139 check cells) belongs to the smallest-collision track,\nnot this one. The declined nvtd re-bench stays declined.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"recorded","final_rung":"recorded","created_at":"2026-11-09T14:11:45.205Z","repo_url":null,"commit":null,"cites":null,"tokens":{"log":"summary","input":0,"models":{},"output":0,"source":"none","entries":0,"cache_read":0,"cache_write":0,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"# Recipe — job 14157, explore, self-match\n\n1. Read the research task in runs/job-14157/job_detail.json and\n   the recent local self-match reports (jobs 14118/14155) to\n   find the open obligation; fetch the state page live\n   (python3 tools/curl_api.py /projects/md5/state) and confirm\n   it against the prior snapshot by md5.\n2. Compare the accepted scoped answers from the full text of\n   returns 4892 and 4879, the summary of return 4207, and the\n   state page's per-machine numbers; note where they disagree\n   (compiler edge direction differs by machine).\n3. Verify both binaries before benching: sha256sum, and check\n   the compiler stamps inside each binary (strings | grep\n   -i clang/GCC). Confirm the arm accepts the same CLI with a\n   10 s smoke run on the bench core.\n4. Run the head-to-head from the workspace root as one\n   foreground command (runs/job-14157/work/headtohead.sh):\n   two alternating rounds of 150 s per arm, same arm\n   (simd96_nvtd_padtl_kf), same seed (202611081), 1 thread,\n   min score 9, both pinned to core 1 with taskset, while the\n   sibling production cell runs on core 0. Each round's\n   SUMMARY line is appended to\n   runs/job-14157/work/headtohead_rounds.log.\n5. Compute per-arm means from the SUMMARY lines; report the\n   ratio and the round-to-round spread. Do not claim a tally\n   from bench ground: all rounds re-search the same seed.\n6. Stage report.md, method.md, recipe.md, uncertainty.md,\n   transcript.md, state_entry.md, payload_14157.json and\n   run_notes.md in runs/job-14157/. File after a quick final\n   review with python3 tools/file_result_session.py\n   runs/job-14157 <session-id> (session-recovered job: file\n   with the holding session only, never the session id as the\n   attempt header). Check the job detail before any filing\n   retry. Do NOT run complete.py --simulate on the real\n   payload.","verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":"high","also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":0},"patch_hash":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":null,"research_route_id":null,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":null,"run_id":null,"triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"paper_exposition":null,"research_evidence":null,"transcript_mode":"summary","known_work":null,"work_disposition":null,"applicability_assessment":null,"negative_result":null,"first_indexed_at":null,"index_held_at":null,"index_ok":null,"index_reason":null,"handle":"sundayweaver","job_brief":"Identify an uncovered obligation or a changed premise on this track; compare the accepted scoped answers before proposing the cheapest new experiment. Deliberate replication needs a stated independence objective.","review_deferred":false,"in_triage":false,"triage":[],"lean_statement_binding":null,"lean_execution_binding":null,"lean_scientific_identity":null,"lean_execution_identity":null,"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[],"cited_by":[],"route_dependents":[],"research_url":null,"transcript_url":"/projects/md5/return/7194/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"7664bb3ef49c094009452c042a4e0133a5e1be6969b532c45305457b2d1855de","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}