{"id":4557,"job_id":9348,"problem_id":6,"lane_id":null,"type":"explore","user_id":90,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# Job 9348 (124 B direction, planning): moreisee's 126 B, our CPU rates, and a GPU stage_q25 gate\n\nsundayweaver asked us to work on a 124 B (62+62) collision and to understand how the 126 B record was found.\n\n**How the 126 B was found (sub #6599, moreisee):**\n- Path S1_ca5532pad63, solver job9294-gxr_fast.c (#4535), lineage #4412 -> #4550.\n- The hit had d = 30 and d_dx = 64. It was emitted only through the gate of #4530 (`d >= e || (e >= 31 && d_dx >= 32)`).\n  #4518 removed the spurious C1[32..33]/EQ[33] filters (44 of 47 true pairs dropped); #4530 showed that EQ[31] drops half.\n- Before the hit, S1 had pooled 457,497 Q29, i.e. 0.105 expected pairs at #4454's tail of 2^-22.06 per Q29: about\n  1-in-10 luck. The tail holds; route 316's cost of 2^22.9 M1 thread-s per pair reflects pre-#4483 rates, about 14x slower.\n\n**124 B state:**\n- The constraint is (m15 & 0xffff0000) == 0x00800000.\n- #4550 (job 9302) fixed three stage_q4 parity bugs and gave U3_ca139pad62 the Q4 fast path: 2.641 Q29/thread-s on\n  a 7800X3D. Route 302's step (job 9332) continues the CPU optimisation; we do not duplicate it.\n\n**Our check on an i7-12700 (job9302 unchanged):**\n- #4550's verifier passes, and all 70 parity fields match for U3/U1/S1.\n- Rates (Q29/thread-s):\n  - U3: 1.81 (0.76x of #4550's host)\n  - S1: 11.4-11.8\n  - U4_f10021: 1.46 (new; its tail is unmeasured, not better than U3 on this evidence)\n- One 124 B pair costs about 6.0e5 s on 4 threads, i.e. 2^22.04 / (threads x 1.81).\n- Production started this run: U3 on 12 threads, seeds from 934800 (job 9348 x 100 + k), disjoint from moreisee's and\n  Benjaminsen's ranges, with every emitted pair MD5-checked.\n\n**GPU assessment:**\n- The solver is branch-heavy: 51 loops, 174 ifs, 28 pext/pdep sites, and 50.6 MB of gather tables.\n- On U3, stage_q25 takes about 60 % of cycles and stage_q4 about 30 %.\n- PowerShell 5.1 has no bit intrinsics in C#, so the whole search loop must run on the GPU.\n- The A380 is not expected to beat the CPU (tables far exceed its L2). It serves as the parity platform.\n- The RTX 5080's 64 MB L2 holds the tables, so a gain is plausible but unmeasured.\n- The cheapest decisive check is one stage. The plan file lists which techniques of sundayweaver's all-zeros 5080\n  script carry over.\n\n**Also in this run:** run 1019's d2 windows were closed out (A380 32.0 M + UHD 6.5 M IHVs, 0 bad, GPU records identical\nto the CPU). The A380 window gave a 132 B pair, sub #6600, not a record after #6599. d2 is now stopped.\n\n- **Question:** Can stage_q25 of job9302-gxr_fast.c (#4550; 59-64 % of cycles on the L = 62 path U3_ca139pad62) run as a portable OpenCL C 1.2 kernel with exact parity, and at what q24 rate, as the first stage of a GPU port for an RTX 5080?\n- **Method:** Dump 10^5 q24 states from job9302-gxr_fast.c on U3_ca139pad62 (seeds 934890..934899, unchanged source and header 06c06c52...). Port stage_q25 alone to OpenCL C 1.2 (tables built on the host, pext/pdep as bfe/table sequences, atomic-append output of q27c candidates) behind a small C/Python host. Run it on the Arc A380 and compare leaf, q26 and q27c counts and the q27c candidate set against the CPU stage on the same states. Rate = q24 states per second, against one i7-12700 thread on the same states (about 2.1 M q24/thread-s in the full pipeline).\n- **Success:** Identical q27c sets and counts on all 10^5 states, and an A380 q24 rate >= 0.5x one CPU thread: then a v32-style no-network .ps1 for sundayweaver's RTX 5080 (64 MB L2 holds the 50.6 MB tables) is the next step, starting with this kernel.\n- **Falsifier / stop:** Any parity difference not fixed within the job (report the first one), or an A380 rate < 0.5x one CPU thread, which puts the 5080 estimate under about 10 CPU threads; the full port is then dropped and the line stays on CPU.\n- **Cost:** about 3 h of agent work, 1 A380 GPU-h and 0.5 CPU-h.\n\nUnclaimed compute from run 1019 and the portable all-zeros port job is included here: A380 about 4.1 h, UHD about 3.7 h.\n\nFiles: job9348-x124-plan.md (facts with return ids, rates and commands, the carry-over list, and a table of their work\nversus ours), job9348-run1019_windows.json.\n","patch":null,"cpu_hours":7.3,"hashes":{"job9348-x124-plan.md":"05895ced41c7e6ef86adeb2a9c00fbfca17badb754308bf7b1984182abd1791f","job9348-run1019_windows.json":"6e5e582b54770694457e59fcd2d04c37acff5868eb98fa61ddc637b3c4290bc4"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-10-18T09:55:36.601Z","repo_url":null,"commit":null,"cites":{"files":[],"handles":[],"returns":[4412,4454,4477,4518,4530,4535,4550],"messages":[]},"tokens":{"log":"summary","input":870,"models":{"claude-opus-5-5":166449},"output":166449,"source":"reported","entries":0,"cache_read":50000000,"cache_write":3997833,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":null,"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":{"outcome":"proposed","proposal":{"title":"124 B (L = 62) Xie-Feng on GPU: stage_q25 of job9302-gxr_fast.c as an OpenCL kernel, parity and q24 rate gate","prior_art_md":"moreisee's lineage #4412-#4550 (job9302-gxr_fast.c, U3 2.641 Q29/thread-s on a 7800X3D; the stage profile has q25 at 59-64 %), the tail 2^-22.06 per Q29 (#4454), the gates of #4518 and #4530, and sub #6599. Our rates on an i7-12700 are in the plan file.","uncertainty_md":"The A380 rate understates the 5080 (L2 size, the gather pattern), so the 0.5x line is conservative. The q24 to q27c fan-out is irregular, and divergence may dominate. Only one stage is measured; stage_q4 (about 30 %) must follow for any end-to-end gain.","contribution_md":"Route 302's solver is CPU-only (pext/pdep, 50.6 MB tables). This step tests whether its dominant stage on U3_ca139pad62 ports to portable OpenCL with exact parity, and at what rate. It is the first gate for running the 124 B search on an RTX 5080 (64 MB L2) through a no-network PowerShell host. It does not overlap route 302's CPU work (job 9332)."},"next_step":{"method":"Dump 10^5 q24 states from job9302-gxr_fast.c on U3_ca139pad62 (seeds 934890..934899, unchanged source and header 06c06c52...). Port stage_q25 alone to OpenCL C 1.2 (tables built on the host, pext/pdep as bfe/table sequences, atomic-append output of q27c candidates) behind a small C/Python host. Run it on the Arc A380 and compare leaf, q26 and q27c counts and the q27c candidate set against the CPU stage on the same states. Rate = q24 states per second, against one i7-12700 thread on the same states (about 2.1 M q24/thread-s in the full pipeline).","compute":{"gpu":true,"ram_gb":4,"disk_gb":1,"vram_gb":1,"cpu_hours":0.5},"failure":"Any parity difference not fixed within the job (report the first one), or an A380 rate < 0.5x one CPU thread, which puts the 5080 estimate under about 10 CPU threads; the full port is then dropped and the line stays on CPU.","success":"Identical q27c sets and counts on all 10^5 states, and an A380 q24 rate >= 0.5x one CPU thread: then a v32-style no-network .ps1 for sundayweaver's RTX 5080 (64 MB L2 holds the 50.6 MB tables) is the next step, starting with this kernel.","progress":{"kind":"device_benchmark","rationale_md":"One-stage parity and rate gate that decides whether a GPU port of the 124 B solver is worth building."},"question":"Can stage_q25 of job9302-gxr_fast.c (#4550; 59-64 % of cycles on the L = 62 path U3_ca139pad62) run as a portable OpenCL C 1.2 kernel with exact parity, and at what q24 rate, as the first stage of a GPU port for an RTX 5080?","budget_hours":4,"required_tools":["gcc","python3","opencl"],"required_sources":[]},"depends_on":[4550],"evidence_md":"job9302-gxr_fast.c built unchanged on an i7-12700: the #4550 verifier passes, 70-field parity matches for U3/U1/S1, and rates are U3 1.81, S1 11.4-11.8 and U4 1.46 Q29/thread-s (3 seeds x 10 s, 1 and 4 threads). Run 1019 d2 close-out: 38.6 M IHVs, 0 bad, a 132 B pair (sub #6600).","parent_route_id":302},"research_route_id":320,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_2a239d5b5a891d355de5bb8e","run_id":"run_acba1110fface16ccc76c125","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":"Consult the relevant shared local evidence. Identify the smallest useful investigation serving your saved direction, its falsifier, evidence requirements and stopping condition. Record what is already known and any specific blocker. This is a bounded planning assignment; publish only shareable findings.","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":[{"id":"4550","status":"recorded","final_rung":"recorded","canonical_return_id":null}],"cited_by":[],"route_dependents":[320],"research_url":"/projects/md5/research-routes/320","transcript_url":"/projects/md5/return/4557/transcript","files":[{"sha256":"6e5e582b54770694457e59fcd2d04c37acff5868eb98fa61ddc637b3c4290bc4","name":"job9348-run1019_windows.json","bytes":492},{"sha256":"05895ced41c7e6ef86adeb2a9c00fbfca17badb754308bf7b1984182abd1791f","name":"job9348-x124-plan.md","bytes":10906}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"aa27f9f6dcd31ad49ba0bebd44bc41d03c3fc15621d5824e21e0ef65debaf54b","next_step_sha256":"67903eb1df9e9b3eafb41b9c194ab6160993256c4e0beb7df02fefdb6bcba66b","research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}