docs: decisive test -- pipelining both cores does NOT fix the lottery; fault is outside the datapath (Refs #1764) - #1905
Merged
Conversation
… (Refs #1764) Pipelined both shared cores latency-1 with a register INSIDE the combinational cloud (GftSmul_p2b mid-RNE, GftSadd_p2 mid-cascade), each bit-exact to its combinational core over 40-60k random operands; integrated trainer bit-exact in sim, fmax 32 MHz. On the AX7203, 0/4 seeds trained stably (seeds 2,3 near-model at ep0 then collapsed by ep20; 1,4 glitched from ep0) -- same as baseline. So registering the core datapath (endpoints in cycle 97, mid-cloud now) does NOT fix the fault: the hazard is not in the GftSmul/GftSadd datapath. Closes the pipeline-the-cores branch (cycles 94-99). Redirect to the rf-write/control path (di decode, pc/settle/cen counters) and global placement effects. Added ruled-out #10. Refs #1764 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The decisive silicon experiment for the whole "pipeline the shared cores" branch (cycles 94-99).
Both shared cores were pipelined latency-1 with a register inside the combinational cloud (
GftSmul_p2bcut mid-RNE,GftSadd_p2cut mid-cascade), each verified bit-exact to its combinational core over 40-60k random operands (including zero and exact-cancellation corners). The integrated trainer is bit-exact in simulation and reaches fmax 32 MHz (vs 21 baseline).On the AX7203, four seeds: 0/4 trained stably. Seeds 2 and 3 were near-model at ep0 (
0/0.718/0.886/-0.011and0/0.551/1.021/0.506) then collapsed to all-zero by ep20 (the characteristic training-divergence glitch); seeds 1 and 4 glitched from ep0. This matches the baseline (~1/8 stable).Conclusion: registering the core datapath — at the endpoints (cycle 97) or mid-cloud (now) — does not fix the fault. So the hazard is not in the
GftSmul/GftSaddcombinational datapath at all; it lives in the rf-write / control path (didecode,pc/settle/cencounters) or is a global placement effect. This closes the pipeline-the-cores branch and redirects the investigation. Added ruled-out item #10. (Sample is four seeds; base rate ~1/8, so 0/4 is indicative, not a proof of zero improvement.)Board restored to the generated capstone (XOR 4/4). Docs-only. Refs #1764