#45: delete BCAST_PIPE — broadcast pipeline lives in skew_lane chain#46
Open
npip99 wants to merge 1 commit into
Open
#45: delete BCAST_PIPE — broadcast pipeline lives in skew_lane chain#46npip99 wants to merge 1 commit into
npip99 wants to merge 1 commit into
Conversation
compute_array.sv held ~302 parent flops on chip clk to register the cmd_unit → cell broadcast (forward push_/drain_/scrub_ pipe) and the symmetric cmd_unit → chip-external reverse pipe. B6 (#40)'s per-skew abutment chain (skew_lane_a/b internal chain_w_s/chain_e_n register) absorbed the forward broadcast pipeline natively, leaving the parent flops as dead weight that violates R1 ("parent design = macros + wires, no logic"). RTL: - compute_array.sv: drop BCAST_PIPE parameter + macro, drop drain/scrub forward pipe, drop reverse output pipe, wire cmd_unit's already- registered status outputs (mma_busy/done, arrive_*, drain_*) and its combinational push_*/drain_en outputs directly to the parent ports and chain head. - pymodel/compute_array.py: drop bcast_pipe= ctor arg, _u_* shadow registers, _fwd_pipe/_out_pipe shift registers — collapse to a single registered-output model. Test + build: - cocotb compute_array regression: PASS 2/2 (same cycle counts as pre-deletion baseline, mma_done at 1665 ns / 9200 ns). - pymodel pytest: 90 PASS. - compute_array/Makefile + test_compute_array.py: drop BCAST_PIPE env. - tech/sky130/Makefile: drop sv2v-bcast-sweep + sv2v-tiny-bcast-sweep targets, add compute_array_tiny.v target (MMA=4, no BCAST_PIPE). - tech/asap7/orfs/compute_array{,_abut}.config.mk: repoint VERILOG_FILES from chip_top_bcast1.v to chip_top.v. - tech/asap7/orfs/compute_array_tiny_bcast0.config.mk: repoint to compute_array_tiny.v (nickname preserved for downstream stability). - tech/asap7/orfs/run.sh: refresh staleness-check fallback target. - Delete 14 sweep configs whose only knob was BCAST_PIPE: compute_array_bcast{1,2,3}, compute_array_tiny_bcast{1,2}, compute_array_tiny_slow{,bal,pipe2,uskew}. Docs: - tech/INVARIANTS.md: strike "mb_pipe/md_pipe etc." from R1 violation list, refresh B1/B2 sv2v staleness-check refs. - tech/asap7/DESIGN.md: rewrite the A2 mitigation section so it describes 2500 ps SDC alone; retire the parent-flop framing. - tech/asap7/problems/A2_hold_timing_rtl.md: closure note on the BCAST_PIPE deletion. - tech/asap7/problems/B2_smem_hold_timing.md: update the compute_array-analog references. - tech/RCA_DISCIPLINE.md: refresh prelaunch-checklist recipe to reference chip_top.v instead of chip_top_bcast1.v. The 40-min compute_array_abut harden is the empirical setup-slack gate the issue calls for (Option 1) — not run here. If it fails to close, the fallback is Option 2 (absorb the pipes inside cmd_unit) which would be a fresh patch on top. Closes #45. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
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
1 participant
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.
Summary
BCAST_PIPEknob incompute_array.svand all~302 parent flops it carried (forward drain/scrub pipe + reverse
output pipe). B6 (Full 32×32 compute_array abutment: re-harden cmd_unit + skew_lanes as abutment-ready tiles #40)'s per-skew abutment chain register in
skew_lane_a/babsorbed the broadcast pipeline natively — the parentflops were dead weight that violated R1 ("parent design = macros +
wires, no logic").
pymodel/compute_array.py, thecompute_array Makefile + cocotb env, sv2v sweep targets, and 14
pure-BCAST_PIPE sweep configs.
compute_array,compute_array_abut,compute_array_tiny_bcast0) fromchip_top_bcast1.v/compute_array_tiny_bcast1.vto the singlechip_top.v/compute_array_tiny.voutputs.Closes #45.
RTL functional check
Cycle counts match the baseline exactly, confirming the deletion is
functionally a no-op when BCAST_PIPE was already 0 (which
chip_top.valready was, and what cocotb / pymodel always tested against).
Acceptance criteria (issue #45)
compute_array.svhas zero flops that aren't part of a macroinstantiation (R1 invariant satisfied)
compute_arrayPASS 2/2 (no functional regression)compute_array_abut32×32: 0 DRC, 0 setup violations, 0 holdviolations — 40-min ORFS harden running separately
tech/INVARIANTS.mdR1 violation list updated (mb_pipe/md_pipeentry struck)
Risk / fallback
The harden is the issue's Option 1 gate. If
compute_array_abutfails to close at full 32×32 without the parent flop break on the
cmd_unit → skew_a[0] / skew_b[0]arc, Option 2 (absorb the pipesinside
cmd_unit) is a fresh patch on top. The cmd_unit→chain-headarc is short by construction (cmd_unit and skew chain-heads abut at
the SW corner per
compute_array_abut.macro_placement.tcl), so theexpectation is that it closes — but the empirical check is the gate.
Test plan
make -C tech/sky130 sv2vandcompute_array_tiny.vbuildsclean; grep confirms zero
BCAST_PIPEsymbols in generatedVerilog.
compute_array_tiny_bcast0ORFS harden: 0 DRC, ≥0 setup/holdslack (sanity check before 32×32).
compute_array_abut32×32 ORFS harden: 0 DRC, ≥0 setup/holdslack on
cmd_unit → chain-headpaths (the issue's acceptance).🤖 Generated with Claude Code