Design 025: batch programs (132's A), the block's facts once at load and only values each batch: knapsack's per-batch cost measured at 0.90 us, 38% of its 11.5 ms (BATCH 1024 against 256 under a sampled profile: re-reading the block about 39%, value kinds 26%, list facts and dependence 22%, dispatch 14%), sum0's 0.33 us, matmul's inner-round dispatch about 13 ms and float-checked arithmetic about 22; five stages (facts at load, static kinds and list facts once a batch, typed kernels, fewer passes, the tails) with bars from those shares; 041 batch-costs data

2 files changed, 90 insertions, 1 deletions

diff --git a/research/experiments/041-batches/2026-10-08-batch-costs.jsonl b/research/experiments/041-batches/2026-10-08-batch-costs.jsonl
new file mode 100644
index 0000000..b3eba50
--- /dev/null
+++ b/research/experiments/041-batches/2026-10-08-batch-costs.jsonl
@@ -0,0 +1,85 @@
+{"note": "b9 core (735ba81), suite-run knapsack, PURE_JIT=0, BATCH 1024 and 256 builds (gcc -O2 -g, module-batch -O3), a 2 kHz sampled profile, ms a round by the in-process C's samples (6.24 ms); measured times 11.49 and 24.56 ms; batches a round about 4,900 and 19,400", "pure_ms": [11.49, 24.56]}
+{"work": "knapsack", "function": "pass.constprop.0", "ms_1024": 1.644, "ms_256": 6.167}
+{"work": "knapsack", "function": "batch_run", "ms_1024": 1.999, "ms_256": 6.003}
+{"work": "knapsack", "function": "binary.constprop.0", "ms_1024": 3.751, "ms_256": 4.882}
+{"work": "knapsack", "function": "rows_in.constprop.0", "ms_1024": 0.149, "ms_256": 0.728}
+{"work": "knapsack", "function": "find.constprop.0", "ms_1024": 0.231, "ms_256": 0.806}
+{"work": "knapsack", "function": "pure_list_cells", "ms_1024": 0.143, "ms_256": 0.687}
+{"work": "knapsack", "function": "affine_of.constprop.0", "ms_1024": 0.141, "ms_256": 0.63}
+{"work": "knapsack", "function": "read_of.part.0.constprop.0", "ms_1024": 0.109, "ms_256": 0.516}
+{"work": "knapsack", "function": "branch.constprop.0", "ms_1024": 0.96, "ms_256": 1.291}
+{"work": "knapsack", "function": "pure_batch_run", "ms_1024": 0.025, "ms_256": 0.326}
+{"work": "knapsack", "function": "_init", "ms_1024": 0.026, "ms_256": 0.268}
+{"work": "knapsack", "function": "cells_at.constprop.0", "ms_1024": 0.163, "ms_256": 0.332}
+{"work": "knapsack", "function": "dense_keys", "ms_1024": 0.042, "ms_256": 0.192}
+{"work": "knapsack", "function": "(other)", "ms_1024": 1.21, "ms_256": 1.343}
+{"work": "knapsack", "function": "batches.constprop.0", "ms_1024": 0.039, "ms_256": 0.13}
+{"work": "knapsack", "function": "pure_list_add_many", "ms_1024": 0.486, "ms_256": 0.562}
+{"work": "knapsack", "function": "pure_list_takes", "ms_1024": 0.209, "ms_256": 0.249}
+{"work": "knapsack", "function": "pure_list_get", "ms_1024": 0.121, "ms_256": 0.156}
+{"work": "knapsack", "function": "pure_list_count", "ms_1024": 0.023, "ms_256": 0.05}
+{"work": "knapsack", "function": "rounds_of.constprop.0", "ms_1024": 0.084, "ms_256": 0.102}
+{"work": "knapsack", "function": "c_knapsack", "ms_1024": 6.24, "ms_256": 6.24}
+{"work": "knapsack", "function": "widen", "ms_1024": 0.105, "ms_256": 0.045}
+{"work": "knapsack", "line": "module-batch.c:1761", "ms_1024": 0.189, "ms_256": 0.688}
+{"work": "knapsack", "line": "module-batch.c:1590", "ms_1024": 0.161, "ms_256": 0.653}
+{"work": "knapsack", "line": "module-batch.c:189", "ms_1024": 0.138, "ms_256": 0.621}
+{"work": "knapsack", "line": "module-batch.c:1087", "ms_1024": 0.164, "ms_256": 0.624}
+{"work": "knapsack", "line": "module-batch.c:589", "ms_1024": 0.108, "ms_256": 0.519}
+{"work": "knapsack", "line": "module-batch.c:1747", "ms_1024": 0.166, "ms_256": 0.501}
+{"work": "knapsack", "line": "module-batch.c:1481", "ms_1024": 0.108, "ms_256": 0.408}
+{"work": "knapsack", "line": "module-batch.c:468", "ms_1024": 0.074, "ms_256": 0.373}
+{"work": "knapsack", "line": "module-batch.c:190", "ms_1024": 0.039, "ms_256": 0.308}
+{"work": "knapsack", "line": "module-batch.c:2735", "ms_1024": 0.006, "ms_256": 0.27}
+{"work": "knapsack", "line": "module-batch.c:695", "ms_1024": 0.086, "ms_256": 0.349}
+{"work": "knapsack", "line": "module-batch.c:1440", "ms_1024": 0.075, "ms_256": 0.336}
+{"work": "knapsack", "line": "module-batch.c:460", "ms_1024": 0.039, "ms_256": 0.291}
+{"work": "knapsack", "line": "module-batch.c:1742", "ms_1024": 0.061, "ms_256": 0.306}
+{"work": "knapsack", "line": "??:0", "ms_1024": 0.026, "ms_256": 0.268}
+{"work": "knapsack", "line": "module-batch.c:2780", "ms_1024": 0.094, "ms_256": 0.318}
+{"work": "knapsack", "line": "module-batch.c:1911", "ms_1024": 0.059, "ms_256": 0.281}
+{"work": "knapsack", "line": "module-batch.c:880", "ms_1024": 0.034, "ms_256": 0.246}
+{"work": "knapsack", "line": "module-batch.c:1480", "ms_1024": 0.056, "ms_256": 0.261}
+{"work": "knapsack", "line": "module-batch.c:1812", "ms_1024": 0.05, "ms_256": 0.255}
+{"work": "knapsack", "line": "module-batch.c:679", "ms_1024": 0.04, "ms_256": 0.233}
+{"work": "knapsack", "line": "module-batch.c:1767", "ms_1024": 0.052, "ms_256": 0.238}
+{"work": "knapsack", "line": "module-batch.c:491", "ms_1024": 0.037, "ms_256": 0.22}
+{"work": "knapsack", "line": "module-batch.c:1589", "ms_1024": 0.064, "ms_256": 0.242}
+{"work": "knapsack", "line": "module-batch.c:1834", "ms_1024": 0.056, "ms_256": 0.229}
+{"work": "knapsack", "line": "module-batch.c:774", "ms_1024": 0.064, "ms_256": 0.235}
+{"work": "knapsack", "line": "natural.h:39", "ms_1024": 4.021, "ms_256": 4.164}
+{"work": "knapsack", "line": "module-batch.c:1439", "ms_1024": 0.036, "ms_256": 0.173}
+{"work": "knapsack", "line": "module-batch.c:1437", "ms_1024": 0.053, "ms_256": 0.188}
+{"work": "knapsack", "line": "module-batch.c:355", "ms_1024": 0.005, "ms_256": 0.14}
+{"work": "knapsack", "line": "module-batch.c:617", "ms_1024": 0.058, "ms_256": 0.188}
+{"work": "knapsack", "line": "module-batch.c:1751", "ms_1024": 0.031, "ms_256": 0.159}
+{"work": "knapsack", "line": "module-list.c:1029", "ms_1024": 0.025, "ms_256": 0.153}
+{"work": "knapsack", "line": "module-batch.c:720", "ms_1024": 0.023, "ms_256": 0.149}
+{"work": "knapsack", "line": "module-batch.c:864", "ms_1024": 0.024, "ms_256": 0.15}
+{"work": "knapsack", "line": "module-batch.c:680", "ms_1024": 0.039, "ms_256": 0.163}
+{"work": "knapsack", "line": "module-batch.c:1147", "ms_1024": 0.038, "ms_256": 0.158}
+{"work": "knapsack", "line": "module-batch.c:1014", "ms_1024": 0.024, "ms_256": 0.141}
+{"work": "knapsack", "line": "module-list.c:1808", "ms_1024": 0.035, "ms_256": 0.152}
+{"work": "knapsack", "line": "module-batch.c:1484", "ms_1024": 0.04, "ms_256": 0.156}
+{"work": "knapsack", "line": "module-batch.c:328", "ms_1024": 0.008, "ms_256": 0.121}
+{"work": "knapsack", "line": "module-batch.c:2206", "ms_1024": 0.048, "ms_256": 0.16}
+{"work": "knapsack", "line": "module-batch.c:2782", "ms_1024": 0.049, "ms_256": 0.16}
+{"work": "knapsack", "line": "module-batch.c:1478", "ms_1024": 0.034, "ms_256": 0.145}
+{"work": "knapsack", "line": "module-batch.c:1723", "ms_1024": 0.088, "ms_256": 0.199}
+{"work": "knapsack", "line": "module-batch.c:1438", "ms_1024": 0.041, "ms_256": 0.148}
+{"work": "knapsack", "line": "module-list.c:1714", "ms_1024": 0.02, "ms_256": 0.119}
+{"work": "knapsack", "line": "module-list.c:1807", "ms_1024": 0.014, "ms_256": 0.109}
+{"work": "knapsack", "line": "module-batch.c:497", "ms_1024": 0.007, "ms_256": 0.101}
+{"work": "knapsack", "line": "module-batch.c:1588", "ms_1024": 0.018, "ms_256": 0.107}
+{"work": "knapsack", "line": "module-batch.c:1120", "ms_1024": 0.009, "ms_256": 0.096}
+{"work": "knapsack", "line": "module-batch.c:1757", "ms_1024": 0.036, "ms_256": 0.124}
+{"work": "knapsack", "line": "module-batch.c:2871", "ms_1024": 0.013, "ms_256": 0.098}
+{"work": "knapsack", "line": "module-batch.c:209", "ms_1024": 0.045, "ms_256": 0.127}
+{"work": "knapsack", "line": "module-batch.c:2783", "ms_1024": 0.02, "ms_256": 0.1}
+{"work": "knapsack", "line": "module-batch.c:1492", "ms_1024": 0.025, "ms_256": 0.104}
+{"work": "knapsack", "line": "module-batch.c:845", "ms_1024": 0.039, "ms_256": 0.115}
+{"work": "knapsack", "line": "module-batch.c:1075", "ms_1024": 0.048, "ms_256": 0.121}
+{"work": "knapsack", "line": "module-batch.c:1101", "ms_1024": 0.038, "ms_256": 0.106}
+{"work": "knapsack", "line": "module-batch.c:1715", "ms_1024": 0.042, "ms_256": 0.103}
+{"note":"time.c sum0 n=200000 (4M rounds), 21 rounds median, b9 BATCH 1024 then 256","module":"sum0.pmod","n":200000,"ms":8.633,"work":44000019,"out":4000000}
+{"note":"time.c sum0 n=200000 (4M rounds), 21 rounds median, b9 BATCH 1024 then 256","module":"sum0.pmod","n":200000,"ms":12.544,"work":44000019,"out":4000000}
diff --git a/research/experiments/041-batches/runs.sh b/research/experiments/041-batches/runs.sh
index 6f3a4e8..876f249 100755
--- a/research/experiments/041-batches/runs.sh
+++ b/research/experiments/041-batches/runs.sh
@@ -47,4 +47,8 @@ esac
 # 2026-10-08-lanes.jsonl (research 132): b9's code with BATCH 1024, 512 and
 # 256 (core/module-batch.c's enum), suite-run 5 rounds twice (ms: the two
 # medians) and costs/time.c's sum0, ops0 and nest under icount 3 rounds:
-# a batch's fixed cost is (the instructions' difference) / (the batches').
+# a batch's fixed cost is (the instructions' difference) / (the batches'),
+# icount's 3 rounds counted: 3,000 and 12,000 batches for sum0 and ops0.
+# 2026-10-08-batch-costs.jsonl (design 025): knapsack's time a round by
+# function and line, b9's code at BATCH 1024 and 256 under a 2 kHz sampled
+# profile (ms by the in-process C's samples); sum0's times at both.