diff options
| author | eunhhu <[email protected]> | 2026-10-08 15:16:39 +0100 |
|---|---|---|
| committer | eunhhu <[email protected]> | 2026-10-08 15:16:39 +0100 |
| commit | 7770574b327238213c07fccc813b575277231c61 (patch) | |
| tree | 9343c511d318cbe76dec626ea9baf52aab538411 /research/experiments/041-batches/runs.sh | |
| parent | 02925df6da43b6f737671c1957739102cdb49b71 (diff) | |
| download | pure-7770574b327238213c07fccc813b575277231c61.tar.gz pure-7770574b327238213c07fccc813b575277231c61.zip | |
Research 134: batch plans' stage 1 measured (b10 = a6fabf0 against b9, interleaved): knapsack's per-batch cost 0.90 to 0.68 us and knapsack 11.48 to 10.34 ms, short of 025's bars (0.60 us, 10.3 ms): 025 counted the pass's per-instruction head as lookups; sum0 2,330 to 1,822 instructions a batch (bar 1,550); matmul 40.0 to 38.2 ms with the same instructions and BFS 24.04 to 24.31 with 0.2% fewer (placement); sum0's 8.69 to 6.13 ms mostly the compiler now vectorizing the reductions' first loop; the executor alone the same instructions; what is left a batch is the pass's Val handling (stage 2); 041 runs.sh pairs, pairs and counts data
Diffstat (limited to 'research/experiments/041-batches/runs.sh')
| -rwxr-xr-x | research/experiments/041-batches/runs.sh | 17 |
1 files changed, 16 insertions, 1 deletions
diff --git a/research/experiments/041-batches/runs.sh b/research/experiments/041-batches/runs.sh index 876f249..4da74fb 100755 --- a/research/experiments/041-batches/runs.sh +++ b/research/experiments/041-batches/runs.sh @@ -10,6 +10,11 @@ # runs.sh counts NAME [WORK ...] user-mode instructions and cycles # of suite-run's 3 rounds, each of # the two executor ways (icount.c) +# runs.sh pairs A B ROUNDS TIMES [WORK ...] the batches way of A then B, +# TIMES turns (interleaved: the +# medians of A's and B's turns) +# (A build of another lane count: CPPFLAGS=-DPURE_BATCH_LANES=256 with +# 040's runs.sh build.) set -e here=research/experiments/041-batches builds=$HOME/.cache/pure-040 @@ -41,7 +46,17 @@ counts) | sed "s/^{/{\"version\":\"$name\",\"way\":\"$way\",\"work\":\"$work\",/" >> "$here/$day-counts.jsonl" done done ;; -*) echo "runs.sh suite NAME ROUNDS [WORK ...] | counts NAME [WORK ...]" >&2 +pairs) + other=$1 rounds=$2 times=$3 + shift 3 + for turn in $(seq "$times"); do + for version in "$name" "$other"; do + quiet + PURE_JIT=0 python3 research/guard.py --most 3072 -- "$builds/$version/suite-run" "$builds/$version" "$rounds" "$@" \ + | grep median | sed "s/^{/{\"version\":\"$version\",\"way\":\"batches\",\"turn\":$turn,/" >> "$here/$day-pairs.jsonl" + done + done ;; +*) echo "runs.sh suite NAME ROUNDS [WORK ...] | counts NAME [WORK ...] | pairs A B ROUNDS TIMES [WORK ...]" >&2 exit 2 ;; esac # 2026-10-08-lanes.jsonl (research 132): b9's code with BATCH 1024, 512 and |
