)]}' { "commit": "c82958326a663f083cc122cfd8aaa5116e1cd798", "tree": "ac1590985a550e8d4380738bd7b957e473a23db2", "parents": [ "c910146a9840f9b82235d60c60f5f74327e4c2ea" ], "author": { "name": "David Chase", "email": "drchase@google.com", "time": "Fri Jun 05 16:16:28 2026 -0400" }, "committer": { "name": "David Chase", "email": "drchase@google.com", "time": "Thu Jun 11 15:41:48 2026 -0700" }, "message": "simd: add some inner-product benchmarks\n\nSome hand-unrolling of important methods\nwas tried, and shown to improve results.\n\nResults from\n\n```\n\nexport GOEXPERIMENT\u003dsimd\n\nfor R in {1..25} ; do\ngo test -bench\u003dB -run\u003dnone -count\u003d1 -ldflags\u003d-randlayout\u003d$R . | tee -a arm64-128.log\nGODEBUG\u003dsimd\u003d0 go test -bench\u003dB -run\u003dnone -count\u003d1 -ldflags\u003d-randlayout\u003d$R . | tee -a arm64-0.log\nGOARCH\u003damd64 go test -bench\u003dB -run\u003dnone -count\u003d1 -ldflags\u003d-randlayout\u003d$R . | tee -a amd64-128.log\nGOARCH\u003damd64 GODEBUG\u003dsimd\u003d+256 go test -bench\u003dB -run\u003dnone -count\u003d1 -ldflags\u003d-randlayout\u003d$R . | tee -a amd64-256.log\nGOARCH\u003damd64 GODEBUG\u003dsimd\u003d0 go test -bench\u003dB -run\u003dnone -count\u003d1 -ldflags\u003d-randlayout\u003d$R . | tee -a amd64-0.log\nGOARCH\u003dwasm GOOS\u003dwasip1 go test -bench\u003dB -run\u003dnone -count\u003d1 -ldflags\u003d-randlayout\u003d$R . | tee -a wasm-128.log\nGOARCH\u003dwasm GOOS\u003dwasip1 GODEBUG\u003dsimd\u003d0 go test -bench\u003dB -run\u003dnone -count\u003d1 -ldflags\u003d-randlayout\u003d$R . | tee -a wasm-0.log\necho done with $R\ndone\n```\n\nrun on a quiet-ish Apple laptop\n\nthen cleaned with\n```\nfor i in *.log ; do sed -e \u00271,$s/-10//\u0027 -i \"\" $i; done\n```\n\n```\nbenchstat -ignore goarch,cpu,goos arm64*log amd64*log wasm*log\npkg: simd\n │ arm64-0.log │ arm64-128.log │ amd64-0.log │ amd64-128.log │ amd64-256.log │ wasm-0.log │ wasm-128.log │\n │ sec/op │ sec/op vs base │ sec/op vs base │ sec/op vs base │ sec/op vs base │ sec/op vs base │ sec/op vs base │\nIPFMA 465.99µ ± 1% 68.17µ ± 0% -85.37% (p\u003d0.000 n\u003d25) 567.46µ ± 3% +21.78% (p\u003d0.000 n\u003d25) 126.51µ ± 0% -72.85% (p\u003d0.000 n\u003d25) 166.29µ ± 0% -64.31% (p\u003d0.000 n\u003d25) 2839.92µ ± 1% +509.44% (p\u003d0.000 n\u003d25) 156.27µ ± 0% -66.47% (p\u003d0.000 n\u003d25)\nIP 533.06µ ± 2% 47.64µ ± 0% -91.06% (p\u003d0.000 n\u003d25) 645.00µ ± 0% +21.00% (p\u003d0.000 n\u003d25) 50.16µ ± 0% -90.59% (p\u003d0.000 n\u003d25) 49.39µ ± 0% -90.73% (p\u003d0.000 n\u003d25) 3702.28µ ± 1% +594.53% (p\u003d0.000 n\u003d25) 156.15µ ± 0% -70.71% (p\u003d0.000 n\u003d25)\nIPUnroll 521.14µ ± 1% 27.92µ ± 0% -94.64% (p\u003d0.000 n\u003d25) 634.97µ ± 3% +21.84% (p\u003d0.000 n\u003d25) 37.38µ ± 1% -92.83% (p\u003d0.000 n\u003d25) 38.85µ ± 0% -92.54% (p\u003d0.000 n\u003d25) 4543.60µ ± 1% +771.86% (p\u003d0.000 n\u003d25) 97.75µ ± 0% -81.24% (p\u003d0.000 n\u003d25)\nIPUnrollMore 531.67µ ± 2% 27.92µ ± 0% -94.75% (p\u003d0.000 n\u003d25) 634.78µ ± 2% +19.40% (p\u003d0.000 n\u003d25) 37.29µ ± 0% -92.99% (p\u003d0.000 n\u003d25) 42.76µ ± 0% -91.96% (p\u003d0.000 n\u003d25) 4679.65µ ± 1% +780.19% (p\u003d0.000 n\u003d25) 96.80µ ± 0% -81.79% (p\u003d0.000 n\u003d25)\nIPnosimd0 722.6µ ± 1% 738.3µ ± 1% +2.18% (p\u003d0.000 n\u003d25) 930.1µ ± 1% +28.73% (p\u003d0.000 n\u003d25) 938.6µ ± 1% +29.89% (p\u003d0.000 n\u003d25) 931.8µ ± 1% +28.95% (p\u003d0.000 n\u003d25) 525.2µ ± 0% -27.32% (p\u003d0.000 n\u003d25) 525.2µ ± 0% -27.32% (p\u003d0.000 n\u003d25)\nIPnosimd1 522.1µ ± 1% 526.0µ ± 1% +0.76% (p\u003d0.011 n\u003d25) 1049.7µ ± 1% +101.05% (p\u003d0.000 n\u003d25) 1049.5µ ± 1% +101.02% (p\u003d0.000 n\u003d25) 1049.8µ ± 1% +101.07% (p\u003d0.000 n\u003d25) 744.4µ ± 0% +42.58% (p\u003d0.000 n\u003d25) 744.7µ ± 0% +42.64% (p\u003d0.000 n\u003d25)\nIPnosimdA 640.2µ ± 0% 643.0µ ± 1% ~ (p\u003d0.088 n\u003d25) 968.7µ ± 1% +51.31% (p\u003d0.000 n\u003d25) 964.3µ ± 1% +50.63% (p\u003d0.000 n\u003d25) 966.5µ ± 1% +50.97% (p\u003d0.000 n\u003d25) 541.7µ ± 0% -15.38% (p\u003d0.000 n\u003d25) 542.0µ ± 0% -15.34% (p\u003d0.000 n\u003d25)\nIPnosimdAnotBloop 278.6µ ± 0% 280.1µ ± 0% +0.57% (p\u003d0.022 n\u003d25) 241.4µ ± 0% -13.34% (p\u003d0.000 n\u003d25) 241.4µ ± 0% -13.33% (p\u003d0.000 n\u003d25) 241.4µ ± 0% -13.35% (p\u003d0.000 n\u003d25) 392.2µ ± 0% +40.78% (p\u003d0.000 n\u003d25) 392.1µ ± 0% +40.78% (p\u003d0.000 n\u003d25)\ngeomean 510.7µ 143.2µ -71.96% 654.5µ +28.15% 194.3µ -61.96% 205.0µ -59.87% 1.441m +182.09% 257.3µ -49.63%\n```\n\nChange-Id: I09b912a3b68431972eeb91200f77a3dd95327d91\nReviewed-on: https://go-review.googlesource.com/c/go/+/787561\nReviewed-by: Junyang Shao \u003cshaojunyang@google.com\u003e\nLUCI-TryBot-Result: golang-scoped@luci-project-accounts.iam.gserviceaccount.com \u003cgolang-scoped@luci-project-accounts.iam.gserviceaccount.com\u003e\n", "tree_diff": [ { "type": "modify", "old_id": "41b6417d16716d7879c7538b2e72192fc9ca2983", "old_mode": 33188, "old_path": "src/simd/internal/bridge/simd_emulated.go", "new_id": "4a82d166c9956a4af2bc4ea3c0189e2b5969bf7c", "new_mode": 33188, "new_path": "src/simd/internal/bridge/simd_emulated.go" }, { "type": "modify", "old_id": "90aea587dabd81c378165bf2d8322da0fb5ca644", "old_mode": 33188, "old_path": "src/simd/ip_test.go", "new_id": "b88c499cadcbb1be4ca539a37eed7b067650f125", "new_mode": 33188, "new_path": "src/simd/ip_test.go" }, { "type": "modify", "old_id": "1ad562ee7c0b66e0c8c2e028fcf08278b265be7e", "old_mode": 33188, "old_path": "src/simd/simd_emulated.go", "new_id": "48f09d5cc164575179cd172b6d0ca513fc576392", "new_mode": 33188, "new_path": "src/simd/simd_emulated.go" } ] }