diff --git a/README.md b/README.md index 05a2c2a..b34f89d 100644 --- a/README.md +++ b/README.md @@ -19,7 +19,7 @@ describes**: | Source | Published under | | --- | --- | -| branch `main`, with `params.ndarray.version: v0.43.1` in `hugo.yaml` | `https://go-ndarray.github.io/docs//`, here `/docs/0.43/` | +| branch `main`, with `params.ndarray.version: v0.44.0` in `hugo.yaml` | `https://go-ndarray.github.io/docs//`, here `/docs/0.44/` | | the newest version | also `https://go-ndarray.github.io/docs/latest/` | `https://go-ndarray.github.io/docs/` redirects to `latest/`. Each push to diff --git a/content/en/_index.md b/content/en/_index.md index b97cb96..f571a3b 100644 --- a/content/en/_index.md +++ b/content/en/_index.md @@ -38,7 +38,7 @@ WebAssembly, with nothing to install — build a pipeline step by step (creation | --- | --- | | Dtypes ([page](dtypes.md)) | bool, int8…int64, uint8…uint64, float32/float64, complex64/complex128 with NumPy 2 promotion: `AsType`, `ZerosOf`, `FromSlice`, `Data[T]`/`Item[T]`, weak Go scalars with `Scalar` | | Creation | `New`, `Zeros`, `Ones`, `Full`, `Arange`, `Linspace`, `Eye`, `Identity`, `FromData` | -| Views & shape | `Slice` (`All`/`R`/`Rng`/`From`/`To`/`Step`), `Reshape`, `Ravel`, `Transpose`, `Copy` | +| Views & shape | `Slice` (`All`/`R`/`Rng`/`From`/`To`/`Step`), `Reshape`, `Ravel` (a view when the elements are one row-major run, as `numpy.ravel`), `Transpose`, `Copy`, `Flatten` | | Elementwise | `Add`/`Sub`/`Mul`/`Div` (+`*Scalar`, +`*Into`), `Map`, `Neg`, `Abs`, comparisons (`Greater`/`Equal`/… → Bool masks), `Maximum`/`Minimum`, `Logical*`, `Real`/`Imag`/`Conj`/`Angle` | | Ufuncs | `Sqrt`, `Exp`, `Log`/`Log2`/`Log10`, `Sin`/`Cos`/`Tan`, `Floor`/`Ceil`, `Round` (half to even, as `np.round`; `RoundHalfAway` keeps the old rule), `Square`, `Power` | | Reductions | `Sum`, `Mean`, `Max`/`Min`, `Prod`, `ArgMax`/`ArgMin`, `CumSum`/`CumProd`, `Clip`, `Where` (+ per-axis) | @@ -115,8 +115,8 @@ is wrong on subnormals ([Performance](performance.md)): one row per public function (or a named exemption), against the pinned NumPy 2.5.4, `GOMAXPROCS=1` and one BLAS thread, interleaved rounds, refused at a 1-minute load of 8 or more, with the -allocation count of every row checked in CI. The table measured at v0.41.0 has -188 of 256 rows slower than NumPy and 54 faster ([ndarray#118](https://github.com/go-ndarray/ndarray/issues/118) +allocation count of every row checked in CI. The table measured at `c198892` (the code of v0.44.0) has +186 of 256 rows slower than NumPy, 51 faster and 19 tied ([ndarray#118](https://github.com/go-ndarray/ndarray/issues/118) tracks the slower rows by cause). The earlier multicore comparison with single-threaded NumPy is kept as [Performance history](performance-history.md) for its kernel-level analysis; its ratios are not a verdict. diff --git a/content/en/accuracy.md b/content/en/accuracy.md index 8576b02..ebc8022 100644 --- a/content/en/accuracy.md +++ b/content/en/accuracy.md @@ -8,7 +8,7 @@ description: "What go-ndarray promises about its results: a stated error bound, tags: [accuracy, numpy, math/big] --- -> Mirrored from [`README.md`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/README.md#accuracy-contract) at v0.43.1. +> Mirrored from [`README.md`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/README.md#accuracy-contract) at v0.44.0. The contract is **correct, and as fast as possible**. "Correct" means a stated error bound, checked against an exact oracle. Equality with NumPy's bits is not the contract: wherever it holds today, it is how a function was ported, and a faster default that stays within its bound may replace it. diff --git a/content/en/dtypes.md b/content/en/dtypes.md index ba08ef3..abb4224 100644 --- a/content/en/dtypes.md +++ b/content/en/dtypes.md @@ -8,7 +8,7 @@ description: "Why the dtype is a run-time property of one Array type, how arrays tags: [dtypes, numpy] --- -> Mirrored from [`docs/design-dtypes.md`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/design-dtypes.md) at v0.43.1. +> Mirrored from [`docs/design-dtypes.md`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/design-dtypes.md) at v0.44.0. > Status: implemented (Phase 2). Supersedes decision 4 of > [plan-ndarray.md](roadmap.md) ("`float64` first"). diff --git a/content/en/linalg.md b/content/en/linalg.md index 877e08d..6b0cb17 100644 --- a/content/en/linalg.md +++ b/content/en/linalg.md @@ -8,9 +8,9 @@ description: "Solve, Inv, Det, Cholesky, QR, SVD, Eig and the rest of numpy.lina tags: [linalg, numpy] --- -> Mirrored from [`docs/linalg.md`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/linalg.md) at v0.43.1. +> Mirrored from [`docs/linalg.md`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/linalg.md) at v0.44.0. -Package [`linalg`](https://pkg.go.dev/github.com/go-ndarray/ndarray@v0.43.1/linalg) gives ndarray arrays the functions of +Package [`linalg`](https://pkg.go.dev/github.com/go-ndarray/ndarray@v0.44.0/linalg) gives ndarray arrays the functions of `numpy.linalg`, written in Go without cgo or assembly of its own: the matrix products it relies on are the packed SIMD GEMM of [go-blas/blas](https://github.com/go-blas/blas), which ndarray's `MatMul` uses too (the same kernels, moved there in v0.42.0, #122). diff --git a/content/en/performance.md b/content/en/performance.md index 01c45e5..1187526 100644 --- a/content/en/performance.md +++ b/content/en/performance.md @@ -8,11 +8,11 @@ description: "go-ndarray against the pinned NumPy on one core, both sides: the m tags: [performance, numpy, refbench] --- -> Mirrored from [`docs/perf.md`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/perf.md) at v0.43.1. +> Mirrored from [`docs/perf.md`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/perf.md) at v0.44.0. -The table is [perf-table.md](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/perf-table.md), generated by +The table is [perf-table.md](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/perf-table.md), generated by [refbench](https://github.com/go-scikit/refbench) (go-scikit/refbench v0.3.0) -from [perf.json](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/perf.json); nothing in it is edited by hand. The rules are +from [perf.json](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/perf.json); nothing in it is edited by hand. The rules are those of the go-scikit performance campaign ([go-scikit/roadmap#27](https://github.com/go-scikit/roadmap/issues/27)): @@ -36,7 +36,7 @@ those of the go-scikit performance campaign `TestRefSmoke` runs every NumPy call once. - **One case per public function** (a function with several shapes of use has several rows, `Array.SumAxis/axis0` and `/axis1`), or a named exemption with - its reason in [perf-exempt.md](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/perf-exempt.md). CI fails on a public function + its reason in [perf-exempt.md](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/perf-exempt.md). CI fails on a public function with neither. ## What a row times @@ -72,27 +72,77 @@ go test -run 'TestDataParity|TestRefSmoke' -perf.parity . # needs uv ## Status -The table is **measured** at v0.41.0 (`de3f41ac`) on one pinned core of cfarm420 -(AMD EPYC 7773X, Zen 3). The 1-minute load over the rounds was 4.12 to 5.56, -and all 256 rows are valid. +The table is **measured** at `c198892` (the head of the `feat/ravel-real-views` +branch, clean) on 2026-11-15, on one pinned core of cfarm420 (AMD EPYC 7773X, +Zen 3), NumPy 2.5.4. The 1-minute load over the rounds was 3.01 to 3.86, and +all 256 rows are valid. It replaces the table measured at v0.41.0 +(`de3f41ac`, 2026-11-14, load 4.12 to 5.56), and the section below lists what +differs. -**188 rows are slower than NumPy and 54 are faster.** Every slower row is an +**186 rows are slower than NumPy, 51 are faster and 19 tie** (v0.41.0: 188, 54 +and 14). Every slower row is an open defect. They are tracked by cause in [#118](https://github.com/go-ndarray/ndarray/issues/118) rather than in one issue per row: -- result allocation: the allocating `Add` is 3.1× slower, while `AddInto` ties (1.03); -- elementwise operations without a vector kernel: `Ceil` is 20× slower, `Abs` 12×; -- copies where NumPy returns a view or wraps the buffer; -- `linalg` at 64 × 64 (1.4–4×); -- the `fft` shifts; +- result allocation: the allocating `Add` is 3.7× slower, while `AddInto` is + within 7% (1.07); +- elementwise operations: the unary `Ceil` (5.4×), `Floor` (5.3×), `Square` + (4.9×), `Round`, `Abs` and `Neg` (4.6×) now run on vector kernels and what + remains is the result allocation; `Maximum`, `Minimum` and `Power` + (5.3–5.5×) and `Log2` (4.2×) are the worst of the others; +- copies where NumPy returns a view or wraps the buffer (`Flatten` 5.0×, + `Copy` 4.8×, `Imag` 4.9×); +- `linalg` at 64 × 64 (1.5–4×); +- the `fft` shifts (9.7× and 9.9×); - the `random` draws built on `log`/`exp`. -Faster, for example: `SqrtInto` (0.51), `fft.FFT` (0.54), `Sum` (0.70), `Exp` (0.89). - -### Since the table - -Fixed after v0.41.0, measured by `go test -bench` on the same core, not yet -by a new table run. On 10⁵ float64 at load 2.2–2.4, over three interleaved +Faster, for example: `SqrtInto` (0.52), `fft.FFT` (0.64), `Sum` (0.76), `Log10` +(0.57). `Exp` (0.92) ties. + +### Since the v0.41.0 table + +The changes below were made after v0.41.0 and are included in the table above. +The ratios of this table against the v0.41.0 table (Go time / NumPy time, 10⁵ +float64): + +| row | v0.41.0 | now | allocations | +|---|---:|---:|---| +| `Ravel` | 566.8 | 0.35 | 4 → 3 | +| `Real` | 565.9 | 0.57 | 4 → 3 | +| `Ceil` | 20.4 | 5.44 | 5 | +| `Floor` | 19.2 | 5.28 | 5 | +| `Round` | 13.8 | 4.63 | 5 | +| `Abs` | 11.9 | 4.61 | 5 | +| `Square` | 11.0 | 4.88 | 5 | +| `Neg` | 10.3 | 4.59 | 5 | +| `AddScalar` | 8.56 | 5.40 | 8 | +| `Full` | 8.00 | 5.96 | 5 | +| `Ones` | 7.34 | 5.83 | 5 | +| `fft.RFFT2` | 1.41 | 1.14 | 31 → 30 | +| `fft.RFFT2With` | 1.50 | 1.28 | 31 → 30 | +| `fft.IHFFT` | 1.33 | 1.10 | 35 → 34 | +| `fft.IHFFTWith` | 1.22 | 1.20 | 35 → 34 | +| `fft.RFFTN` | 0.95 | 0.89 | 32 → 31 | +| `fft.RFFTNWith` | 0.92 | 0.81 | 32 → 31 | +| `fft.RFFT` | 0.90 | 0.84 | 30 → 29 | +| `fft.RFFTWith` | 0.95 | 0.97 | 30 → 29 | + +`Ravel` and `Real` are now faster than NumPy, and the `fft` real-input +transforms each save one allocation because they call `x.Real()`, which is a +view. The unary rows remain slower than NumPy, by about 5× instead of 10–20×: +they still allocate the result, which `AddInto` shows to be the main part of +the remaining gap (`Add` 3.74, `AddInto` 1.07). + +Treat small ratio changes with care: the two tables were measured at different +loads, and rows whose code did not change moved by as much as 0.2 to 0.3 +(`Sqrt` 1.01 to 1.20, `fft.IRFFT` 0.72 to 1.06, `AddInto` 1.03 to 1.07, +`fft.HFFT` 1.01 to 1.11). Only the rows above whose +change comes from a known code change, and whose size exceeds that spread, are +claimed as effects. The `fft` ratios in the table above (0.03 to 0.2 lower) are +within that spread, so only the allocation counts are attributed to `Real`. + +The measurements that follow were made at the time by `go test -bench` on the +same core. On 10⁵ float64 at load 2.2–2.4, over three interleaved rounds: - The scalar forms (`AddScalar` … `DivScalar`, `AddInto` … `DivInto`, and any Float64 operation with a one-element operand against a contiguous @@ -131,8 +181,14 @@ rounds: `Round` was measured in a second run of the same kind (four interleaved rounds, load 2.25 to 2.42), v0.42.0 against this change. `RoundHalfAway`, the control on the scalar path, takes 324–345 µs in the same run. These are - times of the Go code against itself; no NumPy ratio is claimed here, as the - new table run has not been made. + times of the Go code against itself; the NumPy ratios are in the table above. +- `Ravel` of a contiguous array and `Real` of a real array return views, as + NumPy does, instead of copying (the table's rows were 566× slower than + NumPy, which returns views). On 10⁵ float64, cfarm420, one pinned core, three + interleaved rounds at load 2.1 to 2.3: `Ravel` went from 83–90 µs to 93–101 + ns and `Real` from 84–91 µs to 101–104 ns, 803 kB to 128 B per call. The + new table measures them against NumPy at 61 ns and 90 ns against 174 ns and + 157 ns (ratios 0.35 and 0.57). This is a behaviour change (see the README). What remains of the allocation cause is Go's own cost. A CPU profile of a bare `make([]float64, 100000)` shows about 60% garbage-collector marking (assists @@ -140,7 +196,7 @@ in the allocating goroutine) and about 23% zeroing. NumPy frees each result when its last reference goes and reuses that memory warm. In Go, the way to avoid both costs is to not allocate: use the `*Into` forms or a `Workspace`. -The run was made from the release's own code. The test binary was -cross-compiled from a clean v0.41.0 worktree, the tag was shipped to the host -as a git bundle, and the binary ran from outside the clone so that the tree -stayed clean. Done that way, the recorded commit is the tag itself. +The run was made from the measured commit's own code. The test binary was +cross-compiled from a clean checkout of `c198892`, the commit was shipped to +the host as a git bundle, and the binary ran from outside the clone so that the +tree stayed clean. Done that way, the recorded commit is `c198892` itself. diff --git a/content/en/random.md b/content/en/random.md index e306469..5752874 100644 --- a/content/en/random.md +++ b/content/en/random.md @@ -8,9 +8,9 @@ description: "numpy.random with NumPy's numbers for the same seed: the bit gener tags: [random, numpy] --- -> Mirrored from [`docs/random.md`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/random.md) at v0.43.1. +> Mirrored from [`docs/random.md`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/random.md) at v0.44.0. -Package [`random`](https://pkg.go.dev/github.com/go-ndarray/ndarray@v0.43.1/random) is `numpy.random` over go-ndarray arrays. Its goal +Package [`random`](https://pkg.go.dev/github.com/go-ndarray/ndarray@v0.44.0/random) is `numpy.random` over go-ndarray arrays. Its goal is migration: a notebook moved from NumPy to Go keeps its numbers for the same seed. @@ -24,7 +24,7 @@ u, _ := random.Rand(3) // np.random.rand(3) ``` `Example`, `ExampleSeed` and `ExampleGenerator_ChoiceN` in -[`example_test.go`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/random/example_test.go) print NumPy's own numbers. +[`example_test.go`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/random/example_test.go) print NumPy's own numbers. ## What is implemented @@ -142,7 +142,7 @@ random arguments. Used as-is, that made 3% of `np.random.randn` values and 10% of `lognormal` values one ULP off. So `random` rounds `log`, `log1p`, `exp`, `expm1` and `pow` **correctly** -([`libm.go`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/random/libm.go)): a fast evaluation to about 2⁻⁶⁸ with Ziv's +([`libm.go`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/random/libm.go)): a fast evaluation to about 2⁻⁶⁸ with Ziv's rounding test, and a double-double fallback (about 2⁻¹⁰⁰) for the one argument in ~2¹⁵ that needs it. Checked against mpmath at 300 bits on 63,000 arguments (0 misrounded; the committed `libm.json` keeps 8,400), and the fast paths diff --git a/content/en/roadmap.md b/content/en/roadmap.md index 8bc2e71..60744f3 100644 --- a/content/en/roadmap.md +++ b/content/en/roadmap.md @@ -8,7 +8,7 @@ description: "The plan for go-ndarray and what ships today." tags: [roadmap] --- -> Mirrored from [`docs/plan-ndarray.md`](https://github.com/go-ndarray/ndarray/blob/v0.43.1/docs/plan-ndarray.md) at v0.43.1. +> Mirrored from [`docs/plan-ndarray.md`](https://github.com/go-ndarray/ndarray/blob/v0.44.0/docs/plan-ndarray.md) at v0.44.0. > Goal: a **pure-Go (CGO=0)** NumPy-style N-dimensional array library, as a > **standalone, reusable** module with correct scalar kernels now and a clean @@ -118,7 +118,7 @@ suite (plus the per-arch CI jobs already wired in `.github/workflows/ci.yml`). NumPy. What follows is the history of the levers behind the kernel seam, measured then against NumPy 2.2.4 (see [docs/perf.md](performance.md) and the repo's - [BENCHMARKS.md](https://github.com/go-ndarray/ndarray/blob/v0.43.1/BENCHMARKS.md)): + [BENCHMARKS.md](https://github.com/go-ndarray/ndarray/blob/v0.44.0/BENCHMARKS.md)): - **Multicore.** `Add/Sub/Mul/Div/Map`, `Sum/Max/Min`, the axis reductions, `Concatenate`/`Stack`, and the GEMM fan out across `GOMAXPROCS` above a size threshold (small arrays stay serial). Plus an allocation fix: the diff --git a/content/en/security.md b/content/en/security.md index 092d225..7a80fb5 100644 --- a/content/en/security.md +++ b/content/en/security.md @@ -8,7 +8,7 @@ description: "How to report a vulnerability, what the library guarantees, and th tags: [security] --- -> Mirrored from [`SECURITY.md`](https://github.com/go-ndarray/ndarray/blob/main/SECURITY.md) at v0.43.1. +> Mirrored from [`SECURITY.md`](https://github.com/go-ndarray/ndarray/blob/main/SECURITY.md) at v0.44.0. ## Reporting diff --git a/hugo.yaml b/hugo.yaml index fadf2f3..ea86c86 100644 --- a/hugo.yaml +++ b/hugo.yaml @@ -6,7 +6,7 @@ # have used since the site was built with MkDocs and mike. The CI sets it # through --baseURL, from params.ndarray.version below. This default is the # same build. -baseURL: "https://go-ndarray.github.io/docs/0.43/" +baseURL: "https://go-ndarray.github.io/docs/0.44/" title: "go-ndarray" # themes/hextra is the tannevaled/hextra fork, pinned at a release tag: # upstream Hextra plus opt-in features (partial menu items, combos, @@ -100,7 +100,7 @@ params: # The ndarray release these pages describe. Its minor version is the # documentation's version label (v0.9.1 -> 0.9): the CI reads it here. ndarray: - version: "v0.43.1" + version: "v0.44.0" navbar: displayTitle: true