Sync the landing page with ndarray v0.2.0

hugo.toml

@@ -6,15 +6,15 @@ title   = 'go-ndarray'
 disableKinds = ['taxonomy', 'term', 'rss', 'sitemap']
 
 [params]
-  description = "go-ndarray is a pure-Go (CGO=0) NumPy-style n-dimensional array library for float64 — slicing/views, broadcasting, ufuncs, reductions and linear algebra, multicore + SIMD, faster than NumPy on the parallelizable core."
+  description = "go-ndarray is a pure-Go (CGO=0) NumPy-style n-dimensional array library for float64 — slicing/views, broadcasting, ufuncs, reductions and linear algebra, multicore + SIMD on amd64 and arm64, with a Workspace arena for loops."
   brand       = "go-ndarray"
   github      = "go-ndarray"
   heroAccent  = "go-ndarray"
   heroTagline = "NumPy-style n-dimensional arrays in pure Go. float64, multicore + SIMD."
 
   heroIntro = [
     "<strong>go-ndarray</strong> is the <code>numpy</code> equivalent for Go: a float64 n-dimensional array with creation routines, <strong>strided views that share data</strong>, broadcasting elementwise ops and ufuncs, reductions (incl. arg / cumulative / clip / where), manipulation, and linear algebra (a panel-packed, cache-blocked GEMM) — all in <strong>pure Go, with cgo disabled</strong>.",
-    "Ruby has no cgo-free ndarray (<code>Numo::NArray</code> and <code>NMatrix</code> are C extensions) and <code>gonum</code>'s optimized assembly is amd64-only. go-ndarray pairs a portable scalar core with multicore fan-out and go-asmgen SIMD kernels, beating single-threaded NumPy on the parallelizable core. 100% test coverage is the bar, differentially checked against numpy 2.2."
+    "Ruby has no cgo-free ndarray (<code>Numo::NArray</code> and <code>NMatrix</code> are C extensions) and <code>gonum</code>'s optimized assembly is amd64-only. go-ndarray pairs a portable pure-Go core with multicore fan-out and, on amd64 and arm64, go-asmgen SIMD kernels; a <code>Workspace</code> arena takes the garbage collector out of loops. Measured against NumPy 2.5.3 on a 16-core Zen 3 it is faster on whole-array and row reductions, on <code>Exp</code> from 256 Ki elements and on <code>MatMul</code> against single-threaded OpenBLAS, and slower against 16-thread BLAS. 100% test coverage is the bar, differentially checked against NumPy."
   ]
 
 [menu]

layouts/index.html

@@ -18,8 +18,10 @@ <h2>Why go-ndarray</h2>
     brings the <strong>n-dimensional array</strong> vocabulary &mdash; shapes, strided
     views, broadcasting, ufuncs, reductions and linear algebra &mdash; to Go with
     <strong>no cgo</strong>, cross-compiling to a static binary everywhere. The hot paths
-    are multicore + SIMD, so it <strong>beats single-threaded NumPy</strong> on the
-    parallelizable core. It is the cgo-free ndarray the Go and Ruby ecosystems lacked.
+    are multicore, and SIMD on amd64 and arm64, so it <strong>beats NumPy</strong> on
+    whole-array and row reductions, large <code>Exp</code> and <code>MatMul</code> against
+    single-threaded BLAS; a <code>Workspace</code> arena removes the garbage-collector cost
+    of temporaries in loops. It is the cgo-free ndarray the Go and Ruby ecosystems lacked.
   </p>
 </section>
 
@@ -28,11 +30,11 @@ <h2>Repositories</h2>
   <div class="grid">
     <div class="card">
       <h4><a href="https://github.com/go-ndarray/ndarray">ndarray</a></h4>
-      <p>The library. Creation (<code>Zeros</code>/<code>Arange</code>/<code>Linspace</code>/…), strided <strong>views</strong>, broadcasting elementwise + ufuncs, reductions (<code>Sum</code>/<code>Max</code>/<code>ArgMax</code>/<code>CumSum</code>/<code>Clip</code>/<code>Where</code>), manipulation (<code>Reshape</code>/<code>Concatenate</code>/<code>Stack</code>), and linalg (<code>MatMul</code>/<code>Dot</code>) via a panel-packed cache-blocked GEMM with a go-asmgen SIMD-FMA micro-kernel.</p>
+      <p>The library. Creation (<code>Zeros</code>/<code>Arange</code>/<code>Linspace</code>/…), strided <strong>views</strong>, broadcasting elementwise + ufuncs, reductions (<code>Sum</code>/<code>Max</code>/<code>ArgMax</code>/<code>CumSum</code>/<code>Clip</code>/<code>Where</code>), manipulation (<code>Reshape</code>/<code>Concatenate</code>/<code>Stack</code>), linalg (<code>MatMul</code>/<code>Dot</code>) via a panel-packed cache-blocked GEMM with a go-asmgen SIMD-FMA micro-kernel (NEON 4&times;8 on arm64, AVX2/FMA 6&times;8 on amd64), and a <code>Workspace</code> arena for loops (<code>Use</code>/<code>Reset</code>/<code>Detach</code>).</p>
     </div>
     <div class="card">
       <h4><a href="https://go-ndarray.github.io/docs/">docs</a></h4>
-      <p>Versioned documentation site (MkDocs Material): API reference, the roadmap, and honest benchmark pages versus NumPy 2.2 / OpenBLAS. <a href="https://github.com/go-ndarray/docs">Source →</a></p>
+      <p>Versioned documentation site (MkDocs Material): API reference, the roadmap, and honest benchmark pages versus NumPy 2.5.3 / OpenBLAS 0.3.34. <a href="https://github.com/go-ndarray/docs">Source →</a></p>
     </div>
     <div class="card">
       <h4><a href="https://github.com/go-ndarray/brand">brand</a></h4>
@@ -45,10 +47,11 @@ <h4><a href="https://github.com/go-ndarray/brand">brand</a></h4>
   <h2>Quality bar</h2>
   <ul class="features">
     <li><strong>Pure Go.</strong> <code>CGO_ENABLED=0</code> across the org. No cgo, no BLAS shim.</li>
-    <li><strong>100% test coverage.</strong> Gated in CI, differentially checked against <code>numpy</code> 2.2.</li>
-    <li><strong>Faster than NumPy</strong> on the parallelizable core (Add/Mul/Sum/Sqrt/Max) — multicore fan-out + SIMD. The packed GEMM reaches <strong>tuned-BLAS parity at 1024&sup2;</strong> (&approx;1.00&times; single-threaded vecLib, ~373 GFLOP/s) and beats the pure-Go <code>gonum</code> <strong>4&ndash;10&times;</strong>; <code>Dot</code> wins at parity. Honest where tuned BLAS still leads at small n.</li>
-    <li><strong>Multi-arch.</strong> Portable scalar core on all six 64-bit targets; go-asmgen SIMD kernels where the Go assembler allows it.</li>
-    <li><strong>BSD-3-Clause</strong> on every source file.</li>
+    <li><strong>100% test coverage.</strong> Gated in CI, differentially checked against <code>numpy</code>.</li>
+    <li><strong>Measured, wins and losses.</strong> Against NumPy 2.5.3 + OpenBLAS 0.3.34 on a 16-core AMD Zen 3: <strong>faster</strong> on whole-array <code>Sum</code>/<code>Mean</code>/<code>Max</code> at 4&nbsp;Mi (2.3&ndash;4.5&times;), <code>SumAxis(1)</code> (2.2&times;), <code>MaxAxis(1)</code> (1.7&times;), <code>Exp</code> from 256&nbsp;Ki elements (2.4&ndash;5.6&times;) and <code>MatMul</code> against single-threaded OpenBLAS (3.7&ndash;5.9&times; from 512&sup2;); <strong>slower</strong> against 16-thread OpenBLAS (<code>MatMul</code> 0.55&times; at 1024&sup2;, <code>Dot</code>/mat&middot;vec 0.16&times;) and on allocating elementwise forms outside a <code>Workspace</code>. In a <code>Workspace</code>, <code>x + y</code> on 1&nbsp;024 elements drops from 2.6 to 0.38&nbsp;&micro;s (2.1&times; NumPy). On Apple silicon the packed GEMM reaches <strong>tuned-BLAS parity at 1024&sup2;</strong> and beats the pure-Go <code>gonum</code> <strong>4&ndash;10&times;</strong>.</li>
+    <li><strong>Accurate <code>Exp</code>/<code>Log</code>.</strong> Ports of Arm's optimized-routines: worst 0.504 / 0.508 ULP against a 300-bit reference, and correct where Go's amd64 <code>math.Exp</code> returns +Inf (<a href="https://github.com/golang/go/issues/81995">golang/go#81995</a>) and <code>math.Log</code> is wrong on subnormals (<a href="https://github.com/golang/go/issues/56600">golang/go#56600</a>).</li>
+    <li><strong>Multi-arch.</strong> go-asmgen SIMD kernels on amd64 (SSE2, AVX2/FMA picked at run time) and arm64 (NEON); riscv64, loong64, ppc64le, s390x and the 32-bit targets run the same pure-Go code those kernels are tested against. CI runs the suite on amd64, arm64 and 386 natively and on riscv64/loong64/ppc64le/s390x/arm under qemu, on Linux, macOS and Windows, and compiles it for every <code>GOOS/GOARCH</code> pair Go supports; v0.2.0 was also run on real amd64, arm64, ppc64le, riscv64 and loong64 hardware.</li>
+    <li><strong>BSD-3-Clause</strong> licensed; <code>Exp</code> and <code>Log</code> are MIT-licensed ports of Arm's optimized-routines.</li>
   </ul>
 </section>
 {{ end }}