basedpl 0.1.63__tar.gz → 0.1.67__tar.gz

{basedpl-0.1.63 → basedpl-0.1.67}/.cargo/config.toml

@@ -1,7 +1,8 @@
-# rustc runs LLVM's straight-line (SLP) vectorizer only at opt-level 3. `passes=slp-vectorizer` adds it to our opt-level 2 builds.
-# A target's own `rustflags` replace these. The x86-64 and wasm entries repeat the pass.
+# rustc runs LLVM's straight-line (SLP) vectorizer only at opt-level 3. `passes` adds it to our opt-level 2 builds.
+# With LTO, rustc runs these passes before the link step, and the loop vectorizer only at it. `loop-vectorize` runs first. Otherwise SLP packs the lanes of a short inner loop, and the loop vectorizer then leaves the outer loop alone.
+# A target's own `rustflags` replace these. The x86-64 and wasm entries repeat the passes.
 [build]
-rustflags = ["-C", "passes=slp-vectorizer"]
+rustflags = ["-C", "passes=loop-vectorize,slp-vectorizer"]
 
 # rustymath uses Accelerate's per-thread BLAS controls on macOS.
 [env]
@@ -9,11 +10,11 @@ MACOSX_DEPLOYMENT_TARGET = "15.0"
 
 # x86-64 builds need a CPU with AVX2 (x86-64-v3). The kernels rely on its rounding and wider vector instructions.
 [target.'cfg(target_arch = "x86_64")']
-rustflags = ["-C", "target-cpu=x86-64-v3", "-C", "passes=slp-vectorizer"]
+rustflags = ["-C", "target-cpu=x86-64-v3", "-C", "passes=loop-vectorize,slp-vectorizer"]
 
 # The browser build uses wasm's 128-bit SIMD, which current browsers all support.
 [target.wasm32-unknown-unknown]
-rustflags = ["-C", "target-feature=+simd128", "-C", "passes=slp-vectorizer"]
+rustflags = ["-C", "target-feature=+simd128", "-C", "passes=loop-vectorize,slp-vectorizer"]
 
 [alias]
 # Builds the npm package into `wasm/pkg/`. `xtask` builds optimised, because its wasm-bindgen pass runs on every call.

{basedpl-0.1.63 → basedpl-0.1.67}/.github/workflows/speed.yml

@@ -18,6 +18,8 @@ jobs:
       - run: pip install 'fastws-cli>=0.0.25' && cargo stage --profile dist
       - run: grep -m1 'model name' /proc/cpuinfo
       - run: wheel/scripts/bpl --costs tests/speed.txt
+      # `bpl-tests thread_scaling` times a large product, a 512×512 SVD and Hermitian eigendecomposition, a stack of QR decompositions and a large elementwise function on one thread and on the runner's cores, and fails when one speeds up less than 1.5 times. Speed budgets are single-thread measurements, so this is the check of multithreaded speed. It comes last, so that a failure here still leaves every cost logged.
+      - run: pip install -e . && bpl-tests thread_scaling --bpl wheel/scripts/bpl
 
   wasm:
     runs-on: ubicloud-standard-4

{basedpl-0.1.63 → basedpl-0.1.67}/Cargo.lock

@@ -93,7 +93,7 @@ checksum = "ac07cdecf99051d9a5238b80f35af32cdeba5b336e55d957b318b50137e18da5"
 
 [[package]]
 name = "basedpl"
-version = "0.1.63"
+version = "0.1.67"
 dependencies = [
  "base64 0.23.1",
  "bytemuck",
@@ -137,7 +137,7 @@ dependencies = [
 
 [[package]]
 name = "basedpl-cli"
-version = "0.1.63"
+version = "0.1.67"
 dependencies = [
  "async-trait",
  "basedpl",
@@ -150,7 +150,7 @@ dependencies = [
 
 [[package]]
 name = "basedpl-py"
-version = "0.1.63"
+version = "0.1.67"
 dependencies = [
  "basedpl",
  "num-bigint",
@@ -165,7 +165,7 @@ dependencies = [
 
 [[package]]
 name = "basedpl-wasm"
-version = "0.1.63"
+version = "0.1.67"
 dependencies = [
  "basedpl",
  "js-sys",
@@ -867,9 +867,9 @@ dependencies = [
 
 [[package]]
 name = "kernmini"
-version = "0.1.20"
+version = "0.1.21"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4ba8e9bfee7c661fa640e38a41f878edfd3052387d54f51611148713efa6941e"
+checksum = "9d391ffc3e7d540ca47206110b11ad619095d1fb8911604ca2abc1e43fb2c488"
 dependencies = [
  "async-trait",
  "bytes",
@@ -1413,9 +1413,9 @@ dependencies = [
 
 [[package]]
 name = "rustymath"
-version = "0.1.40"
+version = "0.1.43"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4dc5f83f6e77929139aa2a0080425cb483f6fedae447ae40bc788031fca80a01"
+checksum = "47219acf94cb55feca56bf23818d8d1c013a17fd63023721e6edac118c065629"
 dependencies = [
  "foldhash",
  "half",
@@ -1429,13 +1429,14 @@ dependencies = [
  "rand",
  "rayon",
  "strength_reduce",
+ "web-time",
 ]
 
 [[package]]
 name = "rustymath-buffer"
-version = "0.1.40"
+version = "0.1.43"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "74411856ed0b141e1e64c63eea50ce5bb7cbef0a30f9b310e19fe81d1e83d72d"
+checksum = "6991cc5072aadb4af88d88526fb6479f59f9641b86fd6f10b7a407b8e02bd6e4"
 dependencies = [
  "pyo3",
  "rustymath",
@@ -1699,9 +1700,9 @@ dependencies = [
 
 [[package]]
 name = "tokio"
-version = "1.54.0"
+version = "1.54.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a0c168a474b95bfecfe908a65d4bdffc523144b649a74c09c286afb35d53e0bc"
+checksum = "b52fd3255258d26fac7264ab5a76b53c13367136dc011d3d6ad4f8ed070396a3"
 dependencies = [
  "bytes",
  "libc",
@@ -2085,7 +2086,7 @@ checksum = "3ad82d2a33cdc9674dc7465672f271e096168fcdbe0f799d9e6db8c5892679dc"
 
 [[package]]
 name = "xtask"
-version = "0.1.63"
+version = "0.1.67"
 dependencies = [
  "anyhow",
  "rustymath",

{basedpl-0.1.63 → basedpl-0.1.67}/Cargo.toml

@@ -5,14 +5,14 @@ members = ["py"]
 join_comments = true
 
 [workspace.package]
-version = "0.1.63"
+version = "0.1.67"
 edition = "2024"
 rust-version = "1.98"
 license = "Apache-2.0"
 
 [workspace.dependencies]
-rustymath = { version = ">=0.1.40, <1", default-features = false }
-rustymath-buffer = ">=0.1.40, <1"
+rustymath = { version = ">=0.1.43, <1", default-features = false }
+rustymath-buffer = ">=0.1.43, <1"
 pyo3 = ">=0.29.2, <1"
 pyo3-build-config = ">=0.29.2, <1"
 
@@ -34,8 +34,8 @@ inherits = "release"
 
 [profile.release]
 opt-level = 2
-# Without LTO, an edit recompiles only the codegen units whose code changed.
-lto = "off"
+# Thin LTO across the crate's own codegen units, Cargo's default. Incremental builds put `#[inline]` helpers such as iterator adapters in codegen units of their own, where only LTO can inline them.
+lto = false
 overflow-checks = false
 debug-assertions = false
 incremental = true
@@ -45,10 +45,13 @@ debug = false
 [profile.dist]
 inherits = "release"
 incremental = false
-# Cargo's default LTO: thin LTO across the crate's codegen units, which inlines between them.
-lto = false
 strip = true
 
+# A local run before each release (`cargo test --profile checked`): debug assertions add `session()`'s turn check and the standard library's checks in unsafe code, at 12 to 55% per call.
+[profile.checked]
+inherits = "test"
+debug-assertions = true
+
 [features]
 default = ["native"]
 # rustymath's own linear algebra, beside Accelerate on macOS, for `RUSTYMATH_NATIVE=1`. The Mac release builds turn it off.