From f396f2e65eb0ca5f12c9fcb29774c69de94a2a71 Mon Sep 17 00:00:00 2001
From: Ilia Gyrdymov <ilgyrd@gmail.com>
Date: Fri, 7 Jun 2024 00:48:51 +0300
Subject: [PATCH] Speed up Vector.filled (#177)

---
 CHANGELOG.md                                  |  4 +++
 .../vector_initializing/float32x4_filled.dart | 25 +++++++++++++++++++
 lib/src/vector/float32x4_vector.dart          | 14 ++++++++---
 lib/src/vector/float64x2_vector.g.dart        | 14 ++++++++---
 pubspec.yaml                                  |  2 +-
 ...filled_constructor_test_group_factory.dart | 12 +++++++++
 6 files changed, 62 insertions(+), 9 deletions(-)
 create mode 100644 benchmark/vector/float32/vector_initializing/float32x4_filled.dart

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 0421f38e..9c336ad5 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,9 @@
 # Changelog
 
+## 13.12.4
+- `Vector.filled`:
+  - factory-constructor speed-up
+
 ## 13.12.3
 - `Vector.fromList`:
   - factory-constructor speed-up
diff --git a/benchmark/vector/float32/vector_initializing/float32x4_filled.dart b/benchmark/vector/float32/vector_initializing/float32x4_filled.dart
new file mode 100644
index 00000000..c050da1c
--- /dev/null
+++ b/benchmark/vector/float32/vector_initializing/float32x4_filled.dart
@@ -0,0 +1,25 @@
+// Approx. 0.23 second (MacBook Pro 2019), Dart version: 2.16.0
+
+import 'package:benchmark_harness/benchmark_harness.dart';
+import 'package:ml_linalg/vector.dart';
+
+const amountOfElements = 10000000;
+
+class Float32x4VectorFilledBenchmark extends BenchmarkBase {
+  Float32x4VectorFilledBenchmark()
+      : super('Vector initialization (filled), '
+            '$amountOfElements elements');
+
+  static void main() {
+    Float32x4VectorFilledBenchmark().report();
+  }
+
+  @override
+  void run() {
+    Vector.filled(amountOfElements, 10000);
+  }
+}
+
+void main() {
+  Float32x4VectorFilledBenchmark.main();
+}
diff --git a/lib/src/vector/float32x4_vector.dart b/lib/src/vector/float32x4_vector.dart
index 4cd26874..6a170c0d 100644
--- a/lib/src/vector/float32x4_vector.dart
+++ b/lib/src/vector/float32x4_vector.dart
@@ -69,12 +69,18 @@ class Float32x4Vector with IterableMixin<double> implements Vector {
 
   Float32x4Vector.filled(
       this.length, num value, this._cache, this._simdHelper) {
+    if (length < 0) {
+      throw ArgumentError('Length cannot be negative');
+    }
+
     _numOfBuckets = _getNumOfBuckets(length, _bucketSize);
-    final byteData = ByteData(_numOfBuckets * _bytesPerSimdElement);
-    _buffer = byteData.buffer;
+    final list = Float32x4List(_numOfBuckets);
+    _buffer = list.buffer;
 
-    for (var i = 0; i < length; i++) {
-      byteData.setFloat32(_bytesPerElement * i, value.toDouble(), Endian.host);
+    final simdValue = Float32x4.splat(value.toDouble());
+
+    for (var i = 0; i < _numOfBuckets; i++) {
+      list[i] = simdValue;
     }
   }
 
diff --git a/lib/src/vector/float64x2_vector.g.dart b/lib/src/vector/float64x2_vector.g.dart
index 0aebc49f..b235e825 100644
--- a/lib/src/vector/float64x2_vector.g.dart
+++ b/lib/src/vector/float64x2_vector.g.dart
@@ -72,12 +72,18 @@ class Float64x2Vector with IterableMixin<double> implements Vector {
 
   Float64x2Vector.filled(
       this.length, num value, this._cache, this._simdHelper) {
+    if (length < 0) {
+      throw ArgumentError('Length cannot be negative');
+    }
+
     _numOfBuckets = _getNumOfBuckets(length, _bucketSize);
-    final byteData = ByteData(_numOfBuckets * _bytesPerSimdElement);
-    _buffer = byteData.buffer;
+    final list = Float64x2List(_numOfBuckets);
+    _buffer = list.buffer;
 
-    for (var i = 0; i < length; i++) {
-      byteData.setFloat64(_bytesPerElement * i, value.toDouble(), Endian.host);
+    final simdValue = Float64x2.splat(value.toDouble());
+
+    for (var i = 0; i < _numOfBuckets; i++) {
+      list[i] = simdValue;
     }
   }
 
diff --git a/pubspec.yaml b/pubspec.yaml
index 5c63533f..00fc69bb 100644
--- a/pubspec.yaml
+++ b/pubspec.yaml
@@ -1,6 +1,6 @@
 name: ml_linalg
 description: SIMD-based linear algebra and statistics, efficient manipulation with numeric data
-version: 13.12.3
+version: 13.12.4
 homepage: https://github.com/gyrdym/ml_linalg
 
 environment:
diff --git a/test/vector/constructors/filled_constructor/vector_filled_constructor_test_group_factory.dart b/test/vector/constructors/filled_constructor/vector_filled_constructor_test_group_factory.dart
index 786ea88c..d7949c26 100644
--- a/test/vector/constructors/filled_constructor/vector_filled_constructor_test_group_factory.dart
+++ b/test/vector/constructors/filled_constructor/vector_filled_constructor_test_group_factory.dart
@@ -14,5 +14,17 @@ void vectorFilledConstructorTestGroupFactory(DType dtype) =>
           expect(vector.length, equals(10));
           expect(vector.dtype, dtype);
         });
+
+        test(
+            'should provide correct size of vectors when summing up, length is 5',
+            () {
+          final vector1 = Vector.filled(5, 2.0, dtype: dtype);
+          final vector2 = Vector.filled(5, 3.0, dtype: dtype);
+          final vector = vector1 + vector2;
+
+          expect(vector, equals([5.0, 5.0, 5.0, 5.0, 5.0]));
+          expect(vector.length, equals(5));
+          expect(vector.dtype, dtype);
+        });
       });
     });