released 2.0.7

Author: gbenroscience

.gitattributes

@@ -0,0 +1 @@
+*.mp4 filter=lfs diff=lfs merge=lfs -text

LATEST.md

@@ -1,6 +1,11 @@
 # ParserNG
 
 
+
+### ParserNG 2.0.7 is out on maven-central! 
+Full migration of all inbuilt functions and UDFs to the Vector API(not stats functions), leading to 2.0x to 20x speedup in bulk evaluations
+
+
 ### ParserNG 2.0.6 is out on maven-central! 
 Very stable release. Lots of bug fixes in bulk gelu, swiglu etc
 

README.md

@@ -3,12 +3,12 @@
   <parent>
     <artifactId>parser-ng-parent</artifactId>
     <groupId>com.github.gbenroscience</groupId>
-    <version>2.0.6</version>
+    <version>2.0.7</version>
   </parent>
   <modelVersion>4.0.0</modelVersion>
   <artifactId>parser-ng-bench</artifactId>
   <name>parser-ng-bench</name>
-  <version>2.0.6</version>
+  <version>2.0.7</version>
   <build>
     <plugins>
       <plugin>
@@ -24,7 +24,7 @@
               <finalName>benchmarks</finalName>
               <transformers>
                 <transformer>
-                  <mainClass>org.openjdk.jmh.Main</mainClass>
+                  <mainClass>com.github.gbenroscience.parser.ng.bench.ParserNgBench</mainClass>
                 </transformer>
               </transformers>
               <filters>

parser-ng-bench/dependency-reduced-pom.xml

@@ -4,11 +4,11 @@
     <parent>
         <groupId>com.github.gbenroscience</groupId>
         <artifactId>parser-ng-parent</artifactId>
-        <version>2.0.6</version>
+        <version>2.0.7</version>
     </parent>
     <artifactId>parser-ng-bench</artifactId>
     <name>parser-ng-bench</name>
-    <version>2.0.6</version>
+    <version>2.0.7</version>
     <packaging>jar</packaging>
     <properties>
         <exec.mainClass>com.github.gbenroscience.parser.ng.bench.ParserNgBench</exec.mainClass>
@@ -31,7 +31,7 @@
                             <finalName>benchmarks</finalName>
                             <transformers>
                                 <transformer implementation="org.apache.maven.plugins.shade.resource.ManifestResourceTransformer">
-                                    <mainClass>org.openjdk.jmh.Main</mainClass>
+                                    <mainClass>com.github.gbenroscience.parser.ng.bench.ParserNgBench</mainClass>
                                 </transformer>
                             </transformers>
                             <filters>

parser-ng-bench/pom.xml

@@ -136,7 +136,7 @@ public static final String[] getVars(String e) {
 
     private double[] vars;
     private VectorTurboBench.JaninoMathFunction fastEvaluator;
-
+  
     SIMDVectorTurboEvaluator.SIMDVectorCompositeExpression simdVectorTurbo;
 
     private int varCount;

parser-ng-bench/src/main/java/com/github/gbenroscience/parser/ng/bench/ParserNgBench.java

@@ -1,14 +1,14 @@
 # parser-ng-simd
 
-### ParserNG 2.0.5
+### ParserNG 2.0.7
 
 High-performance, hardware-accelerated mathematical kernels for Java. **No JNI. No native binaries.** Powered entirely by the JDK Vector API and optimized for C2 loop fusion.
 
 `parser-ng-simd` provides an architectural extension for ParserNG, enabling zero-dependency, tile-based bulk evaluations via `jdk.incubator.vector`. It compiles mathematical expression trees into vectorized loops optimized for L1/L2 cache residency, achieving near-native throughput directly on the JVM without the configuration overhead of complex native toolchains (such as ND4J/`libnd4j`).
 
 For examples and more information:
 
-- [SIMD-EXAMPLES.md](../parser-ng/BULK.md) — what’s new in 2.0.5
+- [SIMD-EXAMPLES.md](../parser-ng/BULK.md) — what’s new in 2.0.7
 
 ---
 
@@ -18,30 +18,28 @@ For examples and more information:
 
 Executed on an Intel Core i5-1135G7 environment running a JDK 24 Early Access build.
 
-| Operation | Matrix: $70 \times 70$ | Matrix: $100 \times 100$ | Matrix: $200 \times 200$ |
-| --- | --- | --- | --- |
-| **GELU** | 10.7 ns/elt ($52.5\,\mu\text{s}$) | 7.46 ns/elt ($74.6\,\mu\text{s}$) | 7.30 ns/elt ($292\,\mu\text{s}$) |
-| **SwiGLU** | — | — | 13.30 ns/elt ($530\,\mu\text{s}$) |
 
-> 📊 **Note on Peak Efficiency:** An execution speed of `7.3 ns/element` for GELU translates to roughly 26 CPU cycles. This safely hits the theoretical throughput ceiling of what AVX2 combined with hardware `vdexp` intrinsics can process within a managed runtime.
+| Operation | Matrix Scale: 70 x 70 | Matrix Scale: 100 x 100 | Matrix Scale: 200 x 200 |
+| :--- | :--- | :--- | :--- |
+| **GELU** | 2.01 ns/elt (9.85μs) | 1.95 ns/elt (19.5μs) | **1.96 ns/elt** (78.4μs) |
+| **SwiGLU** | 1.57 ns | 1.57 ns | **1.54 ns/elt** (61.6μs) |
 
+> 📊 **Peak Hardware Efficiency:** A measurement of `2.00 ns/element` for a complex activation function like GELU equates to roughly 8 raw CPU cycles. This securely reaches the physical throughput ceiling of what AVX2 vector units combined with hardware `vdexp` primitives can achieve in a managed code layer.
 #### Ecosystem Comparison ($40\text{k}$ Elements, 2 Cores)
 
 | Library / Engine | GELU Execution Profile | Architectural Mechanism |
 | --- | --- | --- |
-| **parser-ng-simd** | **292 µs (7.3 ns/elt)** | **Pure Java Vector API (Direct SIMD)** |
+| **parser-ng-simd** | **292 µs (2.00 ns/elt)** | **Pure Java Vector API (Direct SIMD)** |
 
 
-*Data Scaling Boundary:* At extreme workloads (e.g., 67M+ records), computation throughput bounds shift toward physical DRAM bandwidth, where execution stabilizes at approximately `1.3x` faster than optimized scalar bytecode compilers.
-
 ---
 
 ### Key Architectural Pillars
 
-1. **Tile-Based Workload Dispatch:** Compute tasks are split into deterministic block matrices (from $70 \times 70$ up to $200 \times 200$), ensuring data tiles remain completely resident within L1/L2 CPU caches while minimizing loop-bound orchestration overhead.
-2. **C2 Loop Fusion:** The engine collapses composite abstract syntax trees (ASTs) into unified native vector loops. Sequential steps like `exp + tanh + FMA` resolve into 3 to 4 hyper-optimized AVX2 machine instructions.
-3. **Pure Java Intrinsic Pipelines:** Built directly on JDK Vector API concepts (`FloatVector`, `VectorMask`), utilizing hardware-level primitive acceleration without relying on unsafe memory access blocks or third-party wrappers.
-4. **Zero-Allocation Hot Paths:** Execution tracks pre-allocated internal memory frames. Buffer reuse prevents heap churning, entirely insulating math execution paths from Garbage Collection (GC) pauses.
+
+1. **C2 Loop Fusion:** The engine collapses composite abstract syntax trees (ASTs) into unified native vector loops. Sequential steps like `exp + tanh + FMA` resolve into 3 to 4 hyper-optimized AVX2 machine instructions.
+2. **Pure Java Intrinsic Pipelines:** Built directly on JDK Vector API concepts (`FloatVector`, `VectorMask`), utilizing hardware-level primitive acceleration without relying on unsafe memory access blocks or third-party wrappers.
+3. **Zero-Allocation Hot Paths:** Execution tracks pre-allocated internal memory frames. Buffer reuse prevents heap churning, entirely insulating math execution paths from Garbage Collection (GC) pauses.
 
 ---
 
@@ -68,13 +66,13 @@ Include both the core module and the SIMD engine extension in your `pom.xml`:
     <dependency>
         <groupId>com.github.gbenroscience</groupId>
         <artifactId>parser-ng</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.7</version>
     </dependency>
 
     <dependency>
         <groupId>com.github.gbenroscience</groupId>
         <artifactId>parser-ng-simd</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.7</version>
     </dependency>
 </dependencies>
 ```
@@ -90,7 +88,7 @@ Out-of-the-box support includes: `gelu`, `swiglu`, `silu`, `tanh`, `exp`, `log`,
 To execute microbenchmarks and verify SIMD registration on your host architecture, use the following profile goal:
 
 ```bash
-mvn clean test -Dtest=VectorTurboEvaluatorTest
+mvn clean test -Dtest=SIMDTurboEvaluatorTest
 
 ```
 

parser-ng-simd/README.md

@@ -3,14 +3,14 @@
   <parent>
     <artifactId>parser-ng-parent</artifactId>
     <groupId>com.github.gbenroscience</groupId>
-    <version>2.0.6</version>
+    <version>2.0.7</version>
   </parent>
   <modelVersion>4.0.0</modelVersion>
   <artifactId>parser-ng-simd</artifactId>
   <name>ParserNG SIMD Extension</name>
   <description>Hardware-accelerated SIMD vector calculation pipelines for ParserNG
         Rich and Performant, Cross Platform Java Library(100% Java).No native dependencies.(e.g. JNI) 
-        Version 2.0.6 features updates to GeLU and SwiGLU for correctness
+        Version 2.0.7 features updates to GeLU and SwiGLU for correctness
         For library support, customizations or donations, reach me at gbenroscience@gmail.com|+2348165779034</description>
   <url>https://github.com/gbenroscience/ParserNG</url>
   <build>

parser-ng-simd/dependency-reduced-pom.xml

@@ -8,7 +8,7 @@
     <parent>
         <groupId>com.github.gbenroscience</groupId>
         <artifactId>parser-ng-parent</artifactId>
-        <version>2.0.6</version>
+        <version>2.0.7</version>
         <relativePath>../pom.xml</relativePath>
     </parent>
 
@@ -18,7 +18,7 @@
     <description>
         Hardware-accelerated SIMD vector calculation pipelines for ParserNG
         Rich and Performant, Cross Platform Java Library(100% Java).No native dependencies.(e.g. JNI) 
-        Version 2.0.6 features updates to GeLU and SwiGLU for correctness
+        Version 2.0.7 features updates to GeLU and SwiGLU for correctness
         For library support, customizations or donations, reach me at gbenroscience@gmail.com|+2348165779034
     </description>
     <url>https://github.com/gbenroscience/ParserNG</url>

parser-ng-simd/pom.xml

@@ -1,4 +1,4 @@
-# ParserNG 2.0.5
+# ParserNG 2.0.7
 
 An ultra-high-performance mathematical expression parsing and evaluation engine for Java. Version 2.0.x introduces a breakthrough **Block Vectorization Engine** (`VectorTurboEvaluator`) and `SIMDVectorTurboEvaluator` powered by the Java **Vector API** (`jdk.incubator.vector`). 
 
@@ -39,12 +39,12 @@ To use SIMDVectorTurboEvaluator(e.g. on modern servers(JDK 21+) and laptops(JDK2
 <dependency>
     <groupId>com.github.gbenroscience</groupId>
     <artifactId>parser-ng</artifactId>
-    <version>2.0.5</version>
+    <version>2.0.7</version>
 </dependency>
 <dependency>
     <groupId>com.github.gbenroscience</groupId>
     <artifactId>parser-ng-simd</artifactId>
-    <version>2.0.5</version>
+    <version>2.0.7</version>
 </dependency>
 ```
 
@@ -53,7 +53,7 @@ To use VectorTurboEvaluator(e.g. on Android and legacy systems supporting <JDK21
 <dependency>
     <groupId>com.github.gbenroscience</groupId>
     <artifactId>parser-ng</artifactId>
-    <version>2.0.5</version>
+    <version>2.0.7</version>
 </dependency>
 ```