Merge pull request #130 from sk-ai-net/feature/97-memory-layout

#97 Memory layout

Author: michalharakal

gradle.properties

@@ -1,5 +1,5 @@
 GROUP=sk.ainet.core
-VERSION_NAME=0.0.1.1
+VERSION_NAME=0.0.1.2
 
 POM_DESCRIPTION=SKaiNET
 

skainet-lang/skainet-lang-api/src/commonMain/kotlin/sk/ainet/lang/tensor/Shape.kt

@@ -8,7 +8,7 @@ public data class Shape(val dimensions: IntArray) {
     }
 
     val volume: Int
-        get() = dimensions.fold(1) { a, x -> a * x }
+        get() = dimensions.fold(if (dimensions.isNotEmpty()) 1 else 0) { a, x -> a * x }
 
     val rank: Int
         get() = dimensions.size

skainet-lang/skainet-lang-api/src/commonMain/kotlin/sk/ainet/lang/tensor/Tensor.kt

@@ -7,82 +7,82 @@ import sk.ainet.lang.types.DType
 /**
  * The core tensor abstraction in the SKaiNET framework, representing a fundamental
  * architectural decision to compose tensors from data and operations.
- * 
+ *
  * ## Fundamental Architectural Decision: Tensor Composition
- * 
+ *
  * This interface embodies a key architectural principle in the SKaiNET framework:
  * **tensors are composed of two distinct, complementary components:**
- * 
+ *
  * 1. **Data Component (`TensorData`)** - Responsible for:
  *    - Multi-dimensional data storage and indexing
  *    - Memory layout and access patterns
  *    - Shape and dimensional metadata
  *    - Type-safe element access
- * 
+ *
  * 2. **Operations Component (`TensorOps`)** - Responsible for:
  *    - Mathematical operations and transformations
  *    - Computational algorithms
  *    - Operation chaining and composition
  *    - Performance-optimized implementations
- * 
+ *
  * ## Benefits of This Compositional Architecture
- * 
+ *
  * **Separation of Concerns**: Data management is cleanly separated from computational
  * logic, making the codebase more maintainable and easier to understand.
- * 
+ *
  * **Flexibility**: Different data storage strategies (dense, sparse, distributed) can
  * be combined with different operation implementations (CPU, GPU, specialized hardware)
  * without tight coupling.
- * 
+ *
  * **Performance Optimization**: Data layout can be optimized independently from
  * computational algorithms, enabling targeted performance improvements.
- * 
+ *
  * **Extensibility**: New data formats or operation types can be added without
  * modifying existing code, following the open-closed principle.
- * 
+ *
  * **Testability**: Data and operations can be tested independently, improving
  * test coverage and reducing complexity.
- * 
+ *
  * ## Usage Pattern
- * 
+ *
  * The tensor acts as a unified interface that delegates data access to the `data`
  * component and computational operations to the `ops` component, providing a
  * seamless experience while maintaining the benefits of compositional design.
- * 
+ *
  * @param T the data type constraint extending DType, defining the numerical precision
  * @param V the actual value type that will be stored and accessed
  */
 public interface Tensor<T : DType, V> {
     /**
      * The data component responsible for storage, indexing, and memory management.
-     * 
+     *
      * This component handles all aspects related to data storage and access,
-     * including multi-dimensional indexing, shape management, and memory layout
+     * including multidimensional indexing, shape management, and memory layout
      * optimization. It provides the foundation for the tensor's data model.
      */
     public val data: TensorData<T, V>
-    
+
     /**
      * The operations component responsible for computational algorithms and transformations.
-     * 
+     *
      * This component provides all mathematical operations that can be performed
      * on the tensor, from basic arithmetic to complex neural network operations.
      * It leverages the data component for element access while encapsulating
      * all computational logic.
      */
-    public val ops: TensorOps<T, V>
+    public val ops: TensorOps<V>
 
     /**
      * The data type descriptor defining the numerical precision and value representation.
-     * 
+     *
      * This property provides metadata about the tensor's data type, enabling
      * type-safe operations and appropriate memory allocation strategies.
      */
     public val dtype: T
 
     /**
      * The shape descriptor inherited from the data component.
-     * 
+     *
      * This property delegates to the data component's shape, maintaining consistency
      * between the tensor interface and its underlying data representation.
      */
@@ -91,7 +91,7 @@ public interface Tensor<T : DType, V> {
 
     /**
      * The total number of elements in the tensor.
-     * 
+     *
      * This computed property provides quick access to the tensor's total size,
      * calculated from the shape's volume. Useful for memory allocation and
      * iteration operations.
@@ -101,7 +101,7 @@ public interface Tensor<T : DType, V> {
 
     /**
      * The number of dimensions in the tensor.
-     * 
+     *
      * This computed property provides quick access to the tensor's dimensionality,
      * derived from the shape's rank. Essential for dimension-aware operations
      * and broadcasting compatibility checks.

skainet-lang/skainet-lang-api/src/commonMain/kotlin/sk/ainet/lang/tensor/data/TensorData.kt

@@ -7,14 +7,14 @@ import sk.ainet.lang.types.DType
  * A fundamental data structure interface that provides indexed access to elements.
  * 
  * This interface serves as the base for all data structures that need to provide
- * element access through multi-dimensional indexing. It is designed to support
+ * element access through multidimensional indexing. It is designed to support
  * efficient access patterns commonly used in neural network computations.
  * 
  * @param T the type of elements that can be accessed
  */
 public interface ItemsAccessor<T> {
     /**
-     * Retrieves an element at the specified multi-dimensional indices.
+     * Retrieves an element at the specified multidimensional indices.
      * 
      * This operator function allows accessing elements using bracket notation
      * with variable number of indices, supporting tensors of any dimensionality.
@@ -24,6 +24,12 @@ public interface ItemsAccessor<T> {
      * @throws IndexOutOfBoundsException if any index is out of bounds
      */
     public operator fun get(vararg indices: Int): T
+
+
+    /**
+     * Setter
+     */
+    public operator fun set(vararg indices: Int, value: T)
 }
 
 /**
@@ -61,4 +67,6 @@ public interface TensorData<T : DType, V> : ItemsAccessor<V> {
      * @return the Shape object describing this tensor's dimensional structure
      */
     public val shape: Shape
+
+    public companion object
 }

skainet-lang/skainet-lang-api/src/commonMain/kotlin/sk/ainet/lang/tensor/data/pprint.kt

@@ -0,0 +1,63 @@
+package sk.ainet.lang.tensor.data
+
+import sk.ainet.lang.types.DType
+
+
+public fun <T: DType,V> TensorData<T,V>.pprint(): String {
+    return when (this.shape.rank) {
+        0 -> {
+            // Scalar - single value
+            this.toString()
+        }
+        1 -> {
+            // Vector - horizontal representation with parentheses
+            val sb = StringBuilder()
+            sb.append("[ ")
+            for (i in 0 until this.shape[0]) {
+                if (i > 0) sb.append(", ")
+                sb.append(this[i].toString())
+            }
+            sb.append(" ]")
+            sb.toString()
+        }
+        2 -> {
+            // Matrix - vertical representation with Unicode brackets
+            val sb = StringBuilder()
+            val rows = this.shape[0]
+            val cols = this.shape[1]
+
+            for (i in 0 until rows) {
+                // Left bracket
+                val leftBracket = when {
+                    rows == 1 -> "( "
+                    i == 0 -> "⎛ "
+                    i == rows - 1 -> "⎝ "
+                    else -> "⎜ "
+                }
+                sb.append(leftBracket)
+
+                // Matrix elements
+                for (j in 0 until cols) {
+                    if (j > 0) sb.append(", ")
+                    sb.append(this[i, j].toString())
+                }
+
+                // Right bracket
+                val rightBracket = when {
+                    rows == 1 -> " )"
+                    i == 0 -> " ⎞"
+                    i == rows - 1 -> " ⎠"
+                    else -> " ⎟"
+                }
+                sb.append(rightBracket)
+
+                if (i < rows - 1) sb.append("\n")
+            }
+            sb.toString()
+        }
+        else -> {
+            // Higher rank tensors - use toString
+            this.toString()
+        }
+    }
+}