Merge pull request #428 from hakymulla/reducesumsq

Reduce Sum Square Operator

docs/CHANGELOG.md

@@ -4,6 +4,11 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased] - 2023-11-06
+
+## Added
+- Reduce Sum Square Operator.
+
 ## [Unreleased] - 2023-11-05
 
 ### Added

docs/SUMMARY.md

@@ -97,6 +97,7 @@
     * [tensor.bitwise_and](framework/operators/tensor/tensor.bitwise_and.md)
     * [tensor.round](framework/operators/tensor/tensor.round.md)
     * [tensor.scatter](framework/operators/tensor/tensor.scatter.md)
+    * [tensor.reduce\_sum\_square](framework/operators/tensor/tensor.reduce\_sum\_square.md)
     * [tensor.reduce\_l2](framework/operators/tensor/tensor.reduce\_l2.md)
     * [tensor.reduce\_l1](framework/operators/tensor/tensor.reduce\_l1.md)
   * [Neural Network](framework/operators/neural-network/README.md)

docs/framework/compatibility.md

@@ -75,6 +75,7 @@ You can see below the list of current supported ONNX Operators:
 |             [Round](operators/tensor/tensor.round.md)               | :white\_check\_mark: |
 |      [MaxInTensor](operators/tensor/tensor.max\_in\_tensor.md)      | :white\_check\_mark: |
 |                 [Max](operators/tensor/tensor.max.md)               | :white\_check\_mark: |
+|  [ReduceSumSquare](operators/tensor/tensor.reduce\_sum\_square.md)  | :white\_check\_mark: |
 |                 [Trilu](operators/tensor/tensor.trilu.md)               | :white\_check\_mark: |
 |             [Scatter](operators/tensor/scatter.max.md)              | :white\_check\_mark: |
 |          [ReduceL1](operators/tensor/tensor.reduce\_l1.md)          | :white\_check\_mark: |

docs/framework/operators/tensor/README.md

@@ -94,6 +94,7 @@ use orion::operators::tensor::TensorTrait;
 | [`tensor.round`](tensor.round.md) | Computes the round value of all elements in the input tensor. |
 | [`tensor.trilu`](tensor.trilu.md) | Returns the upper or lower triangular part of a tensor or a batch of 2D matrices. |
 | [`tensor.scatter`](tensor.scatter.md) | Produces a copy of input data, and updates value to values specified by updates at specific index positions specified by indices. |
+| [`tensor.reduce_sum_square`](tensor.reduce\_sum\_square.md) | Computes the sum square of the input tensor's elements along the provided axes.  |
 | [`tensor.reduce_l1`](tensor.reduce\_l1.md) | Computes the L1 norm of the input tensor's elements along the provided axes. |
 
 ## Arithmetic Operations

docs/framework/operators/tensor/tensor.reduce_sum_square.md

@@ -0,0 +1,38 @@
+## tensor.reduce_sum_square
+
+```rust 
+   fn reduce_sum_square(self: @Tensor<T>, axis: usize, keepdims: bool) -> Tensor<T>;
+```
+
+Computes the sum square of the input tensor's elements along the provided axes. 
+## Args
+
+* `self`(`@Tensor<T>`) - The input tensor.
+* `axis`(`usize`) - The dimension to reduce.
+* `keepdims`(`bool`) - If true, retains reduced dimensions with length 1.
+
+## Panics 
+
+* Panics if axis is not in the range of the input tensor's dimensions.
+
+## Returns
+
+A new `Tensor<T>` instance with the specified axis reduced by summing its elements.
+
+fn reduce_sum_square_example() -> Tensor<u32> {
+
+    let mut shape = ArrayTrait::<usize>::new();
+    shape.append(2);
+    shape.append(2);
+    let mut data = ArrayTrait::new();
+    data.append(1);
+    data.append(2);
+    data.append(3);
+    data.append(4);
+    let tensor = TensorTrait::<u32>::new(shape.span(), data.span());
+
+    We can call `reduce_sum_square` function as follows.
+    return tensor.reduce_sum_square(axis: 1, keepdims: true);
+}
+>>> [[5, 25]]
+```

nodegen/node/reduce_sum_square.py

@@ -0,0 +1,266 @@
+import numpy as np
+from nodegen.node import RunAll
+from ..helpers import make_node, make_test, to_fp, Tensor, Dtype, FixedImpl
+import numpy as np
+
+
+class Reduce_sum_square(RunAll):
+    @staticmethod
+    def reduce_sum_square_fp8x23():
+        def reduce_sum_square_export_do_not_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.int64)
+            keepdims = False
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int64)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=False).astype(np.int64)
+
+            x = Tensor(Dtype.FP8x23, x.shape, x.flatten())
+            y = Tensor(Dtype.FP8x23, y.shape, y.flatten())
+
+            name = "reduce_sum_square_fp8x23_export_do_not_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, false)", name)
+
+        def reduce_sum_square_export_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.int64)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int64)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int64)
+
+            x = Tensor(Dtype.FP8x23, x.shape, x.flatten())
+            y = Tensor(Dtype.FP8x23, y.shape, y.flatten())
+
+            name = "reduce_sum_square_fp8x23_export_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, true)", name)
+
+        def reduce_sum_square_axis_0():
+            shape = [3, 3, 3]
+            axes = np.array([0], dtype=np.int64)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int64)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int64)
+
+            x = Tensor(Dtype.FP8x23, x.shape, x.flatten())
+            y = Tensor(Dtype.FP8x23, y.shape, y.flatten())
+
+            name = "reduce_sum_square_fp8x23_export_negative_axes_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(0, true)", name)
+
+
+        reduce_sum_square_export_do_not_keepdims()
+        reduce_sum_square_export_keepdims()
+        reduce_sum_square_axis_0()
+
+    @staticmethod
+    def reduce_sum_square_fp16x16():
+        def reduce_sum_square_export_do_not_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.int64)
+            keepdims = False
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int64)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=False).astype(np.int64)
+
+            x = Tensor(Dtype.FP16x16, x.shape, x.flatten())
+            y = Tensor(Dtype.FP16x16, y.shape, y.flatten())
+
+            name = "reduce_sum_square_fp16x16_export_do_not_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, false)", name)
+
+        def reduce_sum_square_export_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.int64)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int64)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int64)
+
+            x = Tensor(Dtype.FP16x16, x.shape, x.flatten())
+            y = Tensor(Dtype.FP16x16, y.shape, y.flatten())
+
+            name = "reduce_sum_square_fp16x16_export_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, true)", name)
+
+        def reduce_sum_square_axis_0():
+            shape = [2, 2, 2]
+            axes = np.array([0], dtype=np.int64)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int64)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int64)
+
+            x = Tensor(Dtype.FP16x16, x.shape, x.flatten())
+            y = Tensor(Dtype.FP16x16, y.shape, y.flatten())
+
+            name = "reduce_sum_square_fp16x16_export_negative_axes_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(0, true)", name)
+
+
+        reduce_sum_square_export_do_not_keepdims()
+        reduce_sum_square_export_keepdims()
+        reduce_sum_square_axis_0()
+
+    @staticmethod
+    def reduce_sum_square_i8():
+        def reduce_sum_square_export_do_not_keepdims():
+            shape = [2, 2, 2]
+            axes = np.array([2], dtype=np.int8)
+            keepdims = False
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int8)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=False).astype(np.int8)
+
+            x = Tensor(Dtype.I8, x.shape, x.flatten())
+            y = Tensor(Dtype.I8, y.shape, y.flatten())
+
+            name = "reduce_sum_square_i8_export_do_not_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, false)", name)
+
+        def reduce_sum_square_export_keepdims():
+            shape = [2, 2, 2]
+            axes = np.array([2], dtype=np.int8)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int8)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int8)
+
+            x = Tensor(Dtype.I8, x.shape, x.flatten())
+            y = Tensor(Dtype.I8, y.shape, y.flatten())
+
+            name = "reduce_sum_square_i8_export_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, true)", name)
+
+        def reduce_sum_square_axis_0():
+            shape = [2, 2, 2]
+            axes = np.array([0], dtype=np.int8)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int8)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int8)
+
+            x = Tensor(Dtype.I8, x.shape, x.flatten())
+            y = Tensor(Dtype.I8, y.shape, y.flatten())
+
+            name = "reduce_sum_square_i8_export_negative_axes_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(0, true)", name)
+
+
+        reduce_sum_square_export_do_not_keepdims()
+        reduce_sum_square_export_keepdims()
+        reduce_sum_square_axis_0()
+
+    @staticmethod
+    def reduce_sum_square_i32():
+        def reduce_sum_square_export_do_not_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.int32)
+            keepdims = False
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int32)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=False).astype(np.int32)
+
+            x = Tensor(Dtype.I32, x.shape, x.flatten())
+            y = Tensor(Dtype.I32, y.shape, y.flatten())
+
+            name = "reduce_sum_square_i32_export_do_not_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, false)", name)
+
+        def reduce_sum_square_export_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.int32)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int32)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int32)
+
+            x = Tensor(Dtype.I32, x.shape, x.flatten())
+            y = Tensor(Dtype.I32, y.shape, y.flatten())
+
+            name = "reduce_sum_square_i32_export_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, true)", name)
+
+        def reduce_sum_square_axis_0():
+            shape = [3, 3, 3]
+            axes = np.array([0], dtype=np.int32)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.int32)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.int32)
+
+            x = Tensor(Dtype.I32, x.shape, x.flatten())
+            y = Tensor(Dtype.I32, y.shape, y.flatten())
+
+            name = "reduce_sum_square_i32_export_negative_axes_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(0, true)", name)
+
+
+        reduce_sum_square_export_do_not_keepdims()
+        reduce_sum_square_export_keepdims()
+        reduce_sum_square_axis_0()
+
+    @staticmethod
+    def reduce_sum_square_u32():
+        def reduce_sum_square_export_do_not_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.uint32)
+            keepdims = False
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.uint32)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=False).astype(np.uint32)
+
+            x = Tensor(Dtype.U32, x.shape, x.flatten())
+            y = Tensor(Dtype.U32, y.shape, y.flatten())
+
+            name = "reduce_sum_square_u32_export_do_not_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, false)", name)
+
+        def reduce_sum_square_export_keepdims():
+            shape = [3, 2, 2]
+            axes = np.array([2], dtype=np.uint32)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.uint32)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.uint32)
+
+            x = Tensor(Dtype.U32, x.shape, x.flatten())
+            y = Tensor(Dtype.U32, y.shape, y.flatten())
+
+            name = "reduce_sum_square_u32_export_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(2, true)", name)
+
+        def reduce_sum_square_axis_0():
+            shape = [3, 3, 3]
+            axes = np.array([0], dtype=np.uint32)
+            keepdims = True
+            x = np.reshape(np.arange(1, np.prod(shape) + 1, dtype=np.float32), shape).astype(np.uint32)
+            y = np.sum(a=np.square(x), axis=tuple(axes), keepdims=True).astype(np.uint32)
+
+            x = Tensor(Dtype.U32, x.shape, x.flatten())
+            y = Tensor(Dtype.U32, y.shape, y.flatten())
+
+            name = "reduce_sum_square_u32_export_negative_axes_keepdims"
+            make_node([x], [y], name)
+            make_test(
+                [x], y, "input_0.reduce_sum_square(0, true)", name)
+
+
+        reduce_sum_square_export_do_not_keepdims()
+        reduce_sum_square_export_keepdims()
+        reduce_sum_square_axis_0()

src/operators/tensor/core.cairo

@@ -91,6 +91,7 @@ impl TensorSerde<T, impl TSerde: Serde<T>, impl TDrop: Drop<T>> of Serde<Tensor<
 /// reduce_l1 - Computes the L1 norm of the input tensor's elements along the provided axes.
 /// trilu - Returns the upper or lower triangular part of a tensor or a batch of 2D matrices.
 /// scatter - Produces a copy of input data, and updates value to values specified by updates at specific index positions specified by indices.
+/// reduce_sum_square - Computes the sum square of the input tensor's elements along the provided axes. 
 /// reduce_l2 - Computes the L2 norm of the input tensor's elements along the provided axes.
 trait TensorTrait<T> {
     /// # tensor.new
@@ -3491,6 +3492,46 @@ trait TensorTrait<T> {
     /// ```
     ///
     fn reduce_l2(self: @Tensor<T>, axis: usize, keepdims: bool) -> Tensor<T>;
+    /// ## tensor.reduce_sum_square
+    ///
+    /// ```rust 
+    ///    fn reduce_sum_square(self: @Tensor<T>, axis: usize, keepdims: bool) -> Tensor<T>;
+    /// ```
+    ///
+    /// Computes the sum square of the input tensor's elements along the provided axes. 
+    /// ## Args
+    ///
+    /// * `self`(`@Tensor<T>`) - The input tensor.
+    /// * `axis`(`usize`) - The dimension to reduce.
+    /// * `keepdims`(`bool`) - If true, retains reduced dimensions with length 1.
+    ///
+    /// ## Panics 
+    /// 
+    /// * Panics if axis is not in the range of the input tensor's dimensions.
+    ///
+    /// ## Returns
+    ///
+    /// A new `Tensor<T>` instance with the specified axis reduced by summing its elements.
+    ///
+    /// fn reduce_sum_square_example() -> Tensor<u32> {
+    ///
+    ///     let mut shape = ArrayTrait::<usize>::new();
+    ///     shape.append(2);
+    ///     shape.append(2);
+    ///     let mut data = ArrayTrait::new();
+    ///     data.append(1);
+    ///     data.append(2);
+    ///     data.append(3);
+    ///     data.append(4);
+    ///     let tensor = TensorTrait::<u32>::new(shape.span(), data.span());
+    /// 
+    ///     We can call `reduce_sum_square` function as follows.
+    ///     return tensor.reduce_sum_square(axis: 1, keepdims: true);
+    /// }
+    /// >>> [[5, 25]]
+    /// ```
+    ///
+    fn reduce_sum_square(self: @Tensor<T>, axis: usize, keepdims: bool) -> Tensor<T>;
 }
 
 /// Cf: TensorTrait::new docstring