From 7377107378d6c26439320fce78a551e85b5ad36a Mon Sep 17 00:00:00 2001
From: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com>
Date: Mon, 25 Mar 2024 15:55:42 +0000
Subject: Scatter GPU Kernel Implementation for 1D tensors.

Resolves: [COMPMID-6891, COMPMID-6892]
Change-Id: I5b094fff1bff4c4c59cc44f7d6beab0e40133d8e
Signed-off-by: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11394
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
---
 tests/validation/reference/ScatterLayer.cpp | 10 +++++-----
 tests/validation/reference/ScatterLayer.h   |  4 ++--
 2 files changed, 7 insertions(+), 7 deletions(-)

(limited to 'tests/validation/reference')

diff --git a/tests/validation/reference/ScatterLayer.cpp b/tests/validation/reference/ScatterLayer.cpp
index 920f2b9990..7543b46bb1 100644
--- a/tests/validation/reference/ScatterLayer.cpp
+++ b/tests/validation/reference/ScatterLayer.cpp
@@ -66,7 +66,7 @@ template float reduce_op(const float &current,const float &update,const ScatterF
 
 // Note : This function currently only supports 1D src, 1D updates, 2D indices, 1D output tensors.
 template <typename T>
-SimpleTensor<T> scatter_layer_internal(const SimpleTensor<T> &src, const SimpleTensor<T> &updates, const SimpleTensor<uint32_t> &indices, const TensorShape &out_shape, const ScatterInfo &info)
+SimpleTensor<T> scatter_layer_internal(const SimpleTensor<T> &src, const SimpleTensor<T> &updates, const SimpleTensor<int32_t> &indices, const TensorShape &out_shape, const ScatterInfo &info)
 {
     SimpleTensor<T> dst{ out_shape, src.data_type(), 1 };
 
@@ -84,14 +84,14 @@ SimpleTensor<T> scatter_layer_internal(const SimpleTensor<T> &src, const SimpleT
     }
 
     // 2. Get max index of output tensor, then iterate over index tensor.
-    const auto x_bound = dst.shape().x();
+    const int x_bound = static_cast<int>(dst.shape().x());
 
 
     for(int i = 0; i < indices.num_elements(); ++i)
     {
         // 3. Check whether index is out of bounds for dst, if not then apply reduce op.
         const auto index = indices[i];
-        if (index < x_bound) // Note : index is always >= 0 as datatype is unsigned.
+        if (index < x_bound && index >= 0) // Note : we ignore negative index values.
         {
             dst[index] = reduce_op(dst[index], updates[i], info.func);
         }
@@ -100,12 +100,12 @@ SimpleTensor<T> scatter_layer_internal(const SimpleTensor<T> &src, const SimpleT
 }
 
 template <typename T>
-SimpleTensor<T> scatter_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &updates, const SimpleTensor<uint32_t> &indices, const TensorShape &out_shape, const ScatterInfo &info)
+SimpleTensor<T> scatter_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &updates, const SimpleTensor<int32_t> &indices, const TensorShape &out_shape, const ScatterInfo &info)
 {
     return scatter_layer_internal<T>(src, updates, indices, out_shape, info);
 }
 
-template SimpleTensor<float> scatter_layer(const SimpleTensor<float> &src, const SimpleTensor<float> &updates, const SimpleTensor<uint32_t> &indices, const TensorShape &out_shape, const ScatterInfo &info);
+template SimpleTensor<float> scatter_layer(const SimpleTensor<float> &src, const SimpleTensor<float> &updates, const SimpleTensor<int32_t> &indices, const TensorShape &out_shape, const ScatterInfo &info);
 
 } // namespace reference
 } // namespace validation
diff --git a/tests/validation/reference/ScatterLayer.h b/tests/validation/reference/ScatterLayer.h
index dc441a8894..97d5e70b0d 100644
--- a/tests/validation/reference/ScatterLayer.h
+++ b/tests/validation/reference/ScatterLayer.h
@@ -37,10 +37,10 @@ namespace validation
 namespace reference
 {
 template <typename T>
-SimpleTensor<T> scatter_layer_internal(const SimpleTensor<T> &src, const SimpleTensor<T> &update, const SimpleTensor<uint32_t> &indices, const TensorShape &shape, const ScatterInfo &info);
+SimpleTensor<T> scatter_layer_internal(const SimpleTensor<T> &src, const SimpleTensor<T> &update, const SimpleTensor<int32_t> &indices, const TensorShape &shape, const ScatterInfo &info);
 
 template <typename T>
-SimpleTensor<T> scatter_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &update, const SimpleTensor<uint32_t> &indices, const TensorShape &shape, const ScatterInfo &info);
+SimpleTensor<T> scatter_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &update, const SimpleTensor<int32_t> &indices, const TensorShape &shape, const ScatterInfo &info);
 } // namespace reference
 } // namespace validation
 } // namespace test
-- 
cgit v1.2.1