Remove usage of valid window region in NHWC CPU kernels - Part1

Replace all calculate_max_window(ValidRegion, ...) with
calculate_max_window(TensorShape, ...) in CPU kernels

Resolves COMPMID-4152 (1/2)

Change-Id: I7403ea6b24b9e7889890839142a06439d6c8a499
Signed-off-by: SiCongLi <sicong.li@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5202
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

9 files changed, 81 insertions, 44 deletions

diff --git a/src/core/NEON/kernels/NELogicalKernel.cpp b/src/core/NEON/kernels/NELogicalKernel.cpp
index d98694ffe1..e1c24da777 100644
--- a/src/core/NEON/kernels/NELogicalKernel.cpp
+++ b/src/core/NEON/kernels/NELogicalKernel.cpp
@@ -287,9 +287,8 @@ void NELogicalKernel::configure(const ITensorInfo *input1, const ITensorInfo *in
     if(op != LogicalOperation::Not)
     {
         ARM_COMPUTE_ERROR_ON_NULLPTR(input2);
-        const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(*input1, *input2);
-        out_shape = broadcast_pair.first;
-        win       = calculate_max_window(broadcast_pair.second, Steps());
+        out_shape = TensorShape::broadcast_shape(input1->tensor_shape(), input2->tensor_shape());
+        win       = calculate_max_window(out_shape, Steps());
     }
     ICPPKernel::configure(win);
 
diff --git a/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp b/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp
index 6661326ea8..b287e18281 100644
--- a/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp
+++ b/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2020 Arm Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -1486,9 +1486,7 @@ void NEPixelWiseMultiplicationKernel::configure(ITensorInfo *input1, ITensorInfo
 
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input1, input2, output, scale, overflow_policy, rounding_policy));
 
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(*input1, *input2);
-    const TensorShape &out_shape    = broadcast_pair.first;
-    const ValidRegion &valid_region = broadcast_pair.second;
+    const TensorShape &out_shape = TensorShape::broadcast_shape(input1->tensor_shape(), input2->tensor_shape());
 
     // Auto initialize output if not initialized
     set_shape_if_empty(*output, out_shape);
@@ -1624,10 +1622,7 @@ void NEPixelWiseMultiplicationKernel::configure(ITensorInfo *input1, ITensorInfo
     }
 
     // Configure kernel window
-    Coordinates coord;
-    coord.set_num_dimensions(output->num_dimensions());
-    output->set_valid_region(valid_region);
-    Window win = calculate_max_window(valid_region, Steps());
+    Window win = calculate_max_window(out_shape);
 
     INEKernel::configure(win);
 }
@@ -1692,19 +1687,14 @@ void NEComplexPixelWiseMultiplicationKernel::configure(ITensorInfo *input1, ITen
     ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_complex(input1, input2, output));
 
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(*input1, *input2);
-    const TensorShape &out_shape    = broadcast_pair.first;
-    const ValidRegion &valid_region = broadcast_pair.second;
+    const TensorShape &out_shape = TensorShape::broadcast_shape(input1->tensor_shape(), input2->tensor_shape());
 
     // Auto initialize output if not initialized
     const TensorInfo out_info(out_shape, input1->num_channels(), input1->data_type());
     auto_init_if_empty(*output, out_info);
 
     // Configure kernel window
-    Coordinates coord;
-    coord.set_num_dimensions(output->num_dimensions());
-    output->set_valid_region(valid_region);
-    Window win = calculate_max_window(valid_region, Steps());
+    Window win = calculate_max_window(out_shape);
 
     INEKernel::configure(win);
 }
diff --git a/src/core/NEON/kernels/NESelectKernel.cpp b/src/core/NEON/kernels/NESelectKernel.cpp
index 1d5f2b61a1..22cd442889 100644
--- a/src/core/NEON/kernels/NESelectKernel.cpp
+++ b/src/core/NEON/kernels/NESelectKernel.cpp
@@ -224,7 +224,7 @@ void NESelectKernel::configure(const ITensor *c, const ITensor *x, const ITensor
         _function = it->second;
     }
 
-    Window win = calculate_max_window(x->info()->valid_region());
+    Window win = calculate_max_window(*x->info());
     INEKernel::configure(win);
 }
 
diff --git a/src/core/cpu/kernels/CpuAddKernel.cpp b/src/core/cpu/kernels/CpuAddKernel.cpp
index 31c7b2af60..fc88a7e22d 100644
--- a/src/core/cpu/kernels/CpuAddKernel.cpp
+++ b/src/core/cpu/kernels/CpuAddKernel.cpp
@@ -249,9 +249,7 @@ Status validate_arguments(const ITensorInfo &src0, const ITensorInfo &src1, cons
 
 std::pair<Status, Window> validate_and_configure_window(const ITensorInfo &src0, const ITensorInfo &src1, ITensorInfo &dst)
 {
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src0, src1);
-    const TensorShape &out_shape    = broadcast_pair.first;
-    const ValidRegion &valid_region = broadcast_pair.second;
+    const TensorShape &out_shape = TensorShape::broadcast_shape(src0.tensor_shape(), src1.tensor_shape());
 
     // Auto initialize dst if not initialized
     {
@@ -287,12 +285,9 @@ std::pair<Status, Window> validate_and_configure_window(const ITensorInfo &src0,
         }
     }
 
-    Window win = calculate_max_window(valid_region, Steps());
+    Window win = calculate_max_window(out_shape, Steps());
 
     // CpuAddKernel doesn't need padding so update_window_and_padding() can be skipped
-    Coordinates coord;
-    coord.set_num_dimensions(dst.num_dimensions());
-    dst.set_valid_region(valid_region);
     return std::make_pair(Status{}, win);
 }
 } // namespace
diff --git a/src/core/cpu/kernels/CpuElementwiseKernel.cpp b/src/core/cpu/kernels/CpuElementwiseKernel.cpp
index ab915b9d72..1ac21acbc0 100644
--- a/src/core/cpu/kernels/CpuElementwiseKernel.cpp
+++ b/src/core/cpu/kernels/CpuElementwiseKernel.cpp
@@ -182,14 +182,12 @@ void CpuElementwiseKernel::configure_common(const ITensorInfo *input1, const ITe
     ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
 
     // Configure kernel window
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(*input1, *input2);
-    const TensorShape &out_shape    = broadcast_pair.first;
-    const ValidRegion &valid_region = broadcast_pair.second;
+    const TensorShape &out_shape = TensorShape::broadcast_shape(input1->tensor_shape(), input2->tensor_shape());
 
     // Auto initialize output if not initialized
     auto_init_if_empty(*output, out_shape, 1, input1->data_type());
 
-    Window win = calculate_max_window(valid_region);
+    Window win = calculate_max_window(out_shape);
 
     ICpuKernel::configure(win);
 }
diff --git a/src/core/cpu/kernels/CpuElementwiseUnaryKernel.cpp b/src/core/cpu/kernels/CpuElementwiseUnaryKernel.cpp
index d2681bb060..2b5c11f8e1 100644
--- a/src/core/cpu/kernels/CpuElementwiseUnaryKernel.cpp
+++ b/src/core/cpu/kernels/CpuElementwiseUnaryKernel.cpp
@@ -113,14 +113,12 @@ void CpuElementwiseUnaryKernel::configure(ElementWiseUnary op, const ITensorInfo
     ARM_COMPUTE_ERROR_THROW_ON(validate(op, input, output));
 
     // Configure kernel window
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(input);
-    const TensorShape &out_shape    = broadcast_pair.first;
-    const ValidRegion &valid_region = broadcast_pair.second;
+    const TensorShape &out_shape = TensorShape::broadcast_shape(input.tensor_shape());
 
     // Auto initialize output if not initialized
     auto_init_if_empty(output, out_shape, 1, input.data_type());
 
-    Window win = calculate_max_window(valid_region);
+    Window win = calculate_max_window(out_shape);
 
     _op = op;
 
diff --git a/src/core/cpu/kernels/CpuSubKernel.cpp b/src/core/cpu/kernels/CpuSubKernel.cpp
index a03dcf2353..d7057bbe2b 100644
--- a/src/core/cpu/kernels/CpuSubKernel.cpp
+++ b/src/core/cpu/kernels/CpuSubKernel.cpp
@@ -201,9 +201,7 @@ void CpuSubKernel::configure(const ITensorInfo *src0, const ITensorInfo *src1, I
     ARM_COMPUTE_ERROR_ON_NULLPTR(src0, src1, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*src0, *src1, *dst, policy));
 
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(*src0, *src1);
-    const TensorShape &out_shape    = broadcast_pair.first;
-    const ValidRegion &valid_region = broadcast_pair.second;
+    const TensorShape &out_shape = TensorShape::broadcast_shape(src0->tensor_shape(), src1->tensor_shape());
 
     // Auto initialize dst if not initialized
     set_shape_if_empty(*dst, out_shape);
@@ -211,10 +209,7 @@ void CpuSubKernel::configure(const ITensorInfo *src0, const ITensorInfo *src1, I
     _policy = policy;
 
     // CpuSubKernel doesn't need padding so update_window_and_padding() can be skipped
-    Coordinates coord;
-    coord.set_num_dimensions(dst->num_dimensions());
-    dst->set_valid_region(valid_region);
-    Window win = calculate_max_window(valid_region, Steps());
+    Window win = calculate_max_window(out_shape, Steps());
 
     ICpuKernel::configure(win);
 }
diff --git a/src/core/helpers/WindowHelpers.cpp b/src/core/helpers/WindowHelpers.cpp
index ba10eb9775..75ffb71b4b 100644
--- a/src/core/helpers/WindowHelpers.cpp
+++ b/src/core/helpers/WindowHelpers.cpp
@@ -1,5 +1,5 @@
 /*
-* Copyright (c) 2020 Arm Limited.
+* Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -79,6 +79,57 @@ Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps,
     return window;
 }
 
+Window calculate_max_window(const TensorShape &shape, const Steps &steps, bool skip_border, BorderSize border_size)
+{
+    if(!skip_border)
+    {
+        border_size = BorderSize(0);
+    }
+
+    Window window;
+
+    window.set(0, Window::Dimension(
+                   // Skip the border left of the image
+                   border_size.left,
+                   // Skip the border right of the image
+                   // Make sure the window width is a multiple of the step size
+                   border_size.left + ceil_to_multiple(std::max(0, static_cast<int>(shape[0]) - static_cast<int>(border_size.left) - static_cast<int>(border_size.right)), steps[0]),
+                   steps[0]));
+
+    size_t n = 1;
+
+    if(shape.num_dimensions() > 1)
+    {
+        window.set(1, Window::Dimension(
+                       // Skip the border above the image
+                       border_size.top,
+                       // Skip the border below the image
+                       border_size.top + ceil_to_multiple(std::max(0, static_cast<int>(shape[1]) - static_cast<int>(border_size.top) - static_cast<int>(border_size.bottom)), steps[1]),
+                       steps[1]));
+
+        ++n;
+    }
+
+    if(shape.num_dimensions() > 2)
+    {
+        window.set(2, Window::Dimension(0, std::max<size_t>(1, shape[2]), steps[2]));
+
+        ++n;
+    }
+
+    for(; n < shape.num_dimensions(); ++n)
+    {
+        window.set(n, Window::Dimension(0, std::max<size_t>(1, shape[n])));
+    }
+
+    for(; n < Coordinates::num_max_dimensions; ++n)
+    {
+        window.set(n, Window::Dimension(0, 1));
+    }
+
+    return window;
+}
+
 Window calculate_max_enlarged_window(const ValidRegion &valid_region, const Steps &steps, BorderSize border_size)
 {
     const Coordinates &anchor = valid_region.anchor;
diff --git a/src/core/helpers/WindowHelpers.h b/src/core/helpers/WindowHelpers.h
index 9bc2135b6d..9216c33f16 100644
--- a/src/core/helpers/WindowHelpers.h
+++ b/src/core/helpers/WindowHelpers.h
@@ -1,5 +1,5 @@
 /*
-* Copyright (c) 2020 Arm Limited.
+* Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -107,6 +107,17 @@ Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps
 
 /** Calculate the maximum window for a given tensor shape and border setting
  *
+ * @param[in] shape       Shape of the tensor space
+ * @param[in] steps       (Optional) Number of elements processed for each step.
+ * @param[in] skip_border (Optional) If true exclude the border region from the window.
+ * @param[in] border_size (Optional) Border size.
+ *
+ * @return The maximum window the kernel can be executed on.
+ */
+Window calculate_max_window(const TensorShape &shape, const Steps &steps = Steps(), bool skip_border = false, BorderSize border_size = BorderSize());
+
+/** Calculate the maximum window for a given tensor shape and border setting
+ *
  * @param[in] info        Tensor info object defining the shape of the object for which the window is created.
  * @param[in] steps       (Optional) Number of elements processed for each step.
  * @param[in] skip_border (Optional) If true exclude the border region from the window.
@@ -116,7 +127,7 @@ Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps
  */
 inline Window calculate_max_window(const ITensorInfo &info, const Steps &steps = Steps(), bool skip_border = false, BorderSize border_size = BorderSize())
 {
-    return calculate_max_window(info.valid_region(), steps, skip_border, border_size);
+    return calculate_max_window(info.tensor_shape(), steps, skip_border, border_size);
 }
 
 /** Calculate the maximum window used by a horizontal kernel for a given tensor shape and border setting