Remove OpenCL padding: CLPixelWiseMultiplicationKernel

- Change kernel's vec_size to 16 / sizeof(output)
- Change ICLKernel.cpp to handle broadcast without padding

Resolve COMPMID-3913

Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: I03e884b250ef5784dc109bff8cf2c96b345d119f
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5450
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

3 files changed, 1 insertions, 78 deletions

diff --git a/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp b/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
index 5ebaf5d122..efebf2b84c 100644
--- a/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
+++ b/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
@@ -25,7 +25,7 @@
 
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLFillBorderKernel.h"
+#include "src/core/CL/ICLKernel.h"
 #include "src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h"
 
 #include <utility>
diff --git a/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.cpp b/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.cpp
index c4f11a4e29..137a0de6a7 100644
--- a/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.cpp
+++ b/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.cpp
@@ -24,7 +24,6 @@
 #include "src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h"
 
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLFillBorderKernel.h"
 #include "src/core/gpu/cl/ClCompileContext.h"
 #include "src/core/gpu/cl/kernels/ClPixelWiseMultiplicationKernel.h"
 
@@ -32,44 +31,12 @@ namespace arm_compute
 {
 namespace opencl
 {
-namespace
-{
-ITensorPack select_border_input(ITensorPack &tensors)
-{
-    ITensorPack pack;
-    if(tensors.get_tensor(TensorType::ACL_DST)->info()->dimension(0) > 1)
-    {
-        if(tensors.get_const_tensor(TensorType::ACL_SRC_1)->info()->dimension(0) == 1)
-        {
-            pack.add_tensor(TensorType::ACL_SRC, tensors.get_const_tensor(TensorType::ACL_SRC_1));
-        }
-        else
-        {
-            pack.add_tensor(TensorType::ACL_SRC, tensors.get_const_tensor(TensorType::ACL_SRC_0));
-        }
-    }
-    return pack;
-}
-} // namespace
-
 void ClPixelWiseMultiplication::configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, float scale,
                                           ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
 {
     auto k = std::make_unique<kernels::ClPixelWiseMultiplicationKernel>();
     k->configure(compile_context, src1, src2, dst, scale, overflow_policy, rounding_policy, act_info);
     _kernel = std::move(k);
-
-    if(dst->dimension(0) > 1)
-    {
-        ITensorInfo *broadcasted_info = (src1->dimension(0) == 1) ? src1 : src2;
-
-        if(broadcasted_info->dimension(0) == 1)
-        {
-            auto b = std::make_unique<CLFillBorderKernel>();
-            b->configure(compile_context, broadcasted_info, _kernel->border_size(), BorderMode::REPLICATE);
-            _border_handler = std::move(b);
-        }
-    }
 }
 
 Status ClPixelWiseMultiplication::validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, float scale,
@@ -78,48 +45,16 @@ Status ClPixelWiseMultiplication::validate(const ITensorInfo *src1, const ITenso
     return kernels::ClPixelWiseMultiplicationKernel::validate(src1, src2, dst, scale, overflow_policy, rounding_policy, act_info);
 }
 
-void ClPixelWiseMultiplication::run(ITensorPack &tensors)
-{
-    if(_border_handler)
-    {
-        auto border_pack = select_border_input(tensors);
-        CLScheduler::get().enqueue_op(*_border_handler, border_pack);
-    }
-    ICLOperator::run(tensors);
-}
-
 void ClComplexPixelWiseMultiplication::configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, const ActivationLayerInfo &act_info)
 {
     auto k = std::make_unique<kernels::ClComplexPixelWiseMultiplicationKernel>();
     k->configure(compile_context, src1, src2, dst, act_info);
     _kernel = std::move(k);
-
-    if(dst->dimension(0) > 1)
-    {
-        ITensorInfo *broadcasted_info = (src1->dimension(0) == 1) ? src1 : src2;
-
-        if(broadcasted_info->dimension(0) == 1)
-        {
-            auto b = std::make_unique<CLFillBorderKernel>();
-            b->configure(compile_context, broadcasted_info, _kernel->border_size(), BorderMode::REPLICATE);
-            _border_handler = std::move(b);
-        }
-    }
 }
 
 Status ClComplexPixelWiseMultiplication::validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, const ActivationLayerInfo &act_info)
 {
     return kernels::ClComplexPixelWiseMultiplicationKernel::validate(src1, src2, dst, act_info);
 }
-
-void ClComplexPixelWiseMultiplication::run(ITensorPack &tensors)
-{
-    if(_border_handler)
-    {
-        auto border_pack = select_border_input(tensors);
-        CLScheduler::get().enqueue_op(*_border_handler, border_pack);
-    }
-    ICLOperator::run(tensors);
-}
 } // namespace opencl
 } // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h b/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h
index e9b3e4a5ef..e1598cb870 100644
--- a/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h
+++ b/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h
@@ -99,12 +99,6 @@ public:
      */
     static Status validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, float scale,
                            ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-
-    // Inherited methods overridden:
-    void run(ITensorPack &tensors) override;
-
-private:
-    std::unique_ptr<ICLKernel> _border_handler{ nullptr };
 };
 
 /** Basic function to run @ref opencl::ClComplexPixelWiseMultiplication. */
@@ -132,12 +126,6 @@ public:
      * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
      */
     static Status validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-
-    // Inherited methods overridden:
-    void run(ITensorPack &tensors) override;
-
-private:
-    std::unique_ptr<ICLKernel> _border_handler{ nullptr };
 };
 } // namespace opencl
 } // namespace arm_compute