Update CpuGemmConv2d and CpuFlatten to use CpuReshape operator
- Following CpuReshapeKernel Optimizations, update the CpuGemmConv2D and CpuFlatten
to use CpuReshape operator instead of CpuReshapeKernel
- Minor changes to comment in NEReorgLayerKernel.h
Resolves COMPMID-6504
Signed-off-by: Anitha Raj <anitha.raj@arm.com>
Change-Id: Ib6ee1fdc313d91249f9fe41c81e73324031c1ff4
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10186
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/NEON/kernels/NEReorgLayerKernel.h b/src/core/NEON/kernels/NEReorgLayerKernel.h
index 38a7d9f..6e67eb3 100644
--- a/src/core/NEON/kernels/NEReorgLayerKernel.h
+++ b/src/core/NEON/kernels/NEReorgLayerKernel.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018-2021 Arm Limited.
+ * Copyright (c) 2018-2021, 2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -21,8 +21,8 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#ifndef ARM_COMPUTE_NEREORGLAYERKERNEL_H
-#define ARM_COMPUTE_NEREORGLAYERKERNEL_H
+#ifndef ACL_SRC_CORE_NEON_KERNELS_NEREORGLAYERKERNEL_H
+#define ACL_SRC_CORE_NEON_KERNELS_NEREORGLAYERKERNEL_H
#include "src/core/NEON/INEKernel.h"
@@ -60,7 +60,7 @@
*/
void configure(const ITensor *input, ITensor *output, int32_t stride);
- /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuReshapeKernel
+ /** Static function to check if given info will lead to a valid configuration
*
* @param[in] input Source tensor info. Data type supported: All
* @param[in] output Destination tensor info. Data type supported: Same as @p input
@@ -80,4 +80,4 @@
int32_t _stride;
};
} // namespace arm_compute
-#endif /*ARM_COMPUTE_NEREORGLAYERKERNEL_H */
+#endif // ACL_SRC_CORE_NEON_KERNELS_NEREORGLAYERKERNEL_H
diff --git a/src/cpu/operators/CpuFlatten.cpp b/src/cpu/operators/CpuFlatten.cpp
index f6ae139..7bab9e4 100644
--- a/src/cpu/operators/CpuFlatten.cpp
+++ b/src/cpu/operators/CpuFlatten.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2021 Arm Limited.
+ * Copyright (c) 2021, 2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -23,7 +23,7 @@
*/
#include "src/cpu/operators/CpuFlatten.h"
-#include "src/cpu/kernels/CpuReshapeKernel.h"
+#include "src/cpu/operators/CpuReshape.h"
#include "src/common/utils/Log.h"
@@ -31,17 +31,28 @@
{
namespace cpu
{
+CpuFlatten::CpuFlatten()
+ : _reshape(nullptr)
+{
+}
+
+CpuFlatten::~CpuFlatten() = default;
+
void CpuFlatten::configure(const ITensorInfo *src, ITensorInfo *dst)
{
ARM_COMPUTE_LOG_PARAMS(src, dst);
- auto k = std::make_unique<kernels::CpuReshapeKernel>();
- k->configure(src, dst);
- _kernel = std::move(k);
+ _reshape = std::make_unique<CpuReshape>();
+ _reshape->configure(src, dst);
}
Status CpuFlatten::validate(const ITensorInfo *src, const ITensorInfo *dst)
{
- return kernels::CpuReshapeKernel::validate(src, dst);
+ return CpuReshape::validate(src, dst);
+}
+
+void CpuFlatten::run(ITensorPack &tensors)
+{
+ _reshape->run(tensors);
}
} // namespace cpu
} // namespace arm_compute
diff --git a/src/cpu/operators/CpuFlatten.h b/src/cpu/operators/CpuFlatten.h
index 0e9fcbd..911760d 100644
--- a/src/cpu/operators/CpuFlatten.h
+++ b/src/cpu/operators/CpuFlatten.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2021 Arm Limited.
+ * Copyright (c) 2021, 2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -21,8 +21,8 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#ifndef ARM_COMPUTE_CPU_FLATTEN_H
-#define ARM_COMPUTE_CPU_FLATTEN_H
+#ifndef ACL_SRC_CPU_OPERATORS_CPUFLATTEN_H
+#define ACL_SRC_CPU_OPERATORS_CPUFLATTEN_H
#include "src/cpu/ICpuOperator.h"
@@ -30,10 +30,15 @@
{
namespace cpu
{
+class CpuReshape;
/** Basic function to flatten a given input */
class CpuFlatten : public ICpuOperator
{
public:
+ /** Constructor */
+ CpuFlatten();
+ /** Destructor */
+ ~CpuFlatten();
/** Configure operator for a given list of arguments
*
* Valid data layouts:
@@ -58,7 +63,13 @@
* @return a status
*/
static Status validate(const ITensorInfo *src, const ITensorInfo *dst);
+
+ // Inherited methods overridden:
+ void run(ITensorPack &tensors) override;
+
+private:
+ std::unique_ptr<CpuReshape> _reshape;
};
} // namespace cpu
} // namespace arm_compute
-#endif /* ARM_COMPUTE_CPU_FLATTEN_H */
+#endif // ACL_SRC_CPU_OPERATORS_CPUFLATTEN_H
diff --git a/src/cpu/operators/CpuGemmConv2d.cpp b/src/cpu/operators/CpuGemmConv2d.cpp
index 7c0e58b..d11e4f0 100644
--- a/src/cpu/operators/CpuGemmConv2d.cpp
+++ b/src/cpu/operators/CpuGemmConv2d.cpp
@@ -35,11 +35,11 @@
#include "src/core/helpers/MemoryHelpers.h"
#include "src/cpu/kernels/CpuCol2ImKernel.h"
#include "src/cpu/kernels/CpuIm2ColKernel.h"
-#include "src/cpu/kernels/CpuReshapeKernel.h"
#include "src/cpu/kernels/CpuWeightsReshapeKernel.h"
#include "src/cpu/operators/CpuGemm.h"
#include "src/cpu/operators/CpuGemmLowpMatrixMultiplyCore.h"
#include "src/cpu/operators/CpuGemmLowpOutputStage.h"
+#include "src/cpu/operators/CpuReshape.h"
#include "src/cpu/utils/CpuAuxTensorHandler.h"
#include <set>
@@ -92,7 +92,7 @@
}
CpuGemmConv2d::CpuGemmConv2d()
- : _weights_reshape_kernel(nullptr), _im2col_kernel(), _mm_gemm(), _mm_gemmlowp(), _col2im_kernel(), _reshape_kernel(), _im2col_output(), _weights_reshaped(), _gemm_output(), _gemm_output_3d(),
+ : _weights_reshape_kernel(nullptr), _im2col_kernel(), _mm_gemm(), _mm_gemmlowp(), _col2im_kernel(), _reshape(), _im2col_output(), _weights_reshaped(), _gemm_output(), _gemm_output_3d(),
_data_layout(DataLayout::NCHW), _skip_im2col(false), _skip_col2im(false), _is_quantized(false), _is_prepared(false), _aux_mem(AuxTensorIdx::Count)
{
}
@@ -379,8 +379,8 @@
else
{
// Configure reshape layer
- _reshape_kernel = std::make_unique<kernels::CpuReshapeKernel>();
- _reshape_kernel->configure(gemm_output_to_use, dst);
+ _reshape = std::make_unique<CpuReshape>();
+ _reshape->configure(gemm_output_to_use, dst);
}
// Check if GEMM transforms weights
@@ -642,7 +642,7 @@
{ TensorType::ACL_SRC, gemm_output_to_use },
{ TensorType::ACL_DST, dst }
};
- NEScheduler::get().schedule_op(_reshape_kernel.get(), Window::DimY, _reshape_kernel->window(), pack);
+ _reshape->run(pack);
}
}
else if(out_has_padding)
@@ -652,7 +652,7 @@
{ TensorType::ACL_SRC, gemm_output_to_use },
{ TensorType::ACL_DST, dst }
};
- NEScheduler::get().schedule_op(_reshape_kernel.get(), Window::DimY, _reshape_kernel->window(), pack);
+ _reshape->run(pack);
}
}
diff --git a/src/cpu/operators/CpuGemmConv2d.h b/src/cpu/operators/CpuGemmConv2d.h
index 81d34ae..61fe63a 100644
--- a/src/cpu/operators/CpuGemmConv2d.h
+++ b/src/cpu/operators/CpuGemmConv2d.h
@@ -21,8 +21,8 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#ifndef ARM_COMPUTE_CPU_GEMM_CONV2D_H
-#define ARM_COMPUTE_CPU_GEMM_CONV2D_H
+#ifndef ACL_SRC_CPU_OPERATORS_CPUGEMMCONV2D_H
+#define ACL_SRC_CPU_OPERATORS_CPUGEMMCONV2D_H
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Types.h"
@@ -38,12 +38,12 @@
class CpuGemm;
class CpuGemmLowpMatrixMultiplyCore;
class CpuGemmLowpOutputStage;
+class CpuReshape;
namespace kernels
{
class CpuWeightsReshapeKernel;
class CpuIm2ColKernel;
class CpuCol2ImKernel;
-class CpuReshapeKernel;
} // namespace kernels
/** Basic function to compute the convolution layer. This function calls the following kernels/functions:
@@ -130,8 +130,8 @@
const bool enable_fast_math = false);
// Inherited methods overridden:
- void run(ITensorPack &tensors) override;
- void prepare(ITensorPack &tensors) override;
+ void run(ITensorPack &tensors) override;
+ void prepare(ITensorPack &tensors) override;
experimental::MemoryRequirements workspace() const override;
private:
@@ -222,7 +222,7 @@
std::unique_ptr<CpuGemm> _mm_gemm;
std::unique_ptr<CpuGemmLowpMatrixMultiplyCore> _mm_gemmlowp;
std::unique_ptr<kernels::CpuCol2ImKernel> _col2im_kernel;
- std::unique_ptr<kernels::CpuReshapeKernel> _reshape_kernel;
+ std::unique_ptr<CpuReshape> _reshape;
TensorInfo _im2col_output;
TensorInfo _weights_reshaped;
@@ -240,4 +240,4 @@
};
} // namespace cpu
} // namespace arm_compute
-#endif /* ARM_COMPUTE_CPU_GEMM_CONV2D_H */
+#endif // ACL_SRC_CPU_OPERATORS_CPUGEMMCONV2D_H