Make ClPixelWiseMultiplicationKernel stateless

Partially resolves: COMPMID-4183

Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: Ibc08d2d84d023ef8b23ed44d534aa1ca24515e4d
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5274
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

3 files changed, 283 insertions, 107 deletions

diff --git a/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp b/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
index a56018b397..5ebaf5d122 100644
--- a/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
+++ b/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2020 Arm Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,111 +26,18 @@
 #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/kernels/CLPixelWiseMultiplicationKernel.h"
+#include "src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h"
 
 #include <utility>
 
 namespace arm_compute
 {
-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
-
-namespace experimental
-{
-CLPixelWiseMultiplication::CLPixelWiseMultiplication()
-    : _border_handler(std::make_unique<CLFillBorderKernel>())
-{
-}
-
-void CLPixelWiseMultiplication::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, float scale,
-                                          ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
-{
-    auto k = std::make_unique<CLPixelWiseMultiplicationKernel>();
-    k->configure(compile_context, input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
-    _kernel = std::move(k);
-
-    if(output->dimension(0) > 1)
-    {
-        ITensorInfo *broadcasted_info = (input1->dimension(0) == 1) ? input1 : input2;
-
-        if(broadcasted_info->dimension(0) == 1)
-        {
-            _border_handler->configure(compile_context, broadcasted_info, _kernel->border_size(), BorderMode::REPLICATE);
-        }
-    }
-}
-
-Status CLPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale,
-                                           ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
-{
-    return CLPixelWiseMultiplicationKernel::validate(input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
-}
-
-void CLPixelWiseMultiplication::run(ITensorPack &tensors)
-{
-    auto border_pack = select_border_input(tensors);
-    CLScheduler::get().enqueue_op(*_border_handler, border_pack);
-    ICLOperator::run(tensors);
-}
-
-CLComplexPixelWiseMultiplication::CLComplexPixelWiseMultiplication()
-    : _border_handler(std::make_unique<CLFillBorderKernel>())
-{
-}
-
-void CLComplexPixelWiseMultiplication::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info)
-{
-    auto k = std::make_unique<CLComplexPixelWiseMultiplicationKernel>();
-    k->configure(compile_context, input1, input2, output, act_info);
-    _kernel = std::move(k);
-
-    if(output->dimension(0) > 1)
-    {
-        ITensorInfo *broadcasted_info = (input1->dimension(0) == 1) ? input1 : input2;
-
-        if(broadcasted_info->dimension(0) == 1)
-        {
-            _border_handler->configure(compile_context, broadcasted_info, _kernel->border_size(), BorderMode::REPLICATE);
-        }
-    }
-}
-
-Status CLComplexPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
-{
-    return CLComplexPixelWiseMultiplicationKernel::validate(input1, input2, output, act_info);
-}
-
-void CLComplexPixelWiseMultiplication::run(ITensorPack &tensors)
-{
-    auto border_pack = select_border_input(tensors);
-    CLScheduler::get().enqueue_op(*_border_handler, border_pack);
-    ICLOperator::run(tensors);
-}
-} // namespace experimental
-
 struct CLPixelWiseMultiplication::Impl
 {
-    const ICLTensor                                         *src_0{ nullptr };
-    const ICLTensor                                         *src_1{ nullptr };
-    ICLTensor                                               *dst{ nullptr };
-    std::unique_ptr<experimental::CLPixelWiseMultiplication> op{ nullptr };
+    const ICLTensor                                   *src_0{ nullptr };
+    const ICLTensor                                   *src_1{ nullptr };
+    ICLTensor                                         *dst{ nullptr };
+    std::unique_ptr<opencl::ClPixelWiseMultiplication> op{ nullptr };
 };
 
 CLPixelWiseMultiplication::CLPixelWiseMultiplication()
@@ -153,14 +60,14 @@ void CLPixelWiseMultiplication::configure(const CLCompileContext &compile_contex
     _impl->src_0 = input1;
     _impl->src_1 = input2;
     _impl->dst   = output;
-    _impl->op    = std::make_unique<experimental::CLPixelWiseMultiplication>();
+    _impl->op    = std::make_unique<opencl::ClPixelWiseMultiplication>();
     _impl->op->configure(compile_context, input1->info(), input2->info(), output->info(), scale, overflow_policy, rounding_policy, act_info);
 }
 
 Status CLPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale,
                                            ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
 {
-    return experimental::CLPixelWiseMultiplication::validate(input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
+    return opencl::ClPixelWiseMultiplication::validate(input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
 }
 
 void CLPixelWiseMultiplication::run()
@@ -175,10 +82,10 @@ void CLPixelWiseMultiplication::run()
 
 struct CLComplexPixelWiseMultiplication::Impl
 {
-    const ICLTensor                                                *src_0{ nullptr };
-    const ICLTensor                                                *src_1{ nullptr };
-    ICLTensor                                                      *dst{ nullptr };
-    std::unique_ptr<experimental::CLComplexPixelWiseMultiplication> op{ nullptr };
+    const ICLTensor                                          *src_0{ nullptr };
+    const ICLTensor                                          *src_1{ nullptr };
+    ICLTensor                                                *dst{ nullptr };
+    std::unique_ptr<opencl::ClComplexPixelWiseMultiplication> op{ nullptr };
 };
 
 CLComplexPixelWiseMultiplication::CLComplexPixelWiseMultiplication()
@@ -199,13 +106,13 @@ void CLComplexPixelWiseMultiplication::configure(const CLCompileContext &compile
     _impl->src_0 = input1;
     _impl->src_1 = input2;
     _impl->dst   = output;
-    _impl->op    = std::make_unique<experimental::CLComplexPixelWiseMultiplication>();
+    _impl->op    = std::make_unique<opencl::ClComplexPixelWiseMultiplication>();
     _impl->op->configure(compile_context, input1->info(), input2->info(), output->info(), act_info);
 }
 
 Status CLComplexPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    return experimental::CLComplexPixelWiseMultiplication::validate(input1, input2, output, act_info);
+    return opencl::ClComplexPixelWiseMultiplication::validate(input1, input2, output, act_info);
 }
 
 void CLComplexPixelWiseMultiplication::run()
diff --git a/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.cpp b/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.cpp
new file mode 100644
index 0000000000..c4f11a4e29
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.cpp
@@ -0,0 +1,125 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#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"
+
+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,
+                                           ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
+{
+    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
new file mode 100644
index 0000000000..e9b3e4a5ef
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClPixelWiseMultiplication.h
@@ -0,0 +1,144 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_PIXELWISE_MULTIPLICATION_H
+#define ARM_COMPUTE_CL_PIXELWISE_MULTIPLICATION_H
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to run @ref opencl::ClPixelWiseMultiplication. */
+class ClPixelWiseMultiplication : public IClOperator
+{
+public:
+    /** Default Constructor */
+    ClPixelWiseMultiplication() = default;
+    /** Initialise the kernel's sources, dst and convertion policy.
+     *
+     * Valid configurations (src1,src2) -> Output :
+     *
+     *   - (U8,U8)                         -> U8
+     *   - (U8,U8)                         -> S16
+     *   - (U8,S16)                        -> S16
+     *   - (S16,U8)                        -> S16
+     *   - (S16,S16)                       -> S16
+     *   - (F16,F16)                       -> F16
+     *   - (F32,F32)                       -> F32
+     *   - (QASYMM8,QASYMM8)               -> QASYMM8
+     *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED
+     *   - (QSYMM16,QSYMM16)               -> QSYMM16
+     *   - (QSYMM16,QSYMM16)               -> S32
+     *
+     * @param[in]      compile_context The compile context to be used.
+     * @param[in, out] src1            An src tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
+     *                                 The src tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
+     * @param[in, out] src2            An src tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
+     *                                 The src tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
+     * @param[out]     dst             The dst tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
+     * @param[in]      scale           Scale to apply after multiplication.
+     *                                 Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15.
+     * @param[in]      overflow_policy Overflow policy. Supported overflow policies: Wrap, Saturate
+     * @param[in]      rounding_policy Rounding policy. Supported rounding modes: to zero, to nearest even.
+     * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
+     */
+    void configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, float scale,
+                   ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+    /** Static function to check if given info will lead to a valid configuration of @ref ClPixelWiseMultiplication
+     *
+     * Valid configurations (src1,src2) -> Output :
+     *
+     *   - (U8,U8)                         -> U8
+     *   - (U8,U8)                         -> S16
+     *   - (U8,S16)                        -> S16
+     *   - (S16,U8)                        -> S16
+     *   - (S16,S16)                       -> S16
+     *   - (F16,F16)                       -> F16
+     *   - (F32,F32)                       -> F32
+     *   - (QASYMM8,QASYMM8)               -> QASYMM8
+     *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED
+     *   - (QSYMM16,QSYMM16)               -> QSYMM16
+     *   - (QSYMM16,QSYMM16)               -> S32
+     *
+     *
+     * @param[in] src1            An src tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
+     * @param[in] src2            An src tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
+     * @param[in] dst             The dst tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
+     * @param[in] scale           Scale to apply after multiplication.
+     *                            Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15.
+     * @param[in] overflow_policy Overflow policy. Supported overflow policies: Wrap, Saturate
+     * @param[in] rounding_policy Rounding policy. Supported rounding modes: to zero, to nearest even.
+     * @param[in] act_info        (Optional) Activation layer information in case of a fused activation.
+     *
+     * @return a status
+     */
+    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. */
+class ClComplexPixelWiseMultiplication : public IClOperator
+{
+public:
+    /** Default Constructor */
+    ClComplexPixelWiseMultiplication() = default;
+    /** Initialise the kernel's sources, dst.
+     *
+     * @param[in]      compile_context The compile context to be used.
+     * @param[in, out] src1            An src tensor info. Data types supported: F16/F32. Number of channels supported: 2.
+     *                                 The src tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
+     * @param[in, out] src2            An src tensor info. Data types supported: same as @p src1. Number of channels supported: same as @p src1.
+     *                                 The src tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
+     * @param[out]     dst             The dst tensor info, Data types supported: same as @p src1. Number of channels supported: same as @p src1.
+     * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
+     */
+    void configure(const CLCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+    /** Static function to check if given info will lead to a valid configuration of @ref ClComplexPixelWiseMultiplication
+     *
+     * @param[in] src1     An src tensor info. Data types supported: F16/F32. Number of channels supported: 2.
+     * @param[in] src2     An src tensor info. Data types supported: same as @p src1. Number of channels supported: same as @p src1.
+     * @param[in] dst      The dst tensor info, Data types supported: same as @p src1. Number of channels supported: same as @p src1.
+     * @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
+#endif /* ARM_COMPUTE_CL_PIXELWISE_MULTIPLICATION_H */
\ No newline at end of file