Add output operator for dynamic fusion

* The output of the fused operator must be explicitly specified
  using GpuOutput operator.
* Any temporary tensors used to connect the output of an operator
  to the input of another operator will be marked as no-alloc
  and won't be allocated as a tensor in the memory.

Resolves: COMPMID-5771
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: I5ae8e800f8f737db23a055a92b01c4f1d78c3bb8
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8794
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 88 insertions, 0 deletions

diff --git a/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h b/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
index deedf62262..25023ff0a1 100644
--- a/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
+++ b/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
@@ -37,6 +37,7 @@ enum class MemoryType
 {
     User      = 0, /**< Memory coming directly from users, e.g. for argument tensors */
     Auxiliary = 1, /**< Additional memory required by the workload tensor, e.g. for temporary tensors */
+    NoAlloc   = 2, /**< Temporary tile which is not allocated as a whole tensor in the memory */
 };
 
 /** Memory information for tensors with @ref MemoryType::Auxiliary.
diff --git a/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuOutput.h b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuOutput.h
new file mode 100644
index 0000000000..2511b0efd5
--- /dev/null
+++ b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuOutput.h
@@ -0,0 +1,87 @@
+/*
+ * Copyright (c) 2022 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_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUOUTPUT
+#define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUOUTPUT
+
+#include "arm_compute/core/ITensorInfo.h"
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+
+/** Forward declaration */
+class GpuWorkloadContext;
+class GpuWorkloadSketch;
+
+/** Operator interface. */
+class GpuOutput final
+{
+public:
+    /** Create an operator and fuse it into the workload sketch.
+     *    @note If @ref validate_op() fails, the creation also fails and may throw an error.
+     *    @note If @ref validate_op() fails, @p sketch remains unchanged and valid.
+     *
+     * Valid data type configurations:
+     *   - Any
+     *
+     * Valid data layouts:
+     *   - Any
+     *
+     * @param[in, out] sketch Workload sketch into which the operator will be fused.
+     * @param[in, out] src    Source tensor info.
+     * @param[in, out] dst    Destination tensor info.
+     *                        If an uninitialized ITensorInfo is passed in, it will be auto-initialized.
+     */
+    static void create_op(GpuWorkloadSketch &sketch,
+                          ITensorInfo       *src,
+                          ITensorInfo       *dst);
+
+    /** Check if the operator configuration is supported, irrespective of fusion.
+     *
+     * @param[in] context Workload context within which the operator is running.
+     * @param[in] src     Source tensor info.
+     * @param[in] dst     Destination tensor info.
+     */
+    static Status is_supported_op(const GpuWorkloadContext &context,
+                                  const ITensorInfo        *src,
+                                  const ITensorInfo        *dst);
+
+    /** Validate the operator and check if the its configuration is supported and if it can be fused into the workload sketch.
+     *
+     *  Similar to @ref GpuOutput::create_op().
+     */
+    static Status validate_op(const GpuWorkloadSketch &sketch,
+                              const ITensorInfo       *src,
+                              const ITensorInfo       *dst);
+};
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUOUTPUT */