Add multiple output support for dynamic fusion

* The dependency graph now can schedule any acyclic graph into
  a sequential list of operators. This is needed as the output
  operators now form branches in the graph.
* Fix the definition of input, output and intermediate tensors
  in GpuKernelComponentGroup to support non-linear but sequential
  list of operators.
* Add constraint on GpuOperatorGroup to enforce strictly linear
  fusion style, but allow output operator as the only form of
  branch.

Resolves: COMPMID-5771
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: I68de3a31a2456145081f0a397e4e61dd66327682
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8823
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

7 files changed, 8 insertions, 10 deletions

diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateActivation.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateActivation.cpp
index c3128ea552..8adf056912 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateActivation.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateActivation.cpp
@@ -125,7 +125,7 @@ TagLUT ClTemplateActivation::get_tag_lut(const GpuKernelVariableTable &vtable, c
     lut["src"] = vtable.get_variable(_src);
     lut["dst"] = vtable.get_variable(_dst);
 
-    const auto dst_argument = vtable.get_variable(comp_group.get_dst_tensors()[0]);
+    const auto dst_argument = vtable.get_variable(comp_group.get_any_dst_tensor());
     lut["arg_dst"]          = dst_argument.uniq_name;
 
     // Local build options
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateCast.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateCast.cpp
index 1ac49406a8..6ab3a68bb0 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateCast.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateCast.cpp
@@ -137,7 +137,7 @@ TagLUT ClTemplateCast::get_tag_lut(const GpuKernelVariableTable &vtable, const C
     lut["src"] = vtable.get_variable(_src);
     lut["dst"] = vtable.get_variable(_dst);
 
-    const auto dst_argument = vtable.get_variable(comp_group.get_dst_tensors()[0]);
+    const auto dst_argument = vtable.get_variable(comp_group.get_any_dst_tensor());
     lut["arg_dst"]          = dst_argument.uniq_name;
 
     // Local build options
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDepthwiseConv2d.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDepthwiseConv2d.cpp
index 389bd5c65f..6fa77aafe3 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDepthwiseConv2d.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDepthwiseConv2d.cpp
@@ -251,7 +251,7 @@ TagLUT ClTemplateDepthwiseConv2d::get_tag_lut(const GpuKernelVariableTable &vtab
     }
     lut["dst"] = vtable.get_variable(_dst);
 
-    const auto dst_argument = vtable.get_variable(comp_group.get_dst_tensors()[0]);
+    const auto dst_argument = vtable.get_variable(comp_group.get_any_dst_tensor());
     lut["arg_dst"]          = dst_argument.uniq_name;
 
     // Local build options
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDirectConv2d.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDirectConv2d.cpp
index aa324ffb54..221addb7b5 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDirectConv2d.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDirectConv2d.cpp
@@ -268,7 +268,7 @@ TagLUT ClTemplateDirectConv2d::get_tag_lut(const GpuKernelVariableTable &vtable,
     }
     lut["dst"] = vtable.get_variable(_dst);
 
-    const auto dst_argument = vtable.get_variable(comp_group.get_dst_tensors()[0]);
+    const auto dst_argument = vtable.get_variable(comp_group.get_any_dst_tensor());
     lut["arg_dst"]          = dst_argument.uniq_name;
 
     // Local build options
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp
index 6c1e0fb1de..39cec6e31c 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp
@@ -194,7 +194,7 @@ TagLUT ClTemplateElementwiseBinary::get_tag_lut(const GpuKernelVariableTable &vt
         lut["lhs"] = vtable.get_variable(_lhs);
         lut["rhs"] = vtable.get_variable(_rhs);
         lut["dst"] = vtable.get_variable(_dst);
-        lut["out"] = vtable.get_variable(comp_group.get_dst_tensors().front());
+        lut["out"] = vtable.get_variable(comp_group.get_any_dst_tensor());
     }
     else
     {
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateStore.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateStore.cpp
index e4b662b3a8..ef4f2f22a1 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateStore.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateStore.cpp
@@ -84,9 +84,9 @@ TagLUT ClTemplateStore::get_tag_lut(const GpuKernelVariableTable &vtable, const
     // Local build options
     lut["meta_kernel_id"]  = id();
     lut["DST_TENSOR_TYPE"] = "BUFFER";
-    const auto dst_info    = comp_group.get_dst_tensors()[0];
-    lut["DST_DATA_TYPE"]   = dst_info->data_type();
+    lut["DST_DATA_TYPE"]   = _dst->data_type();
 
+    ARM_COMPUTE_UNUSED(comp_group);
     return lut;
 }
 
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateWriter.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateWriter.cpp
index 0afd0e7581..eed481f109 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateWriter.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateWriter.cpp
@@ -203,9 +203,7 @@ std::string ClTemplateWriter::write_code()
 }
 std::string ClTemplateWriter::write_global_section() const
 {
-    const auto dst_tensors = _components.get_dst_tensors();
-    ARM_COMPUTE_ERROR_ON_MSG(dst_tensors.size() != 1, "Only one destination tensor per kernel is allowed");
-    const auto dst_info   = dst_tensors[0];
+    const auto dst_info   = _components.get_any_dst_tensor();
     const auto dst_w      = dst_info->dimension(0);
     const auto tile_w     = std::max(1, get_window().x().step());
     const auto tile_h     = std::max(1, get_window().y().step());