commit 002e6530f6218b00a28aef9be8b21efb08cf3602
author Ramy Elgammal <ramelg01@e129512.arm.com> Wed Jan 11 18:48:04 2023 +0000
committer Ramy Elgammal <ramy.elgammal@arm.com> Wed Jan 25 10:24:23 2023 +0000
tree f3e2f9d064b985ffe283512825b34cdd59f29f50
parent cc2877368d5e15d9ea89d31c84ec651fc0fffd13

Implement dynamic fusion softmax operator

 - Return aux tensorInfo by get_aux_tensors() at runtime to init the aux
   tensor with the right size.
 - Keep softmax unfusable for this commit
 - Hence, added Tensor3D to template writer arguments declaration, for sake of
   keeping dynamic fusion softmax componenets' kernels matching their cl
   counterparts.

Resolves: COMPMID-5523
Change-Id: I667f39545db925f667036ef448302c79a0330373
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/483924
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: bsgcomp <bsgcomp@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8986
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

src/dynamic_fusion/runtime/gpu/cl/ClKernelRuntime.cpp

diff --git a/src/dynamic_fusion/runtime/gpu/cl/ClKernelRuntime.cpp b/src/dynamic_fusion/runtime/gpu/cl/ClKernelRuntime.cpp
index 022d468..b3ec393 100644
--- a/src/dynamic_fusion/runtime/gpu/cl/ClKernelRuntime.cpp
+++ b/src/dynamic_fusion/runtime/gpu/cl/ClKernelRuntime.cpp
@@ -41,7 +41,7 @@
     // Create kernel from kernel source string
     opencl::ClKernelLibrary &klib = opencl::ClKernelLibrary::get();
     _kernel                       = static_cast<cl::Kernel>(compile_ctx.create_kernel(code.name(),
-                                                                                      "" /* Program name: Used to as part of a unique string for built kernel cache. Not needed */,
+                                                                                      code.name(), // program name has to be provided to differentiate between different unfusable components' kernels.
                                                                                       code.code(),
                                                                                       klib.kernel_path() /* Kernel path: Used in cases of embedded kernels */,
                                                                                       code.build_options().options(),

src/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.cpp

diff --git a/src/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.cpp b/src/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.cpp
index 7e427fe..cd21b10 100644
--- a/src/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.cpp
+++ b/src/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -355,12 +355,12 @@
     return Status{};
 }
 
-std::vector<std::pair<CLTensor *, AuxMemoryInfo>> ClWorkloadRuntime::get_auxiliary_tensors()
+std::vector<std::tuple<CLTensor *, TensorInfo, AuxMemoryInfo>> ClWorkloadRuntime::get_auxiliary_tensors()
 {
-    std::vector<std::pair<CLTensor *, AuxMemoryInfo>> aux_tensors;
+    std::vector<std::tuple<CLTensor *, TensorInfo, AuxMemoryInfo>> aux_tensors;
     for(const auto &data : _impl->_aux_tensors.get_tensors())
     {
-        aux_tensors.emplace_back(data.tensor, data.memory_info);
+        aux_tensors.emplace_back(data.tensor, data.tensor_info, data.memory_info);
     }
     return aux_tensors;
 }