Rewrite dynamic fusion
The new version introduces the following major changes:
* Change public interface to simplify and standardize the user experience
- Use the term "Workload" uniformly
- Simplify operator interface to be a set of static methods:
validate_op(), create_op()
* Separate the kernel writing into its own component (template_writer).
This is to allow the co-development of GpuKernelWriter, and to allow
easy replacement once GpuKernelWriter is mature.
* Optimize the core fusion algorithm used by the component graph. The
details can be found in GpuKernelComponentGraph::fuse()
* Use Gpu instead of Cl prefixes for most of the Workload interfaces
(except for runtime and kernel components, which have to be language specific)
This allows the potential extension to other Gpu langauges in the
future.
* Refactor runtime memory interface so that auxiliary tensor handling
is separate from the user tensor passing. This is because the former
is less stable and may require extension in the future.
* Hide source code object from the user as it is not required at the
moment
* Deprecate the old prototype entirely by disabling it in SCons build
Resolves COMPMID-5510, COMPMID-5512, COMPMID-5513
Change-Id: If69d2362856f2de4503546b7b6cf48a525cf3079
Signed-off-by: SiCong Li <sicong.li@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8406
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/GpuKernelVariableTable.h b/src/dynamic_fusion/sketch/gpu/template_writer/GpuKernelVariableTable.h
new file mode 100644
index 0000000..4eee396
--- /dev/null
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/GpuKernelVariableTable.h
@@ -0,0 +1,135 @@
+/*
+ * 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 SRC_DYNAMIC_FUSION_SKETCH_GPU_TEMPLATE_WRITER_GPUKERNELVARIABLETABLE
+#define SRC_DYNAMIC_FUSION_SKETCH_GPU_TEMPLATE_WRITER_GPUKERNELVARIABLETABLE
+
+#include "arm_compute/core/ITensorInfo.h"
+#include "src/dynamic_fusion/sketch/gpu/GpuKernelArgument.h"
+#include "support/Requires.h"
+#include "support/StringSupport.h"
+
+#include <set>
+#include <string>
+#include <type_traits>
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** A table of all the variables used in the kernel
+ * Since fusion is restricted to a linear sequence of components in a kernel, only a single "intermediate variable" (the accumulator) is allowed.
+ * Each kernel has exactly one variable table
+ */
+class GpuKernelVariableTable
+{
+public:
+ /** A tensor variable whose main purposes are:
+ * - Hold the newly assigned @ref GpuKernelArgumentInfo for the associated tensor info
+ * - Hold the generated variable name for the associated tensor info
+ */
+ struct TensorVariable
+ {
+ public:
+ TensorVariable() = default;
+ TensorVariable(const TensorVariable &) = default;
+ TensorVariable &operator=(const TensorVariable &) = default;
+ ITensorInfo::Id id{ ITensorInfo::invalid_tensor_id };
+ std::string uniq_name{ "empty" }; // Unique name, also the final variable name used in the built code
+ GpuKernelArgumentInfo kernel_argument_info{};
+ bool has_valid_id() const
+ {
+ return id != ITensorInfo::invalid_tensor_id;
+ }
+ };
+ using VariableList = std::vector<TensorVariable>;
+
+public:
+ /** Declare a @ref TensorVariable for a corresponding tensor info.
+ *
+ * @note: Later re-declaration of the intermediate variable will overwrite the previous association to the @ref ITensorInfo
+ * Therefore, the order of declaration is important. It's assumed that the components declaring the variable is already in correct order
+ *
+ * @param[in] tensor Tensor info with which the new variable is associated
+ * @param[in] argument_info Kernel argument information
+ * @param[in] is_interm If the new variable is an intermediate variable
+ * @param[in] alias Alias for the variable. Will be used as part of the variable name
+ */
+ void declare_variable(const ITensorInfo *tensor, GpuKernelArgumentInfo argument_info, bool is_interm = false, const std::string &alias = "unnamed");
+ /** Get the @ref TensorVariable associated with @p tensor
+ *
+ * @param[in] tensor Tensor info to be queried
+ *
+ * @return TensorVariable
+ */
+ TensorVariable get_variable(const ITensorInfo *tensor) const;
+ /** Get the @ref TensorVariable list associated with @p tensors
+ * @note Empty tensors are skipped
+ *
+ * @param[in] tensors List of tensor infos to be queried
+ *
+ * @return VariableList
+ */
+ VariableList get_variable_list(const std::vector<const ITensorInfo *> &tensors) const;
+
+private:
+ std::map<ITensorInfo::Id, TensorVariable> _vars{}; /**< Non-intermediate (function parameter) variables*/
+ TensorVariable _interm_var{}; /**< Intermediate variable */
+ std::set<ITensorInfo::Id> _interm_tensors{}; /**< Tensors associated with the single intermediate variable */
+};
+
+/** A tag value will substitute a tag in a string template during its instantiation */
+struct TagVal
+{
+ /** Default constructor */
+ TagVal() = default;
+ /** Construct a @ref TagVal from a @ref GpuKernelVariableTable::TensorVariable */
+ TagVal(const GpuKernelVariableTable::TensorVariable &var);
+ /** Construct a @ref TagVal from an integral type */
+ template <typename T, ARM_COMPUTE_REQUIRES_TA(std::is_integral<T>::value)>
+ TagVal(T val)
+ : value{ support::cpp11::to_string(val) }
+ {
+ }
+ /** Construct a @ref TagVal from a string */
+ TagVal(const std::string &val);
+ /** Construct a @ref TagVal from a c-style string */
+ TagVal(const char *val);
+ /** Construct a @ref TagVal from a @ref DataType */
+ TagVal(const DataType &data_type);
+ /** Get the value of the TagVal as a converted string */
+ std::string value{};
+};
+
+/** A tag used in a string template is a placeholder string to be substituted by real values during template instantiation */
+using Tag = std::string;
+
+/** Tag lookup table. It is used to instantiate a string template */
+using TagLUT = std::unordered_map<Tag, TagVal>;
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif /* SRC_DYNAMIC_FUSION_SKETCH_GPU_TEMPLATE_WRITER_GPUKERNELVARIABLETABLE */