COMPMID-2443: OpenCL kernels cache.

Added an example showing how to load prebuilt opencl kernels from
a binary cache file.

Change-Id: I3292b4762270c1c2a6370d13dffc277e2ccc26ea
Signed-off-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1488
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 224 insertions, 0 deletions

diff --git a/examples/cl_cache.cpp b/examples/cl_cache.cpp
new file mode 100644
index 0000000000..87a3058956
--- /dev/null
+++ b/examples/cl_cache.cpp
@@ -0,0 +1,224 @@
+/*
+ * Copyright (c) 2019 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 "arm_compute/runtime/CL/CLFunctions.h"
+
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLHelpers.h"
+#include "arm_compute/runtime/CL/CLScheduler.h"
+#include "utils/Utils.h"
+
+#include <chrono>
+
+using namespace arm_compute;
+using namespace utils;
+
+namespace
+{
+/** This function loads prebuilt opencl kernels from a file
+ *
+ * @param[in] filename Name of the file to be used to load the kernels
+ */
+void restore_program_cache_from_file(const std::string &filename = "cache.bin")
+{
+    std::cout << "Loading kernels from file " << filename << std::endl;
+    std::ifstream cache_file(filename, std::ios::binary);
+    if(cache_file.is_open())
+    {
+        while(!cache_file.eof())
+        {
+            size_t name_len   = 0;
+            size_t binary_len = 0;
+            cache_file.read(reinterpret_cast<char *>(&name_len), sizeof(size_t));
+            cache_file.read(reinterpret_cast<char *>(&binary_len), sizeof(size_t));
+            if(name_len == 0 || binary_len == 0)
+            {
+                break;
+            }
+            std::vector<char>          tmp(name_len);
+            std::vector<unsigned char> binary(binary_len);
+            std::string                name;
+            cache_file.read(tmp.data(), name_len);
+            name.assign(tmp.data(), name_len);
+            tmp.resize(binary_len);
+            cache_file.read(reinterpret_cast<char *>(binary.data()), binary_len);
+            cl::Context             context = arm_compute::CLScheduler::get().context();
+            cl::Program::Binaries   binaries{ binary };
+            std::vector<cl::Device> devices = context.getInfo<CL_CONTEXT_DEVICES>();
+            cl::Program             program(context, devices, binaries);
+            program.build();
+            CLKernelLibrary::get().add_built_program(name, program);
+        }
+        cache_file.close();
+    }
+}
+
+/** This function saves opencl kernels library to a file
+ *
+ * @param[in] filename Name of the file to be used to save the library
+ */
+void save_program_cache_to_file(const std::string &filename = "cache.bin")
+{
+    std::cout << "Saving opencl kernels to " << filename << std::endl;
+    std::ofstream cache_file(filename, std::ios::binary);
+    if(cache_file.is_open())
+    {
+        for(const auto &it : CLKernelLibrary::get().get_built_programs())
+        {
+            std::vector<std::vector<unsigned char>> binaries = it.second.getInfo<CL_PROGRAM_BINARIES>();
+            ARM_COMPUTE_ERROR_ON(binaries.size() != 1);
+            const std::string kernel_name      = it.first;
+            size_t            kernel_name_size = kernel_name.length();
+            size_t            binary_size      = binaries[0].size();
+            cache_file.write(reinterpret_cast<char *>(&kernel_name_size), sizeof(size_t));
+            cache_file.write(reinterpret_cast<char *>(&binary_size), sizeof(size_t));
+            cache_file.write(kernel_name.c_str(), kernel_name_size);
+            cache_file.write(reinterpret_cast<const char *>(binaries[0].data()), binaries[0].size());
+        }
+        cache_file.close();
+    }
+}
+} // namespace
+
+class CLCacheExample : public Example
+{
+public:
+    CLCacheExample() = default;
+
+    bool do_setup(int argc, char **argv) override
+    {
+        std::cout << "Once the program has run and created the file cache.bin, rerun with --restore_cache." << std::endl;
+        CLScheduler::get().default_init();
+        auto start_time = std::chrono::high_resolution_clock::now();
+        if(argc > 1)
+        {
+            std::string argv1 = argv[1];
+            std::transform(argv1.begin(), argv1.end(), argv1.begin(), ::tolower);
+            if(argv1 == "--restore_cache")
+            {
+                // Load the precompiled kernels from a file into the kernel library, in this way the next time they are needed
+                // compilation won't be required.
+                restore_program_cache_from_file();
+            }
+            else
+            {
+                std::cout << "Unkown option " << argv1 << std::endl;
+            }
+        }
+
+        // Initialise shapes
+        init_tensor(TensorShape(8U, 4U, 2U), tensor_nchw, DataType::U8, DataLayout::NCHW);
+        init_tensor(TensorShape(2U, 8U, 4U), tensor_nhwc, DataType::U8, DataLayout::NHWC);
+        init_tensor(TensorShape(8U, 4U, 2U), tensor_nchw_result, DataType::U8, DataLayout::NCHW);
+
+        // Create the permutation vector to turn a NCHW tensor to NHWC.
+        // The input tensor is NCHW, which means that the fastest changing coordinate is W=8U.
+        // For permutation vectors the fastest changing coordinate is the one on the left too.
+        // Each element in the permutation vector specifies a mapping from the source tensor to the destination one, thus if we
+        // use 2U in the permutation vector's first element we are telling the function to move the channels to the fastest
+        // changing coordinate in the destination tensor.
+
+        const PermutationVector vector_nchw_to_nhwc(2U, 0U, 1U);
+        permute_nhwc.configure(&tensor_nchw, &tensor_nhwc, vector_nchw_to_nhwc);
+
+        // Allocate and fill tensors
+        tensor_nhwc.allocator()->allocate();
+        tensor_nchw.allocator()->allocate();
+        fill_tensor(tensor_nchw);
+
+        // Demostrate autoconfigure for the output tensor
+        const PermutationVector vector_nhwc_to_nchw(1U, 2U, 0U);
+        permute_nchw.configure(&tensor_nhwc, &tensor_nchw_result, vector_nhwc_to_nchw);
+        tensor_nchw_result.allocator()->allocate();
+
+        auto end_time        = std::chrono::high_resolution_clock::now();
+        auto time_elapsed    = end_time - start_time;
+        auto time_elapsed_ms = std::chrono::duration_cast<std::chrono::milliseconds>(time_elapsed).count();
+        std::cout << "Configuration time " << time_elapsed_ms << " ms " << std::endl;
+        // Save the opencl kernels to a file
+        save_program_cache_to_file();
+
+        return true;
+    }
+    void do_run() override
+    {
+        permute_nhwc.run();
+        permute_nchw.run();
+    }
+    void do_teardown() override
+    {
+    }
+
+private:
+    void validate_result(CLTensor &reference, CLTensor &result)
+    {
+        reference.map();
+        result.map();
+        Window window;
+        window.use_tensor_dimensions(reference.info()->tensor_shape());
+        Iterator it_ref(&reference, window);
+        Iterator it_res(&result, window);
+        execute_window_loop(window, [&](const Coordinates &)
+        {
+            assert(*reinterpret_cast<unsigned char *>(it_ref.ptr()) == *reinterpret_cast<unsigned char *>(it_res.ptr()));
+        },
+        it_ref, it_res);
+        reference.unmap();
+        result.unmap();
+    }
+
+    void fill_tensor(CLTensor &tensor)
+    {
+        tensor.map();
+        Window window;
+        window.use_tensor_dimensions(tensor.info()->tensor_shape());
+        Iterator      it_tensor(&tensor, window);
+        unsigned char val(0);
+        execute_window_loop(window, [&](const Coordinates &)
+        {
+            *reinterpret_cast<unsigned char *>(it_tensor.ptr()) = val++;
+        },
+        it_tensor);
+        tensor.unmap();
+    }
+    void init_tensor(const TensorShape shape, CLTensor &tensor, DataType type, DataLayout layout)
+    {
+        tensor.allocator()->init(TensorInfo(shape, 1, type).set_data_layout(layout));
+    }
+
+    CLTensor  tensor_nchw{};
+    CLTensor  tensor_nhwc{};
+    CLTensor  tensor_nchw_result{};
+    CLPermute permute_nhwc{};
+    CLPermute permute_nchw{};
+};
+
+/** Main program creating an example that demostrates how to load precompiled kernels from a file.
+ *
+ * @param[in] argc Number of arguments
+ * @param[in] argv Arguments
+ */
+int main(int argc, char **argv)
+{
+    return utils::run_example<CLCacheExample>(argc, argv);
+}