added module12 localmem atomics to oneAPI-essentials (#1099)

* added module12 localmem atomics to oneAPI-essentials

* updated oneAPI Essentials readme and makefile with module12

* removed gamma-correction sample.json

Author: krisrak

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/07_DPCPP_Library/gamma-correction/sample.json

@@ -0,0 +1,8 @@
+Copyright 2020 Intel Corporation
+
+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.
+

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/12_DPCPP_Local_Memory_And_Atomics/License.txt

@@ -0,0 +1,27 @@
+## Title
+SYCL Local Memory and Atomics: This is part 12 of the oneAPI essentials training series
+  
+## Requirements
+| Optimized for                       | Description
+|:---                               |:---
+| OS                                | Linux* Ubuntu 18.04, 20 Windows* 10
+| Hardware                          | Skylake with GEN9 or newer
+| Software                          | Intel® oneAPI DPC++ Compiler, Jupyter Notebooks, Intel Devcloud
+  
+## Purpose
+This hands-on exercise demonstrates SYCL Atomic Operations and Local Memory Usage. You will learn how to use perform reductions using atomic operation and also learn how to use local memory to optimize for performance.
+
+## License  
+Code samples are licensed under the MIT license. See [License.txt](https://github.com/oneapi-src/oneAPI-samples/blob/master/License.txt) for details.
+
+Third party program Licenses can be found here: [third-party-programs.txt](https://github.com/oneapi-src/oneAPI-samples/blob/master/third-party-programs.txt)
+
+## Install Directions
+
+The Jupyter notebooks are tested and can be run on Intel Devcloud.
+Below are the steps to access these Jupyter notebooks on Intel Devcloud
+1. Register on [Intel Devcloud](https://intelsoftwaresites.secure.force.com/devcloud/oneapi)
+2. Go to the "Terminal" in the Intel Devcloud
+3. Type in the below command to download the oneAPI-essentials series notebooks into your Devcloud account
+    /data/oneapi_workshop/get_jupyter_notebooks.sh
+4. Navigate to oneAPI_Essentials folder and open the Welcome.ipynb, click on "DPC++ Reductions" notebook and follow the instructions

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/12_DPCPP_Local_Memory_And_Atomics/LocalMemory_Atomics.ipynb

@@ -0,0 +1,53 @@
+//==============================================================
+// Copyright © Intel Corporation
+//
+// SPDX-License-Identifier: MIT
+// =============================================================
+#include <CL/sycl.hpp>
+
+using namespace sycl;
+
+static constexpr size_t N = 1024; // global size
+
+int main() {
+  queue q;
+  std::cout << "Device : " << q.get_device().get_info<info::device::name>() << "\n";
+
+  auto data = malloc_shared<int>(N, q);
+  for (int i = 0; i < N; i++) data[i] = i;
+  auto min = malloc_shared<int>(1, q);
+  auto max = malloc_shared<int>(1, q);
+  min[0] = 0;
+  max[0] = 0;
+
+  //# Reduction Kernel using atomics 
+  q.parallel_for(N, [=](auto i) {
+    //# STEP 1: create atomic reference for min and max
+
+    //# YOUR CODE GOES HERE
+    
+    
+    
+    
+    //# STEP 2: add atomic operation for min and max computation  
+
+    //# YOUR CODE GOES HERE
+    
+    
+    
+  }).wait();
+
+  auto mid = 0.0;
+  //# STEP 3: Compute mid-range using the min and max 
+
+  //# YOUR CODE GOES HERE
+    
+    
+    
+  
+  std::cout << "Minimum   = " << min[0] << "\n";
+  std::cout << "Maximum   = " << max[0] << "\n";
+  std::cout << "Mid-Range = " << mid << "\n";
+
+  return 0;
+}

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/12_DPCPP_Local_Memory_And_Atomics/Readme.md

@@ -0,0 +1,26 @@
+//==============================================================
+// Copyright © Intel Corporation
+//
+// SPDX-License-Identifier: MIT
+// =============================================================
+#include <CL/sycl.hpp>
+
+using namespace cl::sycl;
+
+int main() {
+  queue q;
+
+  //# Print the device info
+  std::cout << "device name   : " << q.get_device().get_info<info::device::name>() << "\n";
+  std::cout << "local_mem_size: " << q.get_device().get_info<info::device::local_mem_size>() << "\n";
+
+  auto local_mem_type = q.get_device().get_info<info::device::local_mem_type>();
+  if(local_mem_type == info::local_mem_type::local) 
+    std::cout << "local_mem_type: info::local_mem_type::local" << "\n";
+  else if(local_mem_type == info::local_mem_type::global) 
+    std::cout << "local_mem_type: info::local_mem_type::global" << "\n";
+  else if(local_mem_type == info::local_mem_type::none) 
+    std::cout << "local_mem_type: info::local_mem_type::none" << "\n";
+ 
+  return 0;
+}

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/12_DPCPP_Local_Memory_And_Atomics/assets/localmem.png

@@ -0,0 +1,86 @@
+//==============================================================
+// Copyright © 2021 Intel Corporation
+//
+// SPDX-License-Identifier: MIT
+// =============================================================
+
+
+#include <CL/sycl.hpp>
+#include <iomanip>
+
+using namespace sycl;
+
+int main() {
+    
+    size_t N = 16;
+    std::cout << "MATRIX_SIZE    : " << N << "x" << N << std::endl;
+
+    //# Define vectors for matrices
+    std::vector<float> matrix_a(N*N);
+    std::vector<float> matrix_b(N*N);
+    std::vector<float> matrix_c(N*N);
+    std::vector<float> matrix_d(N*N);
+    
+    //# Initialize matrices with values
+    float v1 = 2.f;
+    float v2 = 3.f;
+    for (int i=0; i<N; i++)
+        for (int j=0; j<N; j++){
+            matrix_a[i*N+j] = v1++;
+            matrix_b[i*N+j] = v2++;
+            matrix_c[i*N+j] = 0.f;
+            matrix_d[i*N+j] = 0.f;
+    }
+    
+    //# Define queue with default device for offloading computation
+    queue q;
+    std::cout << "Offload Device : " << q.get_device().get_info<info::device::name>() << std::endl;
+    
+    //# Create buffers for matrices
+    buffer a(matrix_a);
+    buffer b(matrix_b);
+    buffer c(matrix_c);
+
+    //# Submit command groups to execute on device
+    q.submit([&](handler &h){
+        //# Create accessors to copy buffers to the device
+        accessor A(a, h, read_only);
+        accessor B(b, h, read_only);
+        accessor C(c, h, write_only);
+
+        //# Define size for ND-range and work-group size
+        range<2> global_size(N,N);
+        range<2> work_group_size(N,N);
+
+        //# Parallel Compute Matrix Multiplication
+        h.parallel_for(nd_range<2>{global_size, work_group_size}, [=](nd_item<2> item){
+            const int i = item.get_global_id(0);
+            const int j = item.get_global_id(1);
+
+            //# matrix multiplication computation from local memory
+            float temp = 0.f;
+            for (int k = 0; k < N; k++) {
+                temp += A[i*N+k] * B[k*N+j];
+            }
+            C[i*N+j] = temp;
+        });
+    });
+    host_accessor ha(c, read_only);
+    
+    //# Print Output and Verification
+    auto FAIL = 0;
+    for (int i=0; i<N; i++){
+        for (int j=0; j<N; j++){
+            for(int k=0; k<N; k++){
+                matrix_d[i*N+j] += matrix_a[i*N+k] * matrix_b[k*N+j];
+            }
+            if(matrix_d[i*N+j] != matrix_c[i*N+j]) FAIL = 1;
+            std::cout << std::setw(6) << matrix_c[i*N+j] << " ";
+        }
+        std::cout << "\n";
+    }
+    if(FAIL == 1) std::cout << "FAIL\n"; else std::cout << "PASS\n";
+
+    return 0;
+}
+

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/12_DPCPP_Local_Memory_And_Atomics/assets/naive.PNG

@@ -0,0 +1,99 @@
+//==============================================================
+// Copyright © Intel Corporation
+//
+// SPDX-License-Identifier: MIT
+// =============================================================
+
+
+#include <CL/sycl.hpp>
+#include <iomanip>
+
+using namespace sycl;
+
+int main() {
+    
+    size_t N = 16;
+    std::cout << "MATRIX_SIZE    : " << N << "x" << N << std::endl;
+
+    //# Define vectors for matrices
+    std::vector<float> matrix_a(N*N);
+    std::vector<float> matrix_b(N*N);
+    std::vector<float> matrix_c(N*N);
+    std::vector<float> matrix_d(N*N);
+    
+    //# Initialize matrices with values
+    float v1 = 2.f;
+    float v2 = 3.f;
+    for (int i=0; i<N; i++)
+        for (int j=0; j<N; j++){
+            matrix_a[i*N+j] = v1++;
+            matrix_b[i*N+j] = v2++;
+            matrix_c[i*N+j] = 0.f;
+            matrix_d[i*N+j] = 0.f;
+    }
+    
+    //# Define queue with default device for offloading computation
+    queue q;
+    std::cout << "Offload Device : " << q.get_device().get_info<info::device::name>() << std::endl;
+    
+    //# Create buffers for matrices
+    buffer a(matrix_a);
+    buffer b(matrix_b);
+    buffer c(matrix_c);
+
+    //# Submit command groups to execute on device
+    q.submit([&](handler &h){
+        //# Create accessors to copy buffers to the device
+        accessor A(a, h, read_only);
+        accessor B(b, h, read_only);
+        accessor C(c, h, write_only);
+
+        //# Define size for ND-range and work-group size
+        range<2> global_size(N,N);
+        range<2> work_group_size(N,N);
+
+        //# Create local accessors
+        accessor<float, 2, access::mode::read_write, access::target::local> A_local(range<2>(N, N), h);
+        accessor<float, 2, access::mode::read_write, access::target::local> B_local(range<2>(N, N), h);
+
+        //# Parallel Compute Matrix Multiplication
+        h.parallel_for(nd_range<2>{global_size, work_group_size}, [=](nd_item<2> item){
+            const int i = item.get_global_id(0);
+            const int j = item.get_global_id(1);
+            const int x = item.get_local_id(0);
+            const int y = item.get_local_id(1);
+
+            //# copy from global to local memory
+            A_local[x][y] = A[i * N + j];
+            B_local[x][y] = B[i * N + j];
+
+            //# barrier to sychronize local memory copy across all work items
+            group_barrier(item.get_group());
+
+            //# matrix multiplication computation from local memory
+            float temp = 0.f;
+            for (int k = 0; k < N; k++) {
+                temp += A_local[x][k] * B_local[k][y];
+            }
+            C[i*N+j] = temp;
+        });
+    });
+    host_accessor ha(c, read_only);
+    
+    //# Print Output and Verification
+    auto FAIL = 0;
+    for (int i=0; i<N; i++){
+        for (int j=0; j<N; j++){
+            for(int k=0; k<N; k++){
+                matrix_d[i*N+j] += matrix_a[i*N+k] * matrix_b[k*N+j];
+            }
+            if(matrix_d[i*N+j] != matrix_c[i*N+j]) FAIL = 1;
+            std::cout << std::setw(6) << matrix_c[i*N+j] << " ";
+        }
+        std::cout << "\n";
+    }
+    if(FAIL == 1) std::cout << "FAIL\n"; else std::cout << "PASS\n";
+
+    return 0;
+}
+

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/12_DPCPP_Local_Memory_And_Atomics/lab/atomics_lab.cpp

@@ -0,0 +1,38 @@
+//==============================================================
+// Copyright © Intel Corporation
+//
+// SPDX-License-Identifier: MIT
+// =============================================================
+#include <CL/sycl.hpp>
+
+using namespace sycl;
+
+static constexpr size_t N = 1024; // global size
+
+int main() {
+  queue q;
+  std::cout << "Device : " << q.get_device().get_info<info::device::name>() << "\n";
+
+  std::vector<int> data(N);
+  for (int i = 0; i < N; i++) data[i] = i;
+  int sum = 0;
+  {
+    //# create buffers for data and sum
+    buffer buf_data(data);
+    buffer buf_sum(&sum, range(1));
+
+    //# Reduction Kernel using atomics 
+    q.submit([&](auto &h) {
+      accessor data_acc(buf_data, h, sycl::read_only);
+      accessor sum_acc(buf_sum, h);
+
+      h.parallel_for(N, [=](auto i) {
+        auto sum_atomic = atomic_ref<int, memory_order::relaxed, memory_scope::device, access::address_space::global_space>(sum_acc[0]);
+        sum_atomic += data_acc[i];
+      });
+    });
+  }
+  std::cout << "Sum = " << sum << "\n";
+
+  return 0;
+}

DirectProgramming/DPC++/Jupyter/oneapi-essentials-training/12_DPCPP_Local_Memory_And_Atomics/lab/localmem_info.cpp

@@ -0,0 +1,29 @@
+//==============================================================
+// Copyright © Intel Corporation
+//
+// SPDX-License-Identifier: MIT
+// =============================================================
+#include <CL/sycl.hpp>
+
+using namespace sycl;
+
+static constexpr size_t N = 1024; // global size
+
+int main() {
+  queue q;
+  std::cout << "Device : " << q.get_device().get_info<info::device::name>() << "\n";
+
+  auto data = malloc_shared<int>(N, q);
+  for (int i = 0; i < N; i++) data[i] = i;
+  auto sum = malloc_shared<int>(1, q);
+  sum[0] = 0;
+
+  //# Reduction Kernel using atomics 
+  q.parallel_for(N, [=](auto i) {
+    auto sum_atomic = atomic_ref<int, memory_order::relaxed, memory_scope::device, access::address_space::global_space>(sum[0]);
+    sum_atomic += data[i];
+  }).wait();
+
+  std::cout << "Sum = " << sum[0] << "\n";
+  return 0;
+}