Merge pull request #281 from clEsperanto/release-0.10.0

Release-0.10.0
clEsperanto · Apr 17, 2024 · 9f3aace · 9f3aace
2 parents f17d5d4 + c185459
commit 9f3aace
Show file tree

Hide file tree

Showing 280 changed files with 9,348 additions and 1,133 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.20)
 
-project(CLIc VERSION 0.9.1)
+project(CLIc VERSION 0.10.0)
 
 set(BUILD_CUDA_BACKEND ON CACHE BOOL "Build CUDA backend")
 set(BUILD_OPENCL_BACKEND ON CACHE BOOL "Build OpenCL backend")
@@ -24,6 +24,9 @@ configure_file(
 
 # list project source files
 file(GLOB_RECURSE SOURCES "${CMAKE_CURRENT_SOURCE_DIR}/${LIBRARY_NAME_LOWERCASE}/src/*.cpp")
+file(GLOB_RECURSE TIERS "${CMAKE_CURRENT_SOURCE_DIR}/${LIBRARY_NAME_LOWERCASE}/src/**/*.cpp")
+list(APPEND SOURCES ${TIERS})
+
 file(GLOB_RECURSE HEADERS_SRC "${CMAKE_CURRENT_SOURCE_DIR}/${LIBRARY_NAME_LOWERCASE}/include/*.hpp")
 file(GLOB_RECURSE HEADERS_BIN "${CMAKE_CURRENT_BINARY_DIR}/${LIBRARY_NAME_LOWERCASE}/include/*.hpp")
 set(HEADERS ${HEADERS_SRC})

diff --git a/clic/include/array.hpp b/clic/include/array.hpp
@@ -287,6 +287,20 @@ class Array
    */
   Array(const Array &) = default;
 
+  /**
+   * @brief Check if the shared_ptr is null and throw an exception if it is
+   * @param ptr The shared_ptr to check
+   * @param errorMessage The error message to throw
+   */
+  static inline void
+  check_ptr(const Array::Pointer & ptr, const char * errorMessage)
+  {
+    if (!ptr)
+    {
+      throw std::invalid_argument(errorMessage);
+    }
+  }
+
   /**
    * @brief Print the Array as a matrix for debugging
    */
@@ -320,23 +334,30 @@ class Array
 
 template <typename T>
 auto
-print(const Array::Pointer & array) -> void
+print(const Array::Pointer & array, const char * name = "Array::Pointer") -> void
 {
+  if (array == nullptr)
+  {
+    std::cout << "Print Array::Pointer (nullptr)\n";
+    return;
+  }
   std::vector<T> host_data(array->size());
   array->read(host_data.data());
-
-  for (int i = 0; i < array->depth(); i++)
+  std::ostringstream oss;
+  oss << "Print (" << name << ")\n";
+  for (auto i = 0; i < array->depth(); ++i)
   {
-    std::cout << "z = " << i << std::endl;
-    for (int j = 0; j < array->height(); j++)
+    oss << "z = " << i << '\n';
+    for (auto j = 0; j < array->height(); ++j)
     {
-      for (int k = 0; k < array->width(); k++)
+      for (auto k = 0; k < array->width(); ++k)
       {
-        std::cout << (float)host_data[i * array->height() * array->width() + j * array->width() + k] << " ";
+        oss << static_cast<float>(host_data[i * array->height() * array->width() + j * array->width() + k]) << ' ';
       }
-      std::cout << std::endl;
+      oss << '\n';
     }
   }
+  std::cout << oss.str();
 }
 
 } // namespace cle

diff --git a/clic/include/execution.hpp b/clic/include/execution.hpp
@@ -33,6 +33,24 @@ execute(const Device::Pointer & device,
         const RangeArray &      global_range = { 1, 1, 1 },
         const ConstantList &    constants = {}) -> void;
 
+
+/**
+ * @brief Execution function for separable kernel operation
+ * @param device Device pointer
+ * @param kernel Kernel function name and code
+ * @param src Source array
+ * @param dst Destination array
+ * @param sigma Sigma value for the kernel
+ * @param radius Radius value for the kernel
+ */
+auto
+execute_separable(const Device::Pointer &      device,
+                  const KernelInfo &           kernel,
+                  const Array::Pointer &       src,
+                  const Array::Pointer &       dst,
+                  const std::array<float, 3> & sigma,
+                  const std::array<int, 3> &   radius) -> void;
+
 /**
  * @brief Execute a kernel using native framework
  * @param device Device pointer

diff --git a/clic/include/memory.hpp b/clic/include/memory.hpp
@@ -42,7 +42,7 @@ create_like(const Array & arr) -> Array
 static auto
 create_2d_xy(const Array & arr) -> Array
 {
-  return Array{ arr.width(), arr.height(), 1, arr.dtype(), mType::BUFFER, arr.device() };
+  return Array{ arr.width(), arr.height(), 1, 2, arr.dtype(), mType::BUFFER, arr.device() };
 }
 
 /**
@@ -53,7 +53,7 @@ create_2d_xy(const Array & arr) -> Array
 static auto
 create_2d_zy(const Array & arr) -> Array
 {
-  return Array{ arr.depth(), arr.height(), 1, arr.dtype(), mType::BUFFER, arr.device() };
+  return Array{ arr.depth(), arr.height(), 1, 2, arr.dtype(), mType::BUFFER, arr.device() };
 }
 
 /**
@@ -64,7 +64,7 @@ create_2d_zy(const Array & arr) -> Array
 static auto
 create_2d_xz(const Array & arr) -> Array
 {
-  return Array{ arr.width(), arr.depth(), 1, arr.dtype(), mType::BUFFER, arr.device() };
+  return Array{ arr.width(), arr.depth(), 1, 2, arr.dtype(), mType::BUFFER, arr.device() };
 }
 
 /**

diff --git a/clic/include/tier0.hpp b/clic/include/tier0.hpp
@@ -113,23 +113,6 @@ create_xz(const Array::Pointer & src, Array::Pointer & dst, dType type = dType::
 auto
 create_zx(const Array::Pointer & src, Array::Pointer & dst, dType type = dType::UNKNOWN) -> void;
 
-/**
- * @brief Execution function for separable kernel operation
- * @param device Device pointer
- * @param kernel Kernel function name and code
- * @param src Source array
- * @param dst Destination array
- * @param sigma Sigma value for the kernel
- * @param radius Radius value for the kernel
- */
-auto
-execute_separable_func(const Device::Pointer &      device,
-                       const KernelInfo &           kernel,
-                       const Array::Pointer &       src,
-                       const Array::Pointer &       dst,
-                       const std::array<float, 3> & sigma,
-                       const std::array<int, 3> &   radius) -> void;
-
 
 } // namespace cle::tier0