Be able to share compiled OpenCL binary caches among identical GPUs

Baikal/Controllers/clw_scene_controller.cpp

@@ -51,10 +51,11 @@ namespace Baikal
     }
 
 
+    Material::Ptr ClwSceneController::s_default_material = UberV2Material::Create();
+
     ClwSceneController::ClwSceneController(CLWContext context, RadeonRays::IntersectionApi* api, const CLProgramManager *program_manager)
     : m_context(context)
     , m_api(api)
-    , m_default_material(UberV2Material::Create())
     , m_program_manager(program_manager)
     {
         auto acc_type = "fatbvh";
@@ -72,7 +73,7 @@ namespace Baikal
 
     Material::Ptr ClwSceneController::GetDefaultMaterial() const
     {
-        return m_default_material;
+        return s_default_material;
     }
 
     ClwSceneController::~ClwSceneController()
@@ -1275,7 +1276,7 @@ namespace Baikal
 
     int ClwSceneController::GetMaterialIndex(Collector const& collector, Material::Ptr material) const
     {
-        auto m = material ? material : m_default_material;
+        auto m = material ? material : s_default_material;
         return ResolveMaterialPtr(m);
     }
 
@@ -1392,7 +1393,7 @@ namespace Baikal
 
     int ClwSceneController::GetMaterialLayers(Material::Ptr material) const
     {
-        auto m = material ? material : m_default_material;
+        auto m = material ? material : s_default_material;
         return std::static_pointer_cast<UberV2Material>(m)->GetLayers();
     }
 

Baikal/Controllers/clw_scene_controller.h

@@ -129,7 +129,7 @@ namespace Baikal
         // Intersection API
         RadeonRays::IntersectionApi* m_api;
         // Default material
-        Material::Ptr m_default_material;
+        static Material::Ptr s_default_material;
         // CL Program manager
         const CLProgramManager *m_program_manager;
         // Material to device material map

Baikal/Utils/cl_program.cpp

@@ -103,6 +103,9 @@ inline void SaveBinaries(std::string const& name, std::vector<std::uint8_t>& dat
 }
 
 
+std::unordered_map<std::string, std::shared_ptr<std::mutex>> CLProgram::s_binary_cache_names;
+std::mutex CLProgram::s_binary_cache_map_mutex;
+
 CLProgram::CLProgram(const CLProgramManager *program_manager, uint32_t id, CLWContext context,
                      const std::string &program_name, const std::string &cache_path) :
     m_program_manager(program_manager),
@@ -246,8 +249,24 @@ CLWProgram CLProgram::GetCLWProgram(const std::string &opts)
         cached_program_path.append(".bin");
 
         std::vector<std::uint8_t> binary;
-        if (LoadBinaries(cached_program_path, binary))
+
+        std::unique_lock<std::mutex> map_lock(s_binary_cache_map_mutex);
+        auto iter = s_binary_cache_names.find(filename);
+        if (iter == s_binary_cache_names.end())
+            iter = s_binary_cache_names.insert({ filename, std::make_shared<std::mutex>() }).first;
+        auto cache_mutex = iter->second;
+        assert(cache_mutex != nullptr);
+        map_lock.unlock();
+
+        // Other workers with the same binary cache requirement will be blocked here
+        std::unique_lock<std::mutex> cache_lock(*cache_mutex);
+        bool loaded = LoadBinaries(cached_program_path, binary);
+
+        if (loaded)
         {
+            // Binary cache existed, other workers can read it
+            cache_lock.unlock();
+
             // Create from binary
             std::size_t size = binary.size();
             auto binaries = &binary[0];
@@ -262,6 +281,9 @@ CLWProgram CLProgram::GetCLWProgram(const std::string &opts)
             // Save binaries
             result.GetBinaries(0, binary);
             SaveBinaries(cached_program_path, binary);
+
+            // Block other workers until binary cache generated
+            cache_lock.unlock();
         }
     }
 

Baikal/Utils/cl_program.h

@@ -27,6 +27,7 @@ THE SOFTWARE.
 #include <set>
 #include <unordered_map>
 #include <unordered_set>
+#include <mutex>
 #include "CLWProgram.h"
 #include "CLWContext.h"
 
@@ -95,5 +96,7 @@ namespace Baikal
         CLWContext m_context;
         std::set<std::string> m_included_headers; ///< Set of included headers
 
+        static std::unordered_map<std::string, std::shared_ptr<std::mutex>> s_binary_cache_names;
+        static std::mutex s_binary_cache_map_mutex;
     };
 }