Moved all cache-related functions to a separate file; added a ClearCompiledProgramCache function to clear the cache

4 files changed, 98 insertions, 24 deletions

diff --git a/include/clblast.h b/include/clblast.h
index f73acb57..4a3ec9b6 100644
--- a/include/clblast.h
+++ b/include/clblast.h
@@ -539,6 +539,13 @@ StatusCode Trsm(const Layout layout, const Side side, const Triangle triangle, c
                 cl_command_queue* queue, cl_event* event = nullptr);
 
 // =================================================================================================
+
+// CLBlast stores binaries of compiled kernels into a cache in case the same kernel is used later on
+// for the same device. This cache can be cleared to free up system memory or in case of debugging.
+StatusCode ClearCompiledProgramCache();
+
+// =================================================================================================
+
 } // namespace clblast
 
 // CLBLAST_CLBLAST_H_
diff --git a/include/clblast_c.h b/include/clblast_c.h
index 8c0a0792..1ca300ca 100644
--- a/include/clblast_c.h
+++ b/include/clblast_c.h
@@ -1036,6 +1036,12 @@ StatusCode PUBLIC_API CLBlastZtrsm(const Layout layout, const Side side, const T
 
 // =================================================================================================
 
+// CLBlast stores binaries of compiled kernels into a cache in case the same kernel is used later on
+// for the same device. This cache can be cleared to free up system memory or in case of debugging.
+StatusCode PUBLIC_API CLBlastClearCompiledProgramCache();
+
+// =================================================================================================
+
 #ifdef __cplusplus
 } // extern "C"
 #endif
diff --git a/include/internal/cache.h b/include/internal/cache.h
new file mode 100644
index 00000000..44fad68d
--- /dev/null
+++ b/include/internal/cache.h
@@ -0,0 +1,72 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+//   Cedric Nugteren <www.cedricnugteren.nl>
+//
+// This file implements the caching functionality of compiled binaries.
+//
+// =================================================================================================
+
+#ifndef CLBLAST_CACHE_H_
+#define CLBLAST_CACHE_H_
+
+#include <string>
+#include <vector>
+#include <mutex>
+
+#include "internal/utilities.h"
+
+namespace clblast {
+namespace cache {
+// =================================================================================================
+
+// The cache of compiled OpenCL programs, along with some meta-data
+struct ProgramCache {
+  Program program;
+  std::string device_name;
+  Precision precision;
+  std::string routine_name_;
+
+  // Finds out whether the properties match
+  bool MatchInCache(const std::string &ref_device, const Precision &ref_precision,
+                    const std::string &ref_routine) {
+    return (device_name == ref_device &&
+            precision == ref_precision &&
+            routine_name_ == ref_routine);
+  }
+};
+
+// The actual cache, implemented as a vector of the above data-type, and its mutex
+static std::vector<ProgramCache> program_cache_;
+static std::mutex program_cache_mutex_;
+
+// =================================================================================================
+
+// Stores the compiled program in the cache
+void StoreProgramToCache(const Program& program, const std::string &device_name,
+                         const Precision &precision, const std::string &routine_name);
+
+// Queries the cache and retrieves a matching program. Assumes that the match is available, throws
+// otherwise.
+const Program& GetProgramFromCache(const std::string &device_name, const Precision &precision,
+                                   const std::string &routine_name);
+
+// Queries the cache to see whether or not the compiled kernel is already there
+bool ProgramIsInCache(const std::string &device_name, const Precision &precision,
+                      const std::string &routine_name);
+
+// =================================================================================================
+
+// Clears the cache of stored program binaries
+StatusCode ClearCompiledProgramCache();
+
+// =================================================================================================
+} // namespace cache
+} // namespace clblast
+
+// CLBLAST_CACHE_H_
+#endif
diff --git a/include/internal/routine.h b/include/internal/routine.h
index b2b6f622..013769d8 100644
--- a/include/internal/routine.h
+++ b/include/internal/routine.h
@@ -18,8 +18,8 @@
 
 #include <string>
 #include <vector>
-#include <mutex>
 
+#include "internal/cache.h"
 #include "internal/utilities.h"
 #include "internal/database.h"
 
@@ -31,26 +31,6 @@ template <typename T>
 class Routine {
  public:
 
-  // The cache of compiled OpenCL programs, along with some meta-data
-  struct ProgramCache {
-    Program program;
-    std::string device_name;
-    Precision precision;
-    std::string routine_name_;
-
-    // Finds out whether the properties match
-    bool MatchInCache(const std::string &ref_device, const Precision &ref_precision,
-                      const std::string &ref_routine) {
-      return (device_name == ref_device &&
-              precision == ref_precision &&
-              routine_name_ == ref_routine);
-    }
-  };
-
-  // The actual cache, implemented as a vector of the above data-type, and its mutex
-  static std::vector<ProgramCache> program_cache_;
-  static std::mutex program_cache_mutex_;
-
   // Helper functions which check for errors in the status code
   static constexpr bool ErrorIn(const StatusCode s) { return (s != StatusCode::kSuccess); }
 
@@ -103,12 +83,21 @@ class Routine {
                                     const bool do_transpose, const bool do_conjugate,
                                     const bool upper = false, const bool lower = false,
                                     const bool diagonal_imag_zero = false);
-  
+
+  // Stores a newly compiled program into the cache
+  void StoreProgramToCache(const Program& program) const {
+    return cache::StoreProgramToCache(program, device_name_, precision_, routine_name_);
+  }
+
   // Queries the cache and retrieve either a matching program or a boolean whether a match exists.
   // The first assumes that the program is available in the cache and will throw an exception
   // otherwise.
-  const Program& GetProgramFromCache() const;
-  bool ProgramIsInCache() const;
+  const Program& GetProgramFromCache() const {
+    return cache::GetProgramFromCache(device_name_, precision_, routine_name_);
+  }
+  bool ProgramIsInCache() const {
+    return cache::ProgramIsInCache(device_name_, precision_, routine_name_);
+  }
 
   // Non-static variable for the precision. Note that the same variable (but static) might exist in
   // a derived class.