From 628e1e8cce084657df854fa97ba18ad740270b23 Mon Sep 17 00:00:00 2001
From: Kirill Mavreshko <kimavr@gmail.com>
Date: Fri, 26 May 2017 15:04:19 +0500
Subject: Fixes inability to run GEMM on multiple identical GPUs (issue #155)

---
 src/cache.hpp   | 4 ++--
 src/clblast.cpp | 2 +-
 src/routine.cpp | 6 +++---
 3 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/cache.hpp b/src/cache.hpp
index 694de839..207cb2dd 100644
--- a/src/cache.hpp
+++ b/src/cache.hpp
@@ -80,8 +80,8 @@ extern template std::string BinaryCache::Get(const BinaryKeyRef &, bool *) const
 
 // The key struct for the cache of compiled OpenCL programs (context-dependent)
 // Order of fields: context, precision, routine_name (smaller fields first)
-typedef std::tuple<cl_context, Precision, std::string> ProgramKey;
-typedef std::tuple<const cl_context &, const Precision &, const std::string &> ProgramKeyRef;
+typedef std::tuple<cl_context, cl_device_id, Precision, std::string> ProgramKey;
+typedef std::tuple<const cl_context &, const cl_device_id &, const Precision &, const std::string &> ProgramKeyRef;
 
 typedef Cache<ProgramKey, Program> ProgramCache;
 
diff --git a/src/clblast.cpp b/src/clblast.cpp
index 1eb1ddd7..a73b67f3 100644
--- a/src/clblast.cpp
+++ b/src/clblast.cpp
@@ -2470,7 +2470,7 @@ StatusCode OverrideParameters(const cl_device_id device, const std::string &kern
     // Clears the existing program & binary cache for routines with the target kernel
     const auto routine_names = Routine::routines_by_kernel.at(kernel_name);
     for (const auto &routine_name : routine_names) {
-      ProgramCache::Instance().RemoveBySubset<1, 2>(ProgramKey{nullptr, precision, routine_name});
+      ProgramCache::Instance().RemoveBySubset<1, 2>(ProgramKey{nullptr, device, precision, routine_name});
       BinaryCache::Instance().Remove(BinaryKey{precision, routine_name, device_name});
     }
 
diff --git a/src/routine.cpp b/src/routine.cpp
index cb39c7ee..81baa590 100644
--- a/src/routine.cpp
+++ b/src/routine.cpp
@@ -87,7 +87,7 @@ void Routine::InitProgram(std::initializer_list<const char *> source) {
 
   // Queries the cache to see whether or not the program (context-specific) is already there
   bool has_program;
-  program_ = ProgramCache::Instance().Get(ProgramKeyRef{ context_(), precision_, routine_name_ },
+  program_ = ProgramCache::Instance().Get(ProgramKeyRef{ context_(), device_(), precision_, routine_name_ },
                                           &has_program);
   if (has_program) { return; }
 
@@ -106,7 +106,7 @@ void Routine::InitProgram(std::initializer_list<const char *> source) {
   if (has_binary) {
     program_ = Program(device_, context_, binary);
     program_.Build(device_, options);
-    ProgramCache::Instance().Store(ProgramKey{ context_(), precision_, routine_name_ },
+    ProgramCache::Instance().Store(ProgramKey{ context_(), device_(), precision_, routine_name_ },
                                    Program{ program_ });
     return;
   }
@@ -185,7 +185,7 @@ void Routine::InitProgram(std::initializer_list<const char *> source) {
   BinaryCache::Instance().Store(BinaryKey{ precision_, routine_name_, device_name_ },
                                 program_.GetIR());
 
-  ProgramCache::Instance().Store(ProgramKey{ context_(), precision_, routine_name_ },
+  ProgramCache::Instance().Store(ProgramKey{ context_(), device_(), precision_, routine_name_ },
                                  Program{ program_ });
 
   // Prints the elapsed compilation time in case of debugging in verbose mode
-- 
cgit v1.2.3


From 64ba590279f98899e738e223354547f4bf361b58 Mon Sep 17 00:00:00 2001
From: Kirill Mavreshko <kimavr@gmail.com>
Date: Sat, 27 May 2017 10:30:09 +0500
Subject: Fixed comment decribing the order of program cache fields

---
 src/cache.hpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/cache.hpp b/src/cache.hpp
index 207cb2dd..ed693ea3 100644
--- a/src/cache.hpp
+++ b/src/cache.hpp
@@ -79,7 +79,7 @@ extern template std::string BinaryCache::Get(const BinaryKeyRef &, bool *) const
 // =================================================================================================
 
 // The key struct for the cache of compiled OpenCL programs (context-dependent)
-// Order of fields: context, precision, routine_name (smaller fields first)
+// Order of fields: context, device_id, precision, routine_name (smaller fields first)
 typedef std::tuple<cl_context, cl_device_id, Precision, std::string> ProgramKey;
 typedef std::tuple<const cl_context &, const cl_device_id &, const Precision &, const std::string &> ProgramKeyRef;
 
-- 
cgit v1.2.3