Compiles with -pedantic enabled. Added rvalue versions of setArg. Fur…

…ther progress on ParallelFor

Include/clUtil_kernel.h

@@ -70,7 +70,7 @@ namespace clUtil
 
       _Grid();
   };
-};
+}
 
 //Wrapper function to create a grid object of the correct type so you don't
 //have to fuck with template arguments.
@@ -81,11 +81,11 @@ template<typename... Args> clUtil::_Grid<Args...> clUtilGrid(Args... args)
 
 //This function is a hack needed because variadic partial specialization doesn't
 //seem to work quite right as of g++-4.5.2
-
-//void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Memory&& curArg);
 void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Memory* curArg);
 void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Image& curArg);
+void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Image&& curArg);
 void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Buffer& curArg);
+void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Buffer&& curArg);
 
 template<typename T> void setArg_(cl_kernel kernel, size_t argIndex, T curArg)
 {
@@ -124,7 +124,8 @@ void clUtilEnqueueKernel(const char* kernelName,
 {
   cl_int err;
   clUtil::Device& currentDevice = clUtil::Device::GetCurrentDevice();
-  cl_kernel kernel = currentDevice.getKernel(std::string(kernelName));
+  cl_kernel kernel = 
+    currentDevice.getKernel(std::move(std::string(kernelName)));
 
   clUtilSetArgs(kernel,
                 kernelName,

Makefile.inc

@@ -1,3 +1,3 @@
 CXX=g++
-USERFLAGS=-g3 -O0 -Wall -Werror
+USERFLAGS=-g3 -O0 -Wall -Werror -pedantic
 OpenCLInclude=$(AMDAPPSDKROOT)/include

clUtilKernel.cc

@@ -18,6 +18,15 @@ void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Image& curArg)
   clUtilCheckError(err);
 }
 
+void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Image&& curArg)
+{
+  cl_int err;
+  cl_mem memoryHandle = curArg.getMemHandle();
+
+  err = clSetKernelArg(kernel, argIndex, sizeof(memoryHandle), &memoryHandle);
+  clUtilCheckError(err);
+}
+
 void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Buffer& curArg)
 {
   cl_int err;
@@ -27,6 +36,15 @@ void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Buffer& curArg)
   clUtilCheckError(err);
 }
 
+void setArg_(cl_kernel kernel, size_t argIndex, clUtil::Buffer&& curArg)
+{
+  cl_int err;
+  cl_mem memoryHandle = curArg.getMemHandle();
+
+  err = clSetKernelArg(kernel, argIndex, sizeof(memoryHandle), &memoryHandle);
+  clUtilCheckError(err);
+}
+
 void clUtilSetArgs(cl_kernel kernel,
                    const char* kernelName,
                    clUtil::Grid& workGrid,

clUtilParallelFor.cc

@@ -68,6 +68,17 @@ ParallelForPerformanceModel::ParallelForPerformanceModel(size_t numSamples,
       iterationOffset = newTask.EndIndex + 1;
 
       mPendingSampleQueues[curDeviceGroup].push(newTask);
+
+      //Mark remaining work vector
+      if(curDeviceGroup == 0 && curSample > 0)  
+      {
+        mRemainingWork[curSample - 1].End = newTask.StartIndex - 1;
+      }
+
+      if(curDeviceGroup == numDeviceGroups - 1 && curSample < numSamples - 1)
+      {
+        mRemainingWork[curSample].Start = newTask.EndIndex + 1;
+      }
 
 #if 0
       cout << "DeviceGroup: " << curDeviceGroup
@@ -92,6 +103,25 @@ PendingTask ParallelForPerformanceModel::getWork(size_t deviceGroup)
   }
   else //If empty, use model to get work
   {
+    vector<Sample>& curDeviceModel = mModel[deviceGroup];
+    //size_t bestSample = 0;
+    //bool leftOfSample = false;
+
+    for(size_t curSample = 0; curSample < curDeviceModel.size(); curSample++)
+    {
+      //Check to the left of the sample
+      if(curSample > 0 && 
+         mRemainingWork[curSample - 1].Start <= 
+         mRemainingWork[curSample - 1].End)
+      {
+
+      }
+      if(curSample < curDeviceModel.size() - 1) //Check to the right of sample
+      {
+      }
+
+    }
+
     return work;
   }
 }
@@ -172,7 +202,7 @@ void ParallelForPerformanceModel::updateModel(size_t start,
   }
   else //Updating existing sample
   {
-    if(start > curSample.End) //Work appears to left of sample
+    if(start > curSample.End) //Work appears to right of sample
     {
       Sample& nextSample = deviceModel[sampleNumber + 1];
 
@@ -184,7 +214,7 @@ void ParallelForPerformanceModel::updateModel(size_t start,
 
       curSample.End = end;
     }
-    else if(end < curSample.Start) //Work appears to right of sample
+    else if(end < curSample.Start) //Work appears to left of sample
     {
       Sample& previousSample = deviceModel[sampleNumber - 1];
 

clUtilPlatform.cc

@@ -55,18 +55,18 @@ vector<Platform> Platform::GetPlatforms()
               &majorVersion, 
               &minorVersion) != 2)
     {
-      printf("Warning: platform %zu (%s) has malformed platform version string."
+      printf("Warning: platform %u (%s) has malformed platform version string."
              " As such, clUtil will not use it.\n", 
-             curPlatformID, 
+             (unsigned int)curPlatformID, 
              curPlatform.mPlatformInfo.Name.c_str());
       continue;
     }
 
     if(majorVersion < 1 || (majorVersion == 1 && minorVersion < 1))
     {
-      printf("Warning: platform %zu (%s) has version < 1.1."
+      printf("Warning: platform %u (%s) has version < 1.1."
              " As such, clUtil will not use it.\n", 
-             curPlatformID, 
+             (unsigned int)curPlatformID, 
              curPlatform.mPlatformInfo.Name.c_str());
       continue;
     }