P4 to Git Change 1057998 by gandryey@gera-dev-w7 on 2014/07/22 17:15:58

ECR #304775 - Device enqueuing
	- Use atomic fetch for enqueue flags
	- Switch to a multithreaded scheduler
	- Add a workaround for Linux host_multi_queue failures. Linux has only 2 queues, but the test allocates multiple host queues and the same HW ring can be used

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpublit.cpp#106 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#449 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.hpp#127 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuschedcl.cpp#22 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#325 edit

rocclr/runtime/device/gpu/gpuvirtual.cpp

@@ -453,8 +453,22 @@ VirtualGPU::create(
 #endif // !cl_amd_open_video
     {
         if (dev().engines().numComputeRings()) {
-            //!@note: Add 1 to account the device queue for transfers
-            uint idx = (index() + 1) % dev().engines().numComputeRings();
+            uint    idx;
+
+            //! @todo Temporary workaround for Linux, because 2 HW queues only
+            //! Fixes conformance failures with multi queues
+            if ((0 == deviceQueueSize) || IS_WINDOWS) {
+                idx = index() % (dev().engines().numComputeRings() -
+                    gpuDevice_.numDeviceQueues_);
+            }
+            else {
+                gpuDevice_.numDeviceQueues_++;
+                if (gpuDevice_.numDeviceQueues_ >= dev().engines().numComputeRings()) {
+                    return false;
+                }
+                idx = (dev().engines().numComputeRings() - gpuDevice_.numDeviceQueues_)
+                    % dev().engines().numComputeRings();
+            }
 
             // hwRing_ should be set 0 if forced to have single scratch buffer
             hwRing_ = (dev().settings().useSingleScratch_) ? 0 : idx;
@@ -583,6 +597,10 @@ VirtualGPU::~VirtualGPU()
     amd::ScopedLock k(dev().lockAsyncOps());
     amd::ScopedLock lock(dev().vgpusAccess());
 
+    if ((NULL != virtualQueue_) && IS_LINUX) {
+        gpuDevice_.numDeviceQueues_--;
+    }
+
     uint    i;
     // Destroy all kernels
     for (GslKernels::const_iterator it = gslKernels_.begin();