P4 to Git Change 1071796 by xcui@merged_opencl_jxcwin on 2014/08/29 17:10:04

EPR #404341 - disable OPenCL 2.0 for corssfire and multipe devices temporaraly, this is the OPENCL  changes.
	code review:
	http://ocltc.amd.com/reviews/r/5639/

	precheckin testing results:
	http://ocltc.amd.com:8111/viewModification.html?modId=38513&personal=true&init=1&tab=vcsModificationBuilds

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#458 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.hpp#132 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#277 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/include/cal/cal.h#28 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp#98 edit

rocclr/runtime/device/gpu/gpudevice.cpp

@@ -623,13 +623,6 @@ void Device::fillDeviceInfo(
     info_.linkerAvailable_           = CL_TRUE;
 
     info_.executionCapabilities_     = CL_EXEC_KERNEL;
-    if (settings().oclVersion_ >= OpenCL20) {
-        info_.svmCapabilities_       = CL_DEVICE_SVM_COARSE_GRAIN_BUFFER
-                                     | CL_DEVICE_SVM_FINE_GRAIN_BUFFER;
-        if (settings().svmAtomics_) {
-            info_.svmCapabilities_ |= CL_DEVICE_SVM_ATOMICS;
-        }
-    }
     info_.preferredPlatformAtomicAlignment_ = 0;
     info_.preferredGlobalAtomicAlignment_ = 0;
     info_.preferredLocalAtomicAlignment_ = 0;
@@ -713,6 +706,11 @@ void Device::fillDeviceInfo(
     info_.printfBufferSize_ = PrintfDbg::WorkitemDebugSize * info().maxWorkGroupSize_;
 
     if (settings().oclVersion_ >= OpenCL20) {
+        info_.svmCapabilities_ = CL_DEVICE_SVM_COARSE_GRAIN_BUFFER
+            | CL_DEVICE_SVM_FINE_GRAIN_BUFFER;
+        if (settings().svmAtomics_) {
+            info_.svmCapabilities_ |= CL_DEVICE_SVM_ATOMICS;
+        }
         // OpenCL2.0 device info fields
         info_.maxWriteImageArgs_        = MaxReadWriteImage;    //!< For compatibility
         info_.maxReadWriteImageArgs_    = MaxReadWriteImage;
@@ -749,13 +747,14 @@ void Device::fillDeviceInfo(
 extern const char* SchedulerSourceCode;
 
 bool
-Device::create(CALuint ordinal)
+Device::create(CALuint ordinal, CALuint numOfDevices)
 {
     appProfile_.init();
 
     // Open GSL device
+    bool reportOCL12Device = numOfDevices < 2 ? (OPENCL_VERSION < 200) : true;
     if (!open(ordinal, appProfile_.enableHighPerformanceState(),
-        (appProfile_.reportAsOCL12Device() || (OPENCL_VERSION < 200)))) {
+        appProfile_.reportAsOCL12Device() || reportOCL12Device)) {
         return false;
     }
 
@@ -770,7 +769,7 @@ Device::create(CALuint ordinal)
 #if cl_amd_open_video
           , getVideoAttribs()
 #endif // cl_amd_open_video
-          , appProfile_.reportAsOCL12Device()
+          , appProfile_.reportAsOCL12Device() || reportOCL12Device
         )) {
         return false;
     }
@@ -1249,7 +1248,7 @@ Device::init()
     for (; ordinal < numDevices; ++ordinal) {
         // Create the GPU device object
         Device *d = new Device();
-        result = (NULL != d) && d->create(ordinal);
+        result = (NULL != d) && d->create(ordinal, numDevices);
         if (useDeviceList) {
             result &= (requestedDevices.find(ordinal) != requestedDevices.end());
         }

rocclr/runtime/device/gpu/gpudevice.hpp

@@ -363,7 +363,8 @@ public:
     //! Initialise a device (i.e. all parts of the constructor that could
     //! potentially fail)
     bool create(
-        CALuint ordinal     //!< GPU device ordinal index. Starts from 0
+        CALuint ordinal,     //!< GPU device ordinal index. Starts from 0
+        CALuint numOfDevices //!< number of GPU devices in the system
         );
 
     //! Destructor for the physical GPU device

rocclr/runtime/device/gpu/gpusettings.cpp

@@ -287,6 +287,9 @@ Settings::create(
 #if defined(_LP64)
             oclVersion_ = reportAsOCL12Device ? OpenCL12 : XCONCAT(OpenCL,XCONCAT(OPENCL_MAJOR,OPENCL_MINOR));
 #endif
+            if (calAttr.numOfVpu > 1) {
+                oclVersion_ = OpenCL12;
+            }
             if (GPU_FORCE_OCL20_32BIT) {
                 force32BitOcl20_ = true;
                 oclVersion_ = reportAsOCL12Device ? OpenCL12 : XCONCAT(OpenCL,XCONCAT(OPENCL_MAJOR,OPENCL_MINOR));

rocclr/runtime/device/gpu/gslbe/src/include/cal/cal.h

@@ -237,6 +237,7 @@ typedef struct CALdeviceattribsRec {
     CALuint64  vaStart;                             /**< VA start address */
     CALuint64  vaEnd;                               /**< VA end address */
     bool       isWorkstation;                       /**< Whether Device is a Workstation/Server part */
+    CALuint    numOfVpu;                            /**< number of vpu in the device*/
 } CALdeviceattribs;
 
 /** CAL device status */

rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp

@@ -154,6 +154,7 @@ CALGSLDevice::getAttribs_int(gsl::gsCtx* cs)
 
     m_attribs.vaStart = static_cast<CALuint64>(m_adp->pAsicInfo->vaStart);
     m_attribs.vaEnd   = static_cast<CALuint64>(m_adp->pAsicInfo->vaEnd);
+    m_attribs.numOfVpu = m_adp->pAsicInfo->numberOfVPU;
 }
 
 void