add support for CL_DEVICE_CONCURRENT_ACCESS_ATOM_SIZE_KHR

Only a single value is supported for now.  If this gets expanded
to return an array of values, one per SVM type, then this this
implementation will need to be updated.

Author: bashbaug

include/CL/opencl.hpp

@@ -1577,6 +1577,7 @@ inline cl_int getInfoHelper(Func f, cl_uint name, T* param, int, typename T::cl_
     F(cl_platform_info, CL_PLATFORM_SVM_TYPE_CAPABILITIES_KHR, cl::vector<cl_svm_capabilities_khr>) \
     \
     F(cl_device_info, CL_DEVICE_SVM_TYPE_CAPABILITIES_KHR, cl::vector<cl_svm_capabilities_khr>) \
+    F(cl_device_info, CL_DEVICE_SVM_CONCURRENT_ACCESS_ATOM_SIZE_KHR, size_type) \
     \
     F(cl_svm_pointer_info_khr, CL_SVM_INFO_TYPE_INDEX_KHR, cl_uint) \
     F(cl_svm_pointer_info_khr, CL_SVM_INFO_CAPABILITIES_KHR, cl_svm_capabilities_khr) \

layers/99_svmplusplus/emulate.cpp

@@ -248,6 +248,11 @@ struct SLayerContext
         return TypeCapsDevice[device];
     }
 
+    const size_t getConcurrentAccessAtomSize(cl_device_id device)
+    {
+        return ConcurrentAccessAtomSize[device];
+    }
+
     const SUSMFuncs& getUSMFuncs(cl_platform_id platform)
     {
         return USMFuncs[platform];
@@ -429,6 +434,7 @@ struct SLayerContext
 
         for (auto device: devices) {
             std::vector<cl_svm_capabilities_khr> typeCapsDevice;
+            size_t concurrentAccessAtomSize = 0;
 
             for (auto caps: typeCapsPlatform) {
                 if ((caps & CL_SVM_TYPE_MACRO_DEVICE_KHR) == CL_SVM_TYPE_MACRO_DEVICE_KHR) {
@@ -452,9 +458,14 @@ struct SLayerContext
                 else {
                     assert(0 && "unknown platform SVM type");
                 }
+
+                if (!typeCapsDevice.empty() && (typeCapsDevice.back() & CL_SVM_CAPABILITY_CONCURRENT_ACCESS_KHR)) {
+                    concurrentAccessAtomSize = 1;
+                }
             }
 
             TypeCapsDevice[device] = typeCapsDevice;
+            ConcurrentAccessAtomSize[device] = concurrentAccessAtomSize;
         }
     }
 
@@ -497,6 +508,8 @@ struct SLayerContext
     std::map<cl_platform_id, std::vector<cl_svm_capabilities_khr>>  TypeCapsPlatform;
     std::map<cl_device_id, std::vector<cl_svm_capabilities_khr>>    TypeCapsDevice;
 
+    std::map<cl_device_id, size_t>  ConcurrentAccessAtomSize;
+
     std::map<cl_platform_id, SUSMFuncs> USMFuncs;
 
     typedef std::map<const void*, SAllocInfo> CAllocMap;
@@ -1106,6 +1119,17 @@ cl_int CL_API_CALL clGetDeviceInfo_override(
                 ptr);
         }
         break;
+    case CL_DEVICE_SVM_CONCURRENT_ACCESS_ATOM_SIZE_KHR:
+        {
+            const auto concurrentAccessAtomSize = getLayerContext().getConcurrentAccessAtomSize(device);
+            auto ptr = (size_t*)param_value;
+            return writeParamToMemory(
+                param_value_size,
+                concurrentAccessAtomSize,
+                param_value_size_ret,
+                ptr);
+        }
+        break;
     default: break;
     }
 

samples/usvm/00_usvmqueries/main.cpp

@@ -117,6 +117,8 @@ int main(
             printf("Device[%zu]: %s\n",
                 d,
                 devices[d].getInfo<CL_DEVICE_NAME>().c_str());
+            printf("USM Concurrent Access Atom Size: %zu\n",
+                devices[d].getInfo<CL_DEVICE_SVM_CONCURRENT_ACCESS_ATOM_SIZE_KHR>());
 
             std::vector<cl_svm_capabilities_khr> typeCapsDevice =
                 devices[d].getInfo<CL_DEVICE_SVM_TYPE_CAPABILITIES_KHR>();