Merge branch 'privatestaging' of https://github.com/AMDComputeLibraries/HIP-privatestaging into privatestaging

include/hip_runtime_api.h

@@ -98,6 +98,7 @@ typedef struct hipDeviceProp_t {
     int pciBusID;                               ///< PCI Bus ID.
     int pciDeviceID;                            ///< PCI Device ID.
     size_t maxSharedMemoryPerMultiProcessor;    ///< Maximum Shared Memory Per Multiprocessor.
+    int isMultiGpuBoard;                        ///< 1 if device is on a multi-GPU board, 0 if not.
  } hipDeviceProp_t;
 
 
@@ -188,6 +189,7 @@ typedef enum hipDeviceAttribute_t {
     hipDeviceAttributePciBusId,                             ///< PCI Bus ID.
     hipDeviceAttributePciDeviceId,                          ///< PCI Device ID.
     hipDeviceAttributeMaxSharedMemoryPerMultiprocessor,     ///< Maximum Shared Memory Per Multiprocessor.
+    hipDeviceAttributeIsMultiGpuBoard,                      ///< Multiple GPU devices.
 } hipDeviceAttribute_t;
 
 /**

include/nvcc_detail/hip_runtime_api.h

@@ -266,6 +266,8 @@ inline static hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t att
         cdattr = cudaDevAttrPciDeviceId; break;
     case hipDeviceAttributeMaxSharedMemoryPerMultiprocessor:
         cdattr = cudaDevAttrMaxSharedMemoryPerMultiprocessor; break;
+    case hipDeviceAttributeIsMultiGpuBoard:
+        cdattr = cudaDevAttrIsMultiGpuBoard; break;
     default:
         cerror = cudaErrorInvalidValue; break;
     }

samples/1_Utils/hipInfo/hipInfo.cpp

@@ -80,6 +80,7 @@ void printDeviceProp (int deviceId)
     cout << setw(w1) << "pciDeviceID: " << props.pciDeviceID << endl;
     cout << setw(w1) << "multiProcessorCount: " << props.multiProcessorCount << endl;
     cout << setw(w1) << "maxThreadsPerMultiProcessor: " << props.maxThreadsPerMultiProcessor << endl;
+    cout << setw(w1) << "isMultiGpuBoard: " << props.isMultiGpuBoard << endl;
     cout << setw(w1) << "clockRate: " << (float)props.clockRate / 1000.0 << " Mhz" << endl;
 #ifdef USE_ROCR_20
     cout << setw(w1) << "memoryClockRate: " << (float)props.memoryClockRate / 1000.0 << " Mhz" << endl;

src/hip_hcc.cpp

@@ -654,6 +654,21 @@ hsa_status_t get_region_info(hsa_region_t region, void* data)
     return HSA_STATUS_SUCCESS;
 }
 
+// Determines if the given agent is of type HSA_DEVICE_TYPE_GPU and counts it.
+static hsa_status_t countGpuAgents(hsa_agent_t agent, void *data) {
+    if (data == NULL) {
+        return HSA_STATUS_ERROR_INVALID_ARGUMENT;
+    }
+    hsa_device_type_t device_type;
+    hsa_status_t status = hsa_agent_get_info(agent, HSA_AGENT_INFO_DEVICE, &device_type);
+    if (status != HSA_STATUS_SUCCESS) {
+        return status;
+    }
+    if (device_type == HSA_DEVICE_TYPE_GPU) {
+        (*static_cast<int*>(data))++;
+    }
+    return HSA_STATUS_SUCCESS;
+}
 
 // Internal version,
 hipError_t ihipDevice_t::getProperties(hipDeviceProp_t* prop)
@@ -672,6 +687,14 @@ hipError_t ihipDevice_t::getProperties(hipDeviceProp_t* prop)
         return hipErrorInvalidDevice;
     }
 
+    // Iterates over the agents to determine Multiple GPU devices
+    // using the countGpuAgents callback.
+    int gpuAgentsCount = 0;
+    err = hsa_iterate_agents(countGpuAgents, &gpuAgentsCount);
+    if (err == HSA_STATUS_INFO_BREAK) { err = HSA_STATUS_SUCCESS; }
+    DeviceErrorCheck(err);
+    prop->isMultiGpuBoard = 0 ? gpuAgentsCount < 2 : 1;
+
     // Get agent name
     err = hsa_agent_get_info(_hsa_agent, HSA_AGENT_INFO_NAME, &(prop->name));
     DeviceErrorCheck(err);
@@ -1270,6 +1293,8 @@ hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t attr, int device)
             *pi = prop->pciDeviceID; break;
         case hipDeviceAttributeMaxSharedMemoryPerMultiprocessor:
             *pi = prop->maxSharedMemoryPerMultiProcessor; break;
+        case hipDeviceAttributeIsMultiGpuBoard:
+            *pi = prop->isMultiGpuBoard; break;
         default:
             e = hipErrorInvalidValue; break;
         }

tests/src/hipGetDeviceAttribute.cpp

@@ -72,6 +72,7 @@ int main(int argc, char *argv[])
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeMemoryBusWidth, props.memoryBusWidth));
 #endif
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeMultiprocessorCount, props.multiProcessorCount));
+    CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeIsMultiGpuBoard, props.isMultiGpuBoard));
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeComputeMode, props.computeMode));
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeL2CacheSize, props.l2CacheSize));
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeMaxThreadsPerMultiProcessor, props.maxThreadsPerMultiProcessor));

util/vim/hip.vim

@@ -64,6 +64,7 @@ syn keyword hipFunctionName expf __expf exp logf __logf log
 " Runtime Data Types
 syn keyword hipType hipDeviceProp_t 
 syn keyword hipType hipError_t 
+syn keyword hipType hipStream_t
 
 " Runtime functions
 syn keyword hipFunctionName hipBindTexture hipBindTextureToArray 
@@ -154,6 +155,8 @@ syn keyword hipFunctionName hipUnbindTexture
 " HIP Flags
 syn keyword hipFlags hipFilterModePoint 
 syn keyword hipFlags hipMemcpyHostToDevice 
+syn keyword hipFlags hipMemcpyDeviceToDevice 
+syn keyword hipFlags hipMemcpyHostToHost
 syn keyword hipFlags hipMemcpyDeviceToHost 
 syn keyword hipFlags hipMemcpyHostToHost 
 syn keyword hipFlags hipMemcpyDeviceToDevice