BDFID (BusID/DeviceID/FunctionID) support.

Except FunctionID (or DomainID in CUDA) support, because cudaDeviceProp::pciDomainID is not reported by CUDA.


[ROCm/hip commit: 33f60c300d]

projects/hip/include/hip_runtime_api.h

@@ -91,6 +91,8 @@ typedef struct hipDeviceProp_t {
     int clockInstructionRate;           ///< Frequency in khz of the timer used by the device-side "clock*" instructions.  New for HIP.
     hipDeviceArch_t arch;               ///< Architectural feature flags.  New for HIP.
     int concurrentKernels;              ///< Device can possibly execute multiple kernels concurrently.
+    int pciBusID;                       ///< PCI Bus ID.
+    int pciDeviceID;                    ///< PCI Device ID.
  } hipDeviceProp_t;
 
 
@@ -146,6 +148,8 @@ typedef enum hipDeviceAttribute_t {
     hipDeviceAttributeMaxThreadsPerMultiProcessor,    ///< Maximum resident threads per multiprocessor.
     hipDeviceAttributeComputeCapabilityMajor,         ///< Major compute capability version number.
     hipDeviceAttributeComputeCapabilityMinor,         ///< Minor compute capability version number.
+    hipDeviceAttributePciBusId,                       ///< PCI Bus ID.
+    hipDeviceAttributePciDeviceId,                    ///< PCI Device ID.
 } hipDeviceAttribute_t;
 
 /**

projects/hip/include/nvcc_detail/hip_runtime_api.h

@@ -252,8 +252,10 @@ inline static hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t att
         cdattr = cudaDevAttrMaxThreadsPerMultiProcessor; break;
     case hipDeviceAttributeComputeCapabilityMajor:
         cdattr = cudaDevAttrComputeCapabilityMajor; break;
-    case hipDeviceAttributeComputeCapabilityMinor:
-        cdattr = cudaDevAttrComputeCapabilityMinor; break;
+    case hipDeviceAttributePciBusId:
+        cdattr = cudaDevAttrPciBusId; break;
+    case hipDeviceAttributePciDeviceId:
+        cdattr = cudaDevAttrPciDeviceId; break;
     default:
         cerror = cudaErrorInvalidValue; break;
     }

projects/hip/src/hip_hcc.cpp

@@ -310,7 +310,16 @@ hipError_t ihipDevice_t::getProperties(hipDeviceProp_t* prop)
     //prop->clockInstructionRate = counterHz / 1000;
     prop->clockInstructionRate = 100*1000; /* TODO-RT - hard-code until HSART has function to properly report clock */
 
+    // Get Agent BDFID (bus/device/function ID)
+    uint16_t bdf_id = 1;
+    err = hsa_agent_get_info(_hsa_agent, (hsa_agent_info_t)HSA_AMD_AGENT_INFO_BDFID, &bdf_id);
+    DeviceErrorCheck(err);
 
+    // BDFID is 16bit uint: [8bit - BusID | 5bit - Device ID | 3bit - Function/DomainID]
+    // TODO/Clarify: cudaDeviceProp::pciDomainID how to report?
+    // prop->pciDomainID =  bdf_id & 0x7;
+    prop->pciDeviceID =  (bdf_id>>3) & 0x1F;
+    prop->pciBusID =  (bdf_id>>8) & 0xFF;
 
     // Masquerade as a 3.0-level device. This will change as more HW functions are properly supported.
     // Application code should use the arch.has* to do detailed feature detection.
@@ -839,6 +848,10 @@ hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t attr, int device)
             *pi = prop->major; break;
         case hipDeviceAttributeComputeCapabilityMinor:
             *pi = prop->minor; break;
+        case hipDeviceAttributePciBusId:
+            *pi = prop->pciBusID; break;
+        case hipDeviceAttributePciDeviceId:
+            *pi = prop->pciDeviceID; break;
         default:
             e = hipErrorInvalidValue; break;
         }

projects/hip/tests/src/hipGetDeviceAttribute.cpp

@@ -73,7 +73,8 @@ int main(int argc, char *argv[])
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeMaxThreadsPerMultiProcessor, props.maxThreadsPerMultiProcessor));
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeComputeCapabilityMajor, props.major));
     CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributeComputeCapabilityMinor, props.minor));
-
+    CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributePciBusId, props.pciBusID));
+    CHECK(test_hipDeviceGetAttribute(deviceId, hipDeviceAttributePciDeviceId, props.pciDeviceID));
     passed();
 
 };