Merge branch 'amd-develop' into amd-master

Change-Id: I77fa88b460be549bfcf9e18d3212e732ffc045f5


[ROCm/hip-tests commit: 008499801b]

projects/hip-tests/samples/1_Utils/hipCommander/TODO

@@ -1,50 +0,0 @@
-_ Add AQL kernel.
-_ Fix &*kernel command so the kernel name/type is an argument not a new command.
-
-_ Add command to parse only.
-_ Add regression to parse all the hcm files.
-
-_ Partition HCC, HIP, HSA, OpenCL commands into separate files.
-
-
-_ Show time for back-to-back copies.
-_ Add variables.
-  %loopcnt
-
-    ./hipCommander %loopcnt=4  
-
-_ Add datasize command.
-
-
-_ Add ( ) to parsing.
-_ Add argument parsing and checking.
-
-_ Add verbose option to print each step of setup.
-  - print deliniater between setup and run.  Add run start message.
-
-  - print sizes of all buffers.
-  - print each command before running.
-  - show start/stop of timer routine.
-
-_ 
-_ Clear documentation on what each oepration does.
-_ Add time instrumentation for each command.
-_ Add pcie atomic.
-
-
-_ Add tests for negative cases, ie endloop w/o opening loop.
-
-
-README tips
----
-- HIP_API_TRACE combined with -v is useful to track the exact commands generates by hipCommander.
-
-
-Other ideas:
----
-[ ] Perf guide : stream creation very slow on HCC and should be avoided.
-
-
-Scratch:
-
-

projects/hip-tests/samples/1_Utils/hipInfo/hipInfo.cpp

@@ -63,6 +63,14 @@ double bytesToGB(size_t s)
     return (double)s / (1024.0*1024.0*1024.0);
 }
 
+#define printLimit(w1, limit, units) \
+{\
+    size_t val;\
+    cudaDeviceGetLimit(&val, limit);\
+    std::cout << setw(w1) << #limit": " << val << " " << units << std::endl;\
+}
+
+
 void printDeviceProp (int deviceId)
 {
     using namespace std;
@@ -144,6 +152,17 @@ void printDeviceProp (int deviceId)
     cout << endl;
 
 
+#ifdef __HIP_PLATFORM_NVCC__
+    // Limits:
+    cout << endl;
+    printLimit(w1, cudaLimitStackSize, "bytes/thread");
+    printLimit(w1, cudaLimitPrintfFifoSize, "bytes/device");
+    printLimit(w1, cudaLimitMallocHeapSize, "bytes/device");
+    printLimit(w1, cudaLimitDevRuntimeSyncDepth, "grids");
+    printLimit(w1, cudaLimitDevRuntimePendingLaunchCount, "launches");
+#endif
+
+
 
 
     cout << endl;