rpl_run w/o input file; queue create callback; test for n gpus and n threads

Change-Id: I37157c49cf6454de591cae97b5cc43287ea95956

test/app/standalone_test.cpp

@@ -145,8 +145,7 @@ int main() {
   TEST_STATUS(status == HSA_STATUS_SUCCESS);
 
   // Test initialization
-  TestHsa::SetQueue(prof_queue);
-  TestHsa::HsaInstantiate(0);
+  TestHsa::HsaInstantiate();
 
   // Dispatching profiled kernel n-times to collect all counter groups data
   const unsigned group_n = 0;
@@ -157,9 +156,9 @@ int main() {
   for (unsigned ind = 0; ind < 3; ++ind) {
 #if 1
     const unsigned queue_ind = ind % queue_count;
-    TestHsa::SetQueue(queue[queue_ind]);
-    // ret_val = RunKernel<DummyKernel, TestAql>();
-    ret_val = RunKernel<SimpleConvolution, TestAql>();
+    hsa_queue_t* prof_queue = queue[queue_ind];
+    //ret_val = RunKernel<DummyKernel, TestAql>(0, NULL, NULL, prof_queue);
+    ret_val = RunKernel<SimpleConvolution, TestAql>(0, NULL, NULL, prof_queue);
     std::cout << "run kernel, queue " << queue_ind << std::endl;
 #else
     sleep(3);

test/app/test.cpp

@@ -21,20 +21,66 @@ THE SOFTWARE.
 *******************************************************************************/
 
 #include <hsa.h>
+#include <stdio.h>
 #include <string.h>
 #include <iostream>
+#include <thread>
 
 #include "ctrl/run_kernel.h"
 #include "ctrl/test_aql.h"
+#include "dummy_kernel/dummy_kernel.h"
 #include "simple_convolution/simple_convolution.h"
 
+void thread_fun(const int kiter, const int diter, const uint32_t agents_number) {
+  const AgentInfo* agent_info[agents_number];
+  hsa_queue_t* queue[agents_number];
+  HsaRsrcFactory* rsrc = &HsaRsrcFactory::Instance();
+
+  for (uint32_t n = 0; n < agents_number; ++n) {
+    uint32_t agent_id = n % rsrc->GetCountOfGpuAgents();
+    if (rsrc->GetGpuAgentInfo(agent_id, &agent_info[n]) == false) {
+      fprintf(stderr, "AgentInfo failed\n");
+      abort();
+    }
+    if (rsrc->CreateQueue(agent_info[n], 128, &queue[n]) == false) {
+      fprintf(stderr, "CreateQueue failed\n");
+      abort();
+    }
+  }
+
+  for (int i = 0; i < kiter; ++i) {
+    for (uint32_t n = 0; n < agents_number; ++n) {
+      // RunKernel<DummyKernel, TestAql>(0, NULL, agent_info[n], queue[n], diter);
+      RunKernel<SimpleConvolution, TestAql>(0, NULL, agent_info[n], queue[n], diter);
+    }
+  }
+
+  for (uint32_t n = 0; n < agents_number; ++n) {
+    hsa_queue_destroy(queue[n]);
+  }
+}
+
 int main(int argc, char** argv) {
   const char* kiter_s = getenv("ROCP_KITER");
   const char* diter_s = getenv("ROCP_DITER");
+  const char* agents_s = getenv("ROCP_AGENTS");
+  const char* thrs_s = getenv("ROCP_THRS");
+
   const int kiter = (kiter_s != NULL) ? atol(kiter_s) : 1;
   const int diter = (diter_s != NULL) ? atol(diter_s) : 1;
+  const uint32_t agents_number = (agents_s != NULL) ? (uint32_t)atol(agents_s) : 1;
+  const int thrs = (thrs_s != NULL) ? atol(thrs_s) : 1;
+
   TestHsa::HsaInstantiate();
-  for (int i = 0; i < kiter; ++i) RunKernel<SimpleConvolution, TestAql>(argc, argv, diter);
+
+  std::thread t[thrs];
+  for (int n = 0; n < thrs; ++n) {
+    t[n] = std::thread(thread_fun, kiter, diter, agents_number);
+  }
+  for (int n = 0; n < thrs; ++n) {
+    t[n].join();
+  }
+
   TestHsa::HsaShutdown();
   return 0;
 }