Merge branch 'master' into renameGemmToDot

src/targets/gpu/include/migraph/gpu/rocblas.hpp

@@ -11,6 +11,7 @@ namespace gpu {
 using rocblas_handle_ptr = MIGRAPH_MANAGE_PTR(rocblas_handle, rocblas_destroy_handle);
 
 rocblas_handle_ptr create_rocblas_handle_ptr();
+rocblas_handle_ptr create_rocblas_handle_ptr(hipStream_t s);
 
 } // namespace gpu
 

src/targets/gpu/leaky_relu.cpp

@@ -20,7 +20,7 @@ argument miopen_leaky_relu::compute(context& ctx,
     float alpha = 1, beta = 0;
     auto x_desc = make_tensor(args[0].get_shape());
     auto y_desc = make_tensor(output_shape);
-    miopenActivationForward(ctx.handle.get(),
+    miopenActivationForward(ctx.get_stream().get_miopen(),
                             ad.get(),
                             &alpha,
                             x_desc.get(),

src/targets/gpu/pooling.cpp

@@ -21,7 +21,7 @@ argument miopen_pooling::compute(context& ctx,
 
     float alpha = 1, beta = 0;
 
-    miopenPoolingForward(ctx.handle.get(),
+    miopenPoolingForward(ctx.get_stream().get_miopen(),
                          pd.get(),
                          &alpha,
                          x_desc.get(),

src/targets/gpu/relu.cpp

@@ -20,7 +20,7 @@ argument miopen_relu::compute(context& ctx,
     float alpha = 1, beta = 0;
     auto x_desc = make_tensor(args[0].get_shape());
     auto y_desc = make_tensor(output_shape);
-    miopenActivationForward(ctx.handle.get(),
+    miopenActivationForward(ctx.get_stream().get_miopen(),
                             ad.get(),
                             &alpha,
                             x_desc.get(),

src/targets/gpu/rocblas.cpp

@@ -10,6 +10,13 @@ rocblas_handle_ptr create_rocblas_handle_ptr()
     return rocblas_handle_ptr{handle};
 }
 
+rocblas_handle_ptr create_rocblas_handle_ptr(hipStream_t s)
+{
+    rocblas_handle_ptr rb = create_rocblas_handle_ptr();
+    rocblas_set_stream(rb.get(), s);
+    return rb;
+}
+
 } // namespace gpu
 
 } // namespace migraph

src/targets/gpu/softmax.cpp

@@ -20,7 +20,7 @@ argument miopen_softmax::compute(context& ctx,
     float alpha = 1, beta = 0;
     auto x_desc = make_tensor(args[0].get_shape());
     auto y_desc = make_tensor(output_shape);
-    miopenSoftmaxForward(ctx.handle.get(),
+    miopenSoftmaxForward(ctx.get_stream().get_miopen(),
                          &alpha,
                          x_desc.get(),
                          args[0].implicit(),

src/targets/gpu/target.cpp

@@ -54,10 +54,6 @@ std::vector<pass> target::get_passes(migraph::context& gctx) const
 
 std::string target::name() const { return "miopen"; }
 
-migraph::context target::get_context() const
-{
-    return context{
-        share(make_obj<miopen_handle>(&miopenCreate)), share(create_rocblas_handle_ptr()), {}};
-}
+migraph::context target::get_context() const { return context{}; }
 } // namespace gpu
 } // namespace migraph

test/CMakeLists.txt

@@ -7,7 +7,7 @@ find_package(Threads REQUIRED)
 include(ProcessorCount)
 ProcessorCount(N)
 set(CTEST_PARALLEL_LEVEL ${N} CACHE STRING "CTest parallel level")
-add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -j ${CTEST_PARALLEL_LEVEL} -C ${CMAKE_CFG_INTDIR})
+add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -j ${CTEST_PARALLEL_LEVEL} -C ${CMAKE_CFG_INTDIR} --timeout 1500)
 add_custom_target(tests)
 
 find_program(MIGRAPH_GDB gdb)
@@ -103,7 +103,10 @@ if(MIGRAPH_ENABLE_GPU)
         get_filename_component(BASE_NAME ${TEST} NAME_WE)
         add_test_executable(test_gpu_${BASE_NAME} ${TEST})
         rocm_clang_tidy_check(test_gpu_${BASE_NAME})
-        set_tests_properties(test_gpu_${BASE_NAME} PROPERTIES COST 10)
+        set_tests_properties(test_gpu_${BASE_NAME} PROPERTIES 
+            COST 10 
+            RESOURCE_LOCK gpu
+        )
         target_link_libraries(test_gpu_${BASE_NAME} migraph_gpu)
     endforeach()
 endif()

test/gpu/miopen.cpp

@@ -129,6 +129,7 @@ template <class V>
 void verify_program()
 {
     auto_print::set_terminate_handler(migraph::get_type_name<V>());
+    // std::cout << migraph::get_type_name<V>() << std::endl;
     migraph::program cpu_prog;
     migraph::program gpu_prog;
     auto cpu_arg_f = detach_async([&] { return run_cpu<V>(cpu_prog); });