improve gtensor int tests, add readme

integration_tests/gtensor/CMakeLists.txt

@@ -7,6 +7,7 @@ if(GTENSOR_DEVICE STREQUAL "cuda")
   find_package(CUDAToolkit REQUIRED)
 endif()
 
+option(USE_THAPI_CTL "use thapi-ctl to start/stop tracing" ON)
 set(THAPI_PATH "/opt/thapi" CACHE STRING "Path to thapi installation")
 
 message(STATUS "${PROJECT_NAME}: THAPI_PATH=${THAPI_PATH}")
@@ -19,15 +20,19 @@ CPMFindPackage(NAME gtensor
                OPTIONS "GTENSOR_ENABLE_BLAS ON"
                        "GTENSOR_ENABLE_SOLVER ON")
 
-find_package(PkgConfig REQUIRED)
-pkg_check_modules(LTTNG_CTL REQUIRED lttng-ctl)
-
-add_library(thapi_ctl INTERFACE IMPORTED)
-target_include_directories(thapi_ctl INTERFACE "${THAPI_PATH}/include")
-target_link_libraries(thapi_ctl INTERFACE "${THAPI_PATH}/lib/libthapi-ctl.so")
-target_link_libraries(thapi_ctl INTERFACE ${LTTNG_CTL_LDFLAGS})
-
 add_executable(axpy_start_stop)
 target_gtensor_sources(axpy_start_stop PRIVATE axpy_start_stop.cxx)
 target_link_libraries(axpy_start_stop gtensor::gtensor)
-target_link_libraries(axpy_start_stop thapi_ctl)
+
+if (${USE_THAPI_CTL})
+  find_package(PkgConfig REQUIRED)
+  pkg_check_modules(LTTNG_CTL REQUIRED lttng-ctl)
+
+  add_library(thapi_ctl INTERFACE IMPORTED)
+  target_include_directories(thapi_ctl INTERFACE "${THAPI_PATH}/include")
+  target_link_libraries(thapi_ctl INTERFACE "${THAPI_PATH}/lib/libthapi-ctl.so")
+  target_link_libraries(thapi_ctl INTERFACE ${LTTNG_CTL_LDFLAGS})
+  target_link_libraries(axpy_start_stop thapi_ctl)
+else()
+  target_compile_definitions(axpy_start_stop PRIVATE NO_THAPI_CTL)
+endif()

integration_tests/gtensor/README.md

@@ -0,0 +1,13 @@
+# gtensor integration tests
+
+Allows testing cuda, hip, and ze backends with single source file.
+
+# Usage
+
+```
+GTENSOR_DEVICE=cuda bats ./integration_tests/gtensor.bats
+```
+Or "hip" or "sycl"
+
+Note that THAPI hip backend does not yet do device profiling, modifications will need to be
+made for tests to pass on hip.

integration_tests/gtensor/axpy_start_stop.cxx

@@ -3,7 +3,9 @@
 
 #include <gtensor/gtensor.h>
 
+#ifndef NO_THAPI_CTL
 #include "thapi-ctl.h"
+#endif
 
 // provides convenient shortcuts for common gtensor functions, for example
 // underscore ('_') to represent open slice ends.
@@ -73,20 +75,24 @@ int main(int argc, char** argv)
   auto expr = a * d_x + d_y;
 
   for (int i = 0; i < 10; i++) {
+#ifndef NO_THAPI_CTL
     if (i % 2 == 1) {
       thapi_ctl_start();
-      gt::synchronize();
     }
+#endif
 
     d_axpy = gt::eval(expr);
     h_axpy = gt::empty_like(h_x);
     gt::copy(d_axpy, h_axpy);
     gt::synchronize();
 
+    std::cout << "axpy(10) = " << h_axpy(10) << std::endl;
+
+#ifndef NO_THAPI_CTL
     if (i % 2 == 1) {
       thapi_ctl_stop();
-      gt::synchronize();
     }
+#endif
   }
 
   // Define a slice to print a subset of elements for spot checking the