Break srvs. into rsmi & admin srvs. Add VerifyConnection api.

Change-Id: I67567264c37e31f3409062a14e56eba4801cd944


[ROCm/rdc commit: dc6f6f3e9a]
Dieser Commit ist enthalten in:
Chris Freehill
2019-12-22 20:30:58 -06:00
Ursprung bc7f01e992
Commit ba14edbb4d
19 geänderte Dateien mit 316 neuen und 87 gelöschten Zeilen
+5 -6
Datei anzeigen
@@ -41,9 +41,6 @@ endif ()
# Required Defines first:
set(RSMI_INC_DIR ${ROCM_DIR}/include)
set(RSMI_LIB_DIR ${ROCM_DIR}/lib)
message("")
message("Build Configuration:")
message("-----------BuildType: " ${CMAKE_BUILD_TYPE})
@@ -53,7 +50,6 @@ message("--------Proj Src Dir: " ${PROJECT_SOURCE_DIR})
message("--------Proj Bld Dir: " ${PROJECT_BINARY_DIR})
message("--------Proj Lib Dir: " ${PROJECT_BINARY_DIR}/lib)
message("--------Proj Exe Dir: " ${PROJECT_BINARY_DIR}/bin)
message("--------RSMI Lib Dir: " ${RSMI_LIB_DIR})
message("--------RSMI Inc Dir: " ${RSMI_INC_DIR})
message("")
@@ -66,8 +62,6 @@ include(utils)
## Verbose output.
set(CMAKE_VERBOSE_MAKEFILE on)
include_directories("${CMAKE_CURRENT_SOURCE_DIR}/../../client/include"
"${PROTOB_OUT_DIR}" "${RSMI_INC_DIR}")
set(EXAMPLE_SRC_LIST "${SRC_DIR}/rdc_client_test.cc")
message("EXAMPLE_SRC_LIST=${EXAMPLE_SRC_LIST}")
@@ -77,6 +71,11 @@ set(CLIENT_LIB_INC_LIST "${INC_DIR}/rdc_client.h")
set(TEST_CLIENT_EXE "rdc_test_client")
add_executable(${TEST_CLIENT_EXE} "${EXAMPLE_SRC_LIST}")
target_include_directories(${TEST_CLIENT_EXE} PRIVATE
"${CMAKE_CURRENT_SOURCE_DIR}/../../client/include"
"${RSMI_INC_DIR}")
target_link_libraries(${TEST_CLIENT_EXE} rdc_client)
message("&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&")
+23 -11
Datei anzeigen
@@ -31,7 +31,10 @@ THE SOFTWARE.
#define CHK_RET_STATUS(RET) \
if ((RET) != RDC_STATUS_SUCCESS) { \
std::cout << "rdc call returned error: " << (RET) << std::endl; \
const char *err_msg_str; \
(void)rdc_status_string((RET), &err_msg_str); \
std::cout << "rdc call returned error: " << (RET) << ":\"" << \
err_msg_str << "\"" << std::endl; \
}
#define CHK_RET_STATUS_CONT(RET) \
@@ -41,11 +44,8 @@ THE SOFTWARE.
}
int main(int argc, char** argv) {
(void)argc; // ignore for now
(void)argv; // ignore for now
rdc_status_t ret;
rdc_channel_t server;
rdc_channel_t server_ch;
uint64_t num_gpu;
int64_t temperature;
std::string serv_host("localhost");
@@ -61,27 +61,39 @@ int main(int argc, char** argv) {
std::cout << "Attempting to create channel to " << serv_host << ":" <<
serv_port << std::endl;
ret = rdc_channel_create(&server, serv_host.c_str(), serv_port.c_str(),
ret = rdc_channel_create(&server_ch, serv_host.c_str(), serv_port.c_str(),
false);
CHK_RET_STATUS(ret)
std::cout << "Successfully created channel" << std::endl;
std::cout << "Getting number of gpus at server..." << std::endl;
ret = rdc_num_gpus_get(server, &num_gpu);
grpc_connectivity_state ch_state;
ret = rdc_channel_state_get(server_ch, true, &ch_state);
CHK_RET_STATUS(ret)
std::cout << "Number of GPUs at server is " << num_gpu << std::endl;
std::cout << "Current channel state is " << ch_state << std::endl;
std::cout << "Verifying connection to server..." << std::endl;
ret = rdc_channel_connection_verify(server_ch);
CHK_RET_STATUS(ret)
if (ret == RDC_STATUS_SUCCESS) {
std::cout << "Verified connection to server." << std::endl;
}
std::cout << "Getting number of gpus at server..." << std::endl;
ret = rdc_num_gpus_get(server_ch, &num_gpu);
CHK_RET_STATUS(ret)
std::cout << "Number of GPUs at server is " << server_ch <<
num_gpu << std::endl;
for (uint32_t dv_ind = 0; dv_ind < num_gpu; ++dv_ind) {
std::cout << "Info for Device " << dv_ind << ":" << std::endl;
std::cout << "\tGetting temperature..." << std::endl;
ret = rdc_dev_temp_metric_get(server, dv_ind, RSMI_TEMP_TYPE_JUNCTION,
ret = rdc_dev_temp_metric_get(server_ch, dv_ind, RSMI_TEMP_TYPE_JUNCTION,
RSMI_TEMP_CURRENT, &temperature);
CHK_RET_STATUS_CONT(ret)
std::cout << "\t GPU " << dv_ind << " has a temperature of " <<
temperature << std::endl;
}
ret = rdc_channel_destroy(server);
ret = rdc_channel_destroy(server_ch);
CHK_RET_STATUS(ret)
std::cout << "Successfully destroyed channel to " << serv_host << ":" <<
serv_port << std::endl;