AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Updated RDC to use AMD-SMI 24.6.0 structs

Browse Source 
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Change-Id: I9ef0f3cb786c1238e53cf21df5c6afafac829175
This commit is contained in:
Maisam Arif
2024-05-31 08:59:15 -05:00
committed by Galantsev, Dmitrii
parent 53033a5b77
commit 7c6bd4dc1c
5 changed files with 122 additions and 120 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rdc_libs/rdc/src/RdcMetricFetcherImpl.cc
+25 -24
View File
@@ -369,23 +369,23 @@ rdc_status_t RdcMetricFetcherImpl::fetch_smi_field(uint32_t gpu_index, rdc_field
}
const std::unordered_map<rdc_field_t, uint64_t> rdc_field_to_gpu_metrics = {
{RDC_FI_XGMI_0_READ_KB, gpu_metrics.xgmi_read_data_acc[0]},
{RDC_FI_XGMI_1_READ_KB, gpu_metrics.xgmi_read_data_acc[1]},
{RDC_FI_XGMI_2_READ_KB, gpu_metrics.xgmi_read_data_acc[2]},
{RDC_FI_XGMI_3_READ_KB, gpu_metrics.xgmi_read_data_acc[3]},
{RDC_FI_XGMI_4_READ_KB, gpu_metrics.xgmi_read_data_acc[4]},
{RDC_FI_XGMI_5_READ_KB, gpu_metrics.xgmi_read_data_acc[5]},
{RDC_FI_XGMI_6_READ_KB, gpu_metrics.xgmi_read_data_acc[6]},
{RDC_FI_XGMI_7_READ_KB, gpu_metrics.xgmi_read_data_acc[7]},
{RDC_FI_XGMI_0_WRITE_KB, gpu_metrics.xgmi_write_data_acc[0]},
{RDC_FI_XGMI_1_WRITE_KB, gpu_metrics.xgmi_write_data_acc[1]},
{RDC_FI_XGMI_2_WRITE_KB, gpu_metrics.xgmi_write_data_acc[2]},
{RDC_FI_XGMI_3_WRITE_KB, gpu_metrics.xgmi_write_data_acc[3]},
{RDC_FI_XGMI_4_WRITE_KB, gpu_metrics.xgmi_write_data_acc[4]},
{RDC_FI_XGMI_5_WRITE_KB, gpu_metrics.xgmi_write_data_acc[5]},
{RDC_FI_XGMI_6_WRITE_KB, gpu_metrics.xgmi_write_data_acc[6]},
{RDC_FI_XGMI_7_WRITE_KB, gpu_metrics.xgmi_write_data_acc[7]},
{RDC_FI_PCIE_BANDWIDTH, gpu_metrics.pcie_bandwidth_inst},
{RDC_FI_XGMI_0_READ_KB, gpu_metrics.xgmi_read_data_acc[0]},
{RDC_FI_XGMI_1_READ_KB, gpu_metrics.xgmi_read_data_acc[1]},
{RDC_FI_XGMI_2_READ_KB, gpu_metrics.xgmi_read_data_acc[2]},
{RDC_FI_XGMI_3_READ_KB, gpu_metrics.xgmi_read_data_acc[3]},
{RDC_FI_XGMI_4_READ_KB, gpu_metrics.xgmi_read_data_acc[4]},
{RDC_FI_XGMI_5_READ_KB, gpu_metrics.xgmi_read_data_acc[5]},
{RDC_FI_XGMI_6_READ_KB, gpu_metrics.xgmi_read_data_acc[6]},
{RDC_FI_XGMI_7_READ_KB, gpu_metrics.xgmi_read_data_acc[7]},
{RDC_FI_XGMI_0_WRITE_KB, gpu_metrics.xgmi_write_data_acc[0]},
{RDC_FI_XGMI_1_WRITE_KB, gpu_metrics.xgmi_write_data_acc[1]},
{RDC_FI_XGMI_2_WRITE_KB, gpu_metrics.xgmi_write_data_acc[2]},
{RDC_FI_XGMI_3_WRITE_KB, gpu_metrics.xgmi_write_data_acc[3]},
{RDC_FI_XGMI_4_WRITE_KB, gpu_metrics.xgmi_write_data_acc[4]},
{RDC_FI_XGMI_5_WRITE_KB, gpu_metrics.xgmi_write_data_acc[5]},
{RDC_FI_XGMI_6_WRITE_KB, gpu_metrics.xgmi_write_data_acc[6]},
{RDC_FI_XGMI_7_WRITE_KB, gpu_metrics.xgmi_write_data_acc[7]},
{RDC_FI_PCIE_BANDWIDTH, gpu_metrics.pcie_bandwidth_inst},
};
// In gpu_metrics,the max value means not supported
@@ -398,7 +398,7 @@ rdc_status_t RdcMetricFetcherImpl::fetch_smi_field(uint32_t gpu_index, rdc_field
return;
} else {
RDC_LOG(RDC_DEBUG, "The gpu metrics return max value which indicate not supported:"
<< gpu_metrics_value_ite->second);
<< gpu_metrics_value_ite->second);
}
}
value->status = AMDSMI_STATUS_NOT_SUPPORTED;
@@ -460,9 +460,9 @@ rdc_status_t RdcMetricFetcherImpl::fetch_smi_field(uint32_t gpu_index, rdc_field
}
case RDC_FI_GPU_CLOCK:
case RDC_FI_MEM_CLOCK: {
amdsmi_clk_type_t clk_type = CLK_TYPE_SYS;
amdsmi_clk_type_t clk_type = AMDSMI_CLK_TYPE_SYS;
if (field_id == RDC_FI_MEM_CLOCK) {
clk_type = CLK_TYPE_MEM;
clk_type = AMDSMI_CLK_TYPE_MEM;
}
amdsmi_frequencies_t f = {};
value->status = amdsmi_get_clk_freq(processor_handle, clk_type, &f);
@@ -493,16 +493,17 @@ rdc_status_t RdcMetricFetcherImpl::fetch_smi_field(uint32_t gpu_index, rdc_field
case RDC_FI_GPU_TEMP:
case RDC_FI_MEMORY_TEMP: {
int64_t i64 = 0;
amdsmi_temperature_type_t sensor_type = TEMPERATURE_TYPE_EDGE;
amdsmi_temperature_type_t sensor_type = AMDSMI_TEMPERATURE_TYPE_EDGE;
if (field_id == RDC_FI_MEMORY_TEMP) {
sensor_type = TEMPERATURE_TYPE_VRAM;
sensor_type = AMDSMI_TEMPERATURE_TYPE_VRAM;
}
value->status =
amdsmi_get_temp_metric(processor_handle, sensor_type, AMDSMI_TEMP_CURRENT, &i64);
// fallback to hotspot temperature as some card may not have edge temperature.
if (sensor_type == TEMPERATURE_TYPE_EDGE && value->status == AMDSMI_STATUS_NOT_SUPPORTED) {
sensor_type = TEMPERATURE_TYPE_JUNCTION;
if (sensor_type == AMDSMI_TEMPERATURE_TYPE_EDGE &&
value->status == AMDSMI_STATUS_NOT_SUPPORTED) {
sensor_type = AMDSMI_TEMPERATURE_TYPE_JUNCTION;
value->status =
amdsmi_get_temp_metric(processor_handle, sensor_type, AMDSMI_TEMP_CURRENT, &i64);
}
rdc_libs/rdc/src/RdcSmiDiagnosticImpl.cc
+5 -5
View File
@@ -153,11 +153,11 @@ rdc_status_t RdcSmiDiagnosticImpl::check_smi_process_info(uint32_t gpu_index[RDC
std::string RdcSmiDiagnosticImpl::get_temperature_string(amdsmi_temperature_type_t type) const {
switch (type) {
case TEMPERATURE_TYPE_EDGE:
case AMDSMI_TEMPERATURE_TYPE_EDGE:
return "Edge";
case TEMPERATURE_TYPE_JUNCTION:
case AMDSMI_TEMPERATURE_TYPE_JUNCTION:
return "Junction";
case TEMPERATURE_TYPE_VRAM:
case AMDSMI_TEMPERATURE_TYPE_VRAM:
return "Memory";
default:
return "Unknown";
@@ -244,8 +244,8 @@ rdc_status_t RdcSmiDiagnosticImpl::check_smi_param_info(uint32_t gpu_index[RDC_M
for (uint32_t i = 0; i < gpu_count; i++) {
// temperature
for (amdsmi_temperature_type_t sensor_type = TEMPERATURE_TYPE_FIRST;
sensor_type < TEMPERATURE_TYPE__MAX;) {
for (amdsmi_temperature_type_t sensor_type = AMDSMI_TEMPERATURE_TYPE_FIRST;
sensor_type < AMDSMI_TEMPERATURE_TYPE__MAX;) {
auto status = check_temperature_level(gpu_index[i], sensor_type, result->info,
result->gpu_results[i].gpu_result.msg);
// Set to higher error level
rdc_libs/rdc/src/SmiUtils.cc
+1 -1
View File
@@ -99,7 +99,7 @@ amdsmi_status_t get_processor_handle_from_id(uint32_t gpu_id,
for (auto& processor : processors) {
processor_type_t processor_type = {};
ret = amdsmi_get_processor_type(processor, &processor_type);
if (processor_type != AMD_GPU) {
if (processor_type != AMDSMI_PROCESSOR_TYPE_AMD_GPU) {
RDC_LOG(RDC_ERROR, "Expect AMD_GPU device type!");
return AMDSMI_STATUS_NOT_SUPPORTED;
}