AI/rocm-systems
2
0
Derivar 0
Código Questões Pedidos de integração Operações Pacotes Planeamentos Lançamentos Wiki Trabalho

amdsmi library and sample code updated for amdsmi_get_metrics_table

Ver a proveniência 
Change-Id: Ie03c556f5c38fe4a0365743d3a94220e3aa62b23
Este cometimento está contido em:
Deepak Mewar
2023-12-27 10:13:35 -05:00
cometido por Galantsev, Dmitrii
ascendente 4575990ae7
cometimento 9f3a6dbd29
3 ficheiros modificados com 168 adições e 46 eliminações
Descarregar ficheiro patch Descarregar ficheiro diff Expandir todos os ficheiros Colapsar todos os ficheiros
example/amdsmi_esmi_intg_example.cc
+82 -42
Ver ficheiro
@@ -86,51 +86,57 @@ int main(int argc, char **argv) {
// Get all sockets
uint32_t socket_count = 0;
/*ret = amdsmi_get_cpusocket_handles(&socket_count, nullptr);
CHK_AMDSMI_RET(ret)
// Allocate the memory for the sockets
vector<amdsmi_cpusocket_handle> sockets(socket_count);
// Get the sockets of the system
ret = amdsmi_get_cpusocket_handles(&socket_count, &sockets[0]);
CHK_AMDSMI_RET(ret)*/
ret = amdsmi_get_socket_handles(&socket_count, nullptr);
CHK_AMDSMI_RET(ret)
// Allocate the memory for the sockets
vector<amdsmi_socket_handle> sockets(socket_count);
// Get the sockets of the system
ret = amdsmi_get_socket_handles(&socket_count, &sockets[0]);
CHK_AMDSMI_RET(ret)
cout << "Total Socket: " << socket_count << endl;
// For each socket, get identifier and cores
for (uint8_t i = 0; i < socket_count; i++) {
// For each socket, get cpus and cores
for (uint32_t i = 0; i < socket_count; i++) {
cout << endl << "Socket " << i << endl;
uint32_t cpu_count = 0;
uint32_t core_count = 0;
// Set processor type as AMD_CPU
processor_type_t processor_type = AMD_CPU;
ret = amdsmi_get_processor_handles_by_type(sockets[i], processor_type, nullptr, &cpu_count);
CHK_AMDSMI_RET(ret)
// Allocate the memory for the cpus
vector<amdsmi_processor_handle> plist(cpu_count);
// Get the cpus for each socket
ret = amdsmi_get_processor_handles_by_type(sockets[i], processor_type, &plist[0], &cpu_count);
CHK_AMDSMI_RET(ret)
cout << endl;
// Read core count for each sockets
// Set processor type as AMD_CPU_CORE
processor_type = AMD_CPU_CORE;
ret = amdsmi_get_processor_handles_by_type(sockets[i], processor_type, nullptr, &core_count);
CHK_AMDSMI_RET(ret)
// Allocate the memory for the cpu cores
vector<amdsmi_processor_handle> core_list(core_count);
// Get the cpu cores for each socket
ret = amdsmi_get_processor_handles_by_type(sockets[i], processor_type, &core_list[0], &core_count);
CHK_AMDSMI_RET(ret)
for (int index = 0; index < plist.size(); index++) {
socket_count = plist.size();
ret = amdsmi_get_cpu_hsmp_proto_ver(plist[i], &proto_ver);
for (uint32_t index = 0; index < plist.size(); index++) {
ret = amdsmi_get_cpu_hsmp_proto_ver(plist[index], &proto_ver);
CHK_AMDSMI_RET(ret)
cout<<"\n------------------------------------------";
cout<<"\n| HSMP Proto Version | "<< proto_ver <<"\t\t |"<< endl;
cout<<"------------------------------------------\n";
ret = amdsmi_get_cpu_smu_fw_version(plist[i], &smu_fw);
ret = amdsmi_get_cpu_smu_fw_version(plist[index], &smu_fw);
CHK_AMDSMI_RET(ret)
cout<<"\n------------------------------------------";
@@ -144,14 +150,10 @@ int main(int argc, char **argv) {
uint32_t err_bits = 0;
uint32_t prochot;
cout<<"\n-------------------------------------------------";
cout<<"\n| Sensor Name\t\t\t |";
for (uint32_t i = 0; i < socket_count; i++) {
cout<<setprecision(3)<<" Socket "<<i<<"\t|";
}
cout<<setprecision(3)<<" CPU "<<index<<"\t|";
cout<<"\n-------------------------------------------------";
cout<<"\n| ProchotStatus:\t\t |";
ret = amdsmi_get_cpu_prochot_status(plist[i], &prochot);
ret = amdsmi_get_cpu_prochot_status(plist[index], &prochot);
CHK_AMDSMI_RET(ret)
if (!ret) {
cout<<setprecision(7)<< (prochot ? "active" : "inactive")<<"\t|";
@@ -164,11 +166,7 @@ int main(int argc, char **argv) {
size_t len;
char str[SHOWLINESZ] = {};
int retVal = 0;
cout<<"\n-------------------------------------------------";
cout<<"\n| Sensor Name\t\t\t |";
for (uint32_t i = 0; i < socket_count; i++) {
cout<<setprecision(3)<<" Socket "<<i<<"\t|";
}
cout<<setprecision(3)<<" CPU "<<index<<"\t|";
cout<<"\n-------------------------------------------------";
cout<<"\n| fclk (Mhz)\t\t\t |";
retVal = snprintf(str, SHOWLINESZ, "\n| mclk (Mhz)\t\t\t |");
@@ -176,7 +174,7 @@ int main(int argc, char **argv) {
len = strlen(str);
uint32_t fclk, mclk, cclk;
err_bits = 0;
ret = amdsmi_get_cpu_fclk_mclk(plist[i], &fclk, &mclk);
ret = amdsmi_get_cpu_fclk_mclk(plist[index], &fclk, &mclk);
CHK_AMDSMI_RET(ret)
if (!ret) {
cout<<setprecision(7)<<" "<<fclk<<"\t\t|";
@@ -194,15 +192,11 @@ int main(int argc, char **argv) {
cout<<"\n-------------------------------------------------\n";
uint32_t socket_power;
cout<<"\n-------------------------------------------------";
cout<<"\n| Sensor Name\t\t\t |";
for (uint32_t i = 0; i < socket_count; i++) {
cout<<setprecision(3)<<" Socket "<<i<<"\t|";
}
cout<<setprecision(3)<<" CPU "<<index<<"\t|";
cout<<"\n-------------------------------------------------";
cout<<"\n| Power (Watts)\t\t\t | ";
ret = amdsmi_get_cpu_socket_power(plist[i], &socket_power);
ret = amdsmi_get_cpu_socket_power(plist[index], &socket_power);
CHK_AMDSMI_RET(ret)
if (!ret) {
@@ -212,10 +206,10 @@ int main(int argc, char **argv) {
cout<<" NA (Err:" <<ret<<" |";
}
uint32_t power_limit;
uint32_t power_limit = 0;
cout<<"\n| PowerLimit (Watts)\t\t | ";
ret = amdsmi_get_cpu_socket_power_cap(plist[i], &power_limit);
ret = amdsmi_get_cpu_socket_power_cap(plist[index], &power_limit);
CHK_AMDSMI_RET(ret)
if (!ret) {
@@ -225,10 +219,10 @@ int main(int argc, char **argv) {
cout<<" NA (Err:" <<ret<<" |";
}
uint32_t power_max;
uint32_t power_max = 0;
cout<<"\n| PowerLimitMax (Watts)\t\t | ";
ret = amdsmi_get_cpu_socket_power_cap_max(plist[i], &power_max);
ret = amdsmi_get_cpu_socket_power_cap_max(plist[index], &power_max);
CHK_AMDSMI_RET(ret)
if (!ret) {
@@ -243,20 +237,66 @@ int main(int argc, char **argv) {
power_max = 0;
cout<<"\nEnter the max power to be set:\n";
cin>>input_power;
ret = amdsmi_get_cpu_socket_power_cap_max(plist[i], &power_max);
ret = amdsmi_get_cpu_socket_power_cap_max(plist[index], &power_max);
CHK_AMDSMI_RET(ret)
if ((ret == AMDSMI_STATUS_SUCCESS) && (input_power > power_max)) {
cout<<"Input power is more than max power limit,"
" limiting to "<<static_cast<double>(power_max)/1000<<"Watts\n";
input_power = power_max;
}
ret = amdsmi_set_cpu_socket_power_cap(plist[i], input_power);
ret = amdsmi_set_cpu_socket_power_cap(plist[index], input_power);
CHK_AMDSMI_RET(ret)
if (!ret) {
cout<<"Socket["<<i<<"] power_limit set to "
cout<<"CPU ["<<index<<"] power_limit set to "
<<fixed<<setprecision(3)<<static_cast<double>(input_power)/1000<<" Watts successfully\n";
}
power_limit = 0;
cout<<"\n| PowerLimit (Watts) \t\t | ";
ret = amdsmi_get_cpu_socket_power_cap(plist[index], &power_limit);
CHK_AMDSMI_RET(ret)
if (!ret) {
cout<<fixed<<setprecision(3)<<static_cast<double>(power_limit)/1000<<"\t|";
} else {
err_bits |= 1 << ret;
cout<<" NA (Err:" <<ret<<" |";
}
cout<<"\n-------------------------------------------------\n";
double fraction_q10 = 1/pow(2,10);
double fraction_uq10 = fraction_q10;
const char* err_str1;
amdsmi_hsmp_metric_table_t mtbl = {};
ret = amdsmi_get_metrics_table(plist[index], &mtbl);
CHK_AMDSMI_RET(ret)
if (ret != AMDSMI_STATUS_SUCCESS) {
cout<<"Failed to get Metrics Table for CPU["<<index<<"], Err["<<ret<<"]:"
<<*amdsmi_get_esmi_err_msg(ret, &err_str1)<<endl;
} else {
cout<<"\n| METRICS TABLE \t\t\t\t |\n";
cout<<"\n| ACCUMULATOR COUNTER | "<<mtbl.accumulation_counter<<"\t\t|";
cout<<"\n| SOCKET POWER LIMIT | "<<(mtbl.socket_power_limit * fraction_uq10)<<" W\t\t|";
cout<<"\n| MAX SOCKET POWER LIMIT | "<<(mtbl.max_socket_power_limit * fraction_uq10)<<" W\t\t|";
cout<<"\n| SOCKET POWER | "<<(mtbl.socket_power * fraction_uq10)<<" W\t\t|\n";
cout<<"\n| Effective frequency per AID: \t\t\t\t\t\t|";
cout<<"\n-------------------------------------------------------------------------";
cout<<"\n| AID | SOCCLK \t\t| VCLK \t\t| DCLK \t\t| LCLK \t\t|";
cout<<"\n-------------------------------------------------------------------------";
for(uint32_t j = 0; j < 4 ; j++){
cout<<fixed<<setprecision(3)<<"\n| ["<<j<<"] | "
<<(mtbl.socclk_frequency[j] * fraction_uq10)<<"MHz\t| "
<<(mtbl.vclk_frequency[j] * fraction_uq10)<<"MHz\t| "
<<(mtbl.dclk_frequency[j] * fraction_uq10)<<"MHz\t| "
<<(mtbl.lclk_frequency[j] * fraction_uq10)<<"MHz\t| ";
}
cout<<"\n-------------------------------------------------------------------------\n";
cout<<"\n-------------------------------------------------------------------------\n";
}
}
}
// Clean up resources allocated at amdsmi_init
include/amd_smi/amdsmi.h
+83 -1
Ver ficheiro
@@ -1503,6 +1503,88 @@ typedef struct {
uint8_t min_dpm_level; //!< Min LCLK DPM level[7:0](8 bit data)
} amdsmi_dpm_level_t;
/**
* @brief Metrics table (supported only with hsmp proto version 6).
*/
typedef struct {
uint32_t accumulation_counter;
/* TEMPERATURE */
uint32_t max_socket_temperature;
uint32_t max_vr_temperature;
uint32_t max_hbm_temperature;
uint64_t max_socket_temperature_acc;
uint64_t max_vr_temperature_acc;
uint64_t max_hbm_temperature_acc;
/* POWER */
uint32_t socket_power_limit;
uint32_t max_socket_power_limit;
uint32_t socket_power;
/* ENERGY */
uint64_t timestamp;
uint64_t socket_energy_acc;
uint64_t ccd_energy_acc;
uint64_t xcd_energy_acc;
uint64_t aid_energy_acc;
uint64_t hbm_energy_acc;
/* FREQUENCY */
uint32_t cclk_frequency_limit;
uint32_t gfxclk_frequency_limit;
uint32_t fclk_frequency;
uint32_t uclk_frequency;
uint32_t socclk_frequency[4];
uint32_t vclk_frequency[4];
uint32_t dclk_frequency[4];
uint32_t lclk_frequency[4];
uint64_t gfxclk_frequency_acc[8];
uint64_t cclk_frequency_acc[96];
/* FREQUENCY RANGE */
uint32_t max_cclk_frequency;
uint32_t min_cclk_frequency;
uint32_t max_gfxclk_frequency;
uint32_t min_gfxclk_frequency;
uint32_t fclk_frequency_table[4];
uint32_t uclk_frequency_table[4];
uint32_t socclk_frequency_table[4];
uint32_t vclk_frequency_table[4];
uint32_t dclk_frequency_table[4];
uint32_t lclk_frequency_table[4];
uint32_t max_lclk_dpm_range;
uint32_t min_lclk_dpm_range;
/* XGMI */
uint32_t xgmi_width;
uint32_t xgmi_bitrate;
uint64_t xgmi_read_bandwidth_acc[8];
uint64_t xgmi_write_bandwidth_acc[8];
/* ACTIVITY */
uint32_t socket_c0_residency;
uint32_t socket_gfx_busy;
uint32_t dram_bandwidth_utilization;
uint64_t socket_c0_residency_acc;
uint64_t socket_gfx_busy_acc;
uint64_t dram_bandwidth_acc;
uint32_t max_dram_bandwidth;
uint64_t dram_bandwidth_utilization_acc;
uint64_t pcie_bandwidth_acc[4];
/* THROTTLERS */
uint32_t prochot_residency_acc;
uint32_t ppt_residency_acc;
uint32_t socket_thm_residency_acc;
uint32_t vr_thm_residency_acc;
uint32_t hbm_thm_residency_acc;
uint32_t spare;
/* New items at the end to maintain driver compatibility */
uint32_t gfxclk_frequency[8];
} amdsmi_hsmp_metric_table_t;
#endif
/*****************************************************************************/
@@ -5892,7 +5974,7 @@ amdsmi_status_t amdsmi_get_metrics_table_version(amdsmi_processor_handle process
* @return ::amdsmi_status_t | ::AMDSMI_STATUS_SUCCESS on success, non-zero on fail
*/
amdsmi_status_t amdsmi_get_metrics_table(amdsmi_processor_handle processor_handle,
struct hsmp_metric_table *metrics_table);
amdsmi_hsmp_metric_table_t *metrics_table);
/** @} */
src/amd_smi/amd_smi.cc
+3 -3
Ver ficheiro
@@ -3642,7 +3642,7 @@ amdsmi_status_t amdsmi_get_metrics_table_version(amdsmi_processor_handle process
}
amdsmi_status_t amdsmi_get_metrics_table(amdsmi_processor_handle processor_handle,
struct hsmp_metric_table *metrics_table)
amdsmi_hsmp_metric_table_t *metrics_table)
{
amdsmi_status_t status;
struct hsmp_metric_table metrics_tbl;
@@ -3660,11 +3660,11 @@ amdsmi_status_t amdsmi_get_metrics_table(amdsmi_processor_handle processor_handl
sock_ind = (uint8_t)std::stoi(proc_id, NULL, 0);
status = static_cast<amdsmi_status_t>(esmi_metrics_table_get(sock_ind, &metrics_tbl));
*metrics_table = metrics_tbl;
if (status != AMDSMI_STATUS_SUCCESS)
return status;
std::memcpy(metrics_table, &metrics_tbl, sizeof(amdsmi_hsmp_metric_table_t));
return AMDSMI_STATUS_SUCCESS;
}