Skip to content

Commit

Permalink
Fix CUDA builds for CUDA 11.2 - 12.6
Browse files Browse the repository at this point in the history
  • Loading branch information
@TomTheBear
TomTheBear committed Dec 10, 2024
1 parent 7571d11 commit 1e3a7d5
Show file tree
Hide file tree
Showing 2 changed files with 62 additions and 2 deletions.
31 changes: 31 additions & 0 deletions src/includes/nvmon_perfworks.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -230,6 +230,37 @@ typedef struct {

#endif

#if defined(CUDART_VERSION) && CUDART_VERSION < 11060
typedef struct NVPW_MetricsEvaluator_EvaluateToGpuValues_Params
{
/// [in]
size_t structSize;
/// [in] assign to NULL
void* pPriv;
/// [in]
struct NVPW_MetricsEvaluator* pMetricsEvaluator;
/// [in]
const struct NVPW_MetricEvalRequest* pMetricEvalRequests;
/// [in]
size_t numMetricEvalRequests;
/// [in] set to 'NVPW_MetricEvalRequest_STRUCT_SIZE'
size_t metricEvalRequestStructSize;
/// [in] set to sizeof('NVPW_MetricEvalRequest')
size_t metricEvalRequestStrideSize;
/// [in]
const uint8_t* pCounterDataImage;
/// [in]
size_t counterDataImageSize;
/// [in]
size_t rangeIndex;
/// [in]
NVPA_Bool isolated;
/// [inout] 'pMetricValues' is in, '*pMetricValues' is out
double* pMetricValues;
} NVPW_MetricsEvaluator_EvaluateToGpuValues_Params;
#define NVPW_MetricsEvaluator_EvaluateToGpuValues_Params_STRUCT_SIZE NVPA_STRUCT_SIZE(NVPW_MetricsEvaluator_EvaluateToGpuValues_Params, pMetricValues)
#endif

#if defined(CUDART_VERSION) && CUDART_VERSION >= 12060
typedef struct NVPA_MetricsContext NVPA_MetricsContext;
typedef struct NVPW_CUDA_MetricsContext_Create_Params
Expand Down
33 changes: 31 additions & 2 deletions src/sysFeatures_nvml.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,8 @@
#include <math.h>
#include <dlfcn.h>
#include <nvml.h>
#include <cuda.h>
#include <cuda_runtime_api.h>
#include <limits.h>
#include <stdbool.h>
#include <stdlib.h>
Expand Down Expand Up @@ -72,7 +74,6 @@
} while (0)

DECLAREFUNC_NVML(nvmlDeviceGetBAR1MemoryInfo, nvmlDevice_t device, nvmlBAR1Memory_t* bar1Memory);
DECLAREFUNC_NVML(nvmlDeviceGetBusType, nvmlDevice_t device, nvmlBusType_t *type);
DECLAREFUNC_NVML(nvmlDeviceGetClock, nvmlDevice_t device, nvmlClockType_t clockType, nvmlClockId_t clockId, unsigned int* clockMHz);
DECLAREFUNC_NVML(nvmlDeviceGetClockInfo, nvmlDevice_t device, nvmlClockType_t type, unsigned int* clock);
DECLAREFUNC_NVML(nvmlDeviceGetCurrPcieLinkGeneration, nvmlDevice_t device, unsigned int *currLinkGen);
Expand Down Expand Up @@ -123,6 +124,14 @@ static const char *(*nvmlErrorString_ptr)(nvmlReturn_t result);
DECLAREFUNC_NVML(nvmlInit_v2, void);
DECLAREFUNC_NVML(nvmlShutdown, void);

#if defined(CUDART_VERSION) && CUDART_VERSION < 11060
typedef unsigned int nvmlBusType_t;
#define NVML_BUS_TYPE_UNKNOWN 0
#else
DECLAREFUNC_NVML(nvmlDeviceGetBusType, nvmlDevice_t device, nvmlBusType_t *type);
#endif


static const _SysFeatureList nvidia_gpu_feature_list;
static bool nvml_initialized = false;
static void *dl_nvml = NULL;
Expand Down Expand Up @@ -152,7 +161,9 @@ int likwid_sysft_init_nvml(_SysFeatureList *list)
}

DLSYM_AND_CHECK(dl_nvml, nvmlDeviceGetBAR1MemoryInfo);
#if defined(CUDART_VERSION) && CUDART_VERSION >= 11060
DLSYM_AND_CHECK(dl_nvml, nvmlDeviceGetBusType);
#endif
DLSYM_AND_CHECK(dl_nvml, nvmlDeviceGetClock);
DLSYM_AND_CHECK(dl_nvml, nvmlDeviceGetClockInfo);
DLSYM_AND_CHECK(dl_nvml, nvmlDeviceGetCurrPcieLinkGeneration);
Expand Down Expand Up @@ -394,8 +405,10 @@ static int nvidia_gpu_bar1_total_getter(const LikwidDevice_t device, char **valu
static int nvidia_gpu_bus_type_getter(const LikwidDevice_t device, char **value)
{
LWD_TO_NVMLD(device, nvmlDevice);
nvmlBusType_t type;
nvmlBusType_t type = NVML_BUS_TYPE_UNKNOWN;
#if defined(CUDART_VERSION) && CUDART_VERSION >= 11060
NVML_CALL(nvmlDeviceGetBusType, nvmlDevice, &type);
#endif
static const char *types[] = {
"Unknown", "PCI", "PCIe", "FPCI", "AGP",
};
Expand Down Expand Up @@ -757,10 +770,12 @@ static int nvidia_gpu_encoder_capacity_hevc_getter(const LikwidDevice_t device,
return nvidia_gpu_encoder_capacity_getter(device, NVML_ENCODER_QUERY_HEVC, value);
}

#if defined(CUDART_VERSION) && CUDART_VERSION >= 12030
static int nvidia_gpu_encoder_capacity_av1_getter(const LikwidDevice_t device, char **value)
{
return nvidia_gpu_encoder_capacity_getter(device, NVML_ENCODER_QUERY_AV1, value);
}
#endif

static int nvidia_gpu_encoder_usage_getter(const LikwidDevice_t device, char **value)
{
Expand Down Expand Up @@ -869,6 +884,7 @@ static int nvidia_gpu_temp_thresh_gpu_max_getter(const LikwidDevice_t device, ch
return nvidia_gpu_temp_thresh_getter(device, NVML_TEMPERATURE_THRESHOLD_GPU_MAX, value);
}

#if defined(CUDART_VERSION) && CUDART_VERSION >= 11020
static int nvidia_gpu_temp_thresh_acou_min_getter(const LikwidDevice_t device, char **value)
{
return nvidia_gpu_temp_thresh_getter(device, NVML_TEMPERATURE_THRESHOLD_ACOUSTIC_MIN, value);
Expand All @@ -883,11 +899,14 @@ static int nvidia_gpu_temp_thresh_acou_max_getter(const LikwidDevice_t device, c
{
return nvidia_gpu_temp_thresh_getter(device, NVML_TEMPERATURE_THRESHOLD_ACOUSTIC_MAX, value);
}
#endif

#if defined(CUDART_VERSION) && CUDART_VERSION >= 12060
static int nvidia_gpu_temp_thresh_gps_cur_getter(const LikwidDevice_t device, char **value)
{
return nvidia_gpu_temp_thresh_getter(device, NVML_TEMPERATURE_THRESHOLD_GPS_CURR, value);
}
#endif

static int nvidia_gpu_temp_thresh_setter(const LikwidDevice_t device, nvmlTemperatureThresholds_t t, const char *value)
{
Expand Down Expand Up @@ -923,6 +942,7 @@ static int nvidia_gpu_temp_thresh_gpu_max_setter(const LikwidDevice_t device, co
return nvidia_gpu_temp_thresh_setter(device, NVML_TEMPERATURE_THRESHOLD_GPU_MAX, value);
}

#if defined(CUDART_VERSION) && CUDART_VERSION >= 11020
static int nvidia_gpu_temp_thresh_acou_min_setter(const LikwidDevice_t device, const char *value)
{
return nvidia_gpu_temp_thresh_setter(device, NVML_TEMPERATURE_THRESHOLD_ACOUSTIC_MIN, value);
Expand All @@ -937,11 +957,14 @@ static int nvidia_gpu_temp_thresh_acou_max_setter(const LikwidDevice_t device, c
{
return nvidia_gpu_temp_thresh_setter(device, NVML_TEMPERATURE_THRESHOLD_ACOUSTIC_MAX, value);
}
#endif

#if defined(CUDART_VERSION) && CUDART_VERSION >= 12060
static int nvidia_gpu_temp_thresh_gps_cur_setter(const LikwidDevice_t device, const char *value)
{
return nvidia_gpu_temp_thresh_setter(device, NVML_TEMPERATURE_THRESHOLD_GPS_CURR, value);
}
#endif

static int nvidia_gpu_ofa_usage_getter(const LikwidDevice_t device, char **value)
{
Expand Down Expand Up @@ -1090,7 +1113,9 @@ static _SysFeature nvidia_gpu_features[] = {
{"power_usage", "nvml", "Current power usage", nvidia_gpu_power_cur_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "W"},
{"encoder_cap_h264", "nvml", "Video encoder H264 capacity", nvidia_gpu_encoder_capacity_h264_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "%"},
{"encoder_cap_hevc", "nvml", "Video encoder HEVC capacity", nvidia_gpu_encoder_capacity_hevc_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "%"},
#if defined(CUDART_VERSION) && CUDART_VERSION >= 12030
{"encoder_cap_av1", "nvml", "Video encoder AV1 capacity", nvidia_gpu_encoder_capacity_av1_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "%"},
#endif
{"encoder_usage", "nvml", "Video encoder usage", nvidia_gpu_encoder_usage_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "%"},
{"gom_cur", "nvml", "Current GPU operation mode (all_on, compute_only, graphics_only)", nvidia_gpu_gom_cur_getter, nvidia_gpu_gom_setter, DEVICE_TYPE_NVIDIA_GPU, NULL},
{"gom_pend", "nvml", "Pending (after reboot) GPU operation mode (all_on, compute_only, graphics_only)", nvidia_gpu_gom_pend_getter, nvidia_gpu_gom_setter, DEVICE_TYPE_NVIDIA_GPU, NULL},
Expand All @@ -1102,10 +1127,14 @@ static _SysFeature nvidia_gpu_features[] = {
{"temp_thresh_slow", "nvml", "Slowdown Temperature threshold", nvidia_gpu_temp_thresh_slow_getter, nvidia_gpu_temp_thresh_slow_setter, DEVICE_TYPE_NVIDIA_GPU, NULL, "degrees C"},
{"temp_thresh_mem_max", "nvml", "Maximum memory temperature threshold", nvidia_gpu_temp_thresh_mem_max_getter, nvidia_gpu_temp_thresh_mem_max_setter, DEVICE_TYPE_NVIDIA_GPU, NULL, "degrees C"},
{"temp_thresh_gpu_max", "nvml", "Maximum GPU temperature threshold", nvidia_gpu_temp_thresh_gpu_max_getter, nvidia_gpu_temp_thresh_gpu_max_setter, DEVICE_TYPE_NVIDIA_GPU, NULL, "degrees C"},
#if defined(CUDART_VERSION) && CUDART_VERSION >= 11020
{"temp_thresh_acou_min", "nvml", "Minimum acoustic temperature threshold", nvidia_gpu_temp_thresh_acou_min_getter, nvidia_gpu_temp_thresh_acou_min_setter, DEVICE_TYPE_NVIDIA_GPU, NULL, "degrees C"},
{"temp_thresh_acou_cur", "nvml", "Current acoustic temperature threshold", nvidia_gpu_temp_thresh_acou_cur_getter, nvidia_gpu_temp_thresh_acou_cur_setter, DEVICE_TYPE_NVIDIA_GPU, NULL, "degrees C"},
{"temp_thresh_acou_max", "nvml", "Maximum acoustic temperature threshold", nvidia_gpu_temp_thresh_acou_max_getter, nvidia_gpu_temp_thresh_acou_max_setter, DEVICE_TYPE_NVIDIA_GPU, NULL, "degrees C"},
#endif
#if defined(CUDART_VERSION) && CUDART_VERSION >= 12060
{"temp_thresh_gps_cur", "nvml", "Current GPS temperature threshold", nvidia_gpu_temp_thresh_gps_cur_getter, nvidia_gpu_temp_thresh_gps_cur_setter, DEVICE_TYPE_NVIDIA_GPU, NULL, "degrees C"},
#endif
{"ofa_usage", "nvml", "Optical Flow Accelerator usage", nvidia_gpu_ofa_usage_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "%"},
{"energy", "nvml", "Energy consumed since driver reload", nvidia_gpu_energy_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "J"},
{"gpu_usage", "nvml", "GPU usage", nvidia_gpu_gpu_usage_getter, NULL, DEVICE_TYPE_NVIDIA_GPU, NULL, "%"},
Expand Down

0 comments on commit 1e3a7d5

Please sign in to comment.