diff options
| author | Justin Martin <jaming@protonmail.com> | 2023-11-19 23:54:15 -0500 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-11-19 23:54:15 -0500 |
| commit | e4a6e751ec60f0f2363e0a3493a4f15c1ddc1d85 (patch) | |
| tree | 5ebc213ddec3f83464deaa3e88b9021e8374226f | |
| parent | 5df66006d8df0ff03d839b15ac9ba328d58d3252 (diff) | |
feature: Add gpu proc info (#1276)
* Add gpu util, power and procs.
Consolidated gpu calls into `data_harvester`.
Changed config flag from `enable_gpu_memory` to `enable_gpu`.
Added GPU utilization to the cpu widget.
Added GPU process memory usage and utilization percentage to the proc widget.
Added key binds for gpu process toggling.
Added GPU power usage to the battery widget.
Added bounds check to battery widget header.
Show battery widget header when `gpu_enable`.
Added feature flag `legacy-functions` to `nvml-wrapper`.
updated config file(s).
updated help text.
updated docs.
* Code Review:
Remove GPU util from cpu widget
Remove GPU power from battery widget
Use reference for gpu widgets_to_harvest
Extract match arm to function for feature gate
* Code Review: add gmem% toggle
* Do not poll gpu temp when filtered
* Code Review Two Changes:
adjust doc wordings
remove extra references
remove extra widget harvest checks
init proc gpu values
use convert_temp_unit for gpu temp
30 files changed, 565 insertions, 224 deletions
@@ -93,7 +93,7 @@ indexmap = "2.0.0" itertools = "0.11.0" kstring = { version = "2.0.0", features = ["arc"] } log = { version = "0.4.20", optional = true } -nvml-wrapper = { version = "0.9.0", optional = true } +nvml-wrapper = { version = "0.9.0", optional = true, features = ["legacy-functions"] } once_cell = "1.18.0" regex = "1.9.4" serde = { version = "=1.0.188 ", features = ["derive"] } diff --git a/docs/content/configuration/command-line-flags.md b/docs/content/configuration/command-line-flags.md index e0a7192f..17f78f8a 100644 --- a/docs/content/configuration/command-line-flags.md +++ b/docs/content/configuration/command-line-flags.md @@ -20,7 +20,7 @@ see information on these flags by running `btm -h`, or run `btm --help` to displ | --disable_click | Disables mouse clicks. | | -m, --dot_marker | Uses a dot marker for graphs. | | --enable_cache_memory | Enable collecting and displaying cache and buffer memory. | -| --enable_gpu_memory | Enable collecting and displaying GPU memory usage. | +| --enable_gpu | Enable collecting and displaying GPU usage. | | -e, --expanded | Expand the default widget upon starting the app. | | -f, --fahrenheit | Sets the temperature type to Fahrenheit. | | -g, --group | Groups processes with the same name by default. | diff --git a/docs/content/configuration/config-file/flags.md b/docs/content/configuration/config-file/flags.md index eeca23a3..ed9d0aef 100644 --- a/docs/content/configuration/config-file/flags.md +++ b/docs/content/configuration/config-file/flags.md @@ -38,7 +38,7 @@ each time: | `network_use_binary_prefix` | Boolean | Displays the network widget with binary prefixes. | | `network_use_bytes` | Boolean | Displays the network widget using bytes. | | `network_use_log` | Boolean | Displays the network widget with a log scale. | -| `enable_gpu_memory` | Boolean | Shows the GPU memory widget. | +| `enable_gpu` | Boolean | Shows the GPU widgets. | | `retention` | String (human readable time, such as "10m", "1h", etc.) | How much data is stored at once in terms of time. | | `unnormalized_cpu` | Boolean | Show process CPU% without normalizing over the number of cores. | | `expanded_on_startup` | Boolean | Expand the default widget upon starting the app. | diff --git a/docs/content/configuration/config-file/processes.md b/docs/content/configuration/config-file/processes.md index 2871e1cd..8a2c57b2 100644 --- a/docs/content/configuration/config-file/processes.md +++ b/docs/content/configuration/config-file/processes.md @@ -7,5 +7,5 @@ You can configure which columns are shown by the process widget by setting the ` ```toml [processes] # Pick which columns you want to use in any order. -columns = ["cpu%", "mem%", "pid", "name", "read", "write", "tread", "twrite", "state", "user", "time"] +columns = ["cpu%", "mem%", "pid", "name", "read", "write", "tread", "twrite", "state", "user", "time", "gmem%", "gpu%"] ``` diff --git a/docs/content/usage/widgets/memory.md b/docs/content/usage/widgets/memory.md index 5059d2bf..4304a3f7 100644 --- a/docs/content/usage/widgets/memory.md +++ b/docs/content/usage/widgets/memory.md @@ -13,7 +13,7 @@ If the total RAM or swap available is 0, then it is automatically hidden from th One can also adjust the displayed time range through either the keyboard or mouse, with a range of 30s to 600s. -This widget can also be configured to display Nvidia GPU memory usage (`--enable_gpu_memory`) or cache memory usage (`--enable_cache_memory`). +This widget can also be configured to display Nvidia GPU memory usage (`--enable_gpu` on Linux/Windows) or cache memory usage (`--enable_cache_memory`). ## Key bindings diff --git a/docs/content/usage/widgets/process.md b/docs/content/usage/widgets/process.md index 2a11e3a7..83139da1 100644 --- a/docs/content/usage/widgets/process.md +++ b/docs/content/usage/widgets/process.md @@ -32,6 +32,12 @@ It can also additionally display the following columns: - Process running time +With the feature flag (`--enable_gpu` on Linux/Windows) and gpu process columns enabled in the configuration: + +- GPU memory use percentage +- GPU core utilization percentage + + See [the processes configuration page](../../configuration/config-file/processes.md) on how to customize which columns are shown. @@ -147,6 +153,9 @@ Note all keywords are case-insensitive. To search for a process/command that col | `user` | `user=root` | Matches by user; supports regex | | `state` | `state=running` | Matches by state; supports regex | | `()` | `(<COND 1> AND <COND 2>) OR <COND 3>` | Group together a condition | +| `gmem` | `gmem > 1000 b` | Matches the gpu memory column in terms of bytes; supports comparison operators | +| `gmem%` | `gmem% < 0.5` | Matches the gpu memory column in terms of percent; supports comparison operators| +| `gpu%` | `gpu% > 0` | Matches the gpu usage column in terms of percent; supports comparison operators | #### Comparison operators @@ -207,6 +216,8 @@ Note that key bindings are generally case-sensitive. | ++I++ | Invert the current sort | | ++"%"++ | Toggle between values and percentages for memory usage | | ++t++ , ++f5++ | Toggle tree mode | +| ++M++ | Sort by gpu memory usage, press again to reverse sorting order | +| ++C++ | Sort by gpu usage, press again to reverse sorting order | ### Sort sub-widget diff --git a/docs/content/usage/widgets/temperature.md b/docs/content/usage/widgets/temperature.md index 34d28c98..3cadc9a8 100644 --- a/docs/content/usage/widgets/temperature.md +++ b/docs/content/usage/widgets/temperature.md @@ -10,6 +10,8 @@ The temperature widget provides a table of temperature sensors and their current The temperature widget provides the sensor name as well as its current temperature. +This widget can also be configured to display Nvidia GPU temperatures (`--enable_gpu` on Linux/Windows). + ## Key bindings Note that key bindings are generally case-sensitive. diff --git a/sample_configs/default_config.toml b/sample_configs/default_config.toml index e7b4058e..7d6a9f28 100644 --- a/sample_configs/default_config.toml +++ b/sample_configs/default_config.toml @@ -74,7 +74,7 @@ # Hides advanced options to stop a process on Unix-like systems. #disable_advanced_kill = false # Shows GPU(s) memory -#enable_gpu_memory = false +#enable_gpu = false # Shows cache and buffer memory #enable_cache_memory = false # How much data is stored at once in terms of time. @@ -83,7 +83,7 @@ # These are flags around the process widget. #[processes] -#columns = ["PID", "Name", "CPU%", "Mem%", "R/s", "W/s", "T.Read", "T.Write", "User", "State"] +#columns = ["PID", "Name", "CPU%", "Mem%", "R/s", "W/s", "T.Read", "T.Write", "User", "State", "GMEM%", "GPU%"] # These are all the components that support custom theming. Note that colour support # will depend on terminal support. @@ -103,7 +103,7 @@ #swap_color="LightYellow" # Represents the colour ARC will use in the memory legend and graph. #arc_color="LightCyan" -# Represents the colour the GPU will use in the memory legend and graph. +# Represents the colour the GPU will use in the legend and graph. #gpu_core_colors=["LightGreen", "LightBlue", "LightRed", "Cyan", "Green", "Blue", "Red"] # Represents the colour rx will use in the network legend and graph. #rx_color="LightCyan" @@ -60,7 +60,7 @@ pub struct AppConfigFields { pub use_old_network_legend: bool, pub table_gap: u16, pub disable_click: bool, - pub enable_gpu_memory: bool, + pub enable_gpu: bool, pub enable_cache_memory: bool, pub show_table_scroll_position: bool, pub is_advanced_kill: bool, @@ -1277,6 +1277,30 @@ impl App { disk.set_index(3); } } + #[cfg(feature = "gpu")] + 'M' => { + if let BottomWidgetType::Proc = self.current_widget.widget_type { + if let Some(proc_widget_state) = self + .states + .proc_state + .get_mut_widget_state(self.current_widget.widget_id) + { + proc_widget_state.select_column(ProcWidgetColumn::GpuMem); + } + } + } + #[cfg(feature = "gpu")] + 'C' => { + if let BottomWidgetType::Proc = self.current_widget.widget_type { + if let Some(proc_widget_state) = self + .states + .proc_state + .get_mut_widget_state(self.current_widget.widget_id) + { + proc_widget_state.select_column(ProcWidgetColumn::GpuUtil); + } + } + } '?' => { self.help_dialog_state.is_showing_help = true; self.is_force_redraw = true; @@ -2702,7 +2726,14 @@ impl App { { if (x >= *tlc_x && y >= *tlc_y) && (x <= *brc_x && y <= *brc_y) { - battery_widget_state.currently_selected_battery_index = itx; + if itx >= self.converted_data.battery_data.len() { + // range check to keep within current data + battery_widget_state.currently_selected_battery_index = + self.converted_data.battery_data.len() - 1; + } else { + battery_widget_state.currently_selected_battery_index = + itx; + } break; } } diff --git a/src/app/data_harvester.rs b/src/app/data_harvester.rs index e4571597..d140e29e 100644 --- a/src/app/data_harvester.rs +++ b/src/app/data_harvester.rs @@ -2,7 +2,7 @@ use std::time::{Duration, Instant}; -#[cfg(target_os = "linux")] +#[cfg(any(target_os = "linux", feature = "gpu"))] use hashbrown::HashMap; #[cfg(feature = "battery")] use starship_battery::{Battery, Manager}; @@ -125,6 +125,11 @@ pub struct DataCollector { #[cfg(target_family = "unix")] user_table: processes::UserTable, + + #[cfg(feature = "gpu")] + gpu_pids: Option<Vec<HashMap<u32, (u64, u32)>>>, + #[cfg(feature = "gpu")] + gpus_total_mem: Option<u64>, } impl DataCollector { @@ -153,6 +158,10 @@ impl DataCollector { filters, #[cfg(target_family = "unix")] user_table: Default::default(), + #[cfg(feature = "gpu")] + gpu_pids: None, + #[cfg(feature = "gpu")] + gpus_total_mem: None, } } @@ -288,18 +297,47 @@ impl DataCollector { self.update_cpu_usage(); self.update_memory_usage(); - self.update_processes(); self.update_temps(); - self.update_network_usage(); - self.update_disks(); - #[cfg(feature = "battery")] self.update_batteries(); + #[cfg(feature = "gpu")] + self.update_gpus(); // update_gpus before procs for gpu_pids but after temps for appending + self.update_processes(); + self.update_network_usage(); + self.update_disks(); // Update times for future reference. self.last_collection_time = self.data.collection_time; } + #[cfg(feature = "gpu")] + #[inline] + fn update_gpus(&mut self) { + if self.widgets_to_harvest.use_gpu { + #[cfg(feature = "nvidia")] + if let Some(data) = nvidia::get_nvidia_vecs( + &self.temperature_type, + &self.filters.temp_filter, + &self.widgets_to_harvest, + ) { + if let Some(mut temp) = data.temperature { + if let Some(sensors) = &mut self.data.temperature_sensors { + sensors.append(&mut temp); + } else { + self.data.temperature_sensors = Some(temp); + } + } + if let Some(mem) = data.memory { + self.data.gpu = Some(mem); + } + if let Some(proc) = data.procs { + self.gpu_pids = Some(proc.1); + self.gpus_total_mem = Some(proc.0); + } + } + } + } + #[inline] fn update_cpu_usage(&mut self) { if self.widgets_to_harvest.use_cpu { @@ -365,11 +403,6 @@ impl DataCollector { { self.data.arc = memory::arc::get_arc_usage(); } - - #[cfg(feature = "gpu")] - if self.widgets_to_harvest.use_gpu { - self.data.gpu = memory::gpu::get_gpu_mem_usage(); - } } } diff --git a/src/app/data_harvester/cpu.rs b/src/app/data_harvester/cpu.rs index a827b26c..843df161 100644 --- a/src/app/data_harvester/cpu.rs +++ b/src/app/data_harvester/cpu.rs @@ -1,8 +1,4 @@ //! Data collection for CPU usage and load average. -//! -//! For CPU usage, Linux, macOS, and Windows are handled by Heim, FreeBSD by sysinfo. -//! -//! For load average, macOS and Linux are supported through Heim, FreeBSD by sysinfo. pub mod sysinfo; pub use self::sysinfo::*; diff --git a/src/app/data_harvester/memory.rs b/src/app/data_harvester/memory.rs index 2154b00c..dee65d8d 100644 --- a/src/app/data_harvester/memory.rs +++ b/src/app/data_harvester/memory.rs @@ -15,9 +15,6 @@ cfg_if::cfg_if! { } } -#[cfg(feature = "gpu")] -pub mod gpu; - #[cfg(feature = "zfs")] pub mod arc; diff --git a/src/app/data_harvester/memory/gpu.rs b/src/app/data_harvester/memory/gpu.rs deleted file mode 100644 index 6fd66ba3..00000000 --- a/src/app/data_harvester/memory/gpu.rs +++ /dev/null @@ -1,47 +0,0 @@ -use super::MemHarvest; - -/// Return GPU memory usage. -#[cfg(feature = "gpu")] -pub(crate) fn get_gpu_mem_usage() -> Option<Vec<(String, MemHarvest)>> { - // As we add more support, expand on this. - - #[cfg(feature = "nvidia")] - get_nvidia_mem_usage() -} - -/// Returns the memory usage of NVIDIA cards. -#[inline] -#[cfg(feature = "nvidia")] -fn get_nvidia_mem_usage() -> Option<Vec<(String, MemHarvest)>> { - use crate::data_harvester::nvidia::NVML_DATA; - - if let Ok(nvml) = &*NVML_DATA { - if let Ok(num_gpu) = nvml.device_count() { - let mut results = Vec::with_capacity(num_gpu as usize); - for i in 0..num_gpu { - if let Ok(device) = nvml.device_by_index(i) { - if let (Ok(name), Ok(mem)) = (device.name(), device.memory_info()) { - // add device memory in bytes - results.push(( - name, - MemHarvest { - total_bytes: mem.total, - used_bytes: mem.used, - use_percent: if mem.total == 0 { - None - } else { - Some(mem.used as f64 / mem.total as f64 * 100.0) - }, - }, - )); - } - } - } - Some(results) - } else { - None - } - } else { - None - } -} diff --git a/src/app/data_harvester/nvidia.rs b/src/app/data_harvester/nvidia.rs index 9619da64..19abdb9e 100644 --- a/src/app/data_harvester/nvidia.rs +++ b/src/app/data_harvester/nvidia.rs @@ -1,4 +1,149 @@ +use hashbrown::HashMap; +use nvml_wrapper::enum_wrappers::device::TemperatureSensor; +use nvml_wrapper::enums::device::UsedGpuMemory; use nvml_wrapper::{error::NvmlError, Nvml}; use once_cell::sync::Lazy; +use crate::app::Filter; + +use crate::app::layout_manager::UsedWidgets; +use crate::data_harvester::memory::MemHarvest; +use crate::data_harvester::temperature::{ + convert_temp_unit, is_temp_filtered, TempHarvest, TemperatureType, +}; + pub static NVML_DATA: Lazy<Result<Nvml, NvmlError>> = Lazy::new(Nvml::init); + +pub struct GpusData { + pub memory: Option<Vec<(String, MemHarvest)>>, + pub temperature: Option<Vec<TempHarvest>>, + pub procs: Option<(u64, Vec<HashMap<u32, (u64, u32)>>)>, +} + +/// Returns the GPU data from NVIDIA cards. +#[inline] +pub fn get_nvidia_vecs( + temp_type: &TemperatureType, filter: &Option<Filter>, widgets_to_harvest: &UsedWidgets, +) -> Option<GpusData> { + if let Ok(nvml) = &*NVML_DATA { + if let Ok(num_gpu) = nvml.device_count() { + let mut temp_vec = Vec::with_capacity(num_gpu as usize); + let mut mem_vec = Vec::with_capacity(num_gpu as usize); + let mut proc_vec = Vec::with_capacity(num_gpu as usize); + let mut total_mem = 0; + for i in 0..num_gpu { + if let Ok(device) = nvml.device_by_index(i) { + if let Ok(name) = device.name() { + if widgets_to_harvest.use_mem { + if let Ok(mem) = device.memory_info() { + mem_vec.push(( + name.clone(), + MemHarvest { + total_bytes: mem.total, + used_bytes: mem.used, + use_percent: if mem.total == 0 { + None + } else { + Some(mem.used as f64 / mem.total as f64 * 100.0) + }, + }, + )); + } + } + if widgets_to_harvest.use_temp && is_temp_filtered(filter, &name) { + if let Ok(temperature) = device.temperature(TemperatureSensor::Gpu) { + let temperature = temperature as f32; + let temperature = convert_temp_unit(temperature, temp_type); + temp_vec.push(TempHarvest { + name: name.clone(), + temperature, + }); + } + } + } + if widgets_to_harvest.use_proc { + let mut procs = HashMap::new(); + if let Ok(gpu_procs) = device.process_utilization_stats(None) { + for proc in gpu_procs { + let pid = proc.pid; + let gpu_util = proc.sm_util + proc.enc_util + proc.dec_util; + procs.insert(pid, (0, gpu_util)); + } + } + if let Ok(compute_procs) = device.running_compute_processes() { + for proc in compute_procs { + let pid = proc.pid; + let gpu_mem = match proc.used_gpu_memory { + UsedGpuMemory::Used(val) => val, + UsedGpuMemory::Unavailable => 0, + }; + if let Some(prev) = procs.get(&pid) { + procs.insert(pid, (gpu_mem, prev.1)); + } else { + procs.insert(pid, (gpu_mem, 0)); + } + } + } + // Use the legacy API too but prefer newer API results + if let Ok(graphics_procs) = device.running_graphics_processes_v2() { + for proc in graphics_procs { + let pid = proc.pid; + let gpu_mem = match proc.used_gpu_memory { + UsedGpuMemory::Used(val) => val, + UsedGpuMemory::Unavailable => 0, + }; + if let Some(prev) = procs.get(&pid) { + procs.insert(pid, (gpu_mem, prev.1)); + } else { + procs.insert(pid, (gpu_mem, 0)); + } + } + } + if let Ok(graphics_procs) = device.running_graphics_processes() { + for proc in graphics_procs { + let pid = proc.pid; + let gpu_mem = match proc.used_gpu_memory { + UsedGpuMemory::Used(val) => val, + UsedGpuMemory::Unavailable => 0, + }; + if let Some(prev) = procs.get(&pid) { + procs.insert(pid, (gpu_mem, prev.1)); + } else { + procs.insert(pid, (gpu_mem, 0)); + } + } + } + if !procs.is_empty() { + proc_vec.push(procs); + } + // running total for proc % + if let Ok(mem) = device.memory_info() { + total_mem += mem.total; + } + } + } + } + Some(GpusData { + memory: if !mem_vec.is_empty() { + Some(mem_vec) + } else { + None + }, + temperature: if !temp_vec.is_empty() { + Some(temp_vec) + } else { + None + }, + procs: if !proc_vec.is_empty() { + Some((total_mem, proc_vec)) + } else { + None + }, + }) + } else { + None + } + } else { + None + } +} diff --git a/src/app/data_harvester/processes.rs b/src/app/data_harvester/processes.rs index 5d14930e..7d042635 100644 --- a/src/app/data_harvester/processes.rs +++ b/src/app/data_harvester/processes.rs @@ -83,6 +83,18 @@ pub struct ProcessHarvest { /// This is the process' user. pub user: Cow<'static, str>, + + /// Gpu memory usage as bytes. + #[cfg(feature = "gpu")] + pub gpu_mem: u64, + + /// Gpu memory usage as percentage. + #[cfg(feature = "gpu")] + pub gpu_mem_percent: f32, + + /// Gpu utilization as a percentage. + #[cfg(feature = "gpu")] + pub gpu_util: u32, // TODO: Additional fields // pub rss_kb: u64, // pub virt_kb: u64, @@ -98,6 +110,12 @@ impl ProcessHarvest { self.total_read_bytes += rhs.total_read_bytes; self.total_write_bytes += rhs.total_write_bytes; self.time += rhs.time; + #[cfg(feature = "gpu")] + { + self.gpu_mem += rhs.gpu_mem; + self.gpu_util += rhs.gpu_util; + self.gpu_mem_percent += rhs.gpu_mem_percent; + } } } diff --git a/src/app/data_harvester/processes/linux.rs b/src/app/data_harvester/processes/linux.rs index 8f50823b..0a1ed539 100644 --- a/src/app/data_harvester/processes/linux.rs +++ b/src/app/data_harvester/processes/linux.rs @@ -250,6 +250,12 @@ fn read_proc( uid, user, time, + #[cfg(feature = "gpu")] + gpu_mem: 0, + #[cfg(feature = "gpu")] + gpu_mem_percent: 0.0, + #[cfg(feature = "gpu")] + gpu_util: 0, }, new_process_times, )) @@ -326,7 +332,8 @@ pub(crate) fn linux_process_data( let pid = process.pid; let prev_proc_details = pid_mapping.entry(pid).or_default(); - if let Ok((process_harvest, new_process_times)) = read_proc( + #[allow(unused_mut)] + if let Ok((mut process_harvest, new_process_times)) = read_proc( prev_proc_details, process, cpu_usage, @@ -336,6 +343,23 @@ pub(crate) fn linux_process_data( total_memory, user_table, ) { + #[cfg(feature = "gpu")] + if let Some(gpus) = &collector.gpu_pids { + gpus.iter().for_each(|gpu| { + // add mem/util for all gpus to pid + if let Some((mem, util)) = gpu.get(&(pid as u32)) { + process_harvest.gpu_mem += mem; + process_harvest.gpu_util += util; + } + }); + if let Some(gpu_total_mem) = &collector.gpus_total_mem { + |