Skip to content

Commit f5327dc

Browse files
KinanLakKinanLak
committed
feat: add 'proc_gpu_only' option to filter processes by GPU usage
1 parent bf7f147 commit f5327dc

6 files changed

Lines changed: 122 additions & 25 deletions

File tree

‎src/btop_config.cpp‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -121,6 +121,8 @@ namespace Config {
121121

122122
{"proc_cpu_graphs", "#* Show cpu graph for each process."},
123123

124+
{"proc_gpu_only", "#* Show only processes with active GPU usage or GPU memory allocation in the process list."},
125+
124126
{"proc_info_smaps", "#* Use /proc/[pid]/smaps for memory information in the process info box (very slow but more accurate)"},
125127

126128
{"proc_left", "#* Show proc box on left side of screen instead of right."},
@@ -304,6 +306,7 @@ namespace Config {
304306
{"proc_per_core", false},
305307
{"proc_mem_bytes", true},
306308
{"proc_cpu_graphs", true},
309+
{"proc_gpu_only", false},
307310
{"proc_info_smaps", false},
308311
{"proc_left", false},
309312
{"proc_filter_kernel", false},

‎src/btop_draw.cpp‎

Lines changed: 18 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1890,9 +1890,18 @@ namespace Proc {
18901890

18911891
//? pause, per-core, reverse, tree and sorting
18921892
const auto& sorting = Config::getS("proc_sorting");
1893+
const auto gpu_only = Config::getB("proc_gpu_only");
18931894
const int sort_len = sorting.size();
18941895
const int sort_pos = x + width - sort_len - 8;
18951896

1897+
if (width > 70 + sort_len) {
1898+
fmt::format_to(std::back_inserter(out), "{}{}{}{}{}{}{}{}{}{}{}",
1899+
Mv::to(y, sort_pos - 43), title_left, gpu_only ? Fx::b : "",
1900+
Theme::c("hi_fg"), 'g', Theme::c("title"), "pu-",
1901+
Theme::c("title"), "only",
1902+
Fx::ub, title_right);
1903+
Input::mouse_mappings["g"] = {y, sort_pos - 42, 1, 8};
1904+
}
18961905
if (width > 60 + sort_len) {
18971906
fmt::format_to(std::back_inserter(out), "{}{}{}{}{}{}{}{}{}{}{}",
18981907
Mv::to(y, sort_pos - 32), title_left, pause_proc_list ? Fx::b : "",
@@ -2125,11 +2134,16 @@ namespace Proc {
21252134
}
21262135
string gpu_str;
21272136
if (show_gpu) {
2128-
gpu_str = fmt::format("{:.1f}", clamp(p.gpu_p, 0.0, 100.0));
2129-
if (gpu_str.size() > 4) gpu_str.resize(4);
2130-
if (gpu_str.ends_with('.')) gpu_str.pop_back();
2137+
if (p.gpu_p <= 0.0 and p.gpu_m == 0) {
2138+
gpu_str = "-";
2139+
}
2140+
else {
2141+
gpu_str = fmt::format("{:.1f}", clamp(p.gpu_p, 0.0, 100.0));
2142+
if (gpu_str.size() > 4) gpu_str.resize(4);
2143+
if (gpu_str.ends_with('.')) gpu_str.pop_back();
2144+
}
21312145
}
2132-
const string gpu_mem_str = show_gpu_mem ? floating_humanizer(p.gpu_m, true) : "";
2146+
const string gpu_mem_str = show_gpu_mem ? (p.gpu_p <= 0.0 and p.gpu_m == 0 ? "-" : floating_humanizer(p.gpu_m, true)) : "";
21332147
string mem_str = (mem_bytes ? floating_humanizer(p.mem, true) : "");
21342148
if (not mem_bytes) {
21352149
double mem_p = clamp((double)p.mem * 100 / totalMem, 0.0, 100.0);

‎src/btop_input.cpp‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -377,6 +377,11 @@ namespace Input {
377377
else if (key == "c")
378378
Config::flip("proc_per_core");
379379

380+
else if (key == "g") {
381+
Config::flip("proc_gpu_only");
382+
Config::set("update_following", true);
383+
}
384+
380385
else if (key == "%")
381386
Config::flip("proc_mem_bytes");
382387

‎src/btop_menu.cpp‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -873,6 +873,12 @@ namespace Menu {
873873
"Show cpu graph for each process.",
874874
"",
875875
"True or False"},
876+
{"proc_gpu_only",
877+
"Show only GPU-active processes.",
878+
"",
879+
"When enabled, only processes with",
880+
"non-zero GPU usage or GPU memory",
881+
"allocation are shown."},
876882
{"proc_filter_kernel",
877883
"(Linux) Filter kernel processes from output.",
878884
"",

‎src/btop_shared.cpp‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -181,6 +181,10 @@ bool set_priority(pid_t pid, int priority) {
181181
}
182182

183183
auto matches_filter(const proc_info& proc, const std::string& filter) -> bool {
184+
if (Config::getB("proc_gpu_only") and proc.gpu_p <= 0.0 and proc.gpu_m == 0) {
185+
return false;
186+
}
187+
184188
if (filter.starts_with("!")) {
185189
if (filter.size() == 1) {
186190
return true;

‎src/linux/btop_collect.cpp‎

Lines changed: 86 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -202,6 +202,16 @@ namespace Gpu {
202202
unsigned int computeInstanceId;
203203
unsigned long long usedGpuCcProtectedMemory;
204204
};
205+
struct nvmlProcessInfo_v2_t {
206+
unsigned int pid;
207+
unsigned long long usedGpuMemory;
208+
unsigned int gpuInstanceId;
209+
unsigned int computeInstanceId;
210+
};
211+
struct nvmlProcessInfo_v1_t {
212+
unsigned int pid;
213+
unsigned long long usedGpuMemory;
214+
};
205215
struct proc_stat {
206216
double gpu{};
207217
uint64_t mem{};
@@ -226,7 +236,11 @@ namespace Gpu {
226236
nvmlReturn_t (*nvmlDeviceGetEncoderUtilization)(nvmlDevice_t, unsigned int*, unsigned int*);
227237
nvmlReturn_t (*nvmlDeviceGetDecoderUtilization)(nvmlDevice_t, unsigned int*, unsigned int*);
228238
nvmlReturn_t (*nvmlDeviceGetProcessUtilization)(nvmlDevice_t, nvmlProcessUtilizationSample_t*, unsigned int*, unsigned long long);
239+
nvmlReturn_t (*nvmlDeviceGetGraphicsRunningProcesses_v1)(nvmlDevice_t, unsigned int*, nvmlProcessInfo_v1_t*);
240+
nvmlReturn_t (*nvmlDeviceGetGraphicsRunningProcesses_v2)(nvmlDevice_t, unsigned int*, nvmlProcessInfo_v2_t*);
229241
nvmlReturn_t (*nvmlDeviceGetGraphicsRunningProcesses_v3)(nvmlDevice_t, unsigned int*, nvmlProcessInfo_t*);
242+
nvmlReturn_t (*nvmlDeviceGetComputeRunningProcesses_v1)(nvmlDevice_t, unsigned int*, nvmlProcessInfo_v1_t*);
243+
nvmlReturn_t (*nvmlDeviceGetComputeRunningProcesses_v2)(nvmlDevice_t, unsigned int*, nvmlProcessInfo_v2_t*);
230244
nvmlReturn_t (*nvmlDeviceGetComputeRunningProcesses_v3)(nvmlDevice_t, unsigned int*, nvmlProcessInfo_t*);
231245

232246
//? Data
@@ -1300,15 +1314,20 @@ namespace Gpu {
13001314
LOAD_SYM(nvmlDeviceGetEncoderUtilization);
13011315
LOAD_SYM(nvmlDeviceGetDecoderUtilization);
13021316
nvmlDeviceGetProcessUtilization = (decltype(nvmlDeviceGetProcessUtilization))try_load_nvml_sym("nvmlDeviceGetProcessUtilization");
1317+
nvmlDeviceGetGraphicsRunningProcesses_v1 = (decltype(nvmlDeviceGetGraphicsRunningProcesses_v1))try_load_nvml_sym("nvmlDeviceGetGraphicsRunningProcesses");
1318+
nvmlDeviceGetGraphicsRunningProcesses_v2 = (decltype(nvmlDeviceGetGraphicsRunningProcesses_v2))try_load_nvml_sym("nvmlDeviceGetGraphicsRunningProcesses_v2");
13031319
nvmlDeviceGetGraphicsRunningProcesses_v3 = (decltype(nvmlDeviceGetGraphicsRunningProcesses_v3))try_load_nvml_sym("nvmlDeviceGetGraphicsRunningProcesses_v3");
1320+
nvmlDeviceGetComputeRunningProcesses_v1 = (decltype(nvmlDeviceGetComputeRunningProcesses_v1))try_load_nvml_sym("nvmlDeviceGetComputeRunningProcesses");
1321+
nvmlDeviceGetComputeRunningProcesses_v2 = (decltype(nvmlDeviceGetComputeRunningProcesses_v2))try_load_nvml_sym("nvmlDeviceGetComputeRunningProcesses_v2");
13041322
nvmlDeviceGetComputeRunningProcesses_v3 = (decltype(nvmlDeviceGetComputeRunningProcesses_v3))try_load_nvml_sym("nvmlDeviceGetComputeRunningProcesses_v3");
1305-
if (nvmlDeviceGetGraphicsRunningProcesses_v3 == nullptr)
1306-
nvmlDeviceGetGraphicsRunningProcesses_v3 = (decltype(nvmlDeviceGetGraphicsRunningProcesses_v3))try_load_nvml_sym("nvmlDeviceGetGraphicsRunningProcesses_v2");
1307-
if (nvmlDeviceGetComputeRunningProcesses_v3 == nullptr)
1308-
nvmlDeviceGetComputeRunningProcesses_v3 = (decltype(nvmlDeviceGetComputeRunningProcesses_v3))try_load_nvml_sym("nvmlDeviceGetComputeRunningProcesses_v2");
13091323
process_utilization_function_available = nvmlDeviceGetProcessUtilization != nullptr;
1310-
process_memory_functions_available = nvmlDeviceGetGraphicsRunningProcesses_v3 != nullptr
1311-
and nvmlDeviceGetComputeRunningProcesses_v3 != nullptr;
1324+
process_memory_functions_available =
1325+
nvmlDeviceGetGraphicsRunningProcesses_v3 != nullptr or
1326+
nvmlDeviceGetGraphicsRunningProcesses_v2 != nullptr or
1327+
nvmlDeviceGetGraphicsRunningProcesses_v1 != nullptr or
1328+
nvmlDeviceGetComputeRunningProcesses_v3 != nullptr or
1329+
nvmlDeviceGetComputeRunningProcesses_v2 != nullptr or
1330+
nvmlDeviceGetComputeRunningProcesses_v1 != nullptr;
13121331

13131332
#undef LOAD_SYM
13141333

@@ -1562,30 +1581,72 @@ namespace Gpu {
15621581

15631582
process_stats.clear();
15641583

1565-
auto append_process_memory = [&](nvmlDevice_t device, auto get_processes, std::unordered_map<size_t, uint64_t>& mem_by_pid) {
1566-
unsigned int proc_count = 64;
1584+
auto merge_process_memory = [&](unsigned int pid, unsigned long long used_mem, std::unordered_map<size_t, uint64_t>& mem_by_pid) {
1585+
if (used_mem == NVML_VALUE_NOT_AVAILABLE_ULL) return;
1586+
auto& mem = mem_by_pid[pid];
1587+
mem = max(mem, (uint64_t)used_mem);
1588+
};
1589+
1590+
auto append_process_memory_v3 = [&](nvmlDevice_t device, auto fn, std::unordered_map<size_t, uint64_t>& mem_by_pid) {
1591+
if (fn == nullptr) return;
1592+
unsigned int proc_count = 0;
1593+
auto result = fn(device, &proc_count, nullptr);
1594+
if (result == NVML_ERROR_NOT_FOUND) return;
1595+
if (result != NVML_SUCCESS and result != NVML_ERROR_INSUFFICIENT_SIZE) return;
1596+
if (proc_count == 0) return;
1597+
15671598
std::vector<nvmlProcessInfo_t> processes(proc_count);
1568-
auto result = get_processes(device, &proc_count, processes.data());
1599+
result = fn(device, &proc_count, processes.data());
1600+
if (result != NVML_SUCCESS) return;
15691601

1570-
if (result == NVML_ERROR_INSUFFICIENT_SIZE and proc_count > 0) {
1571-
processes.resize(proc_count);
1572-
result = get_processes(device, &proc_count, processes.data());
1602+
for (unsigned int n = 0; n < proc_count; ++n) {
1603+
merge_process_memory(processes[n].pid, processes[n].usedGpuMemory, mem_by_pid);
15731604
}
1605+
};
1606+
1607+
auto append_process_memory_v2 = [&](nvmlDevice_t device, auto fn, std::unordered_map<size_t, uint64_t>& mem_by_pid) {
1608+
if (fn == nullptr) return;
1609+
unsigned int proc_count = 0;
1610+
auto result = fn(device, &proc_count, nullptr);
1611+
if (result == NVML_ERROR_NOT_FOUND) return;
1612+
if (result != NVML_SUCCESS and result != NVML_ERROR_INSUFFICIENT_SIZE) return;
1613+
if (proc_count == 0) return;
1614+
1615+
std::vector<nvmlProcessInfo_v2_t> processes(proc_count);
1616+
result = fn(device, &proc_count, processes.data());
1617+
if (result != NVML_SUCCESS) return;
1618+
1619+
for (unsigned int n = 0; n < proc_count; ++n) {
1620+
merge_process_memory(processes[n].pid, processes[n].usedGpuMemory, mem_by_pid);
1621+
}
1622+
};
1623+
1624+
auto append_process_memory_v1 = [&](nvmlDevice_t device, auto fn, std::unordered_map<size_t, uint64_t>& mem_by_pid) {
1625+
if (fn == nullptr) return;
1626+
unsigned int proc_count = 0;
1627+
auto result = fn(device, &proc_count, nullptr);
1628+
if (result == NVML_ERROR_NOT_FOUND) return;
1629+
if (result != NVML_SUCCESS and result != NVML_ERROR_INSUFFICIENT_SIZE) return;
1630+
if (proc_count == 0) return;
1631+
1632+
std::vector<nvmlProcessInfo_v1_t> processes(proc_count);
1633+
result = fn(device, &proc_count, processes.data());
15741634
if (result != NVML_SUCCESS) return;
15751635

15761636
for (unsigned int n = 0; n < proc_count; ++n) {
1577-
const auto& proc = processes[n];
1578-
if (proc.usedGpuMemory == NVML_VALUE_NOT_AVAILABLE_ULL) continue;
1579-
auto& mem = mem_by_pid[proc.pid];
1580-
mem = max(mem, (uint64_t)proc.usedGpuMemory);
1637+
merge_process_memory(processes[n].pid, processes[n].usedGpuMemory, mem_by_pid);
15811638
}
15821639
};
15831640

15841641
for (unsigned int i = 0; i < device_count; ++i) {
15851642
if (process_memory_functions_available) {
15861643
std::unordered_map<size_t, uint64_t> mem_by_pid;
1587-
append_process_memory(devices[i], nvmlDeviceGetGraphicsRunningProcesses_v3, mem_by_pid);
1588-
append_process_memory(devices[i], nvmlDeviceGetComputeRunningProcesses_v3, mem_by_pid);
1644+
append_process_memory_v3(devices[i], nvmlDeviceGetGraphicsRunningProcesses_v3, mem_by_pid);
1645+
append_process_memory_v3(devices[i], nvmlDeviceGetComputeRunningProcesses_v3, mem_by_pid);
1646+
append_process_memory_v2(devices[i], nvmlDeviceGetGraphicsRunningProcesses_v2, mem_by_pid);
1647+
append_process_memory_v2(devices[i], nvmlDeviceGetComputeRunningProcesses_v2, mem_by_pid);
1648+
append_process_memory_v1(devices[i], nvmlDeviceGetGraphicsRunningProcesses_v1, mem_by_pid);
1649+
append_process_memory_v1(devices[i], nvmlDeviceGetComputeRunningProcesses_v1, mem_by_pid);
15891650

15901651
for (const auto& [pid, mem] : mem_by_pid) {
15911652
process_stats[pid].mem += mem;
@@ -2926,6 +2987,7 @@ namespace Proc {
29262987
string current_sort;
29272988
string current_filter;
29282989
bool current_rev{};
2990+
bool current_gpu_only{};
29292991
bool is_tree_mode;
29302992

29312993
fs::file_time_type passwd_time;
@@ -3158,14 +3220,17 @@ namespace Proc {
31583220
const auto& sorting = Config::getS("proc_sorting");
31593221
auto reverse = Config::getB("proc_reversed");
31603222
const auto& filter = Config::getS("proc_filter");
3223+
const bool gpu_only = Config::getB("proc_gpu_only");
31613224
auto per_core = Config::getB("proc_per_core");
31623225
auto should_filter_kernel = Config::getB("proc_filter_kernel");
31633226
auto tree = Config::getB("proc_tree");
31643227
auto show_detailed = Config::getB("show_detailed");
31653228
const auto pause_proc_list = Config::getB("pause_proc_list");
31663229
const size_t detailed_pid = Config::getI("detailed_pid");
3167-
bool should_filter = current_filter != filter;
3230+
bool should_filter = current_filter != filter or current_gpu_only != gpu_only;
3231+
if (gpu_only) should_filter = true;
31683232
if (should_filter) current_filter = filter;
3233+
if (should_filter) current_gpu_only = gpu_only;
31693234
bool sorted_change = (sorting != current_sort or reverse != current_rev or should_filter);
31703235
bool tree_mode_change = tree != is_tree_mode;
31713236
if (sorted_change) {
@@ -3509,8 +3574,8 @@ namespace Proc {
35093574
if (should_filter) {
35103575
filter_found = 0;
35113576
for (auto& p : current_procs) {
3512-
if (not tree and not filter.empty()) {
3513-
if (!matches_filter(p, filter)) {
3577+
if ((not tree and (not filter.empty() or gpu_only)) or gpu_only) {
3578+
if (not matches_filter(p, filter)) {
35143579
p.filtered = true;
35153580
filter_found++;
35163581
} else {

0 commit comments

Comments
 (0)