From da9a6b0c4693b24ae94b10a84d0d221ece73cf5e Mon Sep 17 00:00:00 2001
From: ndwarshuis <ndwar@yavin4.ch>
Date: Wed, 27 Sep 2023 00:05:11 -0400
Subject: [PATCH] FIX freq and cpu calculation errors

---
 src/modules/processor.lua |  89 ++++++++++--------
 src/sys.lua               | 193 ++++++++++++--------------------------
 2 files changed, 108 insertions(+), 174 deletions(-)

diff --git a/src/modules/processor.lua b/src/modules/processor.lua
index 9426aa3..0dba35f 100644
--- a/src/modules/processor.lua
+++ b/src/modules/processor.lua
@@ -6,6 +6,7 @@ local cpu = require 'sys'
 local pure = require 'pure'
 
 local __math_floor = math.floor
+local __string_format = string.format
 
 return function(update_freq, main_state, config, common, width, point)
    local dial_inner_radius = 30
@@ -89,6 +90,27 @@ return function(update_freq, main_state, config, common, width, point)
       }
    end
 
+   local read_load = function(core_topology, phy_core_id, thread_id)
+      local i = core_topology[phy_core_id].cpus[thread_id].lgl_cpu_id
+      return mod_state[i].percent_active * 100
+   end
+
+   local get_load_functions = function(cores, nthreads, core_topology)
+      if nthreads == 1 then
+         local update = function(c)
+            dial.set(cores[c].loads, read_load(core_topology, c, 1))
+         end
+         return update, dial.draw_static, dial.draw_dynamic
+      else
+         local update = function(c)
+            for t = 1, nthreads do
+               compound_dial.set(cores[c].loads, t, read_load(core_topology, c, t))
+            end
+         end
+         return update, compound_dial.draw_static, compound_dial.draw_dynamic
+      end
+   end
+
    local mk_core_group = function(group_config, y)
       local nthreads = group_config.threads
       local core_topology = topology[nthreads]
@@ -98,32 +120,8 @@ return function(update_freq, main_state, config, common, width, point)
          create_core, core_cols, y, nthreads, group_config.padding
       )
       local cores = pure.map_n(_create_core, ncores)
-      local group_loads = mod_state[nthreads]
-      local update_loads
-      local draw_static_loads
-      local draw_dynamic_loads
-      if nthreads == 1 then
-         update_loads = function(c)
-            dial.set(
-               cores[c].loads,
-               group_loads[c][1].percent_active * 100
-            )
-         end
-         draw_static_loads = dial.draw_static
-         draw_dynamic_loads = dial.draw_dynamic
-      else
-         update_loads = function(c)
-            for t = 1, nthreads do
-               compound_dial.set(
-                  cores[c].loads,
-                  t,
-                  group_loads[c][t].percent_active * 100
-               )
-            end
-         end
-         draw_static_loads = compound_dial.draw_static
-         draw_dynamic_loads = compound_dial.draw_dynamic
-      end
+      local update_loads, draw_static_loads, draw_dynamic_loads =
+         get_load_functions(cores, nthreads, core_topology)
       local update = function()
          for c = 1, ncores do
             local temp = __math_floor(
@@ -159,13 +157,23 @@ return function(update_freq, main_state, config, common, width, point)
    -- HWP status
 
    local mk_hwp_freq = function(y)
-      local hwp_paths = cpu.get_hwp_paths()
+      local hwp_paths = cpu.get_hwp_paths(topology)
+      local freq_labels
+      local cpu_group_map = cpu.topology_to_cpu_map(topology)
+      local format_label = function(group_id)
+         return __string_format('Ave Freq (%i)', group_id)
+      end
+      if #topology == 1 then
+         freq_labels = {'Ave Freq'}
+      else
+         freq_labels = pure.map_n(format_label, #topology)
+      end
       local cpu_status = common.make_text_rows(
          point.x,
          y,
          width,
          text_spacing,
-         {'HWP Preference', 'Ave Freq'}
+         pure.concat({'HWP Preference'}, freq_labels)
       )
       local update = function()
          -- For some reason this call is slow (querying anything with pstate in
@@ -174,13 +182,18 @@ return function(update_freq, main_state, config, common, width, point)
          if main_state.trigger10 == 0 then
             common.text_rows_set(cpu_status, 1, cpu.read_hwp(hwp_paths))
          end
-         common.text_rows_set(cpu_status, 2, cpu.read_freq())
+         local ave_freqs = cpu.read_ave_freqs(topology, cpu_group_map)
+         local i = 2
+         for group_id, _ in pairs(topology) do
+            common.text_rows_set(cpu_status, i, ave_freqs[group_id])
+            i = i + 1
+         end
       end
       local static = pure.partial(common.text_rows_draw_static, cpu_status)
       local dynamic = pure.partial(common.text_rows_draw_dynamic, cpu_status)
       return common.mk_acc(
          width,
-         text_spacing,
+         text_spacing * #topology,
          update,
          static,
          dynamic
@@ -203,12 +216,8 @@ return function(update_freq, main_state, config, common, width, point)
       )
       local update = function()
          local s = 0
-         for g = 1, #mod_state do
-            for c = 1, #mod_state[g] do
-               for t = 1, #mod_state[g][c] do
-                  s = s + mod_state[g][c][t].percent_active
-               end
-            end
+         for i = 1, ncpus do
+            s = s + mod_state[i].percent_active
          end
          common.tagged_percent_timeseries_set(total_load, s / ncpus * 100)
       end
@@ -269,10 +278,10 @@ return function(update_freq, main_state, config, common, width, point)
       point = point,
       width = width,
       set_state = update_state,
-      top = {
-         table.unpack(pure.map(core_group_section, config.core_groups)),
-         -- {mk_hwp_freq, config.show_stats, sep_spacing},
-      },
+      top = pure.concat(
+         pure.map(core_group_section, config.core_groups),
+         {{mk_hwp_freq, config.show_stats, sep_spacing}}
+      ),
       common.mk_section(
          sep_spacing,
          {mk_load_plot, config.show_plot, geo.table.sec_break},
diff --git a/src/sys.lua b/src/sys.lua
index fa31b46..91cd062 100644
--- a/src/sys.lua
+++ b/src/sys.lua
@@ -172,7 +172,7 @@ end
 --------------------------------------------------------------------------------
 -- cpu
 
--- ASSUME nproc and lscpu will always be available
+-- ASSUME lscpu will always be available
 
 M.get_core_number = function()
    return __tonumber(i_o.read_file('/proc/cpuinfo', 'cpu cores%s+:%s(%d+)'))
@@ -180,13 +180,10 @@ end
 
 M.get_cpu_number = function(topology)
    local n = 0
-   for g = 1, #topology do
-      for c = 1, #topology[g] do
-         n = n + #topology[g][c].cpus
-      end
+   for g, c in pairs(topology) do
+      n = n + g * #c
    end
    return n
-   -- return __tonumber(i_o.execute_cmd('nproc', nil, '*n'))
 end
 
 local get_coretemp_dir = function()
@@ -208,25 +205,6 @@ M.get_core_threads = function()
    return pure.fmap_maybe(make_indexer, i_o.execute_cmd(cmd))
 end
 
-
--- map cores to integer values starting at 1; this is necessary since some cpus
--- don't report their core id's as a sequence of integers starting at 0
--- local get_core_id_indexer = function()
---    local make_indexer = pure.compose(
---       pure.array_to_map,
---       pure.partial(pure.imap, function(i, c) return {__tonumber(c), i} end),
---       pure.partial(gmatch_to_table1, '(%d+)')
---    )
---    return pure.fmap_maybe(
---       make_indexer,
---       i_o.execute_cmd('lscpu -p=CORE | tail -n+5 | sort -k1,1n')
---    )
--- end
-
--- conky_core_idx: the ID of the dial to be drawn for this core
--- conky_thread_idx: the ID of the individual indicator within one dial
---   corresponding to one thread in a core (starting at 1 for each core)
-
 local get_coretemp_mapper = function()
    local d = get_coretemp_dir()
    i_o.assert_exe_exists('grep')
@@ -289,91 +267,54 @@ M.get_core_topology = function()
    return pure.fmap_maybe(f, out)
 end
 
--- for t, k in pairs(get_core_topology()) do
---    print(t)
---    for x, y in pairs(k) do
---       print(x, y.phy_core_id, y.coretemp_path, #y.cpus)
---       -- for _, z in pairs(y.cpus) do
---       --    print(x,z.cpu,z.conky_cpu)
---       -- end
---    end
--- end
-
-local get_core_mappings = function()
-   local core_threads = M.get_core_threads()
-   local assign_cpus = function(x)
-      return {
-         cpu_id = __tonumber(x[1]),
-         core_id = __tonumber(x[2])
-      }
-   end
-   local map_ids = function(indexer)
-      local f = function(acc, next)
-         local cpu_id = __tonumber(next[1]) + 1
-         local core_id = __tonumber(next[2])
-         local conky_core_idx = indexer[core_id]
-         acc.mappings[cpu_id] = {
-            conky_core_idx = conky_core_idx,
-            conky_thread_id = acc.thread_ids[conky_core_idx],
-         }
-         acc.thread_ids[conky_core_idx] = acc.thread_ids[conky_core_idx] + 1
-         return acc
+M.topology_to_cpu_map = function(topology)
+   local r = {}
+   for group_id, group in pairs(topology) do
+      for _, core in pairs(group) do
+         for _, cpu in pairs(core.cpus) do
+            r[cpu.lgl_cpu_id] = group_id
+         end
       end
-      local cpu_to_core_map = pure.maybe(
-         {},
-         pure.partial(gmatch_to_tableN, '(%d+),(%d+)'),
-         i_o.execute_cmd('lscpu -y -p=cpu,core | grep -v \'^#\' | sort -k1,1n')
-      )
-      local init = {mappings = {}, _conky_core_index = 0, _thread_ids = {}}
-      return pure.reduce(f, init, cpu_to_core_map).mappings
    end
-   -- return pure.fmap_maybe(map_ids, )
+   return r
 end
 
-M.get_coretemp_paths = function()
-   local get_paths = function(indexer)
-      local d = get_coretemp_dir()
-      i_o.assert_exe_exists('grep')
-      local get_labels = pure.compose(
-         i_o.execute_cmd,
-         pure.partial(__string_format, 'grep Core %s/temp*_label', true)
-      )
-      local to_tuple = function(m)
-         return {
-            indexer[__tonumber(m[2])],
-            __string_format('%s/%s_input', d, m[1])
-         }
-      end
-      local f = pure.compose(
-         pure.array_to_map,
-         pure.partial(pure.map, to_tuple),
-         pure.partial(gmatch_to_tableN, '/([^/\n]+)_label:Core (%d+)\n')
-      )
-      return pure.maybe({}, f, pure.fmap_maybe(get_labels, d))
-   end
-   return pure.maybe({}, get_paths, get_core_id_indexer())
-end
-
-local match_freq = function(c)
-   local f = 0
-   local n = 0
-   for s in __string_gmatch(c, '(%d+%.%d+)') do
-      f = f + __tonumber(s)
-      n = n + 1
-   end
-   return __string_format('%.0f Mhz', f / n)
-end
-
-M.read_freq = function()
+M.read_ave_freqs = function(topology, cpu_group_map)
    -- NOTE: Using the builtin conky functions for getting cpu freq seems to make
    -- the entire loop jittery due to high variance latency. Querying
    -- scaling_cur_freq in sysfs seems to do the same thing. It appears lscpu
    -- (which queries /proc/cpuinfo) is much faster and doesn't have this jittery
    -- problem.
-   return pure.maybe('N/A', match_freq, i_o.execute_cmd('lscpu -p=MHZ'))
+   local out = i_o.execute_cmd('lscpu -p=MHZ')
+   local init_freqs = function(v)
+      local r = {}
+      for group_id, _ in pairs(topology) do
+         r[group_id] = v
+      end
+      return r
+   end
+   if out == nil then
+      return init_freqs('N/A')
+   else
+      local ave_freqs = init_freqs(0)
+      local cpu_id = 1
+      for s in __string_gmatch(out, '(%d+%.%d+)') do
+         local group_id = cpu_group_map[cpu_id]
+         ave_freqs[group_id] = ave_freqs[group_id] + __tonumber(s)
+         cpu_id = cpu_id + 1
+      end
+      for group_id, _ in pairs(ave_freqs) do
+         ave_freqs[group_id] =
+            __string_format(
+               '%.0f Mhz',
+               ave_freqs[group_id] / (group_id * #topology[group_id])
+            )
+      end
+      return ave_freqs
+   end
 end
 
-M.get_hwp_paths = function()
+M.get_hwp_paths = function(topology)
    -- ASSUME this will never fail
    return pure.map_n(
       function(i)
@@ -381,7 +322,7 @@ M.get_hwp_paths = function()
             .. (i - 1)
             .. '/cpufreq/energy_performance_preference'
       end,
-      M.get_cpu_number()
+      M.get_cpu_number(topology)
    )
 end
 
@@ -410,23 +351,14 @@ M.read_hwp = function(hwp_paths)
 end
 
 M.init_cpu_loads = function(topo)
-   -- -- local m = get_core_mappings()
-   -- local topo = get_core_topology()
+   local ncpus = M.get_cpu_number(topo)
    local cpu_loads = {}
-   for core_group_id, core_group in pairs(topo) do
-      cpu_loads[core_group_id] = {}
-      for lgl_core_id, core in pairs(core_group) do
-         cpu_loads[core_group_id][lgl_core_id] = {}
-         for thread_id = 1, #core.cpus do
-            cpu_loads[core_group_id][lgl_core_id][thread_id] = {
-               active_prev = 0,
-               total_prev = 0,
-               percent_active = 0,
-               -- core_id = lgl_core_id,
-               -- thread_id = thread_id,
-            }
-         end
-      end
+   for lgl_cpu_id = 1, ncpus do
+      cpu_loads[lgl_cpu_id] = {
+         active_prev = 0,
+         total_prev = 0,
+         percent_active = 0,
+      }
    end
    return cpu_loads
 end
@@ -434,24 +366,17 @@ end
 M.read_cpu_loads = function(cpu_loads)
    local iter = io.lines('/proc/stat')
    iter() -- ignore first line
-   for group_id = 1, #cpu_loads do
-      local group = cpu_loads[group_id]
-      for lgl_core_id = 1, #group do
-         local core = group[lgl_core_id]
-         for thread_id = 1, #core do
-            local ln = iter()
-            local user, system, idle =
-               __string_match(ln, '%d+ (%d+) %d+ (%d+) (%d+)', 4)
-            local active = user + system
-            local total = active + idle
-            local thread = core[thread_id]
-            if total > thread.total_prev then -- guard against 1/0 errors
-               thread.percent_active =
-                  (active - thread.active_prev) / (total - thread.total_prev)
-               thread.active_prev = active
-               thread.total_prev = total
-            end
-         end
+   for lgl_cpu_id = 1, #cpu_loads do
+      local ln = iter()
+      local user, system, idle =
+         __string_match(ln, '%d+ (%d+) %d+ (%d+) (%d+)', 4)
+      local active = user + system
+      local total = active + idle
+      local cpu = cpu_loads[lgl_cpu_id]
+      if total > cpu.total_prev then -- guard against 1/0 errors
+         cpu.percent_active = (active - cpu.active_prev) / (total - cpu.total_prev)
+         cpu.active_prev = active
+         cpu.total_prev = total
       end
    end
    return cpu_loads