Fix small matters and improve sysconf()

- Fix mkdeps.com out of memory error
- Remove static memory from __get_cpu_count()
- Add support for passing hyphen to cat in cocmd
- Change more ZipOS errors from ENOTSUP to EROFS
- Specify mem_unit in sysinfo() output on BSD OSes

third_party/ggml/common.cc

@@ -88,7 +88,7 @@ static bool append_file_to_prompt(const char *path, gpt_params & params) {
 }
 
 bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
-    params.n_threads = std::min(20, std::max(1, _getcpucount() >> 1));
+    params.n_threads = std::min(20., (unsigned)__get_cpu_count() * .75);
 
     bool invalid_param = false;
     std::string arg;

third_party/ggml/common.h

@@ -21,7 +21,7 @@
 struct gpt_params {
     int32_t seed          = -1;   // RNG seed
     int32_t verbose       = 0;    // Logging verbosity
-    int32_t n_threads     = std::max(1, _getcpucount() >> 1);
+    int32_t n_threads     = std::max(20., (unsigned)__get_cpu_count() * .75);
     int32_t n_predict     = -1;   // new tokens to predict
     int32_t n_parts       = -1;   // amount of model parts (-1 = determine from model dimensions)
     int32_t n_ctx         = 512;  // context size

third_party/ggml/quantize.cc

@@ -92,7 +92,7 @@ int main(int argc, char ** argv) {
         ftype = (enum llama_ftype)atoi(argv[3]);
     }
 
-    int nthread = argc > 4 ? atoi(argv[4]) : std::min(20, std::max(1, _getcpucount() >> 1));
+    int nthread = argc > 4 ? atoi(argv[4]) : std::min(20, std::max(1, __get_cpu_count() >> 1));
 
     const int64_t t_main_start_us = ggml_time_us();
 

third_party/lua/lunix.c

@@ -3532,7 +3532,7 @@ int LuaUnix(lua_State *L) {
   LuaSetIntField(L, "RUSAGE_CHILDREN", RUSAGE_CHILDREN);
   LuaSetIntField(L, "RUSAGE_BOTH", RUSAGE_BOTH);
 
-  LuaSetIntField(L, "ARG_MAX", __arg_max());
+  LuaSetIntField(L, "ARG_MAX", __get_arg_max());
   LuaSetIntField(L, "BUFSIZ", BUFSIZ);
   LuaSetIntField(L, "CLK_TCK", CLK_TCK);
   LuaSetIntField(L, "NAME_MAX", _NAME_MAX);

third_party/python/Modules/posixmodule.c

@@ -10509,7 +10509,7 @@ os_cpu_count_impl(PyObject *module)
 /*[clinic end generated code: output=5fc29463c3936a9c input=e7c8f4ba6dbbadd3]*/
 {
     int ncpu;
-    ncpu = _getcpucount();
+    ncpu = __get_cpu_count();
     if (ncpu >= 1)
         return PyLong_FromLong(ncpu);
     else

third_party/radpajama/common-gptneox.cc

@@ -39,7 +39,7 @@
 // clang-format off
 
 bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
-    params.n_threads = std::min(20, std::max(1, (int)(_getcpucount() * 0.75)));
+    params.n_threads = std::min(20., (unsigned)__get_cpu_count() * 0.75);
 
     bool invalid_param = false;
     std::string arg;

third_party/radpajama/common-gptneox.h

@@ -19,7 +19,7 @@
 
 struct gpt_params {
     int32_t seed          = -1;   // RNG seed
-    int32_t n_threads     = MIN(4, (int32_t) _getcpucount() * 0.75);
+    int32_t n_threads     = MIN(20., (unsigned) __get_cpu_count() * 0.75);
     int32_t n_predict     = 128;  // new tokens to predict
     int32_t n_parts       = -1;   // amount of model parts (-1 = determine from model dimensions)
     int32_t n_ctx         = 512;  // context size