Propagate --threads to workers spawned with --procs

or --machine-file.

NEWS.md

@@ -54,9 +54,8 @@ Command-line option changes
 ---------------------------
 
   * `-t N`, `--threads N` starts Julia with `N` threads. This option takes precedence over
-    `JULIA_NUM_THREADS`. The specified number of threads only applies to the main Julia
-    process and it does not propagate to worker processes spawned using the `-p`/`--procs`
-    or `--machine-file` command line arguments ([#TBD]).
+    `JULIA_NUM_THREADS`. The specified number of threads also propagates to worker processes
+    spawned using the `-p`/`--procs` or `--machine-file` command line arguments ([#TBD]).
 
 Multi-threading changes
 -----------------------

doc/src/manual/parallel-computing.md

@@ -272,10 +272,11 @@ julia> Threads.threadid()
     Note that this must be done *before* starting Julia.
 
 !!! note
-    The number of threads specified with `-t`/`--threads` only applies to the main process,
-    and, thus, it does not propagate to processes spawned using the `-p`/`--procs` or
-    `--machine-file` command line options. To spawn workers with multiple threads enabled,
-    use [`addprocs`](@ref) and pass `-t`/`--threads` as `exeflags`.
+    The number of threads specified with `-t`/`--threads` is propagated to worker processes
+    that are spawned using the `-p`/`--procs` or `--machine-file` command line options.
+    For example, `julia -p2 -t2` spawns 1 main process with 2 worker processes, and all
+    three processes has 2 threads enabled. For more fine grained control over worker
+    threads use [`addprocs`](@ref) and pass `-t`/`--threads` as `exeflags`.
 
 ## The `@threads` Macro
 

src/jloptions.c

@@ -393,7 +393,7 @@ JL_DLLEXPORT void jl_parse_opts(int *argcp, char ***argvp)
         case 't': // threads
             errno = 0;
             if (!strcmp(optarg,"auto")) {
-                jl_options.nthreads = jl_cpu_threads();
+                jl_options.nthreads = -1;
             }
             else {
                 long nthreads = strtol(optarg, &endptr, 10);

src/threading.c

@@ -407,7 +407,9 @@ void jl_init_threading(void)
     // how many threads available, usable
     int max_threads = jl_cpu_threads();
     jl_n_threads = JULIA_NUM_THREADS;
-    if (jl_options.nthreads != 0)
+    if (jl_options.nthreads < 0) // --threads=auto
+        jl_n_threads = max_threads;
+    else if (jl_options.nthreads > 0) // --threads=N
         jl_n_threads = jl_options.nthreads;
     else if ((cp = getenv(NUM_THREADS_NAME)))
         jl_n_threads = (uint64_t)strtol(cp, NULL, 10);

stdlib/Distributed/src/cluster.jl

@@ -1298,6 +1298,7 @@ end
 
 write_cookie(io::IO) = print(io.in, string(cluster_cookie(), "\n"))
 
+# Starts workers specified by (-n|--procs) and --machine-file command line options
 function process_opts(opts)
     # startup worker.
     # opts.startupfile, opts.load, etc should should not be processed for workers.
@@ -1310,14 +1311,17 @@ function process_opts(opts)
         end
     end
 
+    # Propagate --threads to workers
+    exeflags = opts.nthreads > 0 ? `--threads=$(opts.nthreads)` : ``
+
     # add processors
     if opts.nprocs > 0
-        addprocs(opts.nprocs)
+        addprocs(opts.nprocs; exeflags=exeflags)
     end
 
     # load processes from machine file
     if opts.machine_file != C_NULL
-        addprocs(load_machine_file(unsafe_string(opts.machine_file)))
+        addprocs(load_machine_file(unsafe_string(opts.machine_file)); exeflags=exeflags)
     end
     return nothing
 end

test/cmdlineargs.jl

@@ -209,7 +209,7 @@ let exename = `$(Base.julia_cmd()) --startup-file=no`
     # Combining --threads and --procs: --threads does not propagate
     if cpu_threads > 1; withenv("JULIA_NUM_THREADS"=>nothing) do
         code = "print(Threads.nthreads() + sum(remotecall_fetch(() -> Threads.nthreads(), x) for  x in workers()))"
-        @test read(`$exename -p2 -t2 -e $code`, String) == "4"
+        @test read(`$exename -p2 -t2 -e $code`, String) == "6"
     end end
 
     # --procs