@threads :static for everywhere (#27)

JuliaPerf · Feb 20, 2022 · cb78adb · cb78adb
1 parent 2f3210c
commit cb78adb
Show file tree

Hide file tree

Showing 6 changed files with 16 additions and 16 deletions.
diff --git a/docs/src/examples/perfmon.jl b/docs/src/examples/perfmon.jl
@@ -21,12 +21,12 @@ const NUM_THREADS = 3;
 # Let's pin the first `NUM_THREADS` threads to the first `NUM_THREADS` cores.
 using LIKWID
 cores = 0:NUM_THREADS-1
-@threads for tid in 1:NUM_THREADS
+@threads :static for tid in 1:NUM_THREADS
     LIKWID.pinthread(cores[tid])
 end
 
 # To check that the pinning was successfull, we call [`LIKWID.get_processor_id`](@ref) on each thread.
-@threads for tid in 1:NUM_THREADS
+@threads :static for tid in 1:NUM_THREADS
     core = LIKWID.get_processor_id()
     println("Thread $tid, Core $core")
 end
@@ -64,7 +64,7 @@ PerfMon.init()
 # region. Typically these are done in separate parallel blocks, relying on
 # the implicit barrier at the end of the parallel block. Usually there is
 # a parallel block for initialization and a parallel block for execution.
-@threads for tid in 1:NUM_THREADS
+@threads :static for tid in 1:NUM_THREADS
     Marker.registerregion("Total")
     Marker.registerregion("calc_flops")
 
@@ -93,7 +93,7 @@ function monitor_do_flops(NUM_FLOPS = 100_000_000)
     a = 1.8
     b = 3.2
     c = 1.0
-    @threads for tid in 1:NUM_THREADS
+    @threads :static for tid in 1:NUM_THREADS
         ## Notice that only the first group specified, `FLOPS_DP`, will be measured.
         ## See further below for how to measure multiple groups.
         Marker.startregion("calc_flops")
@@ -108,7 +108,7 @@ monitor_do_flops()
 #
 # To query basic information about the region from all threads
 # we use [`Marker.getregion`](@ref).
-@threads for threadid in 1:NUM_THREADS
+@threads :static for threadid in 1:NUM_THREADS
     nevents, events, time, count = Marker.getregion("calc_flops")
     gid = PerfMon.get_id_of_active_group()
     group_name = PerfMon.get_name_of_group(gid)

diff --git a/docs/src/examples/perfmon.md b/docs/src/examples/perfmon.md
@@ -28,15 +28,15 @@ Let's pin the first `NUM_THREADS` threads to the first `NUM_THREADS` cores.
 ````@example perfmon
 using LIKWID
 cores = 0:NUM_THREADS-1
-@threads for tid in 1:NUM_THREADS
+@threads :static for tid in 1:NUM_THREADS
     LIKWID.pinthread(cores[tid])
 end
 ````
 
 To check that the pinning was successfull, we call [`LIKWID.get_processor_id`](@ref) on each thread.
 
 ````@example perfmon
-@threads for tid in 1:NUM_THREADS
+@threads :static for tid in 1:NUM_THREADS
     core = LIKWID.get_processor_id()
     println("Thread $tid, Core $core")
 end
@@ -84,7 +84,7 @@ the implicit barrier at the end of the parallel block. Usually there is
 a parallel block for initialization and a parallel block for execution.
 
 ````@example perfmon
-@threads for tid in 1:NUM_THREADS
+@threads :static for tid in 1:NUM_THREADS
     Marker.registerregion("Total")
     Marker.registerregion("calc_flops")
 
@@ -120,7 +120,7 @@ function monitor_do_flops(NUM_FLOPS = 100_000_000)
     a = 1.8
     b = 3.2
     c = 1.0
-    @threads for tid in 1:NUM_THREADS
+    @threads :static for tid in 1:NUM_THREADS
         # Notice that only the first group specified, `FLOPS_DP`, will be measured.
         # See further below for how to measure multiple groups.
         Marker.startregion("calc_flops")
@@ -138,7 +138,7 @@ To query basic information about the region from all threads
 we use [`Marker.getregion`](@ref).
 
 ````@example perfmon
-@threads for threadid in 1:NUM_THREADS
+@threads :static for threadid in 1:NUM_THREADS
     nevents, events, time, count = Marker.getregion("calc_flops")
     gid = PerfMon.get_id_of_active_group()
     group_name = PerfMon.get_name_of_group(gid)

diff --git a/docs/src/examples/saxpy.md b/docs/src/examples/saxpy.md
@@ -193,7 +193,7 @@ using Base.Threads: nthreads, @threads
 @assert nthreads() > 1 # multithreading
 
 # Julia threads should be pinned!
-@threads for tid in 1:nthreads()
+@threads :static for tid in 1:nthreads()
     core = LIKWID.get_processor_id()
     println("Thread $tid, Core $core")
 end
@@ -209,7 +209,7 @@ function saxpy_cpu!(z, a, x, y)
 end
 
 function saxpy_threads(zs, a, x, y)
-    @threads for tid in 1:nthreads()
+    @threads :static for tid in 1:nthreads()
         @region "saxpy_cpu!" saxpy_cpu!(zs[tid], a, x, y)
     end
 end

diff --git a/docs/src/likwid-pin.md b/docs/src/likwid-pin.md
@@ -16,7 +16,7 @@ using Base.Threads
 
 glibc_coreid() = @ccall sched_getcpu()::Cint
 
-@threads for i in 1:nthreads()
+@threads :static for i in 1:nthreads()
     println("Thread: $(i), CPU: $(glibc_coreid())")
 end
 ```

diff --git a/examples/perfctr_threads/threads.jl b/examples/perfctr_threads/threads.jl
@@ -6,7 +6,7 @@ using Base.Threads: nthreads, @threads
 @assert nthreads() > 1 # multithreading
 
 # Julia threads should be pinned!
-@threads for tid in 1:nthreads()
+@threads :static for tid in 1:nthreads()
     core = LIKWID.get_processor_id()
     println("Thread $tid, Core $core")
 end
@@ -22,7 +22,7 @@ function saxpy_cpu!(z, a, x, y)
 end
 
 function saxpy_threads(zs, a, x, y)
-    @threads for tid in 1:nthreads()
+    @threads :static for tid in 1:nthreads()
         @region "saxpy_cpu!" saxpy_cpu!(zs[tid], a, x, y)
     end
 end

diff --git a/examples/pin/pin.jl b/examples/pin/pin.jl
@@ -3,6 +3,6 @@ using Base.Threads
 
 glibc_coreid() = @ccall sched_getcpu()::Cint
 
-@threads for i in 1:nthreads()
+@threads :static for i in 1:nthreads()
     println("Thread: $(i), CPU: $(glibc_coreid())")
 end