[rtems commit] testsupport: Determine worker index via processor
Sebastian Huber
sebh at rtems.org
Thu Nov 3 10:24:27 UTC 2016
Module: rtems
Branch: master
Commit: 86d3a2e8a0baf9693928f525afc27f08bc6349e1
Changeset: http://git.rtems.org/rtems/commit/?id=86d3a2e8a0baf9693928f525afc27f08bc6349e1
Author: Sebastian Huber <sebastian.huber at embedded-brains.de>
Date: Thu Nov 3 07:33:11 2016 +0100
testsupport: Determine worker index via processor
Determine worker index via the current processor index to get consistent
job runs with respect to the cache topology.
---
cpukit/libmisc/testsupport/test.h | 14 +++++--
cpukit/libmisc/testsupport/testparallel.c | 62 +++++++++++++------------------
2 files changed, 36 insertions(+), 40 deletions(-)
diff --git a/cpukit/libmisc/testsupport/test.h b/cpukit/libmisc/testsupport/test.h
index 0afd3e9..b4adebf 100644
--- a/cpukit/libmisc/testsupport/test.h
+++ b/cpukit/libmisc/testsupport/test.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2014 embedded brains GmbH. All rights reserved.
+ * Copyright (c) 2014, 2016 embedded brains GmbH. All rights reserved.
*
* embedded brains GmbH
* Dornierstr. 4
@@ -88,6 +88,10 @@ int rtems_test_end(void);
*/
int rtems_test_printf(const char* format, ...) RTEMS_PRINTFLIKE(1, 2);
+#define RTEMS_TEST_PARALLEL_PROCESSOR_MAX 32
+
+typedef struct rtems_test_parallel_job rtems_test_parallel_job;
+
/**
* @brief Internal context for parallel job execution.
*/
@@ -95,8 +99,10 @@ typedef struct {
Atomic_Ulong stop;
SMP_barrier_Control barrier;
size_t worker_count;
- rtems_id worker_ids[32];
+ rtems_id worker_ids[RTEMS_TEST_PARALLEL_PROCESSOR_MAX];
rtems_id stop_worker_timer_id;
+ const struct rtems_test_parallel_job *jobs;
+ size_t job_count;
} rtems_test_parallel_context;
/**
@@ -118,7 +124,7 @@ typedef void (*rtems_test_parallel_worker_setup)(
/**
* @brief Basic parallel job description.
*/
-typedef struct {
+struct rtems_test_parallel_job {
/**
* @brief Job initialization handler.
*
@@ -186,7 +192,7 @@ typedef struct {
* afterwards and incremented step by step until all processors are used).
*/
bool cascade;
-} rtems_test_parallel_job;
+};
/**
* @brief Indicates if a job body should stop its work loop.
diff --git a/cpukit/libmisc/testsupport/testparallel.c b/cpukit/libmisc/testsupport/testparallel.c
index a2f4573..1182242 100644
--- a/cpukit/libmisc/testsupport/testparallel.c
+++ b/cpukit/libmisc/testsupport/testparallel.c
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2013-2015 embedded brains GmbH. All rights reserved.
+ * Copyright (c) 2013, 2016 embedded brains GmbH. All rights reserved.
*
* embedded brains GmbH
* Dornierstr. 4
@@ -50,13 +50,14 @@ static void start_worker_stop_timer(
static void run_tests(
rtems_test_parallel_context *ctx,
const rtems_test_parallel_job *jobs,
- size_t job_count,
- size_t worker_index
+ size_t job_count
)
{
SMP_barrier_State bs = SMP_BARRIER_STATE_INITIALIZER;
size_t i;
+ _SMP_barrier_Wait(&ctx->barrier, &bs, ctx->worker_count);
+
for (i = 0; i < job_count; ++i) {
const rtems_test_parallel_job *job = &jobs[i];
size_t n = rtems_get_processor_count();
@@ -64,6 +65,19 @@ static void run_tests(
while (j < n) {
size_t active_worker = j + 1;
+ size_t worker_index;
+ rtems_interrupt_level level;
+
+ /*
+ * Determine worker index via the current processor index to get
+ * consistent job runs with respect to the cache topology.
+ */
+ rtems_interrupt_local_disable(level);
+ _SMP_barrier_Wait(&ctx->barrier, &bs, ctx->worker_count);
+ worker_index = rtems_get_current_processor();
+ rtems_interrupt_local_enable(level);
+
+ _Assert(worker_index < ctx->worker_count);
if (rtems_test_parallel_is_master_worker(worker_index)) {
rtems_interval duration = (*job->init)(ctx, job->arg, active_worker);
@@ -85,39 +99,27 @@ static void run_tests(
(*job->fini)(ctx, job->arg, active_worker);
}
+ _SMP_barrier_Wait(&ctx->barrier, &bs, ctx->worker_count);
+
++j;
}
}
}
-typedef struct {
- rtems_test_parallel_context *ctx;
- const rtems_test_parallel_job *jobs;
- size_t job_count;
- size_t worker_index;
-} worker_arg;
-
static void worker_task(rtems_task_argument arg)
{
- worker_arg warg = *(worker_arg *) arg;
+ rtems_test_parallel_context *ctx = (rtems_test_parallel_context *) arg;
rtems_status_code sc;
- sc = rtems_event_transient_send(warg.ctx->worker_ids[0]);
- _Assert(sc == RTEMS_SUCCESSFUL);
(void) sc;
- run_tests(warg.ctx, warg.jobs, warg.job_count, warg.worker_index);
+ run_tests(ctx, ctx->jobs, ctx->job_count);
while (true) {
/* Wait for delete by master worker */
}
}
-static char digit(size_t i, size_t pos)
-{
- return '0' + (i / pos) % 10;
-}
-
void rtems_test_parallel(
rtems_test_parallel_context *ctx,
rtems_test_parallel_worker_setup worker_setup,
@@ -133,6 +135,8 @@ void rtems_test_parallel(
_SMP_barrier_Control_initialize(&ctx->barrier);
ctx->worker_count = rtems_get_processor_count();
ctx->worker_ids[0] = rtems_task_self();
+ ctx->jobs = jobs;
+ ctx->job_count = job_count;
if (RTEMS_ARRAY_SIZE(ctx->worker_ids) < ctx->worker_count) {
rtems_fatal_error_occurred(0xdeadbeef);
@@ -156,21 +160,10 @@ void rtems_test_parallel(
}
for (worker_index = 1; worker_index < ctx->worker_count; ++worker_index) {
- worker_arg warg = {
- .ctx = ctx,
- .jobs = jobs,
- .job_count = job_count,
- .worker_index = worker_index
- };
rtems_id worker_id;
sc = rtems_task_create(
- rtems_build_name(
- 'W',
- digit(worker_index, 100),
- digit(worker_index, 10),
- digit(worker_index, 1)
- ),
+ rtems_build_name('W', 'O', 'R', 'K'),
worker_priority,
RTEMS_MINIMUM_STACK_SIZE,
RTEMS_DEFAULT_MODES,
@@ -187,14 +180,11 @@ void rtems_test_parallel(
(*worker_setup)(ctx, worker_index, worker_id);
}
- sc = rtems_task_start(worker_id, worker_task, (rtems_task_argument) &warg);
- _Assert(sc == RTEMS_SUCCESSFUL);
-
- sc = rtems_event_transient_receive(RTEMS_WAIT, RTEMS_NO_TIMEOUT);
+ sc = rtems_task_start(worker_id, worker_task, (rtems_task_argument) ctx);
_Assert(sc == RTEMS_SUCCESSFUL);
}
- run_tests(ctx, jobs, job_count, 0);
+ run_tests(ctx, jobs, job_count);
for (worker_index = 1; worker_index < ctx->worker_count; ++worker_index) {
sc = rtems_task_delete(ctx->worker_ids[worker_index]);
More information about the vc
mailing list