This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] [og9] OpenACC profiling support for AMD GCN
- From: Julian Brown <julian at codesourcery dot com>
- To: <gcc-patches at gcc dot gnu dot org>
- Cc: Thomas Schwinge <thomas at codesourcery dot com>, Andrew Stubbs <ams at codesourcery dot com>
- Date: Fri, 6 Sep 2019 09:02:13 -0700
- Subject: [PATCH] [og9] OpenACC profiling support for AMD GCN
- Ironport-sdr: w3EkFeZT3KJwFUos63GM29YPQ1px4++XaMxh9WBMUoCYgZTQAQ/Lr5ns3lmtyblR6tSzCiRxB3 0W8D9q0Uy8rtGSSf3E+oelFUzq0/ritV2rEBU25ylSjiynsRgz7sVSmWuSbJmsIpo1Ma9n0z0v sPhktM1dWKmMCTs66r15AoVAGrY95U5N3FfJ/7sfITnUHVhbZ/+vPwKKdG36k2Y4Z6T+XzjhYE 9lqgFN+LPf4LM58uGyZRLq5+l03a7bCgEOI/cyDWOcbOTJV7dvyxlC/eYqspU7tudDOARZhfq8 35o=
- Ironport-sdr: Hh7hNmUrl6dav0zG5Pl5lsVUkhaTc2+feNdbx/8uSGll0EHvpGLAHVSDvt48gTLjsECoaEbeaC 4oi1iyHRGxPL00wY7A69XflwzUwSnjfTkO7weiaqpUOIdVvV8P//A4E4yaNGmFWS8KRiSxiTeP 1O7IsEdCH0Y0HEyV1jB2ixFotaA5uAYWtN40DE8PuEI021vM1oVJCDMB0q7NKzZR4W5VZHg3Hd ZNkjQvuveAja2zZDdJCHDxxDGry+P016b7/906iFtC58R6MnEYsXeoh0k1DbAIKISQHgiJOXzp kK4=
- References: <20190906160213.69722-1-julian@codesourcery.com>
This patch adds profiling support to the AMD GCN libgomp plugin, modeled
after the equivalent support in the NVPTX plugin. This gives a positive
test delta in AMD GCN offload testing.
I will apply to the openacc-gcc-9-branch shortly.
Julian
2019-09-06 Julian Brown <julian@codesourcery.com>
libgomp/
* plugin/plugin-gcn.c (GOMP_OFFLOAD_alloc_by_agent,
GOMP_OFFLOAD_free, gcn_exec): Add profiling support.
* testsuite/libgomp.oacc-c-c++-common/acc_prof-init-1.c: Add GCN
support.
* testsuite/libgomp.oacc-c-c++-common/acc_prof-kernels-1.c: Likewise.
* testsuite/libgomp.oacc-c-c++-common/acc_prof-parallel-1.c: Likewise.
---
libgomp/ChangeLog.openacc | 9 ++
libgomp/plugin/plugin-gcn.c | 96 +++++++++++++++++++
.../acc_prof-init-1.c | 2 +
.../acc_prof-kernels-1.c | 4 +
.../acc_prof-parallel-1.c | 12 +++
5 files changed, 123 insertions(+)
diff --git a/libgomp/ChangeLog.openacc b/libgomp/ChangeLog.openacc
index d7a4c7a5f8a..8ed0a10a589 100644
--- a/libgomp/ChangeLog.openacc
+++ b/libgomp/ChangeLog.openacc
@@ -1,3 +1,12 @@
+2019-09-06 Julian Brown <julian@codesourcery.com>
+
+ * plugin/plugin-gcn.c (GOMP_OFFLOAD_alloc_by_agent,
+ GOMP_OFFLOAD_free, gcn_exec): Add profiling support.
+ * testsuite/libgomp.oacc-c-c++-common/acc_prof-init-1.c: Add GCN
+ support.
+ * testsuite/libgomp.oacc-c-c++-common/acc_prof-kernels-1.c: Likewise.
+ * testsuite/libgomp.oacc-c-c++-common/acc_prof-parallel-1.c: Likewise.
+
2019-09-06 Julian Brown <julian@codesourcery.com>
* config/gcn/target.c (omp_pause_resource, omp_pause_resource_all): New
diff --git a/libgomp/plugin/plugin-gcn.c b/libgomp/plugin/plugin-gcn.c
index f0b22ebc3d7..2f273967bad 100644
--- a/libgomp/plugin/plugin-gcn.c
+++ b/libgomp/plugin/plugin-gcn.c
@@ -3024,6 +3024,35 @@ GOMP_OFFLOAD_alloc_by_agent (struct agent_info *agent, size_t size)
return NULL;
}
+ struct goacc_thread *thr = GOMP_PLUGIN_goacc_thread ();
+ bool profiling_dispatch_p
+ = __builtin_expect (thr != NULL && thr->prof_info != NULL, false);
+ if (profiling_dispatch_p)
+ {
+ acc_prof_info *prof_info = thr->prof_info;
+ acc_event_info data_event_info;
+ acc_api_info *api_info = thr->api_info;
+
+ prof_info->event_type = acc_ev_alloc;
+
+ data_event_info.data_event.event_type = prof_info->event_type;
+ data_event_info.data_event.valid_bytes
+ = _ACC_DATA_EVENT_INFO_VALID_BYTES;
+ data_event_info.data_event.parent_construct
+ = acc_construct_parallel;
+ data_event_info.data_event.implicit = 1;
+ data_event_info.data_event.tool_info = NULL;
+ data_event_info.data_event.var_name = NULL;
+ data_event_info.data_event.bytes = size;
+ data_event_info.data_event.host_ptr = NULL;
+ data_event_info.data_event.device_ptr = (void *) ptr;
+
+ api_info->device_api = acc_device_api_other;
+
+ GOMP_PLUGIN_goacc_profiling_dispatch (prof_info, &data_event_info,
+ api_info);
+ }
+
return ptr;
}
@@ -3046,6 +3075,35 @@ GOMP_OFFLOAD_free (int device, void *ptr)
return false;
}
+ struct goacc_thread *thr = GOMP_PLUGIN_goacc_thread ();
+ bool profiling_dispatch_p
+ = __builtin_expect (thr != NULL && thr->prof_info != NULL, false);
+ if (profiling_dispatch_p)
+ {
+ acc_prof_info *prof_info = thr->prof_info;
+ acc_event_info data_event_info;
+ acc_api_info *api_info = thr->api_info;
+
+ prof_info->event_type = acc_ev_free;
+
+ data_event_info.data_event.event_type = prof_info->event_type;
+ data_event_info.data_event.valid_bytes
+ = _ACC_DATA_EVENT_INFO_VALID_BYTES;
+ data_event_info.data_event.parent_construct
+ = acc_construct_parallel;
+ data_event_info.data_event.implicit = 1;
+ data_event_info.data_event.tool_info = NULL;
+ data_event_info.data_event.var_name = NULL;
+ data_event_info.data_event.bytes = 0;
+ data_event_info.data_event.host_ptr = NULL;
+ data_event_info.data_event.device_ptr = (void *) ptr;
+
+ api_info->device_api = acc_device_api_other;
+
+ GOMP_PLUGIN_goacc_profiling_dispatch (prof_info, &data_event_info,
+ api_info);
+ }
+
return true;
}
@@ -3276,6 +3334,35 @@ gcn_exec (struct kernel_info *kernel, size_t mapnum, void **hostaddrs,
{1, 64, 16}
};
+ struct goacc_thread *thr = GOMP_PLUGIN_goacc_thread ();
+ acc_prof_info *prof_info = thr->prof_info;
+ acc_event_info enqueue_launch_event_info;
+ acc_api_info *api_info = thr->api_info;
+ bool profiling_dispatch_p = __builtin_expect (prof_info != NULL, false);
+ if (profiling_dispatch_p)
+ {
+ prof_info->event_type = acc_ev_enqueue_launch_start;
+
+ enqueue_launch_event_info.launch_event.event_type
+ = prof_info->event_type;
+ enqueue_launch_event_info.launch_event.valid_bytes
+ = _ACC_LAUNCH_EVENT_INFO_VALID_BYTES;
+ enqueue_launch_event_info.launch_event.parent_construct
+ = acc_construct_parallel;
+ enqueue_launch_event_info.launch_event.implicit = 1;
+ enqueue_launch_event_info.launch_event.tool_info = NULL;
+ enqueue_launch_event_info.launch_event.kernel_name
+ = (char *) kernel->name;
+ enqueue_launch_event_info.launch_event.num_gangs = kla.gdims[0];
+ enqueue_launch_event_info.launch_event.num_workers = kla.gdims[2];
+ enqueue_launch_event_info.launch_event.vector_length = kla.gdims[1];
+
+ api_info->device_api = acc_device_api_other;
+
+ GOMP_PLUGIN_goacc_profiling_dispatch (prof_info,
+ &enqueue_launch_event_info, api_info);
+ }
+
if (!async)
{
run_kernel (kernel, ind_da, &kla, NULL, false);
@@ -3289,6 +3376,15 @@ gcn_exec (struct kernel_info *kernel, size_t mapnum, void **hostaddrs,
aq->agent->device_id, aq->id, ind_da);
queue_push_callback (aq, gomp_offload_free, ind_da);
}
+
+ if (profiling_dispatch_p)
+ {
+ prof_info->event_type = acc_ev_enqueue_launch_end;
+ enqueue_launch_event_info.launch_event.event_type = prof_info->event_type;
+ GOMP_PLUGIN_goacc_profiling_dispatch (prof_info,
+ &enqueue_launch_event_info,
+ api_info);
+ }
}
void
diff --git a/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-init-1.c b/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-init-1.c
index 6a44e8ffb6a..cf980f1baec 100644
--- a/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-init-1.c
+++ b/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-init-1.c
@@ -224,6 +224,8 @@ static void cb_compute_construct_end (acc_prof_info *prof_info, acc_event_info *
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
diff --git a/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-kernels-1.c b/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-kernels-1.c
index 269b4398478..9c1cfbe292c 100644
--- a/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-kernels-1.c
+++ b/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-kernels-1.c
@@ -107,6 +107,8 @@ static void cb_enqueue_launch_start (acc_prof_info *prof_info, acc_event_info *e
assert (event_info->launch_event.vector_length >= 1);
else if (acc_device_type == acc_device_nvidia) /* ... is special. */
assert (event_info->launch_event.vector_length == 32);
+ else if (acc_device_type == acc_device_gcn) /* ...and so is this. */
+ assert (event_info->launch_event.vector_length == 64);
else
{
#ifdef __OPTIMIZE__
@@ -119,6 +121,8 @@ static void cb_enqueue_launch_start (acc_prof_info *prof_info, acc_event_info *e
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
diff --git a/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-parallel-1.c b/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-parallel-1.c
index 116b9b538a6..5d392511592 100644
--- a/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-parallel-1.c
+++ b/libgomp/testsuite/libgomp.oacc-c-c++-common/acc_prof-parallel-1.c
@@ -265,6 +265,8 @@ static void cb_enter_data_end (acc_prof_info *prof_info, acc_event_info *event_i
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
@@ -319,6 +321,8 @@ static void cb_exit_data_start (acc_prof_info *prof_info, acc_event_info *event_
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
@@ -371,6 +375,8 @@ static void cb_exit_data_end (acc_prof_info *prof_info, acc_event_info *event_in
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
@@ -510,6 +516,8 @@ static void cb_compute_construct_end (acc_prof_info *prof_info, acc_event_info *
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
@@ -573,6 +581,8 @@ static void cb_enqueue_launch_start (acc_prof_info *prof_info, acc_event_info *e
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
@@ -637,6 +647,8 @@ static void cb_enqueue_launch_end (acc_prof_info *prof_info, acc_event_info *eve
if (acc_device_type == acc_device_host)
assert (api_info->device_api == acc_device_api_none);
+ else if (acc_device_type == acc_device_gcn)
+ assert (api_info->device_api == acc_device_api_other);
else
assert (api_info->device_api == acc_device_api_cuda);
assert (api_info->valid_bytes == _ACC_API_INFO_VALID_BYTES);
--
2.22.0