Change-Id: Iea47c25b3c9b6e1eadf097c34323727181975cca


[ROCm/rocprofiler commit: 4d790c8eee]
Этот коммит содержится в:
Evgeny
2018-07-25 14:53:33 -05:00
родитель 379e305ebd
Коммит 1027ae42f5
11 изменённых файлов: 384 добавлений и 281 удалений
+38 -57
Просмотреть файл
@@ -29,6 +29,9 @@ SOFTWARE.
#include <hsa.h>
#include <hsa_ext_amd.h>
#include <unistd.h> // usleep
#include <atomic>
#include <list>
#include <map>
#include <mutex>
#include <vector>
@@ -49,27 +52,7 @@ class Context;
inline unsigned align_size(unsigned size, unsigned alignment) {
return ((size + alignment - 1) & ~(alignment - 1));
}
#if 0
// Block descriptor
struct block_des_t {
uint32_t id;
uint32_t index;
};
// block_des_t less-then functor
struct lt_block_des {
bool operator()(const block_des_t& a1, const block_des_t& a2) const {
return (a1.id < a2.id) || ((a1.id == a2.id) && (a1.index < a2.index));
}
};
// Block status
struct block_status_t {
uint32_t max_counters;
uint32_t counter_index;
uint32_t group_index;
};
#endif
// Metrics arguments
template <class Map> class MetricArgs : public xml::args_cache_t {
public:
@@ -97,6 +80,9 @@ template <class Map> class MetricArgs : public xml::args_cache_t {
// Profiling group
class Group {
public:
typedef uint32_t refs_t;
typedef std::atomic<refs_t> atomic_refs_t;
Group(const util::AgentInfo* agent_info, Context* context, const uint32_t& index)
: pmc_profile_(agent_info),
sqtt_profile_(agent_info),
@@ -146,10 +132,10 @@ class Group {
Context* GetContext() { return context_; }
uint32_t GetIndex() const { return index_; }
void ResetRefs() { refs_ = n_profiles_; }
uint32_t DecrRefs() {
return (refs_ > 0) ? --refs_ : 0;
}
atomic_refs_t* AtomicRefsCount() { return reinterpret_cast<atomic_refs_t*>(&refs_); }
void ResetRefsCount() { AtomicRefsCount()->store(n_profiles_, std::memory_order_release); }
void IncrRefsCount() { AtomicRefsCount()->fetch_add(1, std::memory_order_acq_rel); }
uint32_t FetchDecrRefsCount() { return AtomicRefsCount()->fetch_sub(1, std::memory_order_acq_rel); }
private:
PmcProfile pmc_profile_;
@@ -159,7 +145,7 @@ class Group {
pkt_vector_t stop_vector_;
pkt_vector_t read_vector_;
uint32_t n_profiles_;
uint32_t refs_;
refs_t refs_;
Context* const context_;
const uint32_t index_;
};
@@ -167,7 +153,6 @@ class Group {
// Profiling context
class Context {
public:
typedef std::mutex mutex_t;
typedef std::map<std::string, rocprofiler_feature_t*> info_map_t;
Context(const util::AgentInfo* agent_info, Queue* queue, rocprofiler_feature_t* info,
@@ -180,6 +165,8 @@ class Context {
handler_(handler),
handler_arg_(handler_arg)
{
if (info_count == 0) return;
metrics_ = MetricsDict::Create(agent_info);
if (metrics_ == NULL) EXC_RAISING(HSA_STATUS_ERROR, "MetricsDict create failed");
if (Initialize(info, info_count) == false) {
@@ -192,7 +179,7 @@ class Context {
if (handler != NULL) {
for (unsigned group_index = 0; group_index < set_.size(); ++group_index) {
set_[group_index].ResetRefs();
set_[group_index].ResetRefsCount();
const profile_vector_t profile_vector = GetProfiles(group_index);
for (auto& tuple : profile_vector) {
// Handler for stop packet completion
@@ -307,11 +294,11 @@ class Context {
}
}
void Reset(const uint32_t& group_index) { set_[group_index].ResetRefs(); }
void Reset(const uint32_t& group_index) { set_[group_index].ResetRefsCount(); }
uint32_t GetGroupCount() const { return set_.size(); }
rocprofiler_group_t GetGroupInfo(Group* g) {
inline rocprofiler_group_t GetGroupInfo(Group* g) {
rocprofiler::info_vector_t& info_vector = g->GetInfoVector();
rocprofiler_group_t group = {};
group.index = g->GetIndex();
@@ -320,8 +307,14 @@ class Context {
group.feature_count = info_vector.size();
return group;
}
rocprofiler_group_t GetGroupInfo(const uint32_t& index) {
return GetGroupInfo(&set_[index]);
inline rocprofiler_group_t GetGroupInfo(const uint32_t& index) {
rocprofiler_group_t group = {};
if (set_.empty()) {
group.context = reinterpret_cast<rocprofiler_t*>(this);
} else {
group = GetGroupInfo(&set_[index]);
}
return group;
}
const pkt_vector_t& StartPackets(const uint32_t& group_index) const {
@@ -368,14 +361,7 @@ class Context {
const profile_vector_t profile_vector = GetProfiles(group_index);
for (auto& tuple : profile_vector) {
// Wait for stop packet to complete
const uint64_t timeout = timeout_;
bool complete = false;
while (!complete) {
const hsa_signal_value_t signal_value = hsa_signal_wait_scacquire(tuple.completion_signal, HSA_SIGNAL_CONDITION_LT, 1, timeout,
HSA_WAIT_STATE_BLOCKED);
complete = (signal_value < 1);
if (!complete) WARN_LOGGING("timeout");
}
hsa_rsrc_->SignalWaitRestore(tuple.completion_signal, 1);
for (rocprofiler_feature_t* rinfo : *(tuple.info_vector)) rinfo->data.kind = ROCPROFILER_DATA_KIND_UNINIT;
callback_data_t callback_data{tuple.profile, tuple.info_vector, tuple.info_vector->size(), NULL};
const hsa_status_t status =
@@ -411,8 +397,19 @@ class Context {
}
}
static void SetTimeout(uint64_t timeout) { timeout_ = timeout; }
static uint64_t GetTimeout() { return timeout_; }
static bool Handler(hsa_signal_value_t value, void* arg) {
Group* group = reinterpret_cast<Group*>(arg);
Context* context = group->GetContext();
auto r = group->FetchDecrRefsCount();
if (r == 1) {
const rocprofiler_group_t group_info = context->GetGroupInfo(group);
context->handler_(group_info, context->handler_arg_);
}
return false;
}
Group* GetGroup(const uint32_t& index) { return &set_[index]; }
rocprofiler_handler_t GetHandler(void** arg) const { *arg = handler_arg_; return handler_; }
private:
// Getting profling packets
@@ -425,18 +422,6 @@ class Context {
return vec;
}
static bool Handler(hsa_signal_value_t value, void* arg) {
Group* group = reinterpret_cast<Group*>(arg);
Context* context = group->GetContext();
context->mutex_.lock();
uint32_t r = group->DecrRefs();
context->mutex_.unlock();
if (r == 0) {
return context->handler_(context->GetGroupInfo(group), context->handler_arg_);
}
return false;
}
static hsa_status_t DataCallback(hsa_ven_amd_aqlprofile_info_type_t ainfo_type,
hsa_ven_amd_aqlprofile_info_data_t* ainfo_data, void* data) {
hsa_status_t status = HSA_STATUS_SUCCESS;
@@ -526,9 +511,6 @@ class Context {
return info;
}
// Profiling data waiting timeout
static uint64_t timeout_;
// GPU handel
const hsa_agent_t agent_;
const util::AgentInfo* agent_info_;
@@ -551,7 +533,6 @@ class Context {
// Context completion handler
rocprofiler_handler_t handler_;
void* handler_arg_;
mutex_t mutex_;
};
} // namespace rocprofiler