AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Only check lock if counter collection is used

Browse Source 
Only check the lock if counter collection is being
attempted by the profiler. Not when it is loaded.

SWDEV-474455

Change-Id: Ie14de3c8db57e0cbd279ffca51c333a375ca8654
Signed-off-by: Benjamin Welton <bewelton@amd.com>
This commit is contained in:
Benjamin Welton
2024-07-18 21:55:45 +00:00
committed by Saurabh Verma
parent 95ed584e6d
commit 0be3e07388
5 changed files with 42 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/api/rocprofilerv2.cpp
+23 -1
View File
@@ -325,6 +325,23 @@ ROCPROFILER_API rocprofiler_status_t rocprofiler_create_filter(
// rocprofiler::Exception(ROCPROFILER_STATUS_ERROR_FILTER_DATA_CORRUPTED); if (error_code == 0)
// throw rocprofiler::Exception(ROCPROFILER_STATUS_ERROR_SESSION_FILTER_DATA_MISMATCH);
auto& rocprofiler_singleton = rocprofiler::ROCProfiler_Singleton::GetInstance();
switch (filter_kind) {
case ROCPROFILER_COUNTERS_COLLECTION:
case ROCPROFILER_COUNTERS_SAMPLER:
try{
ProfilingLock::Lock(PROFILER_V2_LOCK);
}catch(std::exception& e){
std::cout << e.what();
abort();
}
break;
case ROCPROFILER_DISPATCH_TIMESTAMPS_COLLECTION:
case ROCPROFILER_PC_SAMPLING_COLLECTION:
case ROCPROFILER_ATT_TRACE_COLLECTION:
case ROCPROFILER_SPM_COLLECTION:
case ROCPROFILER_API_TRACE:
break;
}
if (!rocprofiler_singleton.FindSession(session_id))
throw rocprofiler::Exception(ROCPROFILER_STATUS_ERROR_SESSION_NOT_FOUND);
*filter_id = rocprofiler_singleton
@@ -507,6 +524,12 @@ ROCPROFILER_API rocprofiler_status_t rocprofiler_device_profiling_session_create
int cpu_index, int gpu_index) {
API_METHOD_PREFIX
std::vector<std::string> counters(counter_names, counter_names + num_counters);
try {
ProfilingLock::Lock(PROFILER_V2_LOCK);
} catch (std::exception& e) {
std::cout << e.what();
abort();
}
*session_id =
rocprofiler::ROCProfiler_Singleton::GetInstance().CreateDeviceProfilingSession(counters, cpu_index, gpu_index);
API_METHOD_SUFFIX
@@ -618,7 +641,6 @@ ROCPROFILER_EXPORT bool OnLoad(HsaApiTable* table, uint64_t runtime_version,
uint64_t failed_tool_count, const char* const* failed_tool_names) {
if (started) rocprofiler::fatal("HSA Tool started already!");
started = true;
ProfilingLock::Lock(PROFILER_V2_LOCK);
rocprofiler::HSASupport_Singleton::GetInstance().HSAInitialize(table);
return true;
}
src/core/counters/metrics/derived_counters.xml
+6
View File
@@ -229,6 +229,9 @@
<metric name="TOTAL_16_OPS" expr=(SQ_INSTS_VALU_FMA_F16*2+SQ_INSTS_VALU_ADD_F16+SQ_INSTS_VALU_MUL_F16+SQ_INSTS_VALU_TRANS_F16)*64+((SQ_INSTS_VALU_MFMA_MOPS_F16+SQ_INSTS_VALU_MFMA_MOPS_BF16)*512) descr="The number of 16 bits OPS executed"></metric>
<metric name="TOTAL_32_OPS" expr=(SQ_INSTS_VALU_FMA_F32*2+SQ_INSTS_VALU_INT32+SQ_INSTS_VALU_ADD_F32+SQ_INSTS_VALU_MUL_F32+SQ_INSTS_VALU_TRANS_F32)*64+(SQ_INSTS_VALU_MFMA_MOPS_F32*512) descr="The number of 32 bits OPS executed"></metric>
<metric name="TOTAL_64_OPS" expr=(SQ_INSTS_VALU_FMA_F64*2+SQ_INSTS_VALU_INT64+SQ_INSTS_VALU_ADD_F64+SQ_INSTS_VALU_MUL_F64)*64+(SQ_INSTS_VALU_MFMA_MOPS_F64*512) descr="The number of 64 bits OPS executed"></metric>
<metric name="ACTIVE_CYCLES" expr=GRBM_GUI_ACTIVE descr="Active Cycles"></metric>
<metric name="ELAPSED_CYCLES" expr=GRBM_COUNT descr="Elapsed Cycles"></metric>
<metric name="ACTIVE_WAVES" expr=SQ_WAVES descr="Active Waves"></metric>
<metric name="AggSysCycles" expr=GRBM_GUI_ACTIVE*CU_NUM descr="Unit: cycles"></metric>
## IP Block Utilization Metrics
@@ -482,6 +485,9 @@
<metric name="TOTAL_16_OPS" expr=(SQ_INSTS_VALU_FMA_F16*2+SQ_INSTS_VALU_ADD_F16+SQ_INSTS_VALU_MUL_F16+SQ_INSTS_VALU_TRANS_F16)*64+((SQ_INSTS_VALU_MFMA_MOPS_F16+SQ_INSTS_VALU_MFMA_MOPS_BF16)*512) descr="The number of 16 bits OPS executed"></metric>
<metric name="TOTAL_32_OPS" expr=(SQ_INSTS_VALU_FMA_F32*2+SQ_INSTS_VALU_INT32+SQ_INSTS_VALU_ADD_F32+SQ_INSTS_VALU_MUL_F32+SQ_INSTS_VALU_TRANS_F32)*64+(SQ_INSTS_VALU_MFMA_MOPS_F32*512) descr="The number of 32 bits OPS executed"></metric>
<metric name="TOTAL_64_OPS" expr=(SQ_INSTS_VALU_FMA_F64*2+SQ_INSTS_VALU_INT64+SQ_INSTS_VALU_ADD_F64+SQ_INSTS_VALU_MUL_F64)*64+(SQ_INSTS_VALU_MFMA_MOPS_F64*512) descr="The number of 64 bits OPS executed"></metric>
<metric name="ACTIVE_CYCLES" expr=GRBM_GUI_ACTIVE/XCC_NUM descr="Active Cycles"></metric>
<metric name="ELAPSED_CYCLES" expr=GRBM_COUNT/XCC_NUM descr="Elapsed Cycles"></metric>
<metric name="ACTIVE_WAVES" expr=SQ_WAVES descr="Active Waves"></metric>
</gfx940_expr>
<gfx10_expr>
src/core/profiling_lock.cpp
+9 -7
View File
@@ -9,8 +9,8 @@
#include <sstream>
#include <sstream>
#include <cstring>
#include "util/exception.h"
#include "profiling_lock.h"
#include <stdexcept>
#define ROCPROFILER_LOCK_FILE "/tmp/rocprofiler_process.lock"
#define ROCPROFILER_PID_FILE "/tmp/rocprofiler.pid"
@@ -67,12 +67,12 @@ bool check_process_exists(int pid) {
}
void terminate_current_profiler_instance() {
EXC_RAISING(
0,
"\nA profiling instance already exists! Multiple profiling instances are not "
"allowed.\nCheck "
<< ROCPROFILER_PID_FILE
<< " and kill the process, delete this .pid file and try again.\nTerminating ...\n");
std::stringstream oss;
oss << "\nA profiling instance already exists! Multiple profiling instances are not "
<< "allowed.\nCheck " << ROCPROFILER_PID_FILE
<< " and kill the process, delete this .pid file and try again.\nTerminating "
"...\n";
throw std::runtime_error(oss.str());
}
@@ -95,6 +95,8 @@ void ProfilingLock::Lock(LockMode mode) {
bool is_standalone_mode_v1 = check_standalone_mode() && (mode == PROFILER_V1_LOCK);
ProfilingLock* profiling_lock = Instance();
// Check if we have already locked in this process
if (profiling_lock->already_locked.exchange(true)) return;
if (file_exists(profiling_lock->pid_file)) {
profiling_lock->lock = acquire_lock(profiling_lock->lock_file);
if (profiling_lock->lock < 1) {
src/core/profiling_lock.h
+2
View File
@@ -1,6 +1,7 @@
#ifndef _SRC_CORE_PROFILING_LOCK_H
#define _SRC_CORE_PROFILING_LOCK_H
#include <atomic>
enum LockMode{
PROFILER_V1_LOCK,
@@ -18,6 +19,7 @@ private:
const char *lock_file;
const char *pid_file;
std::atomic<bool> already_locked{false};
int lock;
};
src/core/rocprofiler.cpp
+2 -1
View File
@@ -394,7 +394,6 @@ ROCPROFILER_EXPORT extern const uint32_t HSA_AMD_TOOL_PRIORITY = 25;
PUBLIC_API bool OnLoad(HsaApiTable* table, uint64_t runtime_version, uint64_t failed_tool_count,
const char* const* failed_tool_names) {
ONLOAD_TRACE_BEG();
ProfilingLock::Lock(PROFILER_V1_LOCK);
rocprofiler::SaveHsaApi(table);
rocprofiler::ProxyQueue::InitFactory();
@@ -488,6 +487,7 @@ PUBLIC_API hsa_status_t rocprofiler_open(hsa_agent_t agent, rocprofiler_feature_
uint32_t feature_count, rocprofiler_t** handle,
uint32_t mode, rocprofiler_properties_t* properties) {
API_METHOD_PREFIX
ProfilingLock::Lock(PROFILER_V1_LOCK);
rocprofiler::util::HsaRsrcFactory* hsa_rsrc = &rocprofiler::util::HsaRsrcFactory::Instance();
const rocprofiler::util::AgentInfo* agent_info = hsa_rsrc->GetAgentInfo(agent);
if (agent_info == NULL) {
@@ -673,6 +673,7 @@ rocprofiler_pool_open(hsa_agent_t agent, // GPU handle
rocprofiler_pool_properties_t* properties) // pool properties
{
API_METHOD_PREFIX
ProfilingLock::Lock(PROFILER_V1_LOCK);
rocprofiler::util::HsaRsrcFactory* hsa_rsrc = &rocprofiler::util::HsaRsrcFactory::Instance();
const rocprofiler::util::AgentInfo* agent_info = hsa_rsrc->GetAgentInfo(agent);
if (agent_info == NULL) {