Gbaraldi/att tool (#766)

* Enabling codeobj and thread trace samples

* Updating aqlprofile_v2 header

* Codeobj and thread trace samples with output log files

* Fixing clang format

* Cmake formatting

* Adding coverage to codeobj

* Comment trace sample

* Adding ATT Parser API

* Fixing forwarding to aqlprofile

* Clang formatting

* Clang tidy

* Adding option to print memory kernels

* Clang format

* Remove default from switch case

* Separating  client/main on codeobj sample for ASAn

* Formatting

* Gbaraldi/att tool rebase (#801)

* Enabling codeobj and thread trace samples

* Updating aqlprofile_v2 header

* Codeobj and thread trace samples with output log files

* Fixing clang format

* Cmake formatting

* Adding coverage to codeobj

* Comment trace sample

* Removing python from workflow

* Adding ATT Parser API

* Fixing forwarding to aqlprofile

* Clang formatting

* Clang tidy

* Adding option to print memory kernels

* Clang format

* Remove default from switch case

* Separating  client/main on codeobj sample for ASAn

* Formatting

* Enabling codeobj and thread trace samples

* Updating aqlprofile_v2 header

* Codeobj and thread trace samples with output log files

* Fixing clang format

* Cmake formatting

* Adding coverage to codeobj

* Comment trace sample

* Adding ATT Parser API

* Fixing forwarding to aqlprofile

* Clang formatting

* Clang tidy

* Adding option to print memory kernels

* Clang format

* Remove default from switch case

* Separating  client/main on codeobj sample for ASAn

* Formatting

* Fix codeobj library

* Allow thread trace in parallel with other service

* Zeroing the HSA signals

* Adding exception wrappers in ATT sample

* Removed force configure

* Remove force configure from ISA decode

* Removing codecov flag

* Gbaraldi/att tool tests (#828)

* Adding tests for codeobj ISA decode

* Adding ATT tests

* Adding ATT integration tests

* Formatting

* Changing codeobj binary extension

* Renaming codeobj library spaces

* Fixing samples

* Formatting

* Formatting

* Fixing int test

* Fixing linker error

* Fixing memory fault

* Moving kernel ot inside namespace

* ASAN linking fix

* Removing unecessary headers

* Formatting

* Fixing target_cu

* Remove codeobj binary

* Revert "Remove codeobj binary"

This reverts commit 7d286f89d8096bc36925cd79cd742a5e6d10d179.

* Enable memory snapshot

* adding comgr

---------

Co-authored-by: Ammar ELWazir <ammar.elwazir@amd.com>
This commit is contained in:
Giovanni Lenzi Baraldi
2024-05-03 18:45:47 -03:00
committed by GitHub
parent 6d3fbcffad
commit 099ac7c72d
38 changed files with 1829 additions and 753 deletions
@@ -27,21 +27,10 @@
#include "lib/rocprofiler-sdk/hsa/agent_cache.hpp"
extern "C" {
/**
* @brief Configure buffered dispatch profile Counting Service.
* Collects the counters in dispatch packets and stores them
* in buffer_id. The buffer may contain packets from more than
* one dispatch (denoted by correlation id). Will trigger the
* callback based on the parameters setup in buffer_id_t.
*
* @param [in] context_id context id
* @param [in] buffer_id id of the buffer to use for the counting service
* @param [in] profile profile config to use for dispatch
* @return ::rocprofiler_status_t
*/
rocprofiler_status_t ROCPROFILER_API
rocprofiler_configure_thread_trace_service(rocprofiler_context_id_t context_id,
rocprofiler_att_parameters_t parameters,
rocprofiler_att_parameter_t* parameters,
size_t num_parameters,
rocprofiler_att_dispatch_callback_t dispatch_callback,
rocprofiler_att_shader_data_callback_t shader_callback,
void* callback_userdata)
@@ -49,30 +38,43 @@ rocprofiler_configure_thread_trace_service(rocprofiler_context_id_t
auto* ctx = rocprofiler::context::get_mutable_registered_context(context_id);
if(!ctx) return ROCPROFILER_STATUS_ERROR_CONTEXT_NOT_STARTED;
if(ctx->thread_trace) return ROCPROFILER_STATUS_ERROR_SERVICE_ALREADY_CONFIGURED;
if(parameters.flags.raw != 0) return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED;
auto thread_tracer = std::make_shared<rocprofiler::thread_trace_parameters>();
auto thread_tracer = std::make_shared<rocprofiler::thread_trace_parameter_pack>();
thread_tracer->context_id = context_id;
thread_tracer->dispatch_cb_fn = dispatch_callback;
thread_tracer->shader_cb_fn = shader_callback;
thread_tracer->callback_userdata = callback_userdata;
thread_tracer->flags = parameters.flags;
thread_tracer->buffer_size = parameters.buffer_size;
thread_tracer->target_cu = parameters.target_cu;
thread_tracer->simd_select = parameters.simd_select;
thread_tracer->vmid_mask = parameters.vmid_mask;
for(size_t p = 0; p < num_parameters; p++)
{
const rocprofiler_att_parameter_t& param = parameters[p];
if(param.type > ROCPROFILER_ATT_PARAMETER_LAST)
return ROCPROFILER_STATUS_ERROR_INVALID_ARGUMENT;
thread_tracer->perfcounter_mask = parameters.perfcounter_mask;
thread_tracer->perfcounter_ctrl = parameters.perfcounter_ctrl;
for(int i = 0; i < parameters.perfcounter_num; i++)
thread_tracer->perfcounters.emplace_back(parameters.perfcounter[i]);
thread_tracer->shader_engine_mask = 0;
for(int i = 0; i < parameters.shader_num; i++)
thread_tracer->shader_engine_mask |= 1ul << parameters.shader_ids[i];
switch(param.type)
{
case ROCPROFILER_ATT_PARAMETER_TARGET_CU: thread_tracer->target_cu = param.value; break;
case ROCPROFILER_ATT_PARAMETER_SHADER_ENGINE_MASK:
thread_tracer->shader_engine_mask = param.value;
break;
case ROCPROFILER_ATT_PARAMETER_BUFFER_SIZE:
thread_tracer->buffer_size = param.value;
break;
case ROCPROFILER_ATT_PARAMETER_SIMD_SELECT:
thread_tracer->simd_select = param.value;
break;
case ROCPROFILER_ATT_PARAMETER_OCCUPANCY_MODE_ENABLE:
return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED;
case ROCPROFILER_ATT_PARAMETER_PERFCOUNTERS_CTRL:
return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED;
case ROCPROFILER_ATT_PARAMETER_PERFCOUNTER:
return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED;
case ROCPROFILER_ATT_PARAMETER_LAST: return ROCPROFILER_STATUS_ERROR_INVALID_ARGUMENT;
}
// for(int i = 0; i < parameters.perfcounter_num; i++)
// thread_tracer->perfcounters.emplace_back(parameters.perfcounter[i]);
}
ctx->thread_trace = std::make_shared<rocprofiler::ThreadTracer>(thread_tracer);