rocDecode Buffer Tracing Support (#315)

* Added buffer tracing support for rocdecode and updated tests to work with buffer tracing

* Updated perfetto to output args individually rather than as a string list

* Updated docstrings and operation type, changed OTF2 code to remove warning due to change in operation type

* Updated tests for review comments

* Test args exist and return value

* Updated to use string entry

* Change function name

* Updated PR to reflect review comments

* Updated for PR review comments

* Change function name
This commit is contained in:
Trowbridge, Ian
2025-04-11 16:56:36 -05:00
committed by GitHub
parent 379d760fc1
commit 077723337a
28 changed files with 391 additions and 146 deletions
+2
View File
@@ -503,6 +503,8 @@ Here are the contents of ``rocdecode_api_trace.csv`` file:
:widths: 10,10,10,10,10,20,20
:header-rows: 1
Perfetto will also show rocDeocde API arguments. Pointers will not be dereferenced and only the address will be displayed.
rocJPEG trace
+++++++++++++++
@@ -27,6 +27,8 @@
#include <rocprofiler-sdk/fwd.h>
#include <rocprofiler-sdk/hip/api_args.h>
#include <rocprofiler-sdk/kfd/page_migration_args.h>
#include <rocprofiler-sdk/rocdecode/api_args.h>
#include <rocprofiler-sdk/rocdecode/api_id.h>
#include <stdint.h>
@@ -213,7 +215,7 @@ typedef struct rocprofiler_buffer_tracing_rocdecode_api_record_t
{
uint64_t size; ///< size of this struct
rocprofiler_buffer_tracing_kind_t kind;
rocprofiler_tracing_operation_t operation;
rocprofiler_rocdecode_api_id_t operation;
rocprofiler_correlation_id_t correlation_id; ///< correlation ids for record
rocprofiler_timestamp_t start_timestamp; ///< start time in nanoseconds
rocprofiler_timestamp_t end_timestamp; ///< end time in nanoseconds
@@ -225,6 +227,29 @@ typedef struct rocprofiler_buffer_tracing_rocdecode_api_record_t
/// @brief Specification of the API function, e.g., ::rocprofiler_rocdecode_api_id_t
} rocprofiler_buffer_tracing_rocdecode_api_record_t;
/**
* @brief An extended ROCProfiler rocDecode API Tracer Record which includes function arguments.
* Pointers are not dereferenced.
*/
typedef struct rocprofiler_buffer_tracing_rocdecode_api_ext_record_t
{
uint64_t size; ///< size of this struct
rocprofiler_buffer_tracing_kind_t kind;
rocprofiler_rocdecode_api_id_t operation;
rocprofiler_correlation_id_t correlation_id; ///< correlation ids for record
rocprofiler_timestamp_t start_timestamp; ///< start time in nanoseconds
rocprofiler_timestamp_t end_timestamp; ///< end time in nanoseconds
rocprofiler_thread_id_t thread_id; ///< id for thread generating this record
rocprofiler_rocdecode_api_args_t args; ///< arguments of function call
rocprofiler_rocdecode_api_retval_t retval; ///< return value of function call
/// @var kind
/// @brief ::ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT
/// @var operation
/// @brief Specification of the API function (@see
/// ::rocprofiler_rocdecode_api_id_t)
} rocprofiler_buffer_tracing_rocdecode_api_ext_record_t;
/**
* @brief ROCProfiler Buffer rocJPEG API Record.
*/
@@ -241,7 +266,8 @@ typedef struct rocprofiler_buffer_tracing_rocjpeg_api_record_t
/// @var kind
/// @brief ::ROCPROFILER_CALLBACK_TRACING_ROCJPEG_API
/// @var operation
/// @brief Specification of the API function, e.g., ::rocprofiler_rocjpeg_api_id_t
/// @brief Specification of the API function (@see
/// ::rocprofiler_rocjpeg_api_id_t)
} rocprofiler_buffer_tracing_rocjpeg_api_record_t;
/**
@@ -469,7 +469,7 @@ template <typename ArchiveT>
void
save(ArchiveT& ar, rocprofiler_rocdecode_api_retval_t data)
{
ROCP_SDK_SAVE_DATA_FIELD(rocDecStatus_retval);
ROCP_SDK_SAVE_DATA_FIELD(uint64_t_retval);
}
template <typename ArchiveT>
@@ -605,6 +605,16 @@ save(ArchiveT& ar, rocprofiler_buffer_tracing_rocdecode_api_record_t data)
save_buffer_tracing_api_record(ar, data);
}
template <typename ArchiveT>
void
save(ArchiveT& ar, rocprofiler_buffer_tracing_rocdecode_api_ext_record_t data)
{
save_buffer_tracing_api_record(ar, data);
auto args = sdk::serialization::get_buffer_tracing_args(data);
ROCP_SDK_SAVE_VALUE("args", args);
ROCP_SDK_SAVE_DATA_FIELD(retval);
}
template <typename ArchiveT>
void
save(ArchiveT& ar, rocprofiler_buffer_tracing_rocjpeg_api_record_t data)
+4
View File
@@ -211,6 +211,7 @@ typedef enum rocprofiler_buffer_tracing_kind_t // NOLINT(performance-enum-size)
ROCPROFILER_BUFFER_TRACING_HIP_STREAM, ///< @see ::rocprofiler_hip_stream_operation_t
ROCPROFILER_BUFFER_TRACING_HIP_RUNTIME_API_EXT,
ROCPROFILER_BUFFER_TRACING_HIP_COMPILER_API_EXT,
ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT,
ROCPROFILER_BUFFER_TRACING_LAST,
/// @var ROCPROFILER_BUFFER_TRACING_HIP_RUNTIME_API_EXT
@@ -219,6 +220,9 @@ typedef enum rocprofiler_buffer_tracing_kind_t // NOLINT(performance-enum-size)
/// @var ROCPROFILER_BUFFER_TRACING_HIP_COMPILER_API_EXT
/// @brief Similar to ROCPROFILER_BUFFER_TRACING_HIP_COMPILER_API except the buffer record
/// contains the function argument(s) and return value
/// @var ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT
/// @brief Similar to ROCPROFILER_BUFFER_TRACING_ROCDECODE_API except the buffer record
/// contains the function argument(s) and return value
} rocprofiler_buffer_tracing_kind_t;
/**
@@ -51,6 +51,7 @@ typedef struct rocprofiler_rocdecode_api_no_args
typedef union rocprofiler_rocdecode_api_retval_t
{
uint64_t uint64_t_retval;
int32_t rocDecStatus_retval;
const char* const_charp_retval;
} rocprofiler_rocdecode_api_retval_t;
+1 -1
View File
@@ -160,7 +160,7 @@ using pc_sampling_host_trap_buffered_output_t =
buffered_output<rocprofiler::tool::rocprofiler_tool_pc_sampling_host_trap_record_t,
domain_type::PC_SAMPLING_HOST_TRAP>;
using rocdecode_buffered_output_t =
buffered_output<rocprofiler_buffer_tracing_rocdecode_api_record_t, domain_type::ROCDECODE>;
buffered_output<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t, domain_type::ROCDECODE>;
using rocjpeg_buffered_output_t =
buffered_output<rocprofiler_buffer_tracing_rocjpeg_api_record_t, domain_type::ROCJPEG>;
using kernel_dispatch_buffered_output_with_stream_t =
+4 -4
View File
@@ -735,10 +735,10 @@ generate_csv(const output_config& cfg,
}
void
generate_csv(const output_config& cfg,
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data,
const stats_entry_t& stats)
generate_csv(const output_config& cfg,
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>& data,
const stats_entry_t& stats)
{
if(data.empty()) return;
+4 -4
View File
@@ -88,10 +88,10 @@ generate_csv(const output_config& cfg,
const stats_entry_t& stats);
void
generate_csv(const output_config& cfg,
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data,
const stats_entry_t& stats);
generate_csv(const output_config& cfg,
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>& data,
const stats_entry_t& stats);
void
generate_csv(const output_config& cfg,
+1 -1
View File
@@ -196,7 +196,7 @@ write_json(json_output& json_ar,
generator<rocprofiler_buffer_tracing_scratch_memory_record_t> scratch_memory_gen,
generator<rocprofiler_buffer_tracing_rccl_api_record_t> rccl_api_gen,
generator<rocprofiler_buffer_tracing_memory_allocation_record_t> memory_allocation_gen,
generator<rocprofiler_buffer_tracing_rocdecode_api_record_t> rocdecode_api_gen,
generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t> rocdecode_api_gen,
generator<rocprofiler_buffer_tracing_rocjpeg_api_record_t> rocjpeg_api_gen,
generator<rocprofiler_tool_pc_sampling_host_trap_record_t> pc_sampling_host_trap_gen,
generator<rocprofiler_tool_pc_sampling_stochastic_record_t> pc_sampling_stochastic_gen)
+1 -1
View File
@@ -94,7 +94,7 @@ write_json(json_output& j
generator<rocprofiler_buffer_tracing_scratch_memory_record_t> scratch_memory_gen,
generator<rocprofiler_buffer_tracing_rccl_api_record_t> rccl_api_gen,
generator<rocprofiler_buffer_tracing_memory_allocation_record_t> memory_allocation_gen,
generator<rocprofiler_buffer_tracing_rocdecode_api_record_t> rocdecode_api_gen,
generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t> rocdecode_api_gen,
generator<rocprofiler_buffer_tracing_rocjpeg_api_record_t> rocjpeg_api_gen,
generator<rocprofiler_tool_pc_sampling_host_trap_record_t> pc_sampling_host_trap_gen,
generator<rocprofiler_tool_pc_sampling_stochastic_record_t> pc_sampling_stochastic_gen);
+23 -2
View File
@@ -368,7 +368,7 @@ write_otf2(
std::deque<rocprofiler_buffer_tracing_scratch_memory_record_t>* /*scratch_memory_data*/,
std::deque<rocprofiler_buffer_tracing_rccl_api_record_t>* rccl_api_data,
std::deque<rocprofiler_buffer_tracing_memory_allocation_record_t>* memory_allocation_data,
std::deque<rocprofiler_buffer_tracing_rocdecode_api_record_t>* rocdecode_api_data,
std::deque<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>* rocdecode_api_data,
std::deque<rocprofiler_buffer_tracing_rocjpeg_api_record_t>* rocjpeg_api_data)
{
namespace sdk = ::rocprofiler::sdk;
@@ -618,10 +618,31 @@ write_otf2(
add_event_data(hip_api_data, sdk::category::hip_api{});
add_event_data(marker_api_data, sdk::category::marker_api{});
add_event_data(rccl_api_data, sdk::category::rccl_api{});
add_event_data(rocdecode_api_data, sdk::category::rocdecode_api{});
add_event_data(rocjpeg_api_data, sdk::category::rocjpeg_api{});
}
for(auto itr : *rocdecode_api_data)
{
auto name = buffer_names.at(itr.kind, itr.operation);
_hash_data.emplace(
get_hash_id(name),
region_info{std::string{name}, OTF2_REGION_ROLE_FUNCTION, OTF2_PARADIGM_USER});
auto& _evt_info = thread_event_info.at(itr.thread_id);
_evt_info.event_count += 1;
_data.emplace_back(evt_data{ROCPROFILER_CALLBACK_PHASE_ENTER,
name,
_evt_info.get_location(),
itr.start_timestamp,
get_attr(sdk::category::rocdecode_api{})});
_data.emplace_back(evt_data{ROCPROFILER_CALLBACK_PHASE_EXIT,
name,
_evt_info.get_location(),
itr.end_timestamp,
nullptr});
}
for(auto itr : *memory_copy_data)
{
auto name = buffer_names.at(itr.kind, itr.operation);
+1 -1
View File
@@ -48,7 +48,7 @@ write_otf2(
std::deque<rocprofiler_buffer_tracing_scratch_memory_record_t>* scratch_memory_data,
std::deque<rocprofiler_buffer_tracing_rccl_api_record_t>* rccl_api_data,
std::deque<rocprofiler_buffer_tracing_memory_allocation_record_t>* memory_allocation_data,
std::deque<rocprofiler_buffer_tracing_rocdecode_api_record_t>* rocdecode_api_data,
std::deque<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>* rocdecode_api_data,
std::deque<rocprofiler_buffer_tracing_rocjpeg_api_record_t>* rocjpeg_api_data);
} // namespace tool
} // namespace rocprofiler
+12 -4
View File
@@ -77,7 +77,7 @@ write_perfetto(
const generator<rocprofiler_buffer_tracing_scratch_memory_record_t>& /*scratch_memory_gen*/,
const generator<rocprofiler_buffer_tracing_rccl_api_record_t>& rccl_api_gen,
const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& memory_allocation_gen,
const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& rocdecode_api_gen,
const generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>& rocdecode_api_gen,
const generator<rocprofiler_buffer_tracing_rocjpeg_api_record_t>& rocjpeg_api_gen)
{
namespace sdk = ::rocprofiler::sdk;
@@ -493,8 +493,9 @@ write_perfetto(
for(auto ditr : rocdecode_api_gen)
for(auto itr : rocdecode_api_gen.get(ditr))
{
auto name = buffer_names.at(itr.kind, itr.operation);
auto& track = thread_tracks.at(itr.thread_id);
auto name = buffer_names.at(itr.kind, itr.operation);
auto& track = thread_tracks.at(itr.thread_id);
auto rocdecode_args = sdk::serialization::get_buffer_tracing_args(itr);
TRACE_EVENT_BEGIN(sdk::perfetto_category<sdk::category::rocdecode_api>::name,
::perfetto::StaticString(name.data()),
@@ -516,7 +517,14 @@ write_perfetto(
"corr_id",
itr.correlation_id.internal,
"ancestor_id",
itr.correlation_id.ancestor);
itr.correlation_id.ancestor,
[&](::perfetto::EventContext ctx) {
for(const auto& rocdecode_arg : rocdecode_args)
{
sdk::add_perfetto_annotation(
ctx, rocdecode_arg.name, rocdecode_arg.value);
}
});
TRACE_EVENT_END(sdk::perfetto_category<sdk::category::rocdecode_api>::name,
track,
itr.end_timestamp);
+1 -1
View File
@@ -49,7 +49,7 @@ write_perfetto(
const generator<rocprofiler_buffer_tracing_scratch_memory_record_t>& scratch_memory_gen,
const generator<rocprofiler_buffer_tracing_rccl_api_record_t>& rccl_api_gen,
const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& memory_allocation_gen,
const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& rocdecode_api_gen,
const generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>& rocdecode_api_gen,
const generator<rocprofiler_buffer_tracing_rocjpeg_api_record_t>& rocjpeg_api_gen);
} // namespace tool
} // namespace rocprofiler
+2 -2
View File
@@ -230,8 +230,8 @@ generate_stats(const output_config& /*cfg*/,
stats_entry_t
generate_stats(const output_config& /*cfg*/,
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data)
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>& data)
{
auto rocdecode_stats = stats_map_t{};
for(auto ditr : data)
+3 -3
View File
@@ -77,9 +77,9 @@ generate_stats(const output_config& cfg,
const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& data);
stats_entry_t
generate_stats(const output_config& cfg,
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data);
generate_stats(const output_config& cfg,
const metadata& tool_metadata,
const generator<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t>& data);
stats_entry_t
generate_stats(const output_config& cfg,
+9 -9
View File
@@ -962,9 +962,9 @@ buffered_tracing_callback(rocprofiler_context_id_t /*context*/,
tool::write_ring_buffer(*record, domain_type::RCCL);
}
else if(header->kind == ROCPROFILER_BUFFER_TRACING_ROCDECODE_API)
else if(header->kind == ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT)
{
auto* record = static_cast<rocprofiler_buffer_tracing_rocdecode_api_record_t*>(
auto* record = static_cast<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t*>(
header->payload);
tool::write_ring_buffer(*record, domain_type::ROCDECODE);
@@ -1747,13 +1747,13 @@ tool_init(rocprofiler_client_finalize_t fini_func, void* tool_data)
&get_buffers().rocdecode_api_trace),
"buffer creation");
ROCPROFILER_CALL(
rocprofiler_configure_buffer_tracing_service(get_client_ctx(),
ROCPROFILER_BUFFER_TRACING_ROCDECODE_API,
nullptr,
0,
get_buffers().rocdecode_api_trace),
"buffer tracing service for ROCDecode api configure");
ROCPROFILER_CALL(rocprofiler_configure_buffer_tracing_service(
get_client_ctx(),
ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT,
nullptr,
0,
get_buffers().rocdecode_api_trace),
"buffer tracing service for ROCDecode api configure");
}
if(tool::get_config().rocjpeg_api_trace)
@@ -102,6 +102,7 @@ ROCPROFILER_BUFFER_TRACING_KIND_STRING(ROCJPEG_API)
ROCPROFILER_BUFFER_TRACING_KIND_STRING(HIP_STREAM)
ROCPROFILER_BUFFER_TRACING_KIND_STRING(HIP_RUNTIME_API_EXT)
ROCPROFILER_BUFFER_TRACING_KIND_STRING(HIP_COMPILER_API_EXT)
ROCPROFILER_BUFFER_TRACING_KIND_STRING(ROCDECODE_API_EXT)
template <size_t Idx, size_t... Tail>
std::pair<const char*, size_t>
@@ -302,6 +303,7 @@ rocprofiler_query_buffer_tracing_kind_operation_name(rocprofiler_buffer_tracing_
return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED;
}
case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API:
case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT:
{
val =
rocprofiler::rocdecode::name_by_id<ROCPROFILER_ROCDECODE_TABLE_ID_CORE>(operation);
@@ -451,6 +453,7 @@ rocprofiler_iterate_buffer_tracing_kind_operations(
return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED;
}
case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API:
case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT:
{
ops = rocprofiler::rocdecode::get_ids<ROCPROFILER_ROCDECODE_TABLE_ID_CORE>();
break;
@@ -518,6 +521,14 @@ rocprofiler_iterate_buffer_tracing_record_args(
_payload->operation, _payload->args, callback, user_data);
return ROCPROFILER_STATUS_SUCCESS;
}
case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT:
{
auto* _payload =
static_cast<rocprofiler_buffer_tracing_rocdecode_api_ext_record_t*>(record.payload);
rocprofiler::rocdecode::iterate_args<ROCPROFILER_ROCDECODE_TABLE_ID_CORE>(
_payload->operation, _payload->args, callback, user_data);
return ROCPROFILER_STATUS_SUCCESS;
}
}
return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED;
@@ -574,6 +574,17 @@ rocprofiler_iterate_callback_tracing_kind_operation_args(
user_data);
return ROCPROFILER_STATUS_SUCCESS;
}
case ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API:
{
rocprofiler::rocdecode::iterate_args<ROCPROFILER_ROCDECODE_TABLE_ID_CORE>(
record.operation,
static_cast<rocprofiler_callback_tracing_rocdecode_api_data_t*>(record.payload)
->args,
callback,
max_deref,
user_data);
return ROCPROFILER_STATUS_SUCCESS;
}
case ROCPROFILER_CALLBACK_TRACING_SCRATCH_MEMORY:
case ROCPROFILER_CALLBACK_TRACING_CODE_OBJECT:
case ROCPROFILER_CALLBACK_TRACING_KERNEL_DISPATCH:
@@ -581,7 +592,6 @@ rocprofiler_iterate_callback_tracing_kind_operation_args(
case ROCPROFILER_CALLBACK_TRACING_MEMORY_ALLOCATION:
case ROCPROFILER_CALLBACK_TRACING_RCCL_API:
case ROCPROFILER_CALLBACK_TRACING_RUNTIME_INITIALIZATION:
case ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API:
case ROCPROFILER_CALLBACK_TRACING_ROCJPEG_API:
case ROCPROFILER_CALLBACK_TRACING_HIP_STREAM:
{
@@ -94,9 +94,14 @@
return &base_type::functor<RetT, Args...>; \
} \
\
static std::vector<void*> as_arg_addr(callback_data_type) { return std::vector<void*>{}; } \
static std::vector<void*> as_arg_addr(rocprofiler_rocdecode_api_args_t) \
{ \
return std::vector<void*>{}; \
} \
\
static std::vector<common::stringified_argument> as_arg_list(callback_data_type, int32_t) \
static std::vector<common::stringified_argument> as_arg_list( \
rocprofiler_rocdecode_api_args_t, \
int32_t) \
{ \
return {}; \
} \
@@ -175,16 +180,15 @@
return &base_type::functor<RetT, Args...>; \
} \
\
static std::vector<void*> as_arg_addr(callback_data_type trace_data) \
static std::vector<void*> as_arg_addr(rocprofiler_rocdecode_api_args_t args) \
{ \
return std::vector<void*>{ \
GET_ADDR_MEMBER_FIELDS(get_api_data_args(trace_data.args), __VA_ARGS__)}; \
GET_ADDR_MEMBER_FIELDS(get_api_data_args(args), __VA_ARGS__)}; \
} \
static auto as_arg_list(callback_data_type trace_data, int32_t max_deref) \
static auto as_arg_list(rocprofiler_rocdecode_api_args_t args, int32_t max_deref) \
{ \
return utils::stringize( \
max_deref, \
GET_NAMED_MEMBER_FIELDS(get_api_data_args(trace_data.args), __VA_ARGS__)); \
max_deref, GET_NAMED_MEMBER_FIELDS(get_api_data_args(args), __VA_ARGS__)); \
} \
}; \
} \
@@ -22,11 +22,11 @@
#include "lib/rocprofiler-sdk/rocdecode/rocdecode.hpp"
#include "lib/common/defines.hpp"
#include "lib/common/mpl.hpp"
#include "lib/common/static_object.hpp"
#include "lib/common/string_entry.hpp"
#include "lib/rocprofiler-sdk/buffer.hpp"
#include "lib/rocprofiler-sdk/context/context.hpp"
#include "lib/rocprofiler-sdk/hip/hip.hpp"
#include "lib/rocprofiler-sdk/hip/utils.hpp"
#include "lib/rocprofiler-sdk/registration.hpp"
#include "lib/rocprofiler-sdk/tracing/tracing.hpp"
@@ -35,11 +35,6 @@
#include <rocprofiler-sdk/fwd.h>
#include <rocprofiler-sdk/rocdecode/table_id.h>
#include <hip/driver_types.h>
#include <hip/hip_runtime_api.h>
// must be included after runtime api
#include <hip/hip_deprecated.h>
#include <cstddef>
#include <cstdint>
#include <type_traits>
@@ -101,6 +96,23 @@ template <size_t TableIdx>
auto*
get_table();
template <typename Tp>
decltype(auto)
convert_arg(Tp&& val)
{
using data_type = common::mpl::unqualified_type_t<Tp>;
if constexpr(std::is_same<data_type, const char*>::value)
{
return common::get_string_entry(val)->c_str();
}
else
{
static_assert(!common::mpl::is_string_type<data_type>::value,
"argument type is a string type. preceding if constexpr is incorrect");
return std::forward<Tp>(val);
}
}
} // namespace
template <size_t TableIdx, size_t OpIdx>
@@ -149,6 +161,7 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
using info_type = rocdecode_api_info<TableIdx, OpIdx>;
using callback_api_data_t = typename rocdecode_domain_info<TableIdx>::callback_data_type;
using buffered_api_data_t = typename rocdecode_domain_info<TableIdx>::buffer_data_type;
using buffered_ext_data_t = typename rocdecode_domain_info<TableIdx>::buffered_ext_data_type;
constexpr auto external_corr_id_domain_idx =
rocdecode_domain_info<TableIdx>::external_correlation_id_domain_idx;
@@ -166,6 +179,7 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
auto thr_id = common::get_tid();
auto callback_contexts = tracing::callback_context_data_vec_t{};
auto buffered_contexts = tracing::buffered_context_data_vec_t{};
auto extended_contexts = tracing::buffered_context_data_vec_t{};
auto external_corr_ids = tracing::external_correlation_id_map_t{};
tracing::populate_contexts(info_type::callback_domain_idx,
@@ -174,8 +188,12 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
callback_contexts,
buffered_contexts,
external_corr_ids);
tracing::populate_contexts(info_type::buffered_ext_domain_idx,
info_type::operation_idx,
extended_contexts,
external_corr_ids);
if(callback_contexts.empty() && buffered_contexts.empty())
if(callback_contexts.empty() && buffered_contexts.empty() && extended_contexts.empty())
{
[[maybe_unused]] auto _ret = exec(info_type::get_table_func(), std::forward<Args>(args)...);
if constexpr(!std::is_void<RetT>::value)
@@ -185,6 +203,7 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
}
auto buffer_record = common::init_public_api_struct(buffered_api_data_t{});
auto extended_record = common::init_public_api_struct(buffered_ext_data_t{});
auto tracer_data = common::init_public_api_struct(callback_api_data_t{});
auto* corr_id = tracing::correlation_service::construct(ref_count);
auto internal_corr_id = corr_id->internal;
@@ -196,11 +215,16 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
info_type::operation_idx,
internal_corr_id);
// set the arguments
if(!callback_contexts.empty() || !extended_contexts.empty())
{
set_data_args(info_type::get_api_data_args(tracer_data.args),
convert_arg(std::forward<Args>(args))...);
}
// invoke the callbacks
if(!callback_contexts.empty())
{
set_data_args(info_type::get_api_data_args(tracer_data.args), std::forward<Args>(args)...);
tracing::execute_phase_enter_callbacks(callback_contexts,
thr_id,
internal_corr_id,
@@ -216,7 +240,7 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
external_corr_ids, thr_id, external_corr_id_domain_idx);
// record the start timestamp as close to the function call as possible
if(!buffered_contexts.empty())
if(!buffered_contexts.empty() || !extended_contexts.empty())
{
buffer_record.start_timestamp = common::timestamp_ns();
}
@@ -227,15 +251,18 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
auto _ret = exec(info_type::get_table_func(), std::forward<Args>(args)...);
// record the end timestamp as close to the function call as possible
if(!buffered_contexts.empty())
if(!buffered_contexts.empty() || !extended_contexts.empty())
{
buffer_record.end_timestamp = common::timestamp_ns();
}
if(!callback_contexts.empty())
if(!callback_contexts.empty() || !extended_contexts.empty())
{
set_data_retval(tracer_data.retval, _ret);
}
if(!callback_contexts.empty())
{
tracing::execute_phase_exit_callbacks(callback_contexts,
external_corr_ids,
info_type::callback_domain_idx,
@@ -255,6 +282,23 @@ rocdecode_api_impl<TableIdx, OpIdx>::functor(Args... args)
buffer_record);
}
if(!extended_contexts.empty())
{
extended_record.start_timestamp = buffer_record.start_timestamp;
extended_record.end_timestamp = buffer_record.end_timestamp;
extended_record.args = tracer_data.args;
extended_record.retval = tracer_data.retval;
tracing::execute_buffer_record_emplace(extended_contexts,
thr_id,
internal_corr_id,
external_corr_ids,
ancestor_corr_id,
info_type::buffered_ext_domain_idx,
info_type::operation_idx,
extended_record);
}
// decrement the reference count after usage in the callback/buffers
corr_id->sub_ref_count();
@@ -323,13 +367,13 @@ get_names(std::vector<const char*>& _name_list, std::index_sequence<OpIdx, OpIdx
get_names<TableIdx>(_name_list, std::index_sequence<OpIdxTail...>{});
}
template <size_t TableIdx, typename DataT, size_t OpIdx, size_t... OpIdxTail>
template <size_t TableIdx, typename DataT, typename FuncT, size_t OpIdx, size_t... OpIdxTail>
void
iterate_args(const uint32_t id,
const DataT& data,
rocprofiler_callback_tracing_operation_args_cb_t func,
int32_t max_deref,
void* user_data,
iterate_args(const uint32_t id,
const DataT& data,
FuncT func,
int32_t max_deref,
void* user_data,
std::index_sequence<OpIdx, OpIdxTail...>)
{
if(OpIdx == id)
@@ -339,16 +383,42 @@ iterate_args(const uint32_t id,
auto&& arg_addr = info_type::as_arg_addr(data);
for(size_t i = 0; i < std::min(arg_list.size(), arg_addr.size()); ++i)
{
auto ret = func(info_type::callback_domain_idx, // kind
id, // operation
i, // arg_number
arg_addr.at(i), // arg_value_addr
arg_list.at(i).indirection_level, // indirection
arg_list.at(i).type, // arg_type
arg_list.at(i).name, // arg_name
arg_list.at(i).value.c_str(), // arg_value_str
arg_list.at(i).dereference_count, // num deref in str
user_data);
using return_type = typename common::mpl::function_traits<FuncT>::result_type;
auto ret = return_type{};
if constexpr(std::is_same<FuncT,
rocprofiler_callback_tracing_operation_args_cb_t>::value)
{
ret = func(info_type::callback_domain_idx, // kind
id, // operation
i, // arg_number
arg_addr.at(i), // arg_value_addr
arg_list.at(i).indirection_level, // indirection
arg_list.at(i).type, // arg_type
arg_list.at(i).name, // arg_name
arg_list.at(i).value.c_str(), // arg_value_str
arg_list.at(i).dereference_count, // num deref in str
user_data);
}
else if constexpr(std::is_same<FuncT,
rocprofiler_buffer_tracing_operation_args_cb_t>::value)
{
ret = func(info_type::buffered_ext_domain_idx, // kind
id, // operation
i, // arg_number
arg_addr.at(i), // arg_value_addr
arg_list.at(i).indirection_level, // indirection
arg_list.at(i).type, // arg_type
arg_list.at(i).name, // arg_name
arg_list.at(i).value.c_str(), // arg_value_str
user_data);
}
else
{
static_assert(common::mpl::assert_false<FuncT>::value,
"Error! unsupported callback type");
}
if(ret != 0) break;
}
return;
@@ -361,6 +431,7 @@ iterate_args(const uint32_t id,
bool
should_wrap_functor(rocprofiler_callback_tracing_kind_t _callback_domain,
rocprofiler_buffer_tracing_kind_t _buffered_domain,
rocprofiler_buffer_tracing_kind_t _buffered_ext_domain,
int _operation)
{
// we loop over all the *registered* contexts and see if any of them, at any point in time,
@@ -378,6 +449,11 @@ should_wrap_functor(rocprofiler_callback_tracing_kind_t _callback_domain,
if(itr->buffered_tracer && itr->buffered_tracer->domains(_buffered_domain) &&
itr->buffered_tracer->domains(_buffered_domain, _operation))
return true;
// if there is a buffered tracer enabled for the given domain and op, we need to wrap
if(itr->buffered_tracer && itr->buffered_tracer->domains(_buffered_ext_domain) &&
itr->buffered_tracer->domains(_buffered_ext_domain, _operation))
return true;
}
return false;
}
@@ -436,8 +512,10 @@ update_table(Tp* _orig, std::integral_constant<size_t, OpIdx>)
if(_info.offset() >= _orig->size) return;
// check to see if there are any contexts which enable this operation in the HIP API domain
if(!should_wrap_functor(
_info.callback_domain_idx, _info.buffered_domain_idx, _info.operation_idx))
if(!should_wrap_functor(_info.callback_domain_idx,
_info.buffered_domain_idx,
_info.buffered_ext_domain_idx,
_info.operation_idx))
return;
ROCP_TRACE << "updating table entry for " << _info.name;
@@ -511,11 +589,11 @@ get_names()
template <size_t TableIdx>
void
iterate_args(uint32_t id,
const rocprofiler_callback_tracing_rocdecode_api_data_t& data,
rocprofiler_callback_tracing_operation_args_cb_t callback,
int32_t max_deref,
void* user_data)
iterate_args(uint32_t id,
const rocprofiler_rocdecode_api_args_t& data,
rocprofiler_callback_tracing_operation_args_cb_t callback,
int32_t max_deref,
void* user_data)
{
if(callback)
iterate_args<TableIdx>(id,
@@ -526,6 +604,22 @@ iterate_args(uint32_t id,
std::make_index_sequence<rocdecode_domain_info<TableIdx>::last>{});
}
template <size_t TableIdx>
void
iterate_args(uint32_t id,
const rocprofiler_rocdecode_api_args_t& data,
rocprofiler_buffer_tracing_operation_args_cb_t callback,
void* user_data)
{
if(callback)
iterate_args<TableIdx>(id,
data,
callback,
0,
user_data,
std::make_index_sequence<rocdecode_domain_info<TableIdx>::last>{});
}
template <typename TableT>
void
copy_table(TableT* _orig, uint64_t _tbl_instance)
@@ -547,18 +641,21 @@ update_table(TableT* _orig)
std::make_index_sequence<rocdecode_domain_info<TableIdx>::last>{});
}
using rocdecode_api_data_t = rocprofiler_callback_tracing_rocdecode_api_data_t;
using rocdecode_api_data_t = rocprofiler_rocdecode_api_args_t;
using rocdecode_op_args_cb_t = rocprofiler_callback_tracing_operation_args_cb_t;
using rocdecode_op_args_bf_t = rocprofiler_buffer_tracing_operation_args_cb_t;
#define INSTANTIATE_ROCDECODE_TABLE_FUNC(TABLE_TYPE, TABLE_IDX) \
template void copy_table<TABLE_TYPE>(TABLE_TYPE * _tbl, uint64_t _instv); \
template void update_table<TABLE_TYPE>(TABLE_TYPE * _tbl); \
template const char* name_by_id<TABLE_IDX>(uint32_t); \
template uint32_t id_by_name<TABLE_IDX>(const char*); \
template std::vector<uint32_t> get_ids<TABLE_IDX>(); \
template std::vector<const char*> get_names<TABLE_IDX>(); \
template void iterate_args<TABLE_IDX>( \
uint32_t, const rocdecode_api_data_t&, rocdecode_op_args_cb_t, int32_t, void*);
#define INSTANTIATE_ROCDECODE_TABLE_FUNC(TABLE_TYPE, TABLE_IDX) \
template void copy_table<TABLE_TYPE>(TABLE_TYPE * _tbl, uint64_t _instv); \
template void update_table<TABLE_TYPE>(TABLE_TYPE * _tbl); \
template const char* name_by_id<TABLE_IDX>(uint32_t); \
template uint32_t id_by_name<TABLE_IDX>(const char*); \
template std::vector<uint32_t> get_ids<TABLE_IDX>(); \
template std::vector<const char*> get_names<TABLE_IDX>(); \
template void iterate_args<TABLE_IDX>( \
uint32_t, const rocdecode_api_data_t&, rocdecode_op_args_cb_t, int32_t, void*); \
template void iterate_args<TABLE_IDX>( \
uint32_t, const rocdecode_api_data_t&, rocdecode_op_args_bf_t, void*);
INSTANTIATE_ROCDECODE_TABLE_FUNC(rocdecode_api_func_table_t, ROCPROFILER_ROCDECODE_TABLE_ID_CORE)
} // namespace rocdecode
@@ -36,21 +36,23 @@ namespace rocdecode
template <>
struct rocdecode_domain_info<ROCPROFILER_ROCDECODE_TABLE_ID_LAST>
{
using args_type = rocprofiler_rocdecode_api_args_t;
using retval_type = rocprofiler_rocdecode_api_retval_t;
using callback_data_type = rocprofiler_callback_tracing_rocdecode_api_data_t;
using buffer_data_type = rocprofiler_buffer_tracing_rocdecode_api_record_t;
using args_type = rocprofiler_rocdecode_api_args_t;
using retval_type = rocprofiler_rocdecode_api_retval_t;
using callback_data_type = rocprofiler_callback_tracing_rocdecode_api_data_t;
using buffer_data_type = rocprofiler_buffer_tracing_rocdecode_api_record_t;
using buffered_ext_data_type = rocprofiler_buffer_tracing_rocdecode_api_ext_record_t;
};
template <>
struct rocdecode_domain_info<ROCPROFILER_ROCDECODE_TABLE_ID_CORE>
: rocdecode_domain_info<ROCPROFILER_ROCDECODE_TABLE_ID_LAST>
{
using enum_type = rocprofiler_marker_core_api_id_t;
static constexpr auto callback_domain_idx = ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API;
static constexpr auto buffered_domain_idx = ROCPROFILER_BUFFER_TRACING_ROCDECODE_API;
static constexpr auto none = ROCPROFILER_ROCDECODE_API_ID_NONE;
static constexpr auto last = ROCPROFILER_ROCDECODE_API_ID_LAST;
using enum_type = rocprofiler_marker_core_api_id_t;
static constexpr auto callback_domain_idx = ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API;
static constexpr auto buffered_domain_idx = ROCPROFILER_BUFFER_TRACING_ROCDECODE_API;
static constexpr auto buffered_ext_domain_idx = ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT;
static constexpr auto none = ROCPROFILER_ROCDECODE_API_ID_NONE;
static constexpr auto last = ROCPROFILER_ROCDECODE_API_ID_LAST;
static constexpr auto external_correlation_id_domain_idx =
ROCPROFILER_EXTERNAL_CORRELATION_REQUEST_ROCDECODE_API;
};
@@ -98,11 +98,18 @@ get_ids();
template <size_t TableIdx>
void
iterate_args(uint32_t id,
const rocprofiler_callback_tracing_rocdecode_api_data_t& data,
rocprofiler_callback_tracing_operation_args_cb_t callback,
int32_t max_deref,
void* user_data);
iterate_args(uint32_t id,
const rocprofiler_rocdecode_api_args_t& data,
rocprofiler_callback_tracing_operation_args_cb_t callback,
int32_t max_deref,
void* user_data);
template <size_t TableIdx>
void
iterate_args(uint32_t id,
const rocprofiler_rocdecode_api_args_t& data,
rocprofiler_buffer_tracing_operation_args_cb_t callback,
void* user_data);
template <typename TableT>
void