[Misc] Rework header includes (#311)

* Update header file includes * Fix includes for lib/rocprofiler-sdk/hip/hip.hpp * Minor touch ups * Minor include improvements * Doxygen tweak --------- Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com> [ROCm/rocprofiler-sdk commit: c5a3edc3fa]
2025-04-15 16:02:12 -05:00
@@ -21,6 +21,7 @@
 // SOFTWARE.

 #pragma once
+
 #define ROCPROFILER_VAR_NAME_COMBINE(X, Y) X##Y
 #define ROCPROFILER_VARIABLE(X, Y)         ROCPROFILER_VAR_NAME_COMBINE(X, Y)

@@ -21,6 +21,7 @@
 // SOFTWARE.

 #pragma once
+
 #include <algorithm>
 #include <iostream>
 #include <random>
@@ -28,7 +28,6 @@
 #include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/hsa.h>
 #include <rocprofiler-sdk/internal_threading.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 #include <tuple>

@@ -23,13 +23,23 @@

 #pragma once

+#include <rocprofiler-sdk/agent.h>
 #include <rocprofiler-sdk/buffer.h>
+#include <rocprofiler-sdk/buffer_tracing.h>
 #include <rocprofiler-sdk/callback_tracing.h>
+#include <rocprofiler-sdk/context.h>
+#include <rocprofiler-sdk/counter_config.h>
+#include <rocprofiler-sdk/counters.h>
+#include <rocprofiler-sdk/device_counting_service.h>
+#include <rocprofiler-sdk/dispatch_counting_service.h>
 #include <rocprofiler-sdk/external_correlation.h>
 #include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/hip.h>
+#include <rocprofiler-sdk/hsa.h>
+#include <rocprofiler-sdk/intercept_table.h>
 #include <rocprofiler-sdk/internal_threading.h>
+#include <rocprofiler-sdk/marker.h>
 #include <rocprofiler-sdk/pc_sampling.h>
-#include <rocprofiler-sdk/rocprofiler.h>
 #include <rocprofiler-sdk/cxx/name_info.hpp>
 #include <rocprofiler-sdk/cxx/perfetto.hpp>
 #include <rocprofiler-sdk/cxx/utility.hpp>
@@ -83,5 +83,4 @@ rocprofiler_query_counter_instance_count(rocprofiler_agent_id_t   agent_id,
                                         size_t*                  instance_count) ROCPROFILER_API
    ROCPROFILER_NONNULL(3);

-/** @} */
 ROCPROFILER_EXTERN_C_FINI
@@ -113,6 +113,3 @@
 #define ROCPROFILER_VERSION                                                                        \
    ((10000 * ROCPROFILER_VERSION_MAJOR) + (100 * ROCPROFILER_VERSION_MINOR) +                     \
     ROCPROFILER_VERSION_PATCH)
-
-// include the external version info
-#include "ext_version.h"
@@ -22,7 +22,8 @@

 #include "lib/common/container/record_header_buffer.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/fwd.h>
+
 #include <algorithm>
 #include <atomic>
 #include <new>
@@ -22,10 +22,10 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/container/ring_buffer.hpp"

+#include <rocprofiler-sdk/fwd.h>
+
 #include <atomic>
 #include <limits>
 #include <mutex>
@@ -23,11 +23,11 @@
 #pragma once

 #include <rocprofiler-sdk/agent.h>
+#include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/cxx/serialization.hpp>

 #include <unordered_map>
 #include <vector>
-#include "rocprofiler-sdk/fwd.h"

 namespace rocprofiler
 {
@@ -32,13 +32,13 @@

 #include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/rocprofiler.h>
-#include <cstdint>
 #include <rocprofiler-sdk/cxx/details/tokenize.hpp>

 #include <fmt/core.h>
 #include <fmt/format.h>

 #include <unistd.h>
+#include <cstdint>
 #include <memory>
 #include <vector>

@@ -21,9 +21,11 @@
 // SOFTWARE.

 #pragma once
-#include <memory>
+
 #include "att_decoder.h"

+#include <memory>
+
 namespace rocprofiler
 {
 namespace att_wrapper
@@ -32,7 +32,6 @@

 #include <rocprofiler-sdk/agent.h>
 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>
 #include <rocprofiler-sdk/cxx/details/tokenize.hpp>

 #include <fmt/core.h>
@@ -27,7 +27,6 @@
 #include <fmt/core.h>
 #include <hsa/hsa_ext_amd.h>
 #include "glog/logging.h"
-#include "rocprofiler-sdk/fwd.h"

 #define CHECK_HSA(fn, message)                                                                     \
    {                                                                                              \
@@ -26,15 +26,16 @@
 #include "lib/rocprofiler-sdk/counters/tests/hsa_tables.hpp"
 #include "lib/rocprofiler-sdk/hsa/agent_cache.hpp"
 #include "lib/rocprofiler-sdk/hsa/queue_controller.hpp"
-#include "rocprofiler-sdk/fwd.h"

-#include <unordered_set>
+#include <rocprofiler-sdk/fwd.h>

 #include <gtest/gtest.h>
 #include <hsa/hsa.h>
 #include <hsa/hsa_api_trace.h>
 #include <hsa/hsa_ven_amd_aqlprofile.h>

+#include <unordered_set>
+
 using namespace rocprofiler::counters::test_constants;

 namespace rocprofiler
@@ -33,7 +33,6 @@
 #include "lib/rocprofiler-sdk/registration.hpp"

 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 #include <atomic>
 #include <exception>
@@ -47,7 +47,6 @@
 #include <rocprofiler-sdk/rccl/table_id.h>
 #include <rocprofiler-sdk/rocdecode/table_id.h>
 #include <rocprofiler-sdk/rocjpeg/table_id.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 #include <atomic>
 #include <limits>
@@ -46,7 +46,6 @@
 #include <rocprofiler-sdk/rccl/table_id.h>
 #include <rocprofiler-sdk/rocdecode/table_id.h>
 #include <rocprofiler-sdk/rocjpeg/table_id.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 #include <atomic>
 #include <cstdint>
@@ -22,7 +22,6 @@

 #include <rocprofiler-sdk/context.h>
 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 #include "lib/rocprofiler-sdk/context/context.hpp"
 #include "lib/rocprofiler-sdk/context/domain.hpp"
@@ -20,21 +20,20 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

-#include <rocprofiler-sdk/buffer_tracing.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
+#include "lib/rocprofiler-sdk/context/context.hpp"
 #include "lib/common/container/small_vector.hpp"
 #include "lib/common/container/stable_vector.hpp"
 #include "lib/common/static_object.hpp"
 #include "lib/common/synchronized.hpp"
 #include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/buffer.hpp"
-#include "lib/rocprofiler-sdk/context/context.hpp"
 #include "lib/rocprofiler-sdk/counters/core.hpp"
 #include "lib/rocprofiler-sdk/pc_sampling/service.hpp"
 #include "lib/rocprofiler-sdk/thread_trace/att_core.hpp"

+#include <rocprofiler-sdk/buffer_tracing.h>
+#include <rocprofiler-sdk/fwd.h>
+
 #include <unistd.h>
 #include <atomic>
 #include <cstddef>
@@ -22,10 +22,6 @@

 #pragma once

-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/registration.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/container/small_vector.hpp"
 #include "lib/common/synchronized.hpp"
 #include "lib/rocprofiler-sdk/context/correlation_id.hpp"
@@ -35,7 +31,9 @@
 #include "lib/rocprofiler-sdk/external_correlation.hpp"
 #include "lib/rocprofiler-sdk/pc_sampling/types.hpp"
 #include "lib/rocprofiler-sdk/thread_trace/att_core.hpp"
-#include "rocprofiler-sdk/agent.h"
+
+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/registration.h>

 #include <array>
 #include <cstddef>
@@ -28,7 +28,6 @@
 #include "lib/rocprofiler-sdk/context/context.hpp"

 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 namespace rocprofiler
 {
@@ -22,8 +22,6 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/mpl.hpp"
 #include "lib/rocprofiler-sdk/external_correlation.hpp"

@@ -23,7 +23,6 @@
 #include "lib/rocprofiler-sdk/context/domain.hpp"

 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 #include <limits>

@@ -22,10 +22,10 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/mpl.hpp"

+#include <rocprofiler-sdk/fwd.h>
+
 #include <bitset>
 #include <cstddef>
 #include <cstdint>
@@ -21,7 +21,6 @@
 // SOFTWARE.

 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>
 #include <cstdint>
 #include <unordered_set>

@@ -20,13 +20,6 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

-#include <rocprofiler-sdk/experimental/counters.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-#include <rocprofiler-sdk/cxx/operators.hpp>
-
-#include <fmt/core.h>
-
 #include "lib/common/container/small_vector.hpp"
 #include "lib/common/logging.hpp"
 #include "lib/common/static_object.hpp"
@@ -41,6 +34,13 @@
 #include "lib/rocprofiler-sdk/hsa/queue.hpp"
 #include "lib/rocprofiler-sdk/hsa/queue_controller.hpp"

+#include <rocprofiler-sdk/counters.h>
+#include <rocprofiler-sdk/experimental/counters.h>
+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/cxx/operators.hpp>
+
+#include <fmt/core.h>
+
 namespace rocprofiler
 {
 namespace counters
@@ -21,17 +21,15 @@
 // SOFTWARE.

 #include "lib/rocprofiler-sdk/counters/controller.hpp"
-
-#include <rocprofiler-sdk/agent.h>
-#include <rocprofiler-sdk/dispatch_counting_service.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/rocprofiler-sdk/buffer.hpp"
 #include "lib/rocprofiler-sdk/context/context.hpp"
 #include "lib/rocprofiler-sdk/counters/ioctl.hpp"
 #include "lib/rocprofiler-sdk/counters/metrics.hpp"

+#include <rocprofiler-sdk/agent.h>
+#include <rocprofiler-sdk/dispatch_counting_service.h>
+#include <rocprofiler-sdk/fwd.h>
+
 namespace rocprofiler
 {
 namespace counters
@@ -30,9 +30,9 @@
 #include "lib/rocprofiler-sdk/counters/metrics.hpp"

 #include <rocprofiler-sdk/agent.h>
+#include <rocprofiler-sdk/device_counting_service.h>
 #include <rocprofiler-sdk/dispatch_counting_service.h>
 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 namespace rocprofiler
 {
@@ -33,7 +33,6 @@
 #include "lib/rocprofiler-sdk/kernel_dispatch/profiling_time.hpp"

 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 namespace rocprofiler
 {
@@ -24,9 +24,10 @@

 #include "lib/rocprofiler-sdk/hsa/aql_packet.hpp"

+#include <rocprofiler-sdk/device_counting_service.h>
 #include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/hsa.h>
-#include <rocprofiler-sdk/rocprofiler.h>
+
 #include <cstddef>

 namespace rocprofiler
@@ -22,21 +22,21 @@

 #include "dimensions.hpp"

-#include <cstdint>
-#include <unordered_map>
-#include <vector>
-
-#include <fmt/core.h>
-
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/static_object.hpp"
 #include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/aql/helpers.hpp"
 #include "lib/rocprofiler-sdk/aql/packet_construct.hpp"
 #include "lib/rocprofiler-sdk/counters/evaluate_ast.hpp"

+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
+#include <fmt/core.h>
+
+#include <cstdint>
+#include <unordered_map>
+#include <vector>
+
 namespace rocprofiler
 {
 namespace counters
@@ -23,6 +23,16 @@
 #include "lib/rocprofiler-sdk/counters/evaluate_ast.hpp"
 #include "lib/common/static_object.hpp"
 #include "lib/common/synchronized.hpp"
+#include "lib/common/utility.hpp"
+#include "lib/rocprofiler-sdk/counters/dimensions.hpp"
+#include "lib/rocprofiler-sdk/counters/id_decode.hpp"
+#include "lib/rocprofiler-sdk/counters/parser/reader.hpp"
+
+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
+#include <fmt/core.h>
+#include <fmt/ranges.h>

 #include <algorithm>
 #include <cstdint>
@@ -33,16 +43,6 @@
 #include <unordered_map>
 #include <vector>

-#include <fmt/core.h>
-#include <fmt/ranges.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
-#include "lib/common/utility.hpp"
-#include "lib/rocprofiler-sdk/counters/dimensions.hpp"
-#include "lib/rocprofiler-sdk/counters/id_decode.hpp"
-#include "lib/rocprofiler-sdk/counters/parser/reader.hpp"
-#include "rocprofiler-sdk/fwd.h"
-
 namespace rocprofiler
 {
 namespace counters
@@ -19,9 +19,10 @@
 // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.
+
 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/agent.h>

 namespace rocprofiler
 {
@@ -22,8 +22,6 @@

 #include "metrics.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/filesystem.hpp"
 #include "lib/common/logging.hpp"
 #include "lib/common/static_object.hpp"
@@ -31,9 +29,9 @@
 #include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/agent.hpp"

-#include "glog/logging.h"
+#include <rocprofiler-sdk/fwd.h>

-#include "rocprofiler-sdk/fwd.h"
+#include "glog/logging.h"
 #include "yaml-cpp/exceptions.h"
 #include "yaml-cpp/node/convert.h"
 #include "yaml-cpp/node/detail/impl.h"
@@ -22,7 +22,11 @@

 #pragma once

-#include "rocprofiler-sdk/fwd.h"
+#include <rocprofiler-sdk/fwd.h>
+
+#include <fmt/core.h>
+#include <fmt/ranges.h>
+#include <hsa/hsa_ven_amd_aqlprofile.h>

 #include <cstdint>
 #include <optional>
@@ -31,10 +35,6 @@
 #include <unordered_set>
 #include <vector>

-#include <fmt/core.h>
-#include <fmt/ranges.h>
-#include <hsa/hsa_ven_amd_aqlprofile.h>
-
 namespace rocprofiler
 {
 namespace counters
@@ -35,7 +35,6 @@
 #include "lib/rocprofiler-sdk/kernel_dispatch/profiling_time.hpp"

 #include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/rocprofiler.h>

 namespace rocprofiler
 {
@@ -20,6 +20,17 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

+#include "lib/rocprofiler-sdk/counters/evaluate_ast.hpp"
+#include "lib/rocprofiler-sdk/agent.hpp"
+#include "lib/rocprofiler-sdk/counters/id_decode.hpp"
+#include "lib/rocprofiler-sdk/counters/metrics.hpp"
+#include "lib/rocprofiler-sdk/counters/parser/reader.hpp"
+
+#include <rocprofiler-sdk/fwd.h>
+
+#include <fmt/core.h>
+#include <gtest/gtest.h>
+
 #include <algorithm>
 #include <cstdint>
 #include <tuple>
@@ -27,16 +38,6 @@
 #include <utility>
 #include <vector>

-#include <fmt/core.h>
-#include <gtest/gtest.h>
-
-#include "lib/rocprofiler-sdk/agent.hpp"
-#include "lib/rocprofiler-sdk/counters/evaluate_ast.hpp"
-#include "lib/rocprofiler-sdk/counters/id_decode.hpp"
-#include "lib/rocprofiler-sdk/counters/metrics.hpp"
-#include "lib/rocprofiler-sdk/counters/parser/reader.hpp"
-#include "rocprofiler-sdk/fwd.h"
-
 namespace
 {
 ReduceOperation
@@ -22,20 +22,20 @@

 #include "metrics_test.h"

-#include <gtest/gtest.h>
-
-#include <algorithm>
-#include <cstddef>
-#include <cstdint>
-
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/logging.hpp"
 #include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/agent.hpp"
 #include "lib/rocprofiler-sdk/counters/dimensions.hpp"
 #include "lib/rocprofiler-sdk/counters/metrics.hpp"
-#include "rocprofiler-sdk/fwd.h"
+
+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
+#include <gtest/gtest.h>
+
+#include <algorithm>
+#include <cstddef>
+#include <cstdint>

 namespace
 {
@@ -20,13 +20,14 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/rocprofiler-sdk/context/context.hpp"
 #include "lib/rocprofiler-sdk/counters/core.hpp"
 #include "lib/rocprofiler-sdk/counters/device_counting.hpp"
 #include "lib/rocprofiler-sdk/registration.hpp"
-#include "rocprofiler-sdk/fwd.h"
+
+#include <rocprofiler-sdk/context.h>
+#include <rocprofiler-sdk/device_counting_service.h>
+#include <rocprofiler-sdk/fwd.h>

 #include <string.h>

@@ -23,7 +23,7 @@
 #include "lib/rocprofiler-sdk/counters/controller.hpp"
 #include "lib/rocprofiler-sdk/counters/core.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/dispatch_counting_service.h>

 extern "C" {
 /**
@@ -24,7 +24,7 @@

 #include "lib/rocprofiler-sdk/hip/details/ostream.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/hip.h>

 #include <hip/hip_runtime_api.h>
 // must be included after runtime api
@@ -22,7 +22,7 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/hip.h>

 #include <hip/hip_runtime_api.h>
 // must be included after runtime api
@@ -22,7 +22,9 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/buffer_tracing.h>
+#include <rocprofiler-sdk/callback_tracing.h>
+#include <rocprofiler-sdk/hip.h>

 #include <hip/hip_version.h>
 #include <hip/amd_detail/hip_api_trace.hpp>
@@ -22,7 +22,7 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/hip.h>

 #include <hip/hip_version.h>
 #include <hip/amd_detail/hip_api_trace.hpp>
@@ -25,7 +25,7 @@
 #include "lib/common/container/small_vector.hpp"
 #include "lib/rocprofiler-sdk/aql/aql_profile_v2.h"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/hsa.h>

 #include <hsa/hsa_ext_amd.h>
 #include <hsa/hsa_ven_amd_aqlprofile.h>
@@ -22,8 +22,6 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/rocprofiler-sdk/hsa/hsa.hpp"

 namespace rocprofiler
@@ -22,7 +22,7 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/hsa.h>

 #include <iostream>
 #include <string>
@@ -24,7 +24,9 @@

 #include "lib/rocprofiler-sdk/pc_sampling/defines.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/callback_tracing.h>
+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/hsa.h>

 #include <hsa/hsa_api_trace.h>

@@ -25,7 +25,7 @@
 #include "lib/common/synchronized.hpp"
 #include "lib/rocprofiler-sdk/hsa/rocprofiler_packet.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/fwd.h>

 #include <hsa/hsa.h>
 #include <hsa/hsa_api_trace.h>
@@ -22,8 +22,6 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/rocprofiler-sdk/hsa/hsa.hpp"

 namespace rocprofiler
@@ -22,12 +22,13 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/container/small_vector.hpp"
 #include "lib/rocprofiler-sdk/hsa/hsa_barrier.hpp"
 #include "lib/rocprofiler-sdk/hsa/queue.hpp"

+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/hsa.h>
+
 #include <deque>
 #include <unordered_map>

@@ -22,12 +22,12 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-#include <rocprofiler-sdk/cxx/hash.hpp>
-
 #include "lib/rocprofiler-sdk/hsa/profile_serializer.hpp"
 #include "lib/rocprofiler-sdk/hsa/queue.hpp"

+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/cxx/hash.hpp>
+
 #include <cstdint>
 #include <functional>
 #include <optional>
@@ -24,7 +24,7 @@

 #include "lib/rocprofiler-sdk/hsa/hsa.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/fwd.h>

 #include <cstdint>
 #include <vector>
@@ -20,23 +20,21 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/intercept_table.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/container/stable_vector.hpp"
 #include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/buffer.hpp"
 #include "lib/rocprofiler-sdk/context/context.hpp"
 #include "lib/rocprofiler-sdk/internal_threading.hpp"
-#include "lib/rocprofiler-sdk/registration.hpp"
-
-#include <hsa/hsa_api_trace.h>
-#include <hip/amd_detail/hip_api_trace.hpp>
 #include "lib/rocprofiler-sdk/rccl/rccl.hpp"
+#include "lib/rocprofiler-sdk/registration.hpp"
 #include "lib/rocprofiler-sdk/rocdecode/rocdecode.hpp"
 #include "lib/rocprofiler-sdk/rocjpeg/rocjpeg.hpp"

+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/hip.h>
+#include <rocprofiler-sdk/hsa.h>
+#include <rocprofiler-sdk/intercept_table.h>
+
 #include <cstdint>
 #include <mutex>
 #include <stdexcept>
@@ -20,18 +20,17 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/internal_threading.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
+#include "lib/rocprofiler-sdk/internal_threading.hpp"
 #include "lib/common/container/stable_vector.hpp"
 #include "lib/common/static_object.hpp"
 #include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/buffer.hpp"
 #include "lib/rocprofiler-sdk/context/context.hpp"
-#include "lib/rocprofiler-sdk/internal_threading.hpp"
 #include "lib/rocprofiler-sdk/registration.hpp"

+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/internal_threading.h>
+
 #include <pthread.h>

 #include <cstdint>
@@ -22,9 +22,8 @@

 #pragma once

-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include <rocprofiler-sdk-roctx/api_trace.h>
+#include <rocprofiler-sdk/callback_tracing.h>

 #include <cstdint>
 #include <vector>
@@ -24,7 +24,7 @@

 #include "lib/common/container/small_vector.hpp"

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/fwd.h>

 namespace rocprofiler
 {
@@ -42,7 +42,7 @@
 #    include <rocprofiler-sdk/rccl/details/rccl.h>
 #endif

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/callback_tracing.h>

 #include <cstdint>
 #include <vector>
@@ -53,16 +53,15 @@
 #include "lib/rocprofiler-sdk/runtime_initialization.hpp"

 #include <rocprofiler-sdk/context.h>
+#include <rocprofiler-sdk/ext_version.h>
 #include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/hip.h>
 #include <rocprofiler-sdk/hsa.h>
 #include <rocprofiler-sdk/marker.h>
 #include <rocprofiler-sdk/ompt.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/version.h>

-#include <hsa/hsa_api_trace.h>
-#include <hip/amd_detail/hip_api_trace.hpp>
-
 #include <fmt/format.h>

 #include <dlfcn.h>
@@ -34,7 +34,7 @@
 #    include <rocprofiler-sdk/rocdecode/details/rocparser.h>
 #endif

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/callback_tracing.h>

 #include <cstdint>
 #include <vector>
@@ -32,7 +32,7 @@
 #    include <rocprofiler-sdk/rocjpeg/details/rocjpeg_api_trace.h>
 #endif

-#include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/callback_tracing.h>

 #include <cstdint>
 #include <vector>
@@ -20,11 +20,11 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

+#include "lib/common/utility.hpp"
+
 #include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/rocprofiler.h>

-#include "lib/common/utility.hpp"
-
 namespace rocprofiler
 {
 namespace
@@ -20,10 +20,6 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/intercept_table.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #include "lib/common/container/stable_vector.hpp"
 #include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/agent.hpp"
@@ -33,6 +29,10 @@
 #include "lib/rocprofiler-sdk/internal_threading.hpp"
 #include "lib/rocprofiler-sdk/registration.hpp"

+#include <rocprofiler-sdk/fwd.h>
+#include <rocprofiler-sdk/hsa.h>
+#include <rocprofiler-sdk/intercept_table.h>
+
 #include <hsa/hsa_api_trace.h>

 #include <atomic>
@@ -20,15 +20,16 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 // SOFTWARE.

-#include <glog/logging.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-#include <cstdint>
-
 #include "lib/rocprofiler-sdk/aql/helpers.hpp"
 #include "lib/rocprofiler-sdk/context/context.hpp"
 #include "lib/rocprofiler-sdk/hsa/agent_cache.hpp"
 #include "lib/rocprofiler-sdk/registration.hpp"
-#include "rocprofiler-sdk/amd_detail/thread_trace.h"
+
+#include <rocprofiler-sdk/amd_detail/thread_trace.h>
+
+#include <glog/logging.h>
+
+#include <cstdint>

 using DispatchThreadTracer = rocprofiler::thread_trace::DispatchThreadTracer;
 using DeviceThreadTracer   = rocprofiler::thread_trace::DeviceThreadTracer;