[SDK] Release 1.0 Public API Modifications (#277)

* Make sure all structs/enums can be forward declared * Updates to counter collection - consistency updates and cleanup * Conversion of dimension information to info struct * Added deprecated folder * Testing changes * merge changes * Fix shadowed variable * Source code formatting * Fix shadowed variable * Update rocprofiler_counter_info_v1_t member names * Split version.h into version.h and ext_version.h - ext_version.h contains external version info, e.g. ROCPROFILER_HSA_API_TABLE_MAJOR_VERSION, ROCPROFILER_HSA_RUNTIME_VERSION - this reduces amount of recompilation after a commit since version.h gets updated with the git revision * profile_config -> counter_config * EOF new line * [Samples] Reduce header includes + reorg counter collection samples * Misc compilation fixes - shadowed variables - use of [[deprecated("...")]] in C code - unused variables * Minor misc modifications - use common:: instead of rocprofiler::common:: when inside rocprofiler namespace - counters.cpp - move local anon namespace functions into rocprofiler::counters:: anon namespace - use std::string_view for get_static_string - const ref for get_static_ptr - misc namespace shortening * [Public API] rocprofiler_get_version_triplet + rocprofiler_version_triplet_t - struct rocprofiler_version_triplet_t containing fields for the major, minor, and patch version - public API function: rocprofiler_get_version_triplet - define C++ operators for rocprofiler_version_triplet_t - C++ function compute_version_triplet * [Tests] Improve async-copy-testing test - relax constraints - improve logging * Update counter_config.h doxygen docs * ROCPROFILER_SDK_BETA_COMPAT - ppdef which helps with renaming when set to 1 * Remove spurious include * Fix includes for cxx/version.hpp * Doxygen fixes for rocprofiler_get_version and rocprofiler_get_version_triplet * Public API Experimental Designation - ROCPROFILER_SDK_EXPERIMENTAL added to experimental function - "(experimental)" added to doxygen @brief entries * Fix use of assert instead of static_assert in hip/stream.cpp * Use typedef instead of define for rocprofiler_profile_config_id_t * Use inline rocprofiler_{create,destroy}_profile_config instead of ppdef - added <rocprofiler-sdk/deprecated/profile_config.h> * Doxygen for rocprofiler_{create,destroy}_profile_config * ROCPROFILER_SDK_DEPRECATED_WARNINGS * Temporarily comment out ROCPROFILER_SDK_DEPRECATED_WARNINGS=1 * cmake formatting * Misc variable renaming in samples and tests * Fix declarations of types * Fix hip stream tracing service struct name - rocprofiler_callback_tracing_stream_handle_data_t renamed to rocprofiler_callback_tracing_hip_stream_api_data_t * Rename "HIP_STREAM_API" to "HIP_STREAM" --------- Co-authored-by: Jonathan R. Madsen <jonathanrmadsen@gmail.com> Co-authored-by: Benjamin Welton <bewelton@amd.com>
2025-03-24 12:07:33 +05:30
commit 4cd121e27b
@@ -33,12 +33,6 @@

 #include "client.hpp"

-#include <rocprofiler-sdk/buffer.h>
-#include <rocprofiler-sdk/buffer_tracing.h>
-#include <rocprofiler-sdk/callback_tracing.h>
-#include <rocprofiler-sdk/external_correlation.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/internal_threading.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/rocprofiler.h>

@@ -33,11 +33,10 @@

 #include "client.hpp"

-#include <rocprofiler-sdk/context.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/marker/api_id.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/rocprofiler.h>
+#include <rocprofiler-sdk/cxx/operators.hpp>
+#include <rocprofiler-sdk/cxx/version.hpp>

 #include "common/call_stack.hpp"
 #include "common/defines.hpp"
@@ -318,7 +317,7 @@ stop()

 extern "C" rocprofiler_tool_configure_result_t*
 rocprofiler_configure(uint32_t                 version,
-                      const char*              runtime_version,
+                      const char*              version_string,
                      uint32_t                 priority,
                      rocprofiler_client_id_t* id)
 {
@@ -336,21 +335,26 @@ rocprofiler_configure(uint32_t                 version,
    // generate info string
    auto info = std::stringstream{};
    info << id->name << " (priority=" << priority << ") is using rocprofiler-sdk v" << major << "."
-         << minor << "." << patch << " (" << runtime_version << ")";
+         << minor << "." << patch << " (" << version_string << ")";

    std::clog << info.str() << std::endl;

    // demonstration of alternative way to get the version info
    {
-        auto version_info = std::array<uint32_t, 3>{};
-        ROCPROFILER_CALL(
-            rocprofiler_get_version(&version_info.at(0), &version_info.at(1), &version_info.at(2)),
-            "failed to get version info");
+        auto runtime_version = rocprofiler_version_triplet_t{};
+        ROCPROFILER_CALL(rocprofiler_get_version_triplet(&runtime_version),
+                         "failed to get version info");

-        if(std::array<uint32_t, 3>{major, minor, patch} != version_info)
+        if(rocprofiler_version_triplet_t{major, minor, patch} != runtime_version)
        {
            throw std::runtime_error{"version info mismatch"};
        }
+
+        if(rocprofiler_version_triplet_t{major, minor, patch} !=
+           rocprofiler::sdk::version::compute_version_triplet<100>(version))
+        {
+            throw std::runtime_error{"version triplet incorrectly calculated"};
+        }
    }

    // data passed around all the callbacks
@@ -33,9 +33,6 @@
 * @brief Example rocprofiler client (tool)
 */

-#include <rocprofiler-sdk/buffer.h>
-#include <rocprofiler-sdk/callback_tracing.h>
-#include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/rocprofiler.h>
 #include <rocprofiler-sdk/cxx/codeobj/code_printing.hpp>
@@ -31,9 +31,6 @@
 * @brief Example rocprofiler client (tool)
 */

-#include <rocprofiler-sdk/buffer.h>
-#include <rocprofiler-sdk/callback_tracing.h>
-#include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/rocprofiler.h>

@@ -34,7 +34,7 @@ endif()
 find_package(rocprofiler-sdk REQUIRED)

 add_library(counter-collection-buffer-client SHARED)
-target_sources(counter-collection-buffer-client PRIVATE client.cpp client.hpp)
+target_sources(counter-collection-buffer-client PRIVATE buffered_client.cpp client.hpp)
 target_link_libraries(
    counter-collection-buffer-client
    PUBLIC rocprofiler-sdk::samples-build-flags
@@ -59,10 +59,10 @@ set_tests_properties(
               "${counter-collection-buffer-env}" FAIL_REGULAR_EXPRESSION
               "${ROCPROFILER_DEFAULT_FAIL_REGEX}")

-set_source_files_properties(per_dev_serialization.cpp PROPERTIES LANGUAGE HIP)
+set_source_files_properties(device_serialized_main.cpp PROPERTIES LANGUAGE HIP)
 add_executable(counter-collection-buffer-device-serialization)
 target_sources(counter-collection-buffer-device-serialization
-               PRIVATE per_dev_serialization.cpp)
+               PRIVATE device_serialized_main.cpp)
 target_link_libraries(counter-collection-buffer-device-serialization
                      PRIVATE counter-collection-buffer-client Threads::Threads)

@@ -109,7 +109,7 @@ set_tests_properties(

 add_library(counter-collection-functional-counter-client SHARED)
 target_sources(counter-collection-functional-counter-client
-               PRIVATE print_functional_counters.cpp client.hpp)
+               PRIVATE print_functional_counters_client.cpp client.hpp)
 target_link_libraries(
    counter-collection-functional-counter-client
    PUBLIC rocprofiler-sdk::samples-build-flags
@@ -136,8 +136,8 @@ set_tests_properties(
               "${ROCPROFILER_DEFAULT_FAIL_REGEX}")

 add_library(counter-collection-device-profiling-client SHARED)
-target_sources(counter-collection-device-profiling-client PRIVATE device_counting.cpp
-                                                                  client.hpp)
+target_sources(counter-collection-device-profiling-client
+               PRIVATE device_counting_async_client.cpp client.hpp)
 target_link_libraries(
    counter-collection-device-profiling-client
    PUBLIC rocprofiler-sdk::samples-build-flags
@@ -171,7 +171,7 @@ set_tests_properties(

 add_library(counter-collection-device-profiling-sync-client SHARED)
 target_sources(counter-collection-device-profiling-sync-client
-               PRIVATE device_counting_synchronous.cpp client.hpp)
+               PRIVATE device_counting_sync_client.cpp client.hpp)
 target_link_libraries(
    counter-collection-device-profiling-sync-client
    PUBLIC rocprofiler-sdk::samples-build-flags
@@ -22,6 +22,9 @@

 #include "client.hpp"

+#include <rocprofiler-sdk/registration.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
 #include <cstdint>
 #include <fstream>
 #include <functional>
@@ -33,10 +36,6 @@
 #include <unordered_map>
 #include <vector>

-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/registration.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #define ROCPROFILER_CALL(result, msg)                                                              \
    {                                                                                              \
        rocprofiler_status_t CHECKSTATUS = result;                                                 \
@@ -104,22 +103,15 @@ fill_dimension_cache(rocprofiler_counter_id_t counter)
 {
    assert(*dimension_cache() != nullptr);
    std::vector<rocprofiler_record_dimension_info_t> dims;
-    rocprofiler_available_dimensions_cb_t            cb =
-        [](rocprofiler_counter_id_t,
-           const rocprofiler_record_dimension_info_t* dim_info,
-           size_t                                     num_dims,
-           void*                                      user_data) {
-            std::vector<rocprofiler_record_dimension_info_t>* vec =
-                static_cast<std::vector<rocprofiler_record_dimension_info_t>*>(user_data);
-            for(size_t i = 0; i < num_dims; i++)
-            {
-                vec->push_back(dim_info[i]);
-            }
-            return ROCPROFILER_STATUS_SUCCESS;
-        };
-    ROCPROFILER_CALL(rocprofiler_iterate_counter_dimensions(counter, cb, &dims),
-                     "Could not iterate counter dimensions");
-    (*dimension_cache())->emplace(counter.handle, dims);
+    rocprofiler_counter_info_v1_t                    info;
+    ROCPROFILER_CALL(rocprofiler_query_counter_info(
+                         counter, ROCPROFILER_COUNTER_INFO_VERSION_1, static_cast<void*>(&info)),
+                     "Could not query info for counter");
+
+    (*dimension_cache())
+        ->emplace(counter.handle,
+                  std::vector<rocprofiler_record_dimension_info_t>{
+                      info.dimensions, info.dimensions + info.dimensions_count});
 }

 /**
@@ -187,22 +179,22 @@ buffered_callback(rocprofiler_context_id_t,
 * to select the profile config (and in turn counters) to use when a kernel dispatch
 * is received.
 */
-std::unordered_map<uint64_t, rocprofiler_profile_config_id_t>&
+std::unordered_map<uint64_t, rocprofiler_counter_config_id_t>&
 get_profile_cache()
 {
-    static std::unordered_map<uint64_t, rocprofiler_profile_config_id_t> profile_cache;
+    static std::unordered_map<uint64_t, rocprofiler_counter_config_id_t> profile_cache;
    return profile_cache;
 }

 /**
 * Callback from rocprofiler when an kernel dispatch is enqueued into the HSA queue.
- * rocprofiler_profile_config_id_t* is a return to specify what counters to collect
+ * rocprofiler_counter_config_id_t* is a return to specify what counters to collect
 * for this dispatch (dispatch_packet). This example function creates a profile
 * to collect the counter SQ_WAVES for all kernel dispatch packets.
 */
 void
 dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,
-                  rocprofiler_profile_config_id_t*             config,
+                  rocprofiler_counter_config_id_t*             config,
                  rocprofiler_user_data_t* /*user_data*/,
                  void* /*callback_data_args*/)
 {
@@ -236,7 +228,7 @@ dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,
 * to consturct them once in advance (i.e. in tool_init()) since there are non-trivial
 * costs associated with constructing the profile.
 */
-rocprofiler_profile_config_id_t
+rocprofiler_counter_config_id_t
 build_profile_for_agent(rocprofiler_agent_id_t       agent,
                        const std::set<std::string>& counters_to_collect)
 {
@@ -264,22 +256,22 @@ build_profile_for_agent(rocprofiler_agent_id_t       agent,
    std::vector<rocprofiler_counter_id_t> collect_counters;
    for(auto& counter : gpu_counters)
    {
-        rocprofiler_counter_info_v0_t version;
+        rocprofiler_counter_info_v0_t info;
        ROCPROFILER_CALL(
            rocprofiler_query_counter_info(
-                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&version)),
+                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&info)),
            "Could not query info for counter");
-        if(counters_to_collect.count(std::string(version.name)) > 0)
+        if(counters_to_collect.count(std::string(info.name)) > 0)
        {
-            std::clog << "Counter: " << counter.handle << " " << version.name << "\n";
+            std::clog << "Counter: " << counter.handle << " " << info.name << "\n";
            collect_counters.push_back(counter);
            fill_dimension_cache(counter);
        }
    }

    // Create and return the profile
-    rocprofiler_profile_config_id_t profile = {.handle = 0};
-    ROCPROFILER_CALL(rocprofiler_create_profile_config(
+    rocprofiler_counter_config_id_t profile = {.handle = 0};
+    ROCPROFILER_CALL(rocprofiler_create_counter_config(
                         agent, collect_counters.data(), collect_counters.size(), &profile),
                     "Could not construct profile cfg");

@@ -375,7 +367,7 @@ tool_init(rocprofiler_client_finalize_t, void* user_data)
    // counters to collect by returning a profile config id. In this example, we create the profile
    // configs above and store them in the map get_profile_cache() so we can look them up at
    // dispatch.
-    ROCPROFILER_CALL(rocprofiler_configure_buffered_dispatch_counting_service(
+    ROCPROFILER_CALL(rocprofiler_configure_buffer_dispatch_counting_service(
                         get_client_ctx(), get_buffer(), dispatch_callback, nullptr),
                     "Could not setup buffered service");

@@ -22,6 +22,9 @@

 #include "client.hpp"

+#include <rocprofiler-sdk/registration.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
 #include <fstream>
 #include <functional>
 #include <iostream>
@@ -32,9 +35,6 @@
 #include <unordered_map>
 #include <vector>

-#include <rocprofiler-sdk/registration.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #define ROCPROFILER_CALL(result, msg)                                                              \
    {                                                                                              \
        rocprofiler_status_t CHECKSTATUS = result;                                                 \
@@ -96,13 +96,13 @@ record_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,

 /**
 * Callback from rocprofiler when an kernel dispatch is enqueued into the HSA queue.
- * rocprofiler_profile_config_id_t* is a return to specify what counters to collect
+ * rocprofiler_counter_config_id_t* is a return to specify what counters to collect
 * for this dispatch (dispatch_packet). This example function creates a profile
 * to collect the counter SQ_WAVES for all kernel dispatch packets.
 */
 void
 dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,
-                  rocprofiler_profile_config_id_t*             config,
+                  rocprofiler_counter_config_id_t*             config,
                  rocprofiler_user_data_t* /*user_data*/,
                  void* /*callback_data_args*/)
 {
@@ -114,7 +114,7 @@ dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,
     * set.
     */
    static std::shared_mutex                                             m_mutex       = {};
-    static std::unordered_map<uint64_t, rocprofiler_profile_config_id_t> profile_cache = {};
+    static std::unordered_map<uint64_t, rocprofiler_counter_config_id_t> profile_cache = {};

    auto search_cache = [&]() {
        if(auto pos = profile_cache.find(dispatch_data.dispatch_info.agent_id.handle);
@@ -161,21 +161,21 @@ dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,
    // Look for the counters contained in counters_to_collect in gpu_counters
    for(auto& counter : gpu_counters)
    {
-        rocprofiler_counter_info_v0_t version;
+        rocprofiler_counter_info_v0_t info;
        ROCPROFILER_CALL(
            rocprofiler_query_counter_info(
-                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&version)),
+                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&info)),
            "Could not query info");
-        if(counters_to_collect.count(std::string(version.name)) > 0)
+        if(counters_to_collect.count(std::string(info.name)) > 0)
        {
-            std::clog << "Counter: " << counter.handle << " " << version.name << "\n";
+            std::clog << "Counter: " << counter.handle << " " << info.name << "\n";
            collect_counters.push_back(counter);
        }
    }

    // Create a colleciton profile for the counters
-    rocprofiler_profile_config_id_t profile = {.handle = 0};
-    ROCPROFILER_CALL(rocprofiler_create_profile_config(dispatch_data.dispatch_info.agent_id,
+    rocprofiler_counter_config_id_t profile = {.handle = 0};
+    ROCPROFILER_CALL(rocprofiler_create_counter_config(dispatch_data.dispatch_info.agent_id,
                                                       collect_counters.data(),
                                                       collect_counters.size(),
                                                       &profile),
@@ -22,6 +22,9 @@

 #include "client.hpp"

+#include <rocprofiler-sdk/registration.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
 #include <atomic>
 #include <chrono>
 #include <fstream>
@@ -35,10 +38,6 @@
 #include <unordered_map>
 #include <vector>

-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/registration.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #define ROCPROFILER_CALL(result, msg)                                                              \
    {                                                                                              \
        rocprofiler_status_t CHECKSTATUS = result;                                                 \
@@ -128,23 +127,23 @@ buffered_callback(rocprofiler_context_id_t,
    *output_stream << "[" << __FUNCTION__ << "] " << ss.str() << "\n";
 }

-std::unordered_map<uint64_t, rocprofiler_profile_config_id_t>&
+std::unordered_map<uint64_t, rocprofiler_counter_config_id_t>&
 get_profile_cache()
 {
-    static std::unordered_map<uint64_t, rocprofiler_profile_config_id_t> profile_cache;
+    static std::unordered_map<uint64_t, rocprofiler_counter_config_id_t> profile_cache;
    return profile_cache;
 }

 /**
 * Callback from rocprofiler when an kernel dispatch is enqueued into the HSA queue.
- * rocprofiler_profile_config_id_t* is a return to specify what counters to collect
+ * rocprofiler_counter_config_id_t* is a return to specify what counters to collect
 * for this dispatch (dispatch_packet). This example function creates a profile
 * to collect the counter SQ_WAVES for all kernel dispatch packets.
 */
 void
-set_profile(rocprofiler_context_id_t                 context_id,
-            rocprofiler_agent_id_t                   agent,
-            rocprofiler_agent_set_profile_callback_t set_config,
+set_profile(rocprofiler_context_id_t               context_id,
+            rocprofiler_agent_id_t                 agent,
+            rocprofiler_device_counting_agent_cb_t set_config,
            void*)
 {
    /**
@@ -170,7 +169,7 @@ set_profile(rocprofiler_context_id_t                 context_id,
    }
 }

-rocprofiler_profile_config_id_t
+rocprofiler_counter_config_id_t
 build_profile_for_agent(rocprofiler_agent_id_t agent)
 {
    std::set<std::string>                 counters_to_collect = {"SQ_WAVES"};
@@ -196,20 +195,20 @@ build_profile_for_agent(rocprofiler_agent_id_t agent)
    std::vector<rocprofiler_counter_id_t> collect_counters;
    for(auto& counter : gpu_counters)
    {
-        rocprofiler_counter_info_v0_t version;
+        rocprofiler_counter_info_v0_t info;
        ROCPROFILER_CALL(
            rocprofiler_query_counter_info(
-                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&version)),
+                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&info)),
            "Could not query info for counter");
-        if(counters_to_collect.count(std::string(version.name)) > 0)
+        if(counters_to_collect.count(std::string(info.name)) > 0)
        {
-            std::clog << "Counter: " << counter.handle << " " << version.name << "\n";
+            std::clog << "Counter: " << counter.handle << " " << info.name << "\n";
            collect_counters.push_back(counter);
        }
    }

-    rocprofiler_profile_config_id_t profile = {.handle = 0};
-    ROCPROFILER_CALL(rocprofiler_create_profile_config(
+    rocprofiler_counter_config_id_t profile = {.handle = 0};
+    ROCPROFILER_CALL(rocprofiler_create_counter_config(
                         agent, collect_counters.data(), collect_counters.size(), &profile),
                     "Could not construct profile cfg");

@@ -22,6 +22,9 @@

 #include "client.hpp"

+#include <rocprofiler-sdk/registration.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
 #include <atomic>
 #include <chrono>
 #include <cstdlib>
@@ -38,12 +41,6 @@
 #include <unordered_map>
 #include <vector>

-#include <rocprofiler-sdk/buffer.h>
-#include <rocprofiler-sdk/context.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/registration.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #define ROCPROFILER_CALL(result, msg)                                                              \
    {                                                                                              \
        rocprofiler_status_t CHECKSTATUS = result;                                                 \
@@ -99,15 +96,14 @@ private:
    rocprofiler_agent_id_t          agent_   = {};
    rocprofiler_context_id_t        ctx_     = {};
    rocprofiler_buffer_id_t         buf_     = {};
-    rocprofiler_profile_config_id_t profile_ = {.handle = 0};
+    rocprofiler_counter_config_id_t profile_ = {.handle = 0};

-    std::map<std::vector<std::string>, rocprofiler_profile_config_id_t> cached_profiles_;
+    std::map<std::vector<std::string>, rocprofiler_counter_config_id_t> cached_profiles_;
    std::map<uint64_t, uint64_t>                                        profile_sizes_;
    mutable std::map<uint64_t, std::string>                             id_to_name_;

    // Internal function used to set the profile for the agent when start_context is called
-    void set_profile(rocprofiler_context_id_t                 ctx,
-                     rocprofiler_agent_set_profile_callback_t cb) const;
+    void set_profile(rocprofiler_context_id_t ctx, rocprofiler_device_counting_agent_cb_t cb) const;

    // Get the size of a counter in number of records
    static size_t get_counter_size(rocprofiler_counter_id_t counter);
@@ -153,8 +149,8 @@ counter_sampler::counter_sampler(rocprofiler_agent_id_t agent)
                         agent,
                         [](rocprofiler_context_id_t context_id,
                            rocprofiler_agent_id_t,
-                            rocprofiler_agent_set_profile_callback_t set_config,
-                            void*                                    user_data) {
+                            rocprofiler_device_counting_agent_cb_t set_config,
+                            void*                                  user_data) {
                             if(user_data)
                             {
                                 auto* sampler = static_cast<counter_sampler*>(user_data);
@@ -212,7 +208,7 @@ counter_sampler::sample_counter_values(const std::vector<std::string>&
    if(profile_cached == cached_profiles_.end())
    {
        size_t                                expected_size = 0;
-        rocprofiler_profile_config_id_t       profile       = {};
+        rocprofiler_counter_config_id_t       profile       = {};
        std::vector<rocprofiler_counter_id_t> gpu_counters;
        auto                                  roc_counters = get_supported_counters(agent_);
        for(const auto& counter : counters)
@@ -226,7 +222,7 @@ counter_sampler::sample_counter_values(const std::vector<std::string>&
            gpu_counters.push_back(it->second);
            expected_size += get_counter_size(it->second);
        }
-        ROCPROFILER_CALL(rocprofiler_create_profile_config(
+        ROCPROFILER_CALL(rocprofiler_create_counter_config(
                             agent_, gpu_counters.data(), gpu_counters.size(), &profile),
                         "Could not create profile");
        cached_profiles_.emplace(counters, profile);
@@ -281,8 +277,8 @@ counter_sampler::get_available_agents()
 }

 void
-counter_sampler::set_profile(rocprofiler_context_id_t                 ctx,
-                             rocprofiler_agent_set_profile_callback_t cb) const
+counter_sampler::set_profile(rocprofiler_context_id_t               ctx,
+                             rocprofiler_device_counting_agent_cb_t cb) const
 {
    if(profile_.handle != 0)
    {
@@ -293,22 +289,11 @@ counter_sampler::set_profile(rocprofiler_context_id_t                 ctx,
 size_t
 counter_sampler::get_counter_size(rocprofiler_counter_id_t counter)
 {
-    size_t size = 1;
-    rocprofiler_iterate_counter_dimensions(
-        counter,
-        [](rocprofiler_counter_id_t,
-           const rocprofiler_record_dimension_info_t* dim_info,
-           size_t                                     num_dims,
-           void*                                      user_data) {
-            size_t* s = static_cast<size_t*>(user_data);
-            for(size_t i = 0; i < num_dims; i++)
-            {
-                *s *= dim_info[i].instance_size;
-            }
-            return ROCPROFILER_STATUS_SUCCESS;
-        },
-        static_cast<void*>(&size));
-    return size;
+    rocprofiler_counter_info_v1_t info;
+    ROCPROFILER_CALL(rocprofiler_query_counter_info(
+                         counter, ROCPROFILER_COUNTER_INFO_VERSION_1, static_cast<void*>(&info)),
+                     "Could not query info for counter");
+    return info.instance_ids_count;
 }

 std::unordered_map<std::string, rocprofiler_counter_id_t>
@@ -335,12 +320,12 @@ counter_sampler::get_supported_counters(rocprofiler_agent_id_t agent)
                     "Could not fetch supported counters");
    for(auto& counter : gpu_counters)
    {
-        rocprofiler_counter_info_v0_t version;
+        rocprofiler_counter_info_v0_t info;
        ROCPROFILER_CALL(
            rocprofiler_query_counter_info(
-                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&version)),
+                counter, ROCPROFILER_COUNTER_INFO_VERSION_0, static_cast<void*>(&info)),
            "Could not query info for counter");
-        out.emplace(version.name, counter);
+        out.emplace(info.name, counter);
    }
    return out;
 }
@@ -348,23 +333,12 @@ counter_sampler::get_supported_counters(rocprofiler_agent_id_t agent)
 std::vector<rocprofiler_record_dimension_info_t>
 counter_sampler::get_counter_dimensions(rocprofiler_counter_id_t counter)
 {
-    std::vector<rocprofiler_record_dimension_info_t> dims;
-    rocprofiler_available_dimensions_cb_t            cb =
-        [](rocprofiler_counter_id_t,
-           const rocprofiler_record_dimension_info_t* dim_info,
-           size_t                                     num_dims,
-           void*                                      user_data) {
-            std::vector<rocprofiler_record_dimension_info_t>* vec =
-                static_cast<std::vector<rocprofiler_record_dimension_info_t>*>(user_data);
-            for(size_t i = 0; i < num_dims; i++)
-            {
-                vec->push_back(dim_info[i]);
-            }
-            return ROCPROFILER_STATUS_SUCCESS;
-        };
-    ROCPROFILER_CALL(rocprofiler_iterate_counter_dimensions(counter, cb, &dims),
-                     "Could not iterate counter dimensions");
-    return dims;
+    rocprofiler_counter_info_v1_t info;
+    ROCPROFILER_CALL(rocprofiler_query_counter_info(
+                         counter, ROCPROFILER_COUNTER_INFO_VERSION_1, static_cast<void*>(&info)),
+                     "Could not query info for counter");
+    return std::vector<rocprofiler_record_dimension_info_t>{
+        info.dimensions, info.dimensions + info.dimensions_count};
 }

 std::atomic<bool>&
@@ -22,7 +22,11 @@

 #include "client.hpp"

+#include <rocprofiler-sdk/registration.h>
+#include <rocprofiler-sdk/rocprofiler.h>
+
 #include <unistd.h>
+#include <cstdint>
 #include <map>
 #include <mutex>
 #include <optional>
@@ -30,10 +34,6 @@
 #include <sstream>
 #include <vector>

-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/registration.h>
-#include <rocprofiler-sdk/rocprofiler.h>
-
 #define PRINT_ONLY_FAILING false

 /**
@@ -245,7 +245,7 @@ get_agent_info()

 void
 dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,
-                  rocprofiler_profile_config_id_t*             config,
+                  rocprofiler_counter_config_id_t*             config,
                  rocprofiler_user_data_t* /*user_data*/,
                  void* /*callback_data_args*/)
 {
@@ -282,40 +282,24 @@ dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,

        for(auto& found_counter : counters_needed)
        {
-            rocprofiler_counter_info_v0_t version;
+            rocprofiler_counter_info_v1_t info;

-            ROCPROFILER_CALL(rocprofiler_query_counter_info(found_counter,
-                                                            ROCPROFILER_COUNTER_INFO_VERSION_0,
-                                                            static_cast<void*>(&version)),
-                             "Could not query counter_id");
-            cap.expected_counter_names.emplace(found_counter.handle, std::string(version.name));
-            size_t expected = 0;
-            ROCPROFILER_CALL(rocprofiler_query_counter_instance_count(
-                                 dispatch_data.dispatch_info.agent_id, found_counter, &expected),
-                             "COULD NOT QUERY INSTANCES");
+            ROCPROFILER_CALL(
+                rocprofiler_query_counter_info(
+                    found_counter, ROCPROFILER_COUNTER_INFO_VERSION_1, static_cast<void*>(&info)),
+                "Could not query counter_id");
+            cap.expected_counter_names.emplace(found_counter.handle, std::string(info.name));
            cap.remaining.push_back(found_counter);
-            cap.expected.emplace(found_counter.handle, expected);
+            cap.expected.emplace(found_counter.handle, info.instance_ids_count);

            auto& info_vector =
                cap.expected_data_dims.emplace(found_counter.handle, validate_dim_presence{})
                    .first->second;

-            ROCPROFILER_CALL(rocprofiler_iterate_counter_dimensions(
-                                 found_counter,
-                                 [](rocprofiler_counter_id_t,
-                                    const rocprofiler_record_dimension_info_t* dim_info,
-                                    size_t                                     num_dims,
-                                    void*                                      user_data) {
-                                     validate_dim_presence* dim_presence =
-                                         static_cast<validate_dim_presence*>(user_data);
-                                     for(size_t i = 0; i < num_dims; i++)
-                                     {
-                                         dim_presence->maybe_forward(dim_info[i]);
-                                     }
-                                     return ROCPROFILER_STATUS_SUCCESS;
-                                 },
-                                 static_cast<void*>(&info_vector)),
-                             "Could not fetch dimension info");
+            for(uint64_t i = 0; i < info.dimensions_count; i++)
+            {
+                info_vector.maybe_forward(info.dimensions[i]);
+            }
        }
        if(cap.expected.empty())
        {
@@ -326,10 +310,10 @@ dispatch_callback(rocprofiler_dispatch_counting_service_data_t dispatch_data,
    }
    if(cap.remaining.empty()) return;

-    rocprofiler_profile_config_id_t profile = {.handle = 0};
+    rocprofiler_counter_config_id_t profile = {.handle = 0};

    // Select the next counter to collect.
-    if(rocprofiler_create_profile_config(
+    if(rocprofiler_create_counter_config(
           dispatch_data.dispatch_info.agent_id, &(cap.remaining.back()), 1, &profile) ==
       ROCPROFILER_STATUS_SUCCESS)
    {
@@ -362,7 +346,7 @@ tool_init(rocprofiler_client_finalize_t, void*)

    ROCPROFILER_CALL(rocprofiler_assign_callback_thread(get_buffer(), client_thread),
                     "failed to assign thread for buffer");
-    ROCPROFILER_CALL(rocprofiler_configure_buffered_dispatch_counting_service(
+    ROCPROFILER_CALL(rocprofiler_configure_buffer_dispatch_counting_service(
                         get_client_ctx(), get_buffer(), dispatch_callback, nullptr),
                     "Could not setup buffered service");
    rocprofiler_start_context(get_client_ctx());
@@ -33,14 +33,6 @@

 #include "client.hpp"

-#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/external_correlation.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/hip/runtime_api_id.h>
-#include <rocprofiler-sdk/internal_threading.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/rocprofiler.h>

@@ -33,11 +33,6 @@

 #include "client.hpp"

-#include <rocprofiler-sdk/callback_tracing.h>
-#include <rocprofiler-sdk/context.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/marker/api_id.h>
-#include <rocprofiler-sdk/ompt.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/rocprofiler.h>

@@ -34,9 +34,6 @@
 #include "pcs.hpp"
 #include "utils.hpp"

-#include <rocprofiler-sdk/buffer.h>
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/internal_threading.h>
 #include <rocprofiler-sdk/registration.h>
 #include <rocprofiler-sdk/rocprofiler.h>