Add support for hsa_amd_queue_get_info (#752)

* Add support for hsa_amd_queue_get_info

- HSA_AMD_EXT_API_TABLE_STEP_VERSION == 0x02

* Suppress unused-but-set-parameter warnings

source/include/rocprofiler-sdk/hsa/amd_ext_api_id.h

@@ -103,6 +103,9 @@ typedef enum  // NOLINT(performance-enum-size)
 #    if HSA_AMD_EXT_API_TABLE_STEP_VERSION >= 0x01
     ROCPROFILER_HSA_AMD_EXT_API_ID_hsa_amd_agent_set_async_scratch_limit,
 #    endif
+#    if HSA_AMD_EXT_API_TABLE_STEP_VERSION >= 0x02
+    ROCPROFILER_HSA_AMD_EXT_API_ID_hsa_amd_queue_get_info,
+#    endif
 #endif
 
     ROCPROFILER_HSA_AMD_EXT_API_ID_LAST,

source/include/rocprofiler-sdk/hsa/api_args.h

@@ -1389,6 +1389,14 @@ typedef union rocprofiler_hsa_api_args_t
         size_t      threshold;
     } hsa_amd_agent_set_async_scratch_limit;
 #    endif
+#    if HSA_AMD_EXT_API_TABLE_STEP_VERSION >= 0x02
+    struct
+    {
+        hsa_queue_t*               queue;
+        hsa_queue_info_attribute_t attribute;
+        void*                      value;
+    } hsa_amd_queue_get_info;
+#    endif
 #endif
 } rocprofiler_hsa_api_args_t;
 

source/lib/common/utility.hpp

@@ -49,6 +49,11 @@ namespace rocprofiler
 {
 namespace common
 {
+template <typename... Tp>
+void
+consume_args(Tp&&...)
+{}
+
 uint64_t
 get_clock_period_ns_impl(clockid_t _clk_id);
 

source/lib/rocprofiler-sdk/hsa/hsa.def.cpp

@@ -407,6 +407,15 @@ HSA_API_INFO_DEFINITION_V(ROCPROFILER_HSA_TABLE_ID_AmdExt,
                           agent,
                           threshold)
 #        endif
+#        if HSA_AMD_EXT_API_TABLE_STEP_VERSION >= 0x02
+HSA_API_INFO_DEFINITION_V(ROCPROFILER_HSA_TABLE_ID_AmdExt,
+                          ROCPROFILER_HSA_AMD_EXT_API_ID_hsa_amd_queue_get_info,
+                          hsa_amd_queue_get_info,
+                          hsa_amd_queue_get_info_fn,
+                          queue,
+                          attribute,
+                          value)
+#        endif
 #    endif
 
 #elif defined(ROCPROFILER_LIB_ROCPROFILER_HSA_ASYNC_COPY_CPP_IMPL) &&                              \

source/lib/rocprofiler-sdk/hsa/scratch_memory.cpp

@@ -484,7 +484,8 @@ impl(Args... args)
 
     const auto tid = common::get_tid();
 
-    const auto get_agent_id = [](const hsa_queue_t* hsa_queue) -> rocprofiler_agent_id_t {
+    [[maybe_unused]] const auto get_agent_id =
+        [](const hsa_queue_t* hsa_queue) -> rocprofiler_agent_id_t {
         rocprofiler_agent_id_t _agent_id{static_cast<uint64_t>(-1)};
         bool                   found_agent{false};
 
@@ -635,6 +636,9 @@ update_table(const context_array_t& ctxs, hsa_amd_tool_table_t* _orig)
 
         _func = get_hsa_amd_tool_api_impl<TableIdx, OpIdx>(_func);
     }
+
+    // suppress unused paramter
+    common::consume_args(ctxs, _orig);
 }
 
 template <size_t TableIdx, size_t... OpIdx>

source/lib/rocprofiler-sdk/hsa/types.hpp

@@ -169,8 +169,10 @@ struct table_size<ROCPROFILER_COMPUTE_VERSION(1, 13, 0)>
     // TODO(jomadsen): come up with a better way of handling this
 #            if HSA_AMD_EXT_API_TABLE_STEP_VERSION == 0x00
     static constexpr size_t amd_ext = 552;
-#            else
+#            elif HSA_AMD_EXT_API_TABLE_STEP_VERSION == 0x1
     static constexpr size_t amd_ext = 560;
+#            else
+    static constexpr size_t amd_ext = 568;
 #            endif
 };
 

source/lib/rocprofiler-sdk/pc_sampling.cpp

@@ -23,15 +23,10 @@
 #include <rocprofiler-sdk/fwd.h>
 #include <rocprofiler-sdk/pc_sampling.h>
 
+#include "lib/common/utility.hpp"
 #include "lib/rocprofiler-sdk/registration.hpp"
 
-namespace
-{
-template <typename... Tp>
-auto
-consume_args(Tp&&...)
-{}
-}  // namespace
+using ::rocprofiler::common::consume_args;
 
 extern "C" {
 rocprofiler_status_t