(IMPORTANT: AFAR V ) Hide PC sampling if HSA does not support it. (#881)

* Hide PC sampling if HSA does not support it.

* missing hsa.hpp include

source/lib/rocprofiler-sdk/buffer.cpp

@@ -272,9 +272,11 @@ rocprofiler_create_buffer(rocprofiler_context_id_t        context,
 rocprofiler_status_t
 rocprofiler_flush_buffer(rocprofiler_buffer_id_t buffer_id)
 {
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
     // Drain internal PC sampling buffers, if needed.
     auto status = rocprofiler::pc_sampling::flush_internal_agent_buffers(buffer_id);
     if(status != ROCPROFILER_STATUS_SUCCESS) return status;
+#endif
 
     return rocprofiler::buffer::flush(buffer_id, true);
 }

source/lib/rocprofiler-sdk/context/context.cpp

@@ -323,7 +323,9 @@ start_context(rocprofiler_context_id_t context_id)
     if(cfg->counter_collection) rocprofiler::counters::start_context(cfg);
     if(cfg->thread_trace) cfg->thread_trace->start_context();
     if(cfg->agent_counter_collection) status = rocprofiler::counters::start_agent_ctx(cfg);
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
     if(cfg->pc_sampler) status = rocprofiler::pc_sampling::start_service(cfg);
+#endif
 
     return status;
 }
@@ -360,10 +362,12 @@ stop_context(rocprofiler_context_id_t idx)
                     rocprofiler::counters::stop_agent_ctx(const_cast<context*>(_expected));
                 }
 
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
                 if(_expected->pc_sampler)
                 {
                     rocprofiler::pc_sampling::stop_service(_expected);
                 }
+#endif
 
                 return ROCPROFILER_STATUS_SUCCESS;
             }

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

@@ -371,11 +371,13 @@ WriteInterceptor(const void* packets,
             CreateBarrierPacket(nullptr, nullptr, transformed_packets);
         }
 
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
         if(pc_sampling::is_pc_sample_service_configured(queue.get_agent().get_rocp_agent()->id))
         {
             transformed_packets.emplace_back(pc_sampling::hsa::generate_marker_packet_for_kernel(
                 corr_id, tracing_data_v.external_correlation_ids));
         }
+#endif
 
         transformed_packets.emplace_back(kernel_pkt);
 

source/lib/rocprofiler-sdk/pc_sampling.cpp

@@ -92,6 +92,8 @@ rocprofiler_configure_pc_sampling_service(rocprofiler_context_id_t         conte
     (void) interval;
     (void) buffer_id;
 
+    ROCP_ERROR << "PC sampling unavailable\n";
+
     // ROCr runtime is missing PC sampling.
     return ROCPROFILER_STATUS_ERROR_NOT_AVAILABLE;
 #endif
@@ -118,6 +120,8 @@ rocprofiler_query_pc_sampling_agent_configurations(
     (void) cb;
     (void) user_data;
 
+    ROCP_ERROR << "PC sampling unavailable\n";
+
     // ROCr runtime is missing PC sampling.
     return ROCPROFILER_STATUS_ERROR_NOT_AVAILABLE;
 #endif

source/lib/rocprofiler-sdk/pc_sampling/code_object.cpp

@@ -22,19 +22,21 @@
 
 #include "lib/rocprofiler-sdk/pc_sampling/code_object.hpp"
 
-#include "lib/common/container/operators.hpp"
-#include "lib/common/logging.hpp"
-#include "lib/rocprofiler-sdk/code_object/code_object.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/service.hpp"
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
 
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/pc_sampling.h>
-#include <rocprofiler-sdk/cxx/operators.hpp>
+#    include "lib/common/container/operators.hpp"
+#    include "lib/common/logging.hpp"
+#    include "lib/rocprofiler-sdk/code_object/code_object.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/service.hpp"
 
-#include <glog/logging.h>
-#include <hsa/hsa.h>
-#include <hsa/hsa_api_trace.h>
-#include <hsa/hsa_ven_amd_loader.h>
+#    include <rocprofiler-sdk/fwd.h>
+#    include <rocprofiler-sdk/pc_sampling.h>
+#    include <rocprofiler-sdk/cxx/operators.hpp>
+
+#    include <glog/logging.h>
+#    include <hsa/hsa.h>
+#    include <hsa/hsa_api_trace.h>
+#    include <hsa/hsa_ven_amd_loader.h>
 
 namespace rocprofiler
 {
@@ -188,3 +190,5 @@ finalize()
 }  // namespace code_object
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif

source/lib/rocprofiler-sdk/pc_sampling/code_object.hpp

@@ -22,7 +22,11 @@
 
 #pragma once
 
-#include <hsa/hsa_api_trace.h>
+#include "lib/rocprofiler-sdk/hsa/hsa.hpp"
+
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
+
+#    include <hsa/hsa_api_trace.h>
 
 namespace rocprofiler
 {
@@ -38,3 +42,5 @@ finalize();
 }  // namespace code_object
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif

source/lib/rocprofiler-sdk/pc_sampling/hsa_adapter.cpp

@@ -22,23 +22,25 @@
 
 #include "lib/rocprofiler-sdk/pc_sampling/hsa_adapter.hpp"
 
-#include "lib/common/logging.hpp"
-#include "lib/rocprofiler-sdk/context/context.hpp"
-#include "lib/rocprofiler-sdk/hsa/hsa.hpp"
-#include "lib/rocprofiler-sdk/hsa/queue_controller.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/parser/pc_record_interface.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/service.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/types.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/utils.hpp"
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
 
-#include <hsa/hsa.h>
-#include <hsa/hsa_ext_amd.h>
-#include <hsa/hsa_ven_amd_pc_sampling.h>
+#    include "lib/common/logging.hpp"
+#    include "lib/rocprofiler-sdk/context/context.hpp"
+#    include "lib/rocprofiler-sdk/hsa/hsa.hpp"
+#    include "lib/rocprofiler-sdk/hsa/queue_controller.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/parser/pc_record_interface.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/service.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/types.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/utils.hpp"
 
-#include <mutex>
-#include <optional>
-#include <shared_mutex>
-#include <stdexcept>
+#    include <hsa/hsa.h>
+#    include <hsa/hsa_ext_amd.h>
+#    include <hsa/hsa_ven_amd_pc_sampling.h>
+
+#    include <mutex>
+#    include <optional>
+#    include <shared_mutex>
+#    include <stdexcept>
 
 namespace rocprofiler
 {
@@ -376,3 +378,5 @@ flush_internal_agent_buffers(const PCSAgentSession* agent_session)
 }  // namespace hsa
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif

source/lib/rocprofiler-sdk/pc_sampling/hsa_adapter.hpp

@@ -22,12 +22,16 @@
 
 #pragma once
 
-#include "lib/rocprofiler-sdk/context/context.hpp"
-#include "lib/rocprofiler-sdk/hsa/queue.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/types.hpp"
-#include "lib/rocprofiler-sdk/tracing/fwd.hpp"
+#include "lib/rocprofiler-sdk/hsa/hsa.hpp"
 
-#include <hsa/hsa_api_trace.h>
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
+
+#    include "lib/rocprofiler-sdk/context/context.hpp"
+#    include "lib/rocprofiler-sdk/hsa/queue.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/types.hpp"
+#    include "lib/rocprofiler-sdk/tracing/fwd.hpp"
+
+#    include <hsa/hsa_api_trace.h>
 
 namespace rocprofiler
 {
@@ -54,3 +58,5 @@ flush_internal_agent_buffers(const PCSAgentSession* agent_session);
 }  // namespace hsa
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif

source/lib/rocprofiler-sdk/pc_sampling/service.cpp

@@ -22,10 +22,12 @@
 
 #include "lib/rocprofiler-sdk/pc_sampling/service.hpp"
 
-#include "lib/common/logging.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/hsa_adapter.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/ioctl/ioctl_adapter.hpp"
-#include "lib/rocprofiler-sdk/pc_sampling/utils.hpp"
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
+
+#    include "lib/common/logging.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/hsa_adapter.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/ioctl/ioctl_adapter.hpp"
+#    include "lib/rocprofiler-sdk/pc_sampling/utils.hpp"
 
 namespace rocprofiler
 {
@@ -266,3 +268,5 @@ flush_internal_agent_buffers(rocprofiler_buffer_id_t buffer_id)
 
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif

source/lib/rocprofiler-sdk/pc_sampling/service.hpp

@@ -22,14 +22,18 @@
 
 #pragma once
 
-#include "lib/rocprofiler-sdk/context/context.hpp"
+#include "lib/rocprofiler-sdk/hsa/hsa.hpp"
 
-#include <rocprofiler-sdk/fwd.h>
-#include <rocprofiler-sdk/pc_sampling.h>
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
 
-#include <hsa/hsa_api_trace.h>
+#    include "lib/rocprofiler-sdk/context/context.hpp"
 
-#include <atomic>
+#    include <rocprofiler-sdk/fwd.h>
+#    include <rocprofiler-sdk/pc_sampling.h>
+
+#    include <hsa/hsa_api_trace.h>
+
+#    include <atomic>
 
 namespace rocprofiler
 {
@@ -64,3 +68,5 @@ rocprofiler_status_t
 flush_internal_agent_buffers(rocprofiler_buffer_id_t buffer_id);
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif

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

@@ -1,5 +1,6 @@
 #pragma once
 
+#include "lib/rocprofiler-sdk/hsa/hsa.hpp"
 #include "lib/rocprofiler-sdk/hsa/queue.hpp"
 #include "lib/rocprofiler-sdk/pc_sampling/cid_manager.hpp"
 #include "lib/rocprofiler-sdk/pc_sampling/parser/pc_record_interface.hpp"
@@ -8,7 +9,9 @@
 #include <rocprofiler-sdk/fwd.h>
 
 #include <hsa/hsa.h>
-#include <hsa/hsa_ven_amd_pc_sampling.h>
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
+#    include <hsa/hsa_ven_amd_pc_sampling.h>
+#endif
 
 #include <memory>
 
@@ -28,8 +31,10 @@ struct PCSAgentSession
     rocprofiler_buffer_id_t          buffer_id;
     // hsa relevant information
     std::optional<hsa_agent_t> hsa_agent = std::nullopt;
-    hsa_ven_amd_pcs_t          hsa_pc_sampling;
-    hsa::ClientID              intercept_cb_id{-1};
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
+    hsa_ven_amd_pcs_t hsa_pc_sampling;
+#endif
+    hsa::ClientID intercept_cb_id{-1};
     // ioctl relevant information
     uint32_t ioctl_pcs_id;
     // PC sampling parser

source/lib/rocprofiler-sdk/pc_sampling/utils.cpp

@@ -21,14 +21,15 @@
 // SOFTWARE.
 
 #include "lib/rocprofiler-sdk/pc_sampling/utils.hpp"
-#include "lib/rocprofiler-sdk/agent.hpp"
 
-#include <hsa/hsa_ext_amd.h>
-#include <hsa/hsa_ven_amd_pc_sampling.h>
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
 
-#include <fmt/core.h>
-#include <glog/logging.h>
-#include <vector>
+#    include "lib/rocprofiler-sdk/agent.hpp"
+
+#    include <hsa/hsa_ext_amd.h>
+#    include <hsa/hsa_ven_amd_pc_sampling.h>
+
+#    include <stdexcept>
 
 namespace rocprofiler
 {
@@ -41,10 +42,13 @@ get_matching_hsa_pcs_method(rocprofiler_pc_sampling_method_t method)
 {
     switch(method)
     {
+        case ROCPROFILER_PC_SAMPLING_METHOD_NONE: break;
         case ROCPROFILER_PC_SAMPLING_METHOD_STOCHASTIC: return HSA_VEN_AMD_PCS_METHOD_STOCHASTIC_V1;
         case ROCPROFILER_PC_SAMPLING_METHOD_HOST_TRAP: return HSA_VEN_AMD_PCS_METHOD_HOSTTRAP_V1;
-        default: throw std::runtime_error("Illegal pc sampling method\n");
+        case ROCPROFILER_PC_SAMPLING_METHOD_LAST: break;
     }
+
+    throw std::runtime_error("Illegal pc sampling method\n");
 }
 
 hsa_ven_amd_pcs_units_t
@@ -63,17 +67,8 @@ get_matching_hsa_pcs_units(rocprofiler_pc_sampling_unit_t unit)
 
     throw std::runtime_error("Illegal pc sampling units\n");
 }
-
-uint64_t
-get_unique_correlation_id()
-{
-    // TODO: Remove once we confirmed it is unnecessary.
-    // Also, update the PC sampling parser not to decode correlation ID
-    // (or always 0 for both internal/external correlation IDs)
-    static auto _cnt = std::atomic<int>{0};
-    return ++_cnt;
-}
-
 }  // namespace utils
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif

source/lib/rocprofiler-sdk/pc_sampling/utils.hpp

@@ -20,14 +20,20 @@
 // 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/pc_sampling.h>
+#pragma once
 
-#include <hsa/hsa.h>
-#include <hsa/hsa_ven_amd_pc_sampling.h>
+#include "lib/rocprofiler-sdk/hsa/hsa.hpp"
 
-#include <cstdint>
-#include <optional>
+#if ROCPROFILER_SDK_HSA_PC_SAMPLING > 0
+
+#    include <rocprofiler-sdk/fwd.h>
+#    include <rocprofiler-sdk/pc_sampling.h>
+
+#    include <hsa/hsa.h>
+#    include <hsa/hsa_ven_amd_pc_sampling.h>
+
+#    include <cstdint>
+#    include <optional>
 
 namespace rocprofiler
 {
@@ -54,9 +60,8 @@ get_hsa_pcs_buffer_size()
     // TODO: Find the minimum size of all buffers and use that.
     return 1024 * sizeof(perf_sample_hosttrap_v1_t);
 }
-
-uint64_t
-get_unique_correlation_id();
 }  // namespace utils
 }  // namespace pc_sampling
 }  // namespace rocprofiler
+
+#endif