adding major and minor library versions

[ROCm/rocprofiler commit: 6560364df1]

projects/rocprofiler/inc/rocprofiler.h

@@ -66,10 +66,19 @@
 #include <hsa_ven_amd_aqlprofile.h>
 #include <stdint.h>
 
+#define ROCPROFILER_VERSION_MAJOR 1
+#define ROCPROFILER_VERSION_MINOR 0
+
 #ifdef __cplusplus
 extern "C" {
 #endif  // __cplusplus
 
+////////////////////////////////////////////////////////////////////////////////
+// Returning library version
+
+uint32_t rocprofiler_version_major();
+uint32_t rocprofiler_version_minor();
+
 ////////////////////////////////////////////////////////////////////////////////
 // Returning the error string method
 
@@ -305,6 +314,7 @@ typedef union {
 
 // Profiling info data
 typedef struct {
+  uint32_t agent_idx;
   rocprofiler_info_kind_t kind; // info data kind
   union {
     struct {
@@ -327,20 +337,20 @@ typedef struct {
 
 // Return the info for a given info kind
 hsa_status_t rocprofiler_get_info(
-  hsa_agent_t agent, // GFXIP handle
+  const hsa_agent_t* agent, // [in] GFXIP handle
   rocprofiler_info_kind_t kind, // kind of iterated info
   void *data); // [in/out] returned data
 
 // Iterate over the info for a given info kind, and invoke an application-defined callback on every iteration
 hsa_status_t rocprofiler_iterate_info(
-  hsa_agent_t agent, // GFXIP handle
+  const hsa_agent_t* agent, // [in] GFXIP handle
   rocprofiler_info_kind_t kind, // kind of iterated info
   hsa_status_t (*callback)(const rocprofiler_info_data_t info, void *data), // callback
   void *data); // [in/out] data passed to callback
 
 // Iterate over the info for a given info query, and invoke an application-defined callback on every iteration
 hsa_status_t rocprofiler_query_info(
-  hsa_agent_t agent, // GFXIP handle
+  const hsa_agent_t *agent, // [in] GFXIP handle
   rocprofiler_info_query_t query, // iterated info query
   hsa_status_t (*callback)(const rocprofiler_info_data_t info, void *data), // callback
   void *data); // [in/out] data passed to callback

projects/rocprofiler/src/core/context.h

@@ -190,6 +190,7 @@ class Context {
     // Register input features to not duplicate by features referencing
     for (unsigned i = 0; i < info_count; ++i) {
       rocprofiler_feature_t* info = &info_array[i];
+      if (!info->name) EXC_RAISING(HSA_STATUS_ERROR, "input feature name is NULL");
       info_map_[info->name] = info;
     }
 

projects/rocprofiler/src/core/metrics.h

@@ -199,8 +199,10 @@ class MetricsDict {
             for (unsigned block_index = 0; block_index < query.instance_count; ++block_index) {
               std::ostringstream full_name;
               full_name << name << '[' << block_index << ']';
+              std::ostringstream block_insance;
+              block_insance << block_name << "[" << block_index << "]";
               std::ostringstream alias;
-              alias << block_name << "[" << block_index << "]:" << event_str;
+              alias << block_insance.str() << ":" << event_str;
               const counter_t counter = {full_name.str(), {block_id, block_index, event_id}};
               AddMetric(full_name.str(), alias.str(), counter);
             }
@@ -211,8 +213,10 @@ class MetricsDict {
           }
         } else {
           xml::Expr* expr_obj = new xml::Expr(expr_str, new ExprCache(&cache_));
+#if 0
           std::cout << "# " << descr << std::endl;
           std::cout << name << "=" << expr_obj->String() << "\n" << std::endl;
+#endif
           counters_vec_t counters_vec;
           for (const std::string var : expr_obj->GetVars()) {
             auto it = cache_.find(var);

projects/rocprofiler/src/core/rocprofiler.cpp

@@ -153,6 +153,10 @@ PUBLIC_API bool OnLoad(HsaApiTable* table, uint64_t runtime_version, uint64_t fa
 // HSA-runtime tool on-unload method
 PUBLIC_API void OnUnload() { rocprofiler::RestoreHsaApi(); }
 
+// Returns library vesrion
+PUBLIC_API uint32_t rocprofiler_version_major() { return ROCPROFILER_VERSION_MAJOR; }
+PUBLIC_API uint32_t rocprofiler_version_minor() { return ROCPROFILER_VERSION_MINOR; }
+
 // Returns the last error message
 PUBLIC_API hsa_status_t rocprofiler_error_string(const char** str) {
   API_METHOD_PREFIX
@@ -320,16 +324,17 @@ PUBLIC_API hsa_status_t rocprofiler_iterate_trace_data(
 
 // Return the info for a given info kind
 PUBLIC_API hsa_status_t rocprofiler_get_info(
-  hsa_agent_t agent,
+  const hsa_agent_t *agent,
   rocprofiler_info_kind_t kind,
   void *data)
 {
   API_METHOD_PREFIX
+  if (agent == NULL) EXC_RAISING(HSA_STATUS_ERROR, "NULL agent");
   uint32_t* result_32bit_ptr = reinterpret_cast<uint32_t*>(data);
 
   switch (kind) {
     case ROCPROFILER_INFO_KIND_METRIC_COUNT:
-      *result_32bit_ptr = rocprofiler::GetMetrics(agent)->Size();
+      *result_32bit_ptr = rocprofiler::GetMetrics(*agent)->Size();
       break;
     case ROCPROFILER_INFO_KIND_TRACE_COUNT:
       *result_32bit_ptr = 1;
@@ -342,44 +347,63 @@ PUBLIC_API hsa_status_t rocprofiler_get_info(
 
 // Iterate over the info for a given info kind, and invoke an application-defined callback on every iteration
 PUBLIC_API hsa_status_t rocprofiler_iterate_info(
-  hsa_agent_t agent,
+  const hsa_agent_t* agent,
   rocprofiler_info_kind_t kind,
-  hsa_status_t (*callback)(const rocprofiler_info_data_t info, void *data),
+  hsa_status_t (*callback)(const rocprofiler_info_data_t info, void* data),
-  void *data)
+  void* data)
 {
   API_METHOD_PREFIX
+  rocprofiler::util::HsaRsrcFactory* hsa_rsrc = &rocprofiler::util::HsaRsrcFactory::Instance();
   rocprofiler_info_data_t info{};
   info.kind = kind;
+  uint32_t agent_idx = 0;
+  uint32_t agent_max = 0;
+  const rocprofiler::util::AgentInfo* agent_info = NULL;
 
-  switch (kind) {
+  if (agent != NULL) {
-    case ROCPROFILER_INFO_KIND_METRIC:
+    agent_info = hsa_rsrc->GetAgentInfo(*agent);
-    {
+    agent_idx = agent_info->dev_index;
-      const rocprofiler::MetricsDict* dict = rocprofiler::GetMetrics(agent);
+    agent_max = agent_idx + 1;
-      rocprofiler::MetricsDict::const_iterator_t it = dict->Begin();
+  }
-      rocprofiler::MetricsDict::const_iterator_t end = dict->End();
-      while (it != end) {
-        const rocprofiler::Metric* metric = it->second;
-        std::string name = metric->GetName();
-        const auto* expr = metric->GetExpr();
-        std::string description = "Performance metric " + name + " " + ((expr == NULL) ? "basic" : "= " + expr->String());
-        info.metric.name = strdup(name.c_str());
-        info.metric.description = strdup(description.c_str());
-        status = callback(info, data);
 
-        ++it;
+  while (hsa_rsrc->GetGpuAgentInfo(agent_idx, &agent_info)) {
+    info.agent_idx = agent_idx;
+
+    switch (kind) {
+      case ROCPROFILER_INFO_KIND_METRIC:
+      {
+        const rocprofiler::MetricsDict* dict = rocprofiler::GetMetrics(agent_info->dev_id);
+        rocprofiler::MetricsDict::const_iterator_t it = dict->Begin();
+        rocprofiler::MetricsDict::const_iterator_t end = dict->End();
+        while (it != end) {
+          const rocprofiler::Metric* metric = it->second;
+          std::string name = metric->GetName();
+          //std::string descr = metric->GetDescr();
+          const auto* expr = metric->GetExpr();
+          std::string description = "Performance metric " + name + " " + ((expr == NULL) ? "basic" : "= " + expr->String());
+          info.metric.name = strdup(name.c_str());
+          info.metric.description = strdup(description.c_str());
+          status = callback(info, data);
+          if (status != HSA_STATUS_SUCCESS) break;
+          ++it;
+        }
+        break;
       }
-      break;
+      case ROCPROFILER_INFO_KIND_TRACE:
+      {
+        info.trace.name = strdup("TT");
+        info.trace.description = strdup("Thread Trace");
+        info.trace.parameter_count = 5;
+        status = callback(info, data);
+        if (status != HSA_STATUS_SUCCESS) break;
+        break;
+      }
+      default:
+        EXC_RAISING(HSA_STATUS_ERROR, "unknown info kind(" << kind << ")");
     }
-    case ROCPROFILER_INFO_KIND_TRACE:
+
-    {
+    ++agent_idx;
-      info.trace.name = strdup("TT");
+    if (agent_idx == agent_max) break;
-      info.trace.description = strdup("Thread Trace");
-      info.trace.parameter_count = 5;
-      status = callback(info, data);
-      break;
-    }
-    default:
-      EXC_RAISING(HSA_STATUS_ERROR, "unknown info kind(" << kind << ")");
   }
 
   if (status == HSA_STATUS_INFO_BREAK) status = HSA_STATUS_SUCCESS;
@@ -390,7 +414,7 @@ PUBLIC_API hsa_status_t rocprofiler_iterate_info(
 
 // Iterate over the info for a given info query, and invoke an application-defined callback on every iteration
 PUBLIC_API hsa_status_t rocprofiler_query_info(
-  hsa_agent_t agent,
+  const hsa_agent_t *agent,
   rocprofiler_info_query_t query,
   hsa_status_t (*callback)(const rocprofiler_info_data_t info, void *data),
   void *data)

projects/rocprofiler/src/util/logger.h

@@ -91,6 +91,8 @@ class Logger {
     std::lock_guard<mutex_t> lck(mutex_);
     if (messaging) {
       message_[GetTid()] = "";
+    } else if (streaming_) {
+      Put("\n");
     }
     messaging_ = messaging;
     streaming_ = messaging;

projects/rocprofiler/test/CMakeLists.txt

@@ -30,6 +30,7 @@ target_include_directories ( ${EXE_NAME} PRIVATE ${TEST_DIR} ${ROOT_DIR} ${HSA_R
 target_link_libraries( ${EXE_NAME} ${ROCPROFILER_TARGET} ${HSA_RUNTIME_LIB} c stdc++ dl pthread rt atomic )
 execute_process ( COMMAND sh -xc "cp ${TEST_DIR}/run.sh ${PROJECT_BINARY_DIR}" )
 execute_process ( COMMAND sh -xc "cp ${TEST_DIR}/*.xml ${PROJECT_BINARY_DIR}" )
+execute_process ( COMMAND sh -xc "mkdir -p ${PROJECT_BINARY_DIR}/RESULTS" )
 
 ## Build test library
 set ( TEST_LIB "tool" )

projects/rocprofiler/test/ctrl/tool.cpp

@@ -322,6 +322,11 @@ hsa_status_t dispatch_callback(const rocprofiler_callback_data_t* callback_data,
   return status;
 }
 
+static hsa_status_t info_callback(const rocprofiler_info_data_t info, void * arg) {
+    printf("  gpu-agent%d.%s : %s\n", info.agent_idx, info.metric.name, info.metric.description);
+    return HSA_STATUS_SUCCESS;
+}
+
 // Tool constructor
 CONSTRUCTOR_API void constructor()
 {
@@ -337,6 +342,11 @@ CONSTRUCTOR_API void constructor()
   parameters_dict["HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_TOKEN_MASK2"] =
       HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_TOKEN_MASK2;
 
+  if (getenv("ROCP_INFO") != NULL) {
+    rocprofiler_iterate_info(NULL, ROCPROFILER_INFO_KIND_METRIC, info_callback, NULL);
+    return;
+  }
+
   // Set output file
   result_prefix = getenv("ROCP_OUTPUT_DIR");
   if (result_prefix != NULL) {

projects/rocprofiler/test/input.xml

@@ -1,4 +1,5 @@
-<metric name=SQ_CYCLES,SQ_WAVES,SQ_INSTS_SMEM,SQ_INSTS_VALU,TA_FLAT_WRITE_WAVEFRONTS[0],TA_FLAT_WRITE_WAVEFRONTS[1],CPC_ALWAYS_COUNT,CPC_ME1_STALL_WAIT_ON_RCIU_READ,GPU_BUSY,MEM_BUSY,SFETCH_INSTS,VALU_INSTS,FETCH_SIZE,VWRITE_INSTS ></metric>
+<metric name=SQ_CYCLES,SQ_WAVES,SQ_INSTS_SMEM,SQ_INSTS_VALU,TA_FLAT_WRITE_WAVEFRONTS[0],TA_FLAT_WRITE_WAVEFRONTS[1],CPC_ALWAYS_COUNT,CPC_ME1_STALL_WAIT_ON_RCIU_READ,GPUBusy,VALUBusy,SALUBusy,MemUnitBusy,SFetchInsts,FetchSize,VWriteInsts,WriteSize
+></metric>
 
 <trace name=SQTT copy=true >
   <parameters

projects/rocprofiler/test/metrics.xml

@@ -19,10 +19,8 @@
   <metric name=SQ_WAIT_INST_LDS block=SQ event=61 descr="Number of wave-cycles spent waiting for LDS instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
   <metric name=SQ_ACTIVE_INST_VALU block=SQ event=69 descr="Number of cycles the SQ instruction arbiter is working on a VALU instruction. (per-simd, nondeterministic)"></metric>
   <metric name=SQ_INST_CYCLES_SALU block=SQ event=86 descr="Number of cycles needed to execute non-memory read scalar operations. (per-simd, emulated)"></metric>
-
+  <metric name=SQ_THREAD_CYCLES_VALU block=SQ event=89 descr="Number of thread-cycles used to execute VALU operations (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd)"></metric>
-  <metric name=SQ_THREAD_CYCLES_VALU block=SQ event=89 ></metric>
+  <metric name=SQ_THREAD_CYCLES_VALU_MAX block=SQ event=90 descr="Maximum number of thread-cycles VALU operations that could have been executed given the instruction mix (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd, emulated)"></metric>
-  <metric name=SQ_THREAD_CYCLES_VALU_MAX block=SQ event=90 ></metric>
-
   <metric name=SQ_LDS_BANK_CONFLICT block=SQ event=97 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
 
   <metric name=TA_BUSY block=TA event=15 ></metric>
@@ -83,10 +81,8 @@
   <metric name=SQ_WAIT_INST_LDS block=SQ event=63 descr="Number of wave-cycles spent waiting for LDS instruction issue. In units of 4 cycles. (per-simd, nondeterministic)"></metric>
   <metric name=SQ_ACTIVE_INST_VALU block=SQ event=71 descr="regspec 71? Number of cycles the SQ instruction arbiter is working on a VALU instruction. (per-simd, nondeterministic)"></metric>
   <metric name=SQ_INST_CYCLES_SALU block=SQ event=84 descr="Number of cycles needed to execute non-memory read scalar operations. (per-simd, emulated)"></metric>
-
+  <metric name=SQ_THREAD_CYCLES_VALU block=SQ event=85 descr="Number of thread-cycles used to execute VALU operations (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd)"></metric>
-  <metric name=SQ_THREAD_CYCLES_VALU block=SQ event=85 ></metric>
+  <metric name=SQ_THREAD_CYCLES_VALU_MAX block=SQ event=86 descr="Maximum number of thread-cycles VALU operations that could have been executed given the instruction mix (similar to INST_CYCLES_VALU but multiplied by # of active threads). (per-simd, emulated)"></metric>
-  <metric name=SQ_THREAD_CYCLES_VALU_MAX block=SQ event=86 ></metric>
-
   <metric name=SQ_LDS_BANK_CONFLICT block=SQ event=93 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
 
   <metric name=TA_BUSY block=TA event=15 ></metric>
@@ -95,11 +91,16 @@
 
   <metric name=TCC_CYCLE block=TCC event=1 ></metric>
   <metric name=TCC_REQ block=TCC event=3 ></metric>
-  <metric name=TCC_HIT block=TCC event=17 ></metric>
+  <metric name=TCC_HIT block=TCC event=20 ></metric>
-  <metric name=TCC_MISS block=TCC event=19 ></metric>
+  <metric name=TCC_MISS block=TCC event=22 ></metric>
-  <metric name=TCC_WRITEBACK block=TCC event=22 ></metric>
+  <metric name=TCC_WRITEBACK block=TCC event=25 ></metric>
-  <metric name=TCC_EA_RDREQ block=TCC event=36 ></metric>
+  <metric name=TCC_EA_WRREQ block=TCC event=29 ></metric>
-  <metric name=TCC_EA_RDREQ_32B block=TCC event=37 ></metric>
+  <metric name=TCC_EA_WRREQ_64B block=TCC event=30 ></metric>
+  <metric name=TCC_EA_WRREQ_STALL block=TCC event=33 ></metric>
+  <metric name=TCC_EA_RDREQ block=TCC event=41 ></metric>
+  <metric name=TCC_EA_RDREQ_32B block=TCC event=42 ></metric>
+
+  <metric name=TCP_TA_DATA_STALL_CYCLES block=TCP event=6 descr="TCP stalls TA data interface. Now Windowed."></metric>
 
   <metric name=CPC_ALWAYS_COUNT block=CPC event=0 ></metric>
   <metric name=CPC_ME1_STALL_WAIT_ON_RCIU_READ block=CPC event=8 ></metric>
@@ -127,7 +128,7 @@
   # GPU_BUSY, percentage
   # The percentage of time GPU was busy.
   <metric
-    name=GPU_BUSY
+    name="GPUBusy"
     expr=100*GRBM_GUI_ACTIVE/GRBM_COUNT
     descr="The percentage of time GPU was busy."
   ></metric>
@@ -169,7 +170,7 @@
 
   # VWriteInsts     The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory.
   <metric
-    name=VWriteInsts
+    name="VWriteInsts"
     expr=(SQ_INSTS_VMEM_WR-TA_FLAT_WRITE_WAVEFRONTS_sum)/SQ_WAVES
     descr="The average number of vector write instructions to the video memory executed per work-item (affected by flow control). Excludes FLAT instructions that write to video memory."
   ></metric>
@@ -219,7 +220,7 @@
   # SALUBusy        The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal).
   <metric
     name="SALUBusy"
-    expr=100*SQ_INST_CYCLES_SALU*4/(NUM_SIMDS/NUM_SHADER_ENGINES)/GRBM_GUI_ACTIVE
+    expr=100*SQ_INST_CYCLES_SALU*4/NUM_SIMDS/GRBM_GUI_ACTIVE
     descr="The percentage of GPUTime scalar ALU instructions are processed. Value range: 0% (bad) to 100% (optimal)."
   ></metric>