Merge branch 'rocm_sdk_update' of github.com:RRZE-HPC/likwid into roc…

…m_sdk_update

groups/amd_gpu_v1/GDS.txt

@@ -0,0 +1,15 @@
+SHORT GDS Instructions
+
+EVENTSET
+ROCM0 ROCP_SQ_INSTS_GDS
+ROCM1 ROCP_SQ_WAVES
+
+METRICS
+GPU GDS rw insts per work-item ROCM0/ROCM1
+
+LONG
+Formulas:
+GPU GDS rw insts per work-item = ROCP_SQ_INSTS_GDS/ROCP_SQ_WAVES
+--
+The average number of GDS read or GDS write instructions executed 
+per work item (affected by flow control).

groups/amd_gpu_v1/MEM.txt

@@ -0,0 +1,18 @@
+SHORT Memory utilization
+
+EVENTSET
+ROCM0 ROCP_TA_TA_BUSY
+ROCM1 ROCP_GRBM_GUI_ACTIVE
+ROCM2 ROCP_SE_NUM
+
+METRICS
+GPU memory utilization 100*max(ROCM0,16)/ROCM1/ROCM2
+
+LONG
+Formulas:
+GPU memory utilization = 100*max(ROCP_TA_TA_BUSY,16)/ROCP_GRBM_GUI_ACTIVE/ROCP_SE_NUM
+--
+The percentage of GPUTime the memory unit is active. The result includes 
+the stall time (MemUnitStalled). This is measured with all extra fetches 
+and writes and any cache or memory effects taken into account. 
+Value range: 0% to 100% (fetch-bound).

groups/amd_gpu_v1/PCI.txt

@@ -0,0 +1,23 @@
+SHORT PCI Transfers
+
+EVENTSET
+ROCM0 RSMI_PCI_THROUGHPUT_SENT
+ROCM1 RSMI_PCI_THROUGHPUT_RECEIVED
+
+
+METRICS
+Runtime time
+PCI sent ROCM0
+PCI received ROCM1
+PCI send bandwidth 1E-6*ROCM0/time
+PCI recv bandwidth 1E-6*ROCM1/time
+
+LONG
+Formulas:
+PCI sent = RSMI_PCI_THROUGHPUT_SENT
+PCI received = RSMI_PCI_THROUGHPUT_RECEIVED
+PCI send bandwidth = 1E-6*RSMI_PCI_THROUGHPUT_SENT/runtime
+PCI recv bandwidth = 1E-6*RSMI_PCI_THROUGHPUT_RECEIVED/runtime
+--
+Currently not usable since the RSMI_PCI_THROUGHPUT_* events require
+one second per call, so 2 seconds for both of them.

groups/amd_gpu_v1/POWER.txt

@@ -0,0 +1,21 @@
+SHORT Power, temperature and voltage
+
+EVENTSET
+ROCM0 RSMI_POWER_AVE[0]
+ROCM1 RSMI_TEMP_EDGE
+ROCM2 RSMI_VOLT_VDDGFX
+
+
+METRICS
+Power average 1E-6*ROCM0
+Edge temperature 1E-3*ROCM1
+Voltage 1E-3*ROCM2
+
+LONG
+Formulas:
+Power average = RSMI_POWER_AVE[0]
+Edge temperature = 1E-3*RSMI_TEMP_EDGE
+Voltage = 1E-3*RSMI_VOLT_VDDGFX
+--
+Gets the current average power consumption in watts, the
+temperature in celsius and the voltage in volts.

groups/amd_gpu_v1/SALU.txt

@@ -0,0 +1,15 @@
+SHORT SALU Instructions
+
+EVENTSET
+ROCM0 ROCP_SQ_INSTS_SALU
+ROCM1 ROCP_SQ_WAVES
+
+METRICS
+GPU SALU insts per work-item ROCM0/ROCM1
+
+LONG
+Formulas:
+GPU SALU insts per work-item = ROCP_SQ_INSTS_SALU/ROCP_SQ_WAVES
+--
+The average number of scalar ALU instructions executed per work-item 
+(affected by flow control).

groups/amd_gpu_v1/SFETCH.txt

@@ -0,0 +1,15 @@
+SHORT SFetch Instructions
+
+EVENTSET
+ROCM0 ROCP_SQ_INSTS_SMEM
+ROCM1 ROCP_SQ_WAVES
+
+METRICS
+GPU SFETCH insts per work-item ROCM0/ROCM1
+
+LONG
+Formulas:
+GPU SFETCH insts per work-item = ROCP_SQ_INSTS_SMEM/ROCP_SQ_WAVES
+--
+The average number of scalar fetch instructions from the video memory 
+executed per work-item (affected by flow control).

groups/amd_gpu_v1/STALLED.txt

@@ -0,0 +1,19 @@
+SHORT ALU stalled by LDS
+
+EVENTSET
+ROCM0 ROCP_SQ_WAIT_INST_LDS
+ROCM1 ROCP_SQ_WAVES
+ROCM2 ROCP_GRBM_GUI_ACTIVE
+
+METRICS
+GPU ALD stalled 100*ROCM0*4/ROCM1/ROCM2
+
+LONG
+Formulas:
+GPU ALD stalled = 100*ROCP_SQ_WAIT_INST_LDS*4/ROCP_SQ_WAVES/ROCP_GRBM_GUI_ACTIVE
+--
+The percentage of GPUTime ALU units are stalled by the LDS input queue 
+being full or the output queue being not ready. If there are LDS bank 
+conflicts, reduce them. Otherwise, try reducing the number of LDS 
+accesses if possible. 
+Value range: 0% (optimal) to 100% (bad).

groups/amd_gpu_v1/UTIL.txt

@@ -0,0 +1,18 @@
+SHORT GPU utilization
+
+EVENTSET
+ROCM0 ROCP_GRBM_COUNT
+ROCM1 ROCP_GRBM_GUI_ACTIVE
+
+
+METRICS
+GPU utilization 100*ROCM1/ROCM0
+
+
+LONG
+Formulas:
+GPU utilization = 100*ROCP_GRBM_GUI_ACTIVE/ROCP_GRBM_COUNT
+--
+This group reassembles the 'GPUBusy' metric provided by RocProfiler.
+We should add, that we can select the GPUBusy metric directly and the
+calculations are done internally in case the metric formula changes.

groups/amd_gpu_v1/VALU.txt

@@ -0,0 +1,15 @@
+SHORT VALU Instructions
+
+EVENTSET
+ROCM0 ROCP_SQ_INSTS_VALU
+ROCM1 ROCP_SQ_WAVES
+
+METRICS
+GPU VALU insts per work-item ROCM0/ROCM1
+
+LONG
+Formulas:
+GPU VALU insts per work-item = ROCP_SQ_INSTS_VALU/ROCP_SQ_WAVES
+--
+The average number of vector ALU instructions executed per work-item 
+(affected by flow control).

groups/amd_gpu_v1/WAVE.txt

@@ -0,0 +1,15 @@
+SHORT Wavefronts
+
+EVENTSET
+ROCM0 ROCP_SQ_WAVES
+
+
+METRICS
+GPU wavefronts ROCM0
+
+
+LONG
+Formulas:
+GPU wavefronts = ROCP_SQ_WAVES
+--
+Total Wavefronts

src/includes/rocmon_sdk.h

@@ -397,7 +397,6 @@ _rocmon_sdk_read_buffers(rocprofiler_context_id_t device_context,
         if(h->category == ROCPROFILER_BUFFER_CATEGORY_COUNTERS && h->kind == ROCPROFILER_COUNTER_RECORD_VALUE)
         {
             rocprofiler_record_counter_t* r = h->payload;
-            printf("Counter ID %d Value %f Dispatch %ld\n", r->id, r->counter_value, r->dispatch_id);
             rocprofiler_counter_id_t cid = {.handle = 0};
             (*rocprofiler_query_record_counter_id_ptr)(r->id, &cid);
             for (int j = 0; j < context->numDevices; j++)
@@ -619,8 +618,6 @@ _rocmon_sdk_set_profile(rocprofiler_context_id_t                 context_id,
 
 
 
-
-
 rocprofiler_tool_configure_result_t*
 rocprofiler_configure(uint32_t                 version,
                       const char*              runtime_version,
@@ -658,7 +655,6 @@ rocmon_sdk_init(RocmonContext* context, int numGpus, const int* gpuIds)
     }
     if (rocmon_sdk_initialized)
     {
-
         return 0;
     }
 

src/includes/rocmon_smi.h

@@ -1156,7 +1156,7 @@ void rocmon_smi_finalize(RocmonContext* context)
     }
     ROCMON_DEBUG_PRINT(DEBUGLEV_DEVELOP, Shutdown RSMI);
     RSMI_CALL(rsmi_shut_down, (), {
-        ROCMON_DEBUG_PRINT(DEBUGLEV_DEVELOP, Shutdown SMI);
+        ERROR_PRINT(DEBUGLEV_DEVELOP, Shutdown SMI failed);
         // fall through
     });
     rocmon_smi_initialized = FALSE;

src/includes/rocmon_v1.h

@@ -575,7 +575,7 @@ rocmon_v1_finalize(RocmonContext* context)
     }
 
     ROCM_CALL(hsa_shut_down, (), {
-        //ROCMON_DEBUG_PRINT(DEBUGLEV_DEVELOP, Shutdown HSA);
+        ERROR_PRINT(DEBUGLEV_DEVELOP, Shutdown HSA failed);
         // fall through
     });
 }