rocprofiler: add valu utilization

SWDEV-475242

For the description of "FP32 Engine Activity" and "FP64 Engine Activity" in dcgm,
It seems that we do not have an equivalent to these pipe-utilizations on our hardware.

In rocprofiler, I think VALU Utilization is the closest to what we want.

Change-Id: Ibce8835ef4757084cdfd73258de6fc1606ca0158
Signed-off-by: Chen Gong <curry.gong@amd.com>


[ROCm/rdc commit: 251fcbe49d]

projects/rdc/common/rdc_field.data

@@ -137,6 +137,7 @@ FLD_DESC_ENT(RDC_FI_PROF_EVAL_MEM_W_BW,         "Written to video memory kb / ms
 FLD_DESC_ENT(RDC_FI_PROF_EVAL_FLOPS_16,         "Number of fp16 OPS / ms",               "FLOPS_16",          false)
 FLD_DESC_ENT(RDC_FI_PROF_EVAL_FLOPS_32,         "Number of fp32 OPS / ms",               "FLOPS_32",          false)
 FLD_DESC_ENT(RDC_FI_PROF_EVAL_FLOPS_64,         "Number of fp64 OPS / ms",               "FLOPS_64",          false)
+FLD_DESC_ENT(RDC_FI_PROF_VALU_PIPE_ISSUE_UTIL,  "Percent of Active Pipe VALU",           "VALU_UTILIZATION",  false)
 
 // Events
 FLD_DESC_ENT(RDC_EVNT_XGMI_0_NOP_TX,     "NOPs sent to neighbor 0",                     "XGMI_NOP_0",       false)

projects/rdc/include/rdc/rdc.h

@@ -277,6 +277,7 @@ typedef enum {
   RDC_FI_PROF_EVAL_FLOPS_16,
   RDC_FI_PROF_EVAL_FLOPS_32,
   RDC_FI_PROF_EVAL_FLOPS_64,
+  RDC_FI_PROF_VALU_PIPE_ISSUE_UTIL,
 
   /**
    * @brief Raw XGMI counter events

projects/rdc/python_binding/rdc_bootstrap.py

@@ -138,16 +138,18 @@ class rdc_field_t(c_int):
      RDC_FI_XGMI_7_WRITE_KB = 715
      RDC_FI_XGMI_TOTAL_READ_KB = 716
      RDC_FI_XGMI_TOTAL_WRITE_KB = 717
-     RDC_FI_PROF_MEAN_OCCUPANCY_PER_CU = 800
-     RDC_FI_PROF_MEAN_OCCUPANCY_PER_ACTIVE_CU = 801
-     RDC_FI_PROF_ACTIVE_CYCLES = 802
-     RDC_FI_PROF_ACTIVE_WAVES = 803
-     RDC_FI_PROF_ELAPSED_CYCLES = 804
-     RDC_FI_PROF_EVAL_MEM_R_BW = 805
-     RDC_FI_PROF_EVAL_MEM_W_BW = 806
-     RDC_FI_PROF_EVAL_FLOPS_16 = 807
-     RDC_FI_PROF_EVAL_FLOPS_32 = 808
-     RDC_FI_PROF_EVAL_FLOPS_64 = 809
+     RDC_FI_PROF_OCCUPANCY_PERCENT = 800
+     RDC_FI_PROF_ACTIVE_CYCLES = 801
+     RDC_FI_PROF_ACTIVE_WAVES = 802
+     RDC_FI_PROF_ELAPSED_CYCLES = 803
+     RDC_FI_PROF_TENSOR_ACTIVE_PERCENT = 804
+     RDC_FI_PROF_GPU_UTIL_PERCENT = 805
+     RDC_FI_PROF_EVAL_MEM_R_BW = 806
+     RDC_FI_PROF_EVAL_MEM_W_BW = 807
+     RDC_FI_PROF_EVAL_FLOPS_16 = 808
+     RDC_FI_PROF_EVAL_FLOPS_32 = 809
+     RDC_FI_PROF_EVAL_FLOPS_64 = 810
+     RDC_FI_PROF_VALU_PIPE_ISSUE_UTIL = 811
      RDC_EVNT_XGMI_0_NOP_TX = 1000
      RDC_EVNT_XGMI_0_REQ_TX = 1001
      RDC_EVNT_XGMI_0_RESP_TX = 1002

projects/rdc/rdc_libs/rdc_modules/rdc_rocp/RdcRocpBase.cc

@@ -258,6 +258,7 @@ RdcRocpBase::RdcRocpBase() {
       {RDC_FI_PROF_EVAL_FLOPS_16, "TOTAL_16_OPS"},
       {RDC_FI_PROF_EVAL_FLOPS_32, "TOTAL_32_OPS"},
       {RDC_FI_PROF_EVAL_FLOPS_64, "FP64_ACTIVE"},
+      {RDC_FI_PROF_VALU_PIPE_ISSUE_UTIL, "ValuPipeIssueUtil"},
   };
 
   std::vector<std::string> all_fields;