adding run scripts

Change-Id: Ia7048a0baaa6ea03283aade53a0758d147dc6fcd

CMakeLists.txt

@@ -78,6 +78,22 @@ add_subdirectory ( ${TEST_DIR} ${PROJECT_BINARY_DIR}/test )
 ## Install information
 install ( TARGETS ${ROCPROFILER_TARGET} LIBRARY DESTINATION ${ROCPROFILER_NAME}/lib )
 install ( FILES ${CMAKE_CURRENT_SOURCE_DIR}/inc/rocprofiler.h DESTINATION ${ROCPROFILER_NAME}/include )
+# rpl_run.sh tblextr.py txt2xml.sh
+install ( FILES
+          ${CMAKE_CURRENT_SOURCE_DIR}/bin/rpl_run.sh
+          ${CMAKE_CURRENT_SOURCE_DIR}/bin/txt2xml.sh
+          ${CMAKE_CURRENT_SOURCE_DIR}/bin/tblextr.py
+          DESTINATION ${ROCPROFILER_NAME}/bin
+          PERMISSIONS OWNER_READ OWNER_EXECUTE GROUP_READ GROUP_EXECUTE WORLD_READ WORLD_EXECUTE )
+# gfx_metrics.xml metrics.xml
+install ( FILES
+          ${CMAKE_CURRENT_SOURCE_DIR}/test/tool/metrics.xml
+          ${CMAKE_CURRENT_SOURCE_DIR}/test/tool/gfx_metrics.xml
+          DESTINATION ${ROCPROFILER_NAME}/lib )
+# libtool.so
+install ( FILES ${PROJECT_BINARY_DIR}/test/libtool.so DESTINATION ${ROCPROFILER_NAME}/tool )
+install ( FILES ${PROJECT_BINARY_DIR}/test/ctrl DESTINATION ${ROCPROFILER_NAME}/tool
+          PERMISSIONS OWNER_READ OWNER_EXECUTE GROUP_READ GROUP_EXECUTE WORLD_READ WORLD_EXECUTE )
 
 ## Packaging directives
 set ( CPACK_PACKAGE_NAME "${ROCPROFILER_NAME}-dev" )

bin/rpl_run.sh

@@ -0,0 +1,318 @@
+#!/bin/sh
+time_stamp=`date +%y%m%d_%H%M%S`
+BIN_DIR=`dirname $0`
+BIN_DIR=`cd $BIN_DIR; pwd`
+RUN_DIR=`pwd`
+TMP_DIR="/tmp"
+DATA_PATH=$TMP_DIR
+DATA_DIR="rpl_data_${time_stamp}_$$"
+
+PKG_DIR=`echo $BIN_DIR | sed "s/\/bin\/*//"`
+BIN_DIR=$PKG_DIR/bin
+
+# PATH to custom HSA and OpenCl runtimes
+HSA_PATH=$PKG_DIR/lib/hsa
+
+export LD_LIBRARY_PATH=$PKG_DIR/lib:$PKG_DIR/tool:$HSA_PATH
+export PATH=.:$PATH
+
+# enable error logging
+export HSA_TOOLS_REPORT_LOAD_FAILURE=1
+export HSA_VEN_AMD_AQLPROFILE_LOG=1
+export ROCPROFILER_LOG=1
+
+# ROC Profiler environment
+# Loading of ROC Profiler by HSA runtime
+export HSA_TOOLS_LIB=librocprofiler64.so
+# Loading of the test tool by ROC Profiler
+export ROCP_TOOL_LIB=libtool.so
+# Enabling HSA dispatches intercepting by ROC PRofiler
+export ROCP_HSA_INTERCEPT=1
+# Disabling internal ROC Profiler proxy queue (simple version supported for testing purposes)
+unset ROCP_PROXY_QUEUE
+# ROC Profiler metrics definition
+export ROCP_METRICS=$PKG_DIR/lib/metrics.xml
+# ROC Profiler package path
+export ROCP_PACKAGE_DIR=$PKG_DIR
+
+# error handling
+fatal() {
+  echo "$0: Error: $1"
+  echo ""
+  usage
+}
+
+error() {
+  echo "$0: Error: $1"
+  echo ""
+  exit 1
+}
+
+# usage method
+usage() {
+  bin_name=`basename $0`
+  echo "ROCm Profiling Library (RPL) run script, a part of ROCprofiler library package."
+  echo "Full path: $BIN_DIR/$bin_name"
+  echo "Metrics definition: $PKG_DIR/lib/metrics.xml"
+  echo ""
+  echo "Usage:"
+  echo "  rpl_run.sh [-h] [--list-basic] [--list-derived] [-i <input .txt/.xml file>] [-o <output CSV file>] <app command line>"
+  echo ""
+  echo "Options:"
+  echo "  -h - this help"
+  echo "  --verbose - verbose mode, dumping all base counters used in the input metrics"
+  echo "  --list-basic - to print the list of basic HW counters"
+  echo "  --list-derived - to print the list of derived metrics with formulas"
+  echo ""
+  echo "  -i <.txt|.xml file> - input file"
+  echo "      Input file .txt format, automatically rerun application for every pmc/sqtt line:"
+  echo ""
+  echo "        # Perf counters group 1"
+  echo "        pmc : Wavefronts VALUInsts SALUInsts SFetchInsts FlatVMemInsts LDSInsts FlatLDSInsts GDSInsts VALUUtilization FetchSize"
+  echo "        # Perf counters group 2"
+  echo "        pmc : WriteSize L2CacheHit"
+  echo "        # Filter by dispatches range, GPU index and kernel names"
+  echo "        # supported range formats: \"3:9\", \"3:\", \"3\""
+  echo "        range: 1 : 4"
+  echo "        gpu: 0 1 2 3"
+  echo "        kernel: simple Pass1 simpleConvolutionPass2"
+  echo ""
+  echo "      Input file .xml format, for single profiling run:"
+  echo ""
+  echo "        # Metrics list definition, also the form \"<block-name>:<event-id>\" can be used"
+  echo "        # All defined metrics can be found in the 'metrics.xml'"
+  echo "        # There are basic metrics for raw HW counters and high-level metrics for derived counters"
+  echo "        <metric name=SQ:4,SQ_WAVES,VFetchInsts"
+  echo "        ></metric>"
+  echo ""
+  echo "        # Filter by dispatches range, GPU index and kernel names"
+  echo "        <metric"
+  echo "          # range formats: \"3:9\", \"3:\", \"3\""
+  echo "          range=\"\""
+  echo "          # list of gpu indexes \"0,1,2,3\""
+  echo "          gpu_index=\"\""
+  echo "          # list of matched sub-strings \"Simple1,Conv1,SimpleConvolution\""
+  echo "          kernel=\"\""
+  echo "        ></metric>"
+  echo ""
+  echo "  -o <output file> - output CSV file [<input file base>.csv]"
+  echo "  -d <data directory> - directory where profiler store profiling data including thread treaces [/tmp]"
+  echo "      The data directory is renoving autonatically if the directory is matching the temporary one, which is the default."
+  echo "  -t <temporary directory> - to change the temporary directory [/tmp]"
+  echo "      By changing the temporary directory you can prevent removing the profiling data from /tmp or enable removing from not '/tmp' directory."
+  echo ""
+  echo "  --basenames <on|off> - to turn on/off truncating of the kernel full function names till the base ones [off]"
+  echo "  --timestamp <on|off> - to turn on/off the kernel disoatches timestamps, dispatch/begin/end/complete [off]"
+  echo "  --ctx-limit <max number> - maximum number of outstanding contexts [0 - unlimited]"
+  echo "  --heartbeat <rate sec> - to turn on/off the kernel disoatches timestamps, dispatch/begin/end/complete [0 - disabled]"
+  echo "  --sqtt-size <byte size> - to set SQTT buffer size, aggregate for all SE [0x2000000]"
+  echo "      Can be set in KB (1024B) or MB (1048576) units, examples 20K or 20M respectively."
+  echo "  --sqtt-local <on|off> - to allocate SQTT buffer in local GPU memory [on]"
+  echo ""
+  echo "Configuration file:"
+  echo "  You can set your parameters defaults preferences in the configuration file 'rpl_rc.xml'. The search path sequence: .:${HOME}:<package path>"
+  echo "  First the configuration file is looking in the current directory, then in your home, and then in the package directory."
+  echo "  Configurable options: 'basenames', 'timestamp', 'ctx-limit', 'heartbeat', 'sqtt-size', 'sqtt-local'."
+  echo "  An example of 'rpl_rc.xml':"
+  echo "    <defaults"
+  echo "      basenames=off"
+  echo "      timestamp=off"
+  echo "      ctx-limit=0"
+  echo "      heartbeat=0"
+  echo "      sqtt-size=0x20M"
+  echo "      sqtt-local=on"
+  echo "    ></defaults>"
+  echo ""
+  exit 1
+}
+
+# profiling run method
+OUTPUT_LIST=""
+run() {
+  export ROCP_INPUT="$1"
+  OUTPUT_DIR="$2"
+  shift
+  shift
+  APP_CMD=$*
+
+  if [ "$OUTPUT_DIR" = "-" ] ; then
+    input_tag=`echo $ROCP_INPUT | sed "s/\.xml//"`
+    export ROCP_OUTPUT_DIR=${input_tag}_results_${time_stamp}
+  elif [ "$OUTPUT_DIR" = "--" ] ; then
+    unset ROCP_OUTPUT_DIR
+  else
+    export ROCP_OUTPUT_DIR=$OUTPUT_DIR
+  fi
+  echo "RPL: result dir '$ROCP_OUTPUT_DIR'"
+  
+  if [ ! -e "$ROCP_INPUT" ] ; then
+    error "Input file '$ROCP_INPUT' not found"
+  fi
+  
+  if [ -n "$ROCP_OUTPUT_DIR" ] ; then
+    if [ "$OUTPUT_DIR" = "-" ] ; then
+      if [ -e "$ROCP_OUTPUT_DIR" ] ; then
+        error "generated dir '$ROCP_OUTPUT_DIR' exists"
+      fi
+    fi
+    mkdir -p "$ROCP_OUTPUT_DIR"
+  fi
+  
+  if [ -n "$ROCP_OUTPUT_DIR" ] ; then
+    OUTPUT_LIST="$OUTPUT_LIST $ROCP_OUTPUT_DIR/results.txt"
+    eval "$APP_CMD 2>&1 | tee $ROCP_OUTPUT_DIR/log.txt"
+  else
+    eval "$APP_CMD"
+  fi
+}
+
+# main
+echo "RPL: on '$time_stamp' from '$PKG_DIR' in '$RUN_DIR'"
+# Parsing arguments
+if [ -z "$1" ] ; then
+  usage
+fi
+
+INPUT_FILE=""
+OUTPUT_DIR="-"
+output=""
+csv_output=""
+
+ARG_IN=""
+while [ 1 ] ; do
+  ARG_IN=$1
+  ARG_VAL=1
+  if [ "$1" = "-h" ] ; then
+    usage
+  elif [ "$1" = "-i" ] ; then
+    INPUT_FILE="$2"
+  elif [ "$1" = "-o" ] ; then
+    output="$2"
+  elif [ "$1" = "-d" ] ; then
+    OUTPUT_DIR="$2"
+    DATA_PATH=$OUTPUT_DIR
+  elif [ "$1" = "-t" ] ; then
+    TMP_DIR="$2"
+    if [ "$OUTPUT_DIR" = "-" ] ; then
+      DATA_PATH=$TMP_DIR
+    fi
+  elif [ "$1" = "--list-basic" ] ; then
+    export ROCP_INFO=b
+    eval "$PKG_DIR/tool/ctrl"
+    exit 1
+  elif [ "$1" = "--list-derived" ] ; then
+    export ROCP_INFO=d
+    eval "$PKG_DIR/tool/ctrl"
+    exit 1
+  elif [ "$1" = "--basenames" ] ; then
+    if [ "$2" = "on" ] ; then
+      export ROCP_TRUNCATE_NAMES=1
+    else
+      export ROCP_TRUNCATE_NAMES=0
+    fi
+  elif [ "$1" = "--timestamp" ] ; then
+    if [ "$2" = "on" ] ; then
+      export ROCP_TIMESTAMP_ON=1
+    else
+      export ROCP_TIMESTAMP_ON=0
+    fi
+  elif [ "$1" = "--ctx-limit" ] ; then
+    export ROCP_OUTSTANDING_MAX="$2"
+  elif [ "$1" = "--heartbeat" ] ; then
+    export ROCP_OUTSTANDING_MON="$2"
+  elif [ "$1" = "--sqtt-size" ] ; then
+    size_m=`echo "$2" | sed -n "s/^\(.*\)M$/\1/p"`
+    size_k=`echo "$2" | sed -n "s/^\(.*\)K$/\1/p"`
+    if [ -n "$size_m" ] ; then size_b=$((size_m*1024*1024))
+    elif [ -n "$size_k" ] ; then size_b=$((size_k*1024))
+    else size_b=$2
+    fi
+    export ROCP_SQTT_SIZE=$size_b
+  elif [ "$1" = "--sqtt-local" ] ; then
+    if [ "$2" = "on" ] ; then
+      export ROCP_SQTT_LOCAL=1
+    else
+      export ROCP_SQTT_LOCAL=0
+    fi
+  elif [ "$1" = "--verbose" ] ; then
+    ARG_VAL=0
+    export ROCP_VERBOSE_MODE=1
+  else
+    break
+  fi
+  shift
+  if [ "$ARG_VAL" = 1 ] ; then shift; fi
+done
+
+ARG_CK=`echo $ARG_IN | sed "s/^-.*$/-/"`
+if [ "$ARG_CK" = "-" ] ; then
+  fatal "Wrong option '$ARG_IN'"
+fi
+
+if [ -z "$INPUT_FILE" ] ; then
+  fatal "Need input file"
+fi
+
+input_base=`echo "$INPUT_FILE" | sed "s/^\(.*\)\.\([^\.]*\)$/\1/"`
+input_type=`echo "$INPUT_FILE" | sed "s/^\(.*\)\.\([^\.]*\)$/\2/"`
+if [ -z "${input_base}" -o -z "${input_type}" ] ; then
+  fatal "Bad input file '$INPUT_FILE'"
+fi
+input_base=`basename $input_base`
+
+if [ "$OUTPUT_DIR" = "--" ] ; then
+  fatal "Bad output dir '$OUTPUT_DIR'"
+fi
+
+if [ -n "$output" ] ; then
+  if [ "$output" = "--" ] ; then
+    OUTPUT_DIR="--"
+  else
+    csv_output=$output
+  fi
+else
+  csv_output=$RUN_DIR/${input_base}.csv
+fi
+
+APP_CMD=$*
+
+echo "RPL: profiling '$APP_CMD'"
+echo "RPL: input file '$INPUT_FILE'"
+
+input_list=""
+RES_DIR=""
+if [ "$input_type" = "xml" ] ; then
+  input_list=$INPUT_FILE
+elif [ "$input_type" = "txt" ] ; then
+  OUTPUT_DIR="-"
+  RES_DIR=$DATA_PATH/$DATA_DIR
+  if [ -e $RES_DIR ] ; then
+    error "Rundir '$RES_DIR' exists"
+  fi
+  mkdir -p $RES_DIR
+  echo "RPL: output dir '$RES_DIR'"
+  $BIN_DIR/txt2xml.sh $INPUT_FILE $RES_DIR
+  input_list=`/bin/ls $RES_DIR/input*.xml`
+else
+  fatal "Bad input file type '$INPUT_FILE'"
+fi
+
+for name in $input_list; do
+  run $name $OUTPUT_DIR $APP_CMD
+done
+
+if [ -n "$csv_output" ] ; then
+  python $BIN_DIR/tblextr.py $csv_output $OUTPUT_LIST
+  if [ "$?" = 1 ] ; then
+    error "CSV generation error, profiling results '$RES_DIR'"
+  fi
+  echo "RPL: '$csv_output' is generated"
+fi
+
+if [ "$DATA_PATH" = "$TMP_DIR" ] ; then
+  if [ -e "$RES_DIR" ] ; then
+    rm -rf $RES_DIR
+  fi
+fi
+
+exit 0

bin/tblextr.py

@@ -0,0 +1,97 @@
+#!/usr/bin/python
+import os, sys, re
+
+# Parsing results in the format:
+#dispatch[0], queue_index(0), kernel_name("SimpleConvolution"), time(1048928000311041,1048928006154674,1048928006168274,1048928006170503):
+#  GRBM_GUI_ACTIVE (74332)
+#  SQ_WAVES (4096)
+#  SQ_INSTS_VMEM_RD (36864)
+
+# global vars
+var_list = ['Index', 'KernelName', 'DispatchNs', 'BeginNs', 'EndNs', 'CompleteNs']
+var_table = {}
+#############################################################
+
+def fatal(msg):
+  sys.stderr.write(sys.argv[0] + ": " + msg + "\n");
+  sys.exit(1)
+#############################################################
+
+# parse results method
+def parse_res(infile):
+  if not os.path.isfile(infile): fatal("Error: input file '" + infile + "' not found")
+  inp = open(infile, 'r')
+
+  beg_pattern = re.compile("^dispatch\[(\d*)\], queue_index\(\d*\), kernel_name\(\"([^\"]*)\"\)")
+  ts_pattern = re.compile(", time\((\d*),(\d*),(\d*),(\d*)\)")
+  var_pattern = re.compile("^\s*([^\s]*)\s+\((\d*)\)")
+
+  dispatch_number = 0
+  for line in inp.readlines():
+    record = line[:-1]
+
+    m = var_pattern.match(record)
+    if m:
+      if not dispatch_number in var_table: fatal("Error: dispatch number not unique '" + str(dispatch_number) + "'")
+      var = m.group(1)
+      val = m.group(2)
+      var_table[dispatch_number][m.group(1)] = m.group(2)
+      if not var in var_list: var_list.append(var)
+
+    m = beg_pattern.match(record)
+    if m:
+      dispatch_number = m.group(1)
+      if not dispatch_number in var_table:
+        var_table[dispatch_number] = {
+          'Index': dispatch_number,
+          'KernelName': "\"" + m.group(2) + "\""
+        }
+        m = ts_pattern.search(record)
+        if m:
+          var_table[dispatch_number]['DispatchNs'] = m.group(1)
+          var_table[dispatch_number]['BeginNs'] = m.group(2)
+          var_table[dispatch_number]['EndNs'] = m.group(3)
+          var_table[dispatch_number]['CompleteNs'] = m.group(4)
+
+  inp.close()
+#############################################################
+
+# print results table method
+def print_tbl(outfile):
+  global var_list
+
+  out = open(outfile, 'w')
+
+  keys = var_table.keys()
+  keys.sort(key=int)
+
+  entry = var_table[keys[0]]
+  list1 = []
+  for var in var_list:
+    if var in entry:
+      list1.append(var)
+  var_list = list1
+    
+  for var in var_list: out.write(var + ',')
+  out.write("\n")
+
+  for ind in keys:
+    entry = var_table[ind]
+    dispatch_number = entry['Index']
+    if ind != dispatch_number: fatal("Dispatch #" + ind + " index mismatch (" + dispatch_number + ")\n")
+    for var in var_list: out.write(entry[var] + ',')
+    out.write("\n")
+
+  out.close()
+#############################################################
+
+# main
+if (len(sys.argv) < 3): fatal("Usage: " + sys.argv[0] + " <output CSV file> <input result files list>")
+
+outfile = sys.argv[1]
+infiles = sys.argv[2:]
+for f in infiles :
+  parse_res(f)
+print_tbl(outfile)
+sys.exit(0)
+#############################################################

bin/txt2xml.sh

@@ -0,0 +1,72 @@
+#!/bin/bash
+timestamp=`date +%y%m%d_%H%M%S`
+
+if [ $# = 0 ] ; then
+  echo "Usage: $0 <input text file> [output dir]"
+  exit -1
+fi
+
+input=$1
+outdir=$2
+if [ -z "$outdir" ] ; then
+  outdir="."
+fi
+
+range=""
+kernel=""
+gpu_index=""
+
+parse() {
+  scan="$1"
+  index=0
+  while read -r line ; do
+    line=`echo $line | sed "s/\s*#.*$//"`
+    if [ -z "$line" ] ; then
+      continue
+    fi
+  
+    feature=`echo $line | sed -n "s/^\s*\([a-z]*\)\s*:.*$/\1/p"`
+    line=`echo $line | sed "s/^[^:]*:\s*//"`
+    line=`echo "$line" | sed -e "s/\s*=\s*/=/g" -e "s/\s*:\s*/:/g" -e "s/,\{1,\}/ /g" -e "s/\s\{1,\}/ /g" -e "s/\s*$//"`
+
+    if [ "$scan" = 0 ] ; then
+      line=`echo "$line" | sed -e "s/ /,/g"`
+      if [ "$feature" == "range" ] ; then
+        range=$line
+      fi
+      if [ "$feature" == "kernel" ] ; then
+        kernel=$line
+      fi
+      if [ "$feature" == "gpu" ] ; then
+        gpu_index=$line
+      fi
+    else
+      output=$outdir/input${index}.xml
+      header="# $timestamp '$output' generated with '$0 $*'"
+    
+      if [ "$feature" == "pmc" ] ; then
+        line=`echo "$line" | sed -e "s/ /,/g"`
+        cat >> $output <<EOF
+$header
+<metric range="$range" kernel="$kernel" gpu_index="$gpu_index"></metric>
+<metric name=$line ></metric>
+EOF
+      fi
+    
+      if [ "$feature" == "sqtt" ] ; then
+        cat >> $output <<EOF
+$header
+<metric range="$range" kernel="$kernel" gpu_index="$gpu_index"></metric>
+<trace name="SQTT"><parameters $line ></parameters></trace>
+EOF
+      fi
+    fi
+  
+    index=$((index + 1))
+  done < $input
+}
+
+parse 0
+parse 1
+
+exit 0

test/tool/gfx_metrics.xml

@@ -13,11 +13,11 @@
   <metric name=SQ_INSTS_LDS block=SQ event=34 ></metric>
   <metric name=SQ_INSTS_GDS block=SQ event=35 ></metric>
 
-  <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_LDS_BANK_CONFLICT block=SQ event=97 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
+  <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."></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."></metric>
+  <metric name=SQ_INST_CYCLES_SALU block=SQ event=86 descr="Number of cycles needed to execute non-memory read scalar operations."></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)."></metric>
+  <metric name=SQ_LDS_BANK_CONFLICT block=SQ event=97 descr="Number of cycles LDS is stalled by bank conflicts."></metric>
 
   <metric name=TA_TA_BUSY block=TA event=15 ></metric>
   <metric name=TA_FLAT_READ_WAVEFRONTS block=TA event=101 ></metric>
@@ -26,8 +26,8 @@
   <metric name=TCC_HIT block=TCC event=18 ></metric>
   <metric name=TCC_MISS block=TCC event=19 ></metric>
   <metric name="TCC_MC_RDREQ" block=TCC event=35 ></metric>
-  <metric name="TCC_MC_WRREQ" block=TCC event=26 descr="Number of 32-byte transactions going over the TC_MC_wrreq interface. Atomics may travel over the same interface and are generally classified as write requests."></metric>
-  <metric name="TCC_MC_WRREQ_STALL" block=TCC event=28 descr="Number of cycles a write request was stalled."></metric>
+  <metric name="TCC_MC_WRREQ" block=TCC event=26 ></metric>
+  <metric name="TCC_MC_WRREQ_STALL" block=TCC event=28 ></metric>
 
   <metric name="TCP_TA_DATA_STALL_CYCLES" block=TCP event=3 descr="TCP stalls TA data interface. Now Windowed."></metric>
 </gfx8>
@@ -47,11 +47,11 @@
   <metric name=SQ_INSTS_LDS block=SQ event=34 ></metric>
   <metric name=SQ_INSTS_GDS block=SQ event=35 ></metric>
 
-  <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_LDS_BANK_CONFLICT block=SQ event=93 descr="Number of cycles LDS is stalled by bank conflicts. (emulated)"></metric>
+  <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."></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."></metric>
+  <metric name=SQ_INST_CYCLES_SALU block=SQ event=84 descr="Number of cycles needed to execute non-memory read scalar operations."></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)."></metric>
+  <metric name=SQ_LDS_BANK_CONFLICT block=SQ event=93 descr="Number of cycles LDS is stalled by bank conflicts."></metric>
 
   <metric name=TA_TA_BUSY block=TA event=15 ></metric>
   <metric name=TA_FLAT_READ_WAVEFRONTS block=TA event=101 ></metric>