diff --git a/bin/rpl_run.sh b/bin/rpl_run.sh
index 4f15b97c82..6fff8e77f0 100755
--- a/bin/rpl_run.sh
+++ b/bin/rpl_run.sh
@@ -75,6 +75,8 @@ export ROCP_METRICS=$PKG_DIR/lib/metrics.xml
 export AQLPROFILE_READ_API=0
 # ROC Profiler package path
 export ROCP_PACKAGE_DIR=$PKG_DIR
+# enabled SPM KFD mode
+export ROCP_SPM_KFD_MODE=1
 
 # error handling
 fatal() {
@@ -184,7 +186,6 @@ usage() {
   echo "  --flush-rate <rate> - to enable trace flush rate (time period)"
   echo "    Supported time formats: <number(m|s|ms|us)>"
   echo "  --parallel-kernels - to enable cnocurrent kernels"
-  echo "  --spm-mode <on|off> - to enable collecting time-based sampling of HW counter data [off]"
   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>"
@@ -445,10 +446,6 @@ while [ 1 ] ; do
     ARG_VAL=0
     export ROCP_K_CONCURRENT=1
     export AQLPROFILE_READ_API=1
-  elif [ "$1" = "--spm-mode" ] ; then
-    if [ "$2" = "on" ] ; then
-      export ROCP_SPM_KFD_MODE=1
-    fi
   elif [ "$1" = "--verbose" ] ; then
     ARG_VAL=0
     export ROCP_VERBOSE_MODE=1
diff --git a/bin/txt2xml.sh b/bin/txt2xml.sh
index 126337ed03..50c20926f8 100755
--- a/bin/txt2xml.sh
+++ b/bin/txt2xml.sh
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/bin/bash -x
 
 ################################################################################
 # Copyright (c) 2018 Advanced Micro Devices, Inc. All rights reserved.
@@ -64,7 +64,7 @@ parse() {
         gpu_index=$line
       fi
     else
-      found=$(echo $feature | sed -n "/^\(pmc\|sqtt\|hsa\)$/ p")
+      found=$(echo $feature | sed -n "/^\(pmc\|sqtt\|hip\|hsa\|kfd\)$/ p")
       if [ -n "$found" ] ; then
         output=$outdir/input${index}.xml
         header="# $timestamp '$output' generated with '$0 $*'"
@@ -85,9 +85,21 @@ EOF
 EOF
         fi
 
+        if [ "$feature" == "hip" ] ; then
+          cat >> $output <<EOF
+<trace name="HIP"><parameters api="$line"></parameters></trace>
+EOF
+        fi
+
         if [ "$feature" == "hsa" ] ; then
           cat >> $output <<EOF
 <trace name="HSA"><parameters api="$line"></parameters></trace>
+EOF
+        fi
+
+        if [ "$feature" == "kfd" ] ; then
+          cat >> $output <<EOF
+<trace name="KFD"><parameters api="$line"></parameters></trace>
 EOF
         fi
       fi
diff --git a/src/core/context.h b/src/core/context.h
index c368d42c04..f3c2f2a805 100644
--- a/src/core/context.h
+++ b/src/core/context.h
@@ -323,7 +323,7 @@ class Context {
     for (auto& tuple : profile_vector) {
       if (pcsmp_mode_) const_cast<profile_t*>(tuple.profile)->event_count = UINT32_MAX;
       const hsa_status_t status =
-          api_->hsa_ven_amd_aqlprofile_iterate_data(tuple.profile, callback, data);
+        api_->hsa_ven_amd_aqlprofile_iterate_data(tuple.profile, callback, data);
       if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "context iterate data failed");
     }
   }
diff --git a/test/tool/tool.cpp b/test/tool/tool.cpp
index 71d5603850..c945fa1ac9 100644
--- a/test/tool/tool.cpp
+++ b/test/tool/tool.cpp
@@ -443,7 +443,8 @@ hsa_status_t trace_data_cb(hsa_ven_amd_aqlprofile_info_type_t info_type,
 }
 
 // SPM counter trace start/stop methods
-std::vector<rocprofiler_t*> spm_ctx_vec;
+typedef std::vector<rocprofiler_t*> spm_ctx_vec_t;
+spm_ctx_vec_t *spm_ctx_vec = NULL;
 void spm_ctrl_start(rocprofiler_feature_t* features, uint32_t features_found) {
   // Start SPM trace collection for all GPUs
   uint32_t gpu_count = HsaRsrcFactory::Instance().GetCountOfGpuAgents();
@@ -459,6 +460,14 @@ void spm_ctrl_start(rocprofiler_feature_t* features, uint32_t features_found) {
     rocprofiler_properties_t properties{};
     properties.queue_depth = 256;
 
+    for (rocprofiler_feature_t* p = features; p < features + features_found; ++p) {
+      int val = p->kind;
+      if (val == ROCPROFILER_FEATURE_KIND_METRIC) {
+        val = ROCPROFILER_FEATURE_KIND_TRACE | ROCPROFILER_FEATURE_KIND_SPM_MOD;
+        p->kind = (rocprofiler_feature_kind_t)val;
+      }
+    }
+
     // Creating SPM context
     rocprofiler_t* context = NULL;
     hsa_status_t status = rocprofiler_open(agent, features, features_found, &context,
@@ -472,13 +481,16 @@ void spm_ctrl_start(rocprofiler_feature_t* features, uint32_t features_found) {
     status = rocprofiler_start(context, 0);
     check_status(status);
     // saving the context in the vectort
-    spm_ctx_vec.push_back(context);
+    if (spm_ctx_vec == NULL) spm_ctx_vec = new spm_ctx_vec_t;
+    spm_ctx_vec->push_back(context);
   }
 }
 void spm_ctrl_stop() {
-  for (rocprofiler_t* context : spm_ctx_vec) {
+  for (rocprofiler_t* context : *spm_ctx_vec) {
     hsa_status_t status = rocprofiler_stop(context, 0);
     check_status(status);
+    status = rocprofiler_iterate_trace_data(context, NULL, NULL);
+    check_status(status);
     rocprofiler_close(context);
   }
 }
@@ -1565,12 +1577,6 @@ extern "C" PUBLIC_API void OnLoadToolProp(rocprofiler_settings_t* settings)
   // set a value to indicate tracing mode
   if (settings->k_concurrent != 0) settings->k_concurrent = (traces_found == 0) ? 1 : 2;
 
-  if (is_spm_trace) {
-    for (uint32_t index = 0; index < features_found; index++) {
-      features[index].kind = ROCPROFILER_FEATURE_KIND_TRACE;
-    }
-  }
-
   // Context array aloocation
   context_array = new context_array_t;