SWDEV-405575: Added gfx941 and gfx942

Change-Id: I45a49cd64a76d3ae32c209497c70fe27b5be212b [ROCm/rocprofiler commit: e1285e3fd4]
2023-06-13 15:26:06 -03:00
@@ -617,6 +617,9 @@
    <metric name="TCC_EA0_WRREQ_DRAM" block="TCC" event="103" descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC)."></metric>
  </gfx940>

+  <gfx941 base="gfx940"></gfx941>
+  <gfx942 base="gfx940"></gfx942>
+
  <gfx9>
    <metric name="GRBM_COUNT" block="GRBM" event="0" descr="Tie High - Count Number of Clocks"></metric>
    <metric name="GRBM_GUI_ACTIVE" block="GRBM" event="2" descr="The GUI is Active"></metric>
@@ -386,6 +386,9 @@
    <metric name="WRITE_REQ_32B" expr="TCC_EA0_WRREQ_64B_sum*2+(TCC_EA0_WRREQ_sum-TCC_EA0_WRREQ_64B_sum)" descr="The total number of 32-byte effective memory writes."></metric>
  </gfx940>

+  <gfx941 base="gfx940"></gfx941>
+  <gfx942 base="gfx940"></gfx942>
+
  <gfx10>
    <metric name="GPUBusy" expr="100*GRBM_GUI_ACTIVE/GRBM_COUNT" descr="The percentage of time GPU was busy."></metric>
    <metric name="Wavefronts" expr="SQ_WAVES" descr="Total wavefronts."></metric>
@@ -630,6 +630,9 @@
  <metric name="TCC_EA0_WRREQ_DRAM" block=TCC event=103 descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC)."></metric>
 </gfx940>

+<gfx941 base="gfx940"></gfx941>
+<gfx942 base="gfx940"></gfx942>
+
 <gfx10>
  <metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
  <metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
@@ -733,3 +736,4 @@

 <gfx1101 base="gfx11">
 </gfx1101>
+
@@ -505,6 +505,8 @@
 <gfx90a base="gfx90a_expr"></gfx90a>
 #Mi300
 <gfx940 base="gfx940_expr"></gfx940>
+<gfx941 base="gfx940_expr"></gfx941>
+<gfx942 base="gfx940_expr"></gfx942>
 #Navi21
 <gfx1030 base="gfx1030_expr"></gfx1030>
 <gfx1031 base="gfx1031_expr"></gfx1031>
@@ -39,6 +39,7 @@ THE SOFTWARE.
 #include "expr.h"
 #include "xml.h"
 #include <mutex>
+#include <unordered_set>
 #include "src/core/hardware/hsa_info.h"

 namespace rocprofiler {
@@ -203,14 +204,17 @@ class MetricsDict {
      xml_->AddConst("top.const.metric", "SE_NUM", agent_info->getShaderEngineCount());
      ImportMetrics(agent_info, "const");
      agent_name_ = agent_info->getName();
-      if (agent_name_.substr(0, 6) == "gfx940")
-        agent_name_ =
-            "gfx940";  // To correct the agent_name from "gfx940:forcestoresc1+" -> "gfx940"
-      if (std::string("gfx906") == agent_name_ || std::string("gfx908") == agent_name_ ||
-          std::string("gfx90a") == agent_name_ || std::string("gfx940") == agent_name_ ||
-          std::string("gfx1032") == agent_name_ || std::string("gfx1031") == agent_name_ ||
-          std::string("gfx1030") == agent_name_ || std::string("gfx1100") == agent_name_ ||
-          std::string("gfx1101") == agent_name_) {
+
+      if (agent_name_.find(':') != std::string::npos) // Remove compiler flags from the agent_name
+        agent_name_ = agent_name_.substr(0, agent_name_.find(':'));
+
+      std::unordered_set<std::string> supported_agent_names = {
+          "gfx906", "gfx908" "gfx90a",      // Vega
+          "gfx940", "gfx941", "gfx942",     // Mi300
+          "gfx1030", "gfx1031", "gfx1032",  // Navi2x
+          "gfx1100", "gfx1101"   // Navi3x
+      };
+      if (supported_agent_names.find(agent_name_) != supported_agent_names.end()) {
        ImportMetrics(agent_info, agent_name_);
      } else {
        agent_name_ = agent_info->getGfxip();
@@ -85,13 +85,11 @@ AgentInfo::AgentInfo(const hsa_agent_t agent, ::CoreApiTable* table) : handle_(a
      rocprofiler::fatal("hsa_agent_get_info for PCI info failed");
  }

-  // TODO: (sauverma) use hsa_agent_get_info_fn(HSA_AMD_AGENT_INFO_NUM_XCC)
-  // to get xcc_num once hsa headers are updated from rocr/hsa
-  std::string gpu_name = std::string(name_).substr(0, 6);
-  if (gpu_name == "gfx940")
-    xcc_num_ = 6;
-  else
-    xcc_num_ = 1;
+  // TODO(saurabh, giovanni): Remove this in 5.7
+  if (table->hsa_agent_get_info_fn(agent, 
+    static_cast<hsa_agent_info_t>(HSA_AMD_AGENT_INFO_NUM_XCC), &xcc_num_) != HSA_STATUS_SUCCESS) {
+      xcc_num_ = 1;
+  }
 }

 uint64_t AgentInfo::getIndex() const { return index_; }
@@ -193,7 +193,7 @@ static const kernel_descriptor_t* GetKernelCode(uint64_t kernel_object) {
 static uint32_t arch_vgpr_count(Agent::AgentInfo& info, const kernel_descriptor_t& kernel_code) {
  const std::string_view& name = info.getName();
  std::string info_name(name.data(), name.size());
-  if (strcmp(name.data(), "gfx90a") == 0 || strcmp(name.data(), "gfx940") == 0)
+  if (strcmp(name.data(), "gfx90a") == 0 || strncmp(name.data(), "gfx94", 5) == 0)
    return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc3,
                             AMD_COMPUTE_PGM_RSRC_THREE_ACCUM_OFFSET) +
            1) *
@@ -211,7 +211,7 @@ static uint32_t accum_vgpr_count(Agent::AgentInfo& info, const kernel_descriptor
  const std::string_view& name = info.getName();
  std::string info_name(name.data(), name.size());
  if (strcmp(info_name.c_str(), "gfx908") == 0) return arch_vgpr_count(info, kernel_code);
-  if (strcmp(info_name.c_str(), "gfx90a") == 0 || strcmp(info_name.c_str(), "gfx940") == 0)
+  if (strcmp(info_name.c_str(), "gfx90a") == 0 || strncmp(info_name.c_str(), "gfx94", 5) == 0)
    return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc1,
                             AMD_COMPUTE_PGM_RSRC_ONE_GRANULATED_WORKITEM_VGPR_COUNT) +
            1) *
@@ -33,6 +33,7 @@ THE SOFTWARE.
 #include <list>
 #include <map>
 #include <vector>
+#include <unordered_set>

 #include "core/types.h"
 #include "util/exception.h"
@@ -201,14 +202,17 @@ class MetricsDict {
      xml_->AddConst("top.const.metric", "SE_NUM", agent_info->se_num);
      ImportMetrics(agent_info, "const");
      agent_name_ = agent_info->name;
-      if (agent_name_.substr(0, 6) == "gfx940")
-        agent_name_ =
-            "gfx940";  // To correct the agent_name from "gfx940:forcestoresc1+" -> "gfx940"
-      if (std::string("gfx906") == agent_name_ || std::string("gfx908") == agent_name_ ||
-          std::string("gfx90a") == agent_name_ || std::string("gfx940") == agent_name_ ||
-          std::string("gfx1032") == agent_name_ || std::string("gfx1031") == agent_name_ ||
-          std::string("gfx1030") == agent_name_ || std::string("gfx1100") == agent_name_ ||
-          std::string("gfx1101") == agent_name_) {
+  
+      if (agent_name_.find(':') != std::string::npos) // Remove compiler flags from the agent_name
+        agent_name_ = agent_name_.substr(0, agent_name_.find(':'));
+
+      std::unordered_set<std::string> supported_agent_names = {
+          "gfx906", "gfx908" "gfx90a",      // Vega
+          "gfx940", "gfx941", "gfx942",     // Mi300
+          "gfx1030", "gfx1031", "gfx1032",  // Navi2x
+          "gfx1100", "gfx1101"   // Navi3x
+      };
+      if (supported_agent_names.find(agent_name_) != supported_agent_names.end()) {
        ImportMetrics(agent_info, agent_name_);
      } else {
        agent_name_ = agent_info->gfxip;
@@ -631,6 +631,9 @@
  <metric name="TCC_EA0_WRREQ_DRAM" block=TCC event=103 descr="Number of TCC/EA write requests (either 32-byte of 64-byte) destined for DRAM (MC)."></metric>
 </gfx940>

+<gfx941 base="gfx940"></gfx941>
+<gfx942 base="gfx940"></gfx942>
+
 <gfx10>
  <metric name="GRBM_COUNT" block=GRBM event=0 descr="Tie High - Count Number of Clocks"></metric>
  <metric name="GRBM_GUI_ACTIVE" block=GRBM event=2 descr="The GUI is Active"></metric>
@@ -739,3 +742,4 @@

 <gfx1101 base="gfx11">
 </gfx1101>
+
@@ -426,6 +426,8 @@
 <gfx90a base="gfx90a_expr"></gfx90a>
 #Mi300
 <gfx940 base="gfx940_expr"></gfx940>
+<gfx941 base="gfx940_expr"></gfx941>
+<gfx942 base="gfx940_expr"></gfx942>
 #Navi21
 <gfx1030 base="gfx1030_expr"></gfx1030>

@@ -670,7 +670,7 @@ static const kernel_descriptor_t* GetKernelCode(uint64_t kernel_object) {
 }

 static uint32_t arch_vgpr_count(const AgentInfo &info, const kernel_descriptor_t &kernel_code) {
-  if (strcmp(info.name, "gfx90a") == 0 || strcmp(info.name, "gfx940") == 0)
+  if (strcmp(info.name, "gfx90a") == 0 || strncmp(info.name, "gfx94", 5) == 0)
    return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc3, AMD_COMPUTE_PGM_RSRC_THREE_ACCUM_OFFSET) + 1) * 4;

  return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc1, AMD_COMPUTE_PGM_RSRC_ONE_GRANULATED_WORKITEM_VGPR_COUNT) + 1)
@@ -680,7 +680,7 @@ static uint32_t arch_vgpr_count(const AgentInfo &info, const kernel_descriptor_t
 static uint32_t accum_vgpr_count(const AgentInfo &info, const kernel_descriptor_t &kernel_code) {
  if (strcmp(info.name, "gfx908") == 0)
    return arch_vgpr_count(info, kernel_code);
-  if (strcmp(info.name, "gfx90a") == 0 || strcmp(info.name, "gfx940") == 0)
+  if (strcmp(info.name, "gfx90a") == 0 || strncmp(info.name, "gfx94", 5) == 0)
    return (AMD_HSA_BITS_GET(kernel_code.compute_pgm_rsrc1,
                             AMD_COMPUTE_PGM_RSRC_ONE_GRANULATED_WORKITEM_VGPR_COUNT) + 1) * 8
      - arch_vgpr_count(info, kernel_code);
@@ -737,6 +737,7 @@ void set_kernel_properties(const rocprofiler_callback_data_t* callback_data,
 // Kernel disoatch callback
 hsa_status_t dispatch_callback(const rocprofiler_callback_data_t* callback_data, void* user_data,
                               rocprofiler_group_t* group) {
+  // TODO: return success, make atomic flag
  // Passed tool data
  callbacks_data_t* tool_data = reinterpret_cast<callbacks_data_t*>(user_data);
  // HSA status