From 8e4bc9615ae2fc23e06fd59ee03b1a147b0d0f2f Mon Sep 17 00:00:00 2001
From: wjchuee <wjchuee@foxmail.com>
Date: Thu, 12 Jun 2025 14:14:06 +0800
Subject: [PATCH] Ant msmonitor add start

---
 .../_dynamic_profiler_config_context.py           | 15 +++++++++++++++
 .../_dynamic_profiler_monitor_shm.py              |  1 +
 torch_npu/profiler/_non_intrusive_profile.py      |  4 ++--
 torch_npu/profiler/dynamic_profile.py             |  3 ++-
 4 files changed, 20 insertions(+), 3 deletions(-)

diff --git a/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_config_context.py b/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_config_context.py
index ab8a5abfe6..675da67d7b 100644
--- a/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_config_context.py
+++ b/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_config_context.py
@@ -25,6 +25,7 @@ class ConfigContext:
         self.experimental_config = None
         self._active = 1
         self._start_step = 0
+        self._start = False
         self._warmup = 0
         self.is_valid = False
         self._meta_data = {}
@@ -52,6 +53,7 @@ class ConfigContext:
         self._parse_active(json_data)
         self._parse_warmup(json_data)
         self._parse_start_step(json_data)
+        self._parse_start(json_data)
         self._parse_exp_cfg(json_data)
         self._parse_ranks(json_data)
 
@@ -72,6 +74,16 @@ class ConfigContext:
         DynamicProfilerUtils.out_log("Start step will be set to {}.".format(
             self._start_step), DynamicProfilerUtils.LoggerLevelEnum.INFO)
 
+    def _parse_start(self, json_data: dict):
+        if not self._is_dyno:
+            self._start = json_data.get("start", False)
+        else:
+            start = json_data.get("PROFILE_START", False)
+            if isinstance(start, str):
+                self._start = self.BOOL_MAP.get(start.lower(), False)
+            else:
+                self._start = False
+
     def _parse_prof_dir(self, json_data: dict):
         if not self._is_dyno:
             self.prof_path = json_data.get('prof_dir', self.DEADLINE_PROF_DIR)
@@ -343,6 +355,9 @@ class ConfigContext:
     def start_step(self) -> int:
         return self._start_step
 
+    def start(self) -> bool:
+        return self._start
+
     def experimental_config(self) -> _ExperimentalConfig:
         return self.experimental_config
 
diff --git a/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_monitor_shm.py b/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_monitor_shm.py
index 9ffbce78f7..dd8e4820be 100644
--- a/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_monitor_shm.py
+++ b/torch_npu/profiler/_dynamic_profiler/_dynamic_profiler_monitor_shm.py
@@ -27,6 +27,7 @@ class DynamicProfilerShareMemory:
         "active": 1,
         "warmup": 0,
         "start_step": 0,
+        "start": False,
         "is_rank": False,
         "rank_list": [],
         "experimental_config": {
diff --git a/torch_npu/profiler/_non_intrusive_profile.py b/torch_npu/profiler/_non_intrusive_profile.py
index a60303adec..1d8115b637 100644
--- a/torch_npu/profiler/_non_intrusive_profile.py
+++ b/torch_npu/profiler/_non_intrusive_profile.py
@@ -59,11 +59,11 @@ class _NonIntrusiveProfile:
     @staticmethod
     def init():
         prof_config_path = os.getenv("PROF_CONFIG_PATH", "")
-        dyno_enable_flag = os.getenv("KINETO_USE_DAEMON", 0)
+        dyno_enable_flag = os.getenv("KINETO_USE_DAEMON_NPU", 0)
         try:
             dyno_enable_flag = int(dyno_enable_flag)
         except ValueError:
-            print_error_msg("Environment variable KINETO_USE_DAEMON value not valid, will be set to 0 !")
+            print_error_msg("Environment variable KINETO_USE_DAEMON_NPU value not valid, will be set to 0 !")
             dyno_enable_flag = 0
         if not prof_config_path and dyno_enable_flag != 1:
             return
diff --git a/torch_npu/profiler/dynamic_profile.py b/torch_npu/profiler/dynamic_profile.py
index 99bfd76f72..b5ea180ff6 100644
--- a/torch_npu/profiler/dynamic_profile.py
+++ b/torch_npu/profiler/dynamic_profile.py
@@ -85,7 +85,8 @@ class _DynamicProfile:
                 self.prof = None
                 DynamicProfilerUtils.out_log("Stop Dynamic Profiler at {} step.".format(
                     self.cur_step), DynamicProfilerUtils.LoggerLevelEnum.INFO)
-        elif self.prof is None and self.cfg_ctx is not None and self.cur_step == self.cfg_ctx.start_step():
+        elif self.prof is None and self.cfg_ctx is not None and \
+            (self.cur_step == self.cfg_ctx.start_step() or self.cfg_ctx.start()):
             self.step_num = self.cfg_ctx.active() + self.cfg_ctx.warmup()
             self.enable_prof()
             self.cfg_ctx = None
-- 
Gitee