[0.7.3] Use env for soc during compilation (#606)

### What this PR does / why we need it?

Sometimes SOC can not be correctly detected by npu-smi, it is better to
use env, which can be set by users.

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?

Tested by installing, compiling and UT.

Signed-off-by: Shuqiao Li <celestialli@outlook.com>

Author: celestialli

setup.py

@@ -157,19 +157,22 @@ def configure(self, ext: CMakeExtension) -> None:
             # else specify pybind11 path installed from source code on CI container
             raise RuntimeError(f"CMake configuration failed: {e}")
 
-        # try retrive soc version from npu-smi
-        soc_command = [
-            "bash",
-            "-c",
-            "npu-smi info | grep OK | awk '{print $3}' | head -n 1",
-        ]
-        try:
-            soc_version = subprocess.check_output(soc_command,
-                                                  text=True).strip()
-            soc_version = soc_version.split("-")[0]
-            soc_version = "Ascend" + soc_version
-        except subprocess.CalledProcessError as e:
-            raise RuntimeError(f"Retrive Soc version failed: {e}")
+        if envs.SOC_VERSION is None:
+            # try retrive soc version from npu-smi
+            soc_command = [
+                "bash",
+                "-c",
+                "npu-smi info | grep OK | awk '{print $3}' | head -n 1",
+            ]
+            try:
+                soc_version = subprocess.check_output(soc_command,
+                                                      text=True).strip()
+                soc_version = soc_version.split("-")[0]
+                soc_version = "Ascend" + soc_version
+            except subprocess.CalledProcessError as e:
+                raise RuntimeError(f"Retrive Soc version failed: {e}")
+        else:
+            soc_version = envs.SOC_VERSION
 
         # add SOC_VERSION
         cmake_args += [f"-DSOC_VERSION={soc_version}"]

vllm_ascend/envs.py

@@ -31,7 +31,9 @@
     "CXX_COMPILER":
     lambda: os.getenv("CXX_COMPILER", None),
     "C_COMPILER":
-    lambda: os.getenv("C_COMPILER", None)
+    lambda: os.getenv("C_COMPILER", None),
+    "SOC_VERSION":
+    lambda: os.getenv("SOC_VERSION", None),
 }