rocm&zluda handle apu

installer.py

@@ -493,8 +493,12 @@ def install_rocm_zluda():
                         os.environ.setdefault('HIP_VISIBLE_DEVICES', str(idx))
                         # if os.environ.get('TENSORFLOW_PACKAGE') == 'tensorflow-rocm': # do not use tensorflow-rocm for navi 3x
                         #    os.environ['TENSORFLOW_PACKAGE'] = 'tensorflow==2.13.0'
-                        break
-                    log.debug(f'ROCm: HSA_OVERRIDE_GFX_VERSION auto config skipped for {gpu.name}')
+                        if not device.is_apu:
+                            # although apu was found, there can be a dedicated card. do not break loop.
+                            # if no dedicated card was found, apu will be used.
+                            break
+                    else:
+                        log.debug(f'ROCm: HSA_OVERRIDE_GFX_VERSION auto config skipped for {gpu.name}')
             else:
                 device_id = int(args.device_id)
                 if device_id < len(amd_gpus):

modules/rocm.py

@@ -53,18 +53,20 @@ class MicroArchitecture(Enum):
 class Agent:
     name: str
     arch: MicroArchitecture
+    is_apu: bool
     if sys.platform != "win32":
         blaslt_supported: bool
 
     def __init__(self, name: str):
         self.name = name
-        gfx_version = name[3:7]
-        if len(gfx_version) == 4:
+        gfx = name[3:7]
+        if len(gfx) == 4:
             self.arch = MicroArchitecture.RDNA
-        elif gfx_version in ("908", "90a", "942",):
+        elif gfx in ("908", "90a", "942",):
             self.arch = MicroArchitecture.CDNA
         else:
             self.arch = MicroArchitecture.GCN
+        self.is_apu = gfx in ("801", "902", "90c", "1013", "1033", "1035", "1036", "1103",)
         if sys.platform != "win32":
             self.blaslt_supported = os.path.exists(os.path.join(HIPBLASLT_TENSILE_LIBPATH, f"extop_{name}.co"))