Initial ZLUDA support.

modules/shared.py

@@ -17,6 +17,7 @@ from modules import errors, shared_items, shared_state, cmd_args, theme
 from modules.paths import models_path, script_path, data_path, sd_configs_path, sd_default_config, sd_model_file, default_sd_model_file, extensions_dir, extensions_builtin_dir # pylint: disable=W0611
 from modules.dml import memory_providers, default_memory_provider, directml_do_hijack
 from modules.onnx_impl import initialize_onnx, execution_providers
+from modules.zluda import initialize_zluda
 import modules.interrogate
 import modules.memmon
 import modules.styles
@@ -922,6 +923,8 @@ mem_mon = modules.memmon.MemUsageMonitor("MemMon", devices.device)
 max_workers = 4
 if devices.backend == "directml":
     directml_do_hijack()
+elif devices.backend == "cuda":
+    initialize_zluda()
 initialize_onnx()
 
 

modules/zluda.py

@@ -0,0 +1,8 @@
+import torch
+from modules import shared, devices
+
+
+def initialize_zluda():
+    if devices.cuda_ok and torch.cuda.get_device_name(devices.get_optimal_device()).endswith("[ZLUDA]"):
+        torch.backends.cudnn.enabled = False
+        shared.opts.cross_attention_optimization = "Batch matrix-matrix"