From dc1117233ef8f9b25ff1ac40b158f20b70ba2fcb Mon Sep 17 00:00:00 2001
From: AUTOMATIC <16777216c@gmail.com>
Date: Sat, 8 Oct 2022 17:02:18 +0300
Subject: [PATCH] simplify xfrmers options: --xformers to enable and that's it

---
 launch.py                          |  2 +-
 modules/sd_hijack.py               |  2 +-
 modules/sd_hijack_optimizations.py | 20 +++++++++++++-------
 modules/shared.py                  |  2 +-
 4 files changed, 16 insertions(+), 10 deletions(-)

diff --git a/launch.py b/launch.py
index a592e1ba..61f62096 100644
--- a/launch.py
+++ b/launch.py
@@ -32,7 +32,7 @@ def extract_arg(args, name):
 
 
 args, skip_torch_cuda_test = extract_arg(args, '--skip-torch-cuda-test')
-args, xformers = extract_arg(args, '--xformers')
+xformers = '--xformers' in args
 
 
 def repo_dir(name):
diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py
index 5d93f7f6..91e98c16 100644
--- a/modules/sd_hijack.py
+++ b/modules/sd_hijack.py
@@ -22,7 +22,7 @@ def apply_optimizations():
     undo_optimizations()
 
     ldm.modules.diffusionmodules.model.nonlinearity = silu
-    if not cmd_opts.disable_opt_xformers_attention and not (cmd_opts.opt_split_attention or torch.version.hip) and shared.xformers_available:
+    if cmd_opts.xformers and shared.xformers_available and not torch.version.hip:
         ldm.modules.attention.CrossAttention.forward = sd_hijack_optimizations.xformers_attention_forward
         ldm.modules.diffusionmodules.model.AttnBlock.forward = sd_hijack_optimizations.xformers_attnblock_forward
     elif cmd_opts.opt_split_attention_v1:
diff --git a/modules/sd_hijack_optimizations.py b/modules/sd_hijack_optimizations.py
index 05023b6f..d23d733b 100644
--- a/modules/sd_hijack_optimizations.py
+++ b/modules/sd_hijack_optimizations.py
@@ -1,4 +1,7 @@
 import math
+import sys
+import traceback
+
 import torch
 from torch import einsum
 
@@ -7,13 +10,16 @@ from einops import rearrange
 
 from modules import shared
 
-try:
-    import xformers.ops
-    import functorch
-    xformers._is_functorch_available = True
-    shared.xformers_available = True
-except Exception:
-    print('Cannot find xformers, defaulting to split attention. Try adding --xformers commandline argument to your webui-user file if you wish to install it.')
+if shared.cmd_opts.xformers:
+    try:
+        import xformers.ops
+        import functorch
+        xformers._is_functorch_available = True
+        shared.xformers_available = True
+    except Exception:
+        print("Cannot import xformers", file=sys.stderr)
+        print(traceback.format_exc(), file=sys.stderr)
+
 
 # see https://github.com/basujindal/stable-diffusion/pull/117 for discussion
 def split_cross_attention_forward_v1(self, x, context=None, mask=None):
diff --git a/modules/shared.py b/modules/shared.py
index d68df751..02cb2722 100644
--- a/modules/shared.py
+++ b/modules/shared.py
@@ -43,7 +43,7 @@ parser.add_argument("--realesrgan-models-path", type=str, help="Path to director
 parser.add_argument("--scunet-models-path", type=str, help="Path to directory with ScuNET model file(s).", default=os.path.join(models_path, 'ScuNET'))
 parser.add_argument("--swinir-models-path", type=str, help="Path to directory with SwinIR model file(s).", default=os.path.join(models_path, 'SwinIR'))
 parser.add_argument("--ldsr-models-path", type=str, help="Path to directory with LDSR model file(s).", default=os.path.join(models_path, 'LDSR'))
-parser.add_argument("--disable-opt-xformers-attention", action='store_true', help="force-disables xformers attention optimization")
+parser.add_argument("--xformers", action='store_true', help="enable xformers for cross attention layers")
 parser.add_argument("--opt-split-attention", action='store_true', help="force-enables cross-attention layer optimization. By default, it's on for torch.cuda and off for other torch devices.")
 parser.add_argument("--disable-opt-split-attention", action='store_true', help="force-disables cross-attention layer optimization")
 parser.add_argument("--opt-split-attention-v1", action='store_true', help="enable older version of split attention optimization that does not consume all the VRAM it can find")