From e0c8d37d5e6a94e420919178dee842f776758ce7 Mon Sep 17 00:00:00 2001
From: Disty0 <semihgulec2005@gmail.com>
Date: Sun, 17 Sep 2023 16:14:40 +0300
Subject: [PATCH] Fix DPM SDE and update IPEX defaults

---
 modules/intel/ipex/attention.py | 2 +-
 modules/processing_diffusers.py | 6 +++---
 modules/shared.py               | 2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/modules/intel/ipex/attention.py b/modules/intel/ipex/attention.py
index d7335bfaf..fc4ab6e26 100644
--- a/modules/intel/ipex/attention.py
+++ b/modules/intel/ipex/attention.py
@@ -65,7 +65,7 @@ original_scaled_dot_product_attention = torch.nn.functional.scaled_dot_product_a
 def scaled_dot_product_attention(query, key, value, attn_mask=None, dropout_p=0.0, is_causal=False):
     #ARC GPUs can't allocate more than 4GB to a single block, Slice it:
     shape_one, batch_size_attention, query_tokens, shape_four = query.shape
-    block_multiply = 2.4 if query.dtype == torch.float32 else 1.2
+    block_multiply = 3.6 if query.dtype == torch.float32 else 1.8
     block_size = (shape_one * batch_size_attention * query_tokens * shape_four) / 1024 * block_multiply #MB
     split_slice_size = batch_size_attention
     if block_size >= 4000:
diff --git a/modules/processing_diffusers.py b/modules/processing_diffusers.py
index 9c47ae957..9cd74f162 100644
--- a/modules/processing_diffusers.py
+++ b/modules/processing_diffusers.py
@@ -283,7 +283,7 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
     recompile_model()
 
     is_karras_compatible = shared.sd_model.__class__.__init__.__annotations__.get("scheduler", None) == diffusers.schedulers.scheduling_utils.KarrasDiffusionSchedulers
-    if (not hasattr(shared.sd_model.scheduler, 'name')) or (shared.sd_model.scheduler.name != p.sampler_name) and (p.sampler_name != 'Default') and is_karras_compatible:
+    if ((not hasattr(shared.sd_model.scheduler, 'name')) or (p.sampler_name == 'DPM SDE') or (shared.sd_model.scheduler.name != p.sampler_name)) and (p.sampler_name != 'Default') and is_karras_compatible:
         sampler = sd_samplers.all_samplers_map.get(p.sampler_name, None)
         if sampler is None:
             sampler = sd_samplers.all_samplers_map.get("UniPC")
@@ -380,7 +380,7 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
             if latent_scale_mode is not None or p.hr_force:
                 p.ops.append('hires')
                 recompile_model(hires=True)
-                if (not hasattr(shared.sd_model.scheduler, 'name')) or (shared.sd_model.scheduler.name != p.latent_sampler) and (p.latent_sampler != 'Default') and is_karras_compatible:
+                if ((not hasattr(shared.sd_model.scheduler, 'name')) or (p.latent_sampler == 'DPM SDE') or (shared.sd_model.scheduler.name != p.latent_sampler)) and (p.latent_sampler != 'Default') and is_karras_compatible:
                     sampler = sd_samplers.all_samplers_map.get(p.latent_sampler, None)
                     if sampler is None:
                         sampler = sd_samplers.all_samplers_map.get("UniPC")
@@ -416,7 +416,7 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
             shared.sd_model.to(devices.cpu)
             devices.torch_gc()
 
-        if (not hasattr(shared.sd_refiner.scheduler, 'name')) or (shared.sd_refiner.scheduler.name != p.latent_sampler) and (p.latent_sampler != 'Default'):
+        if ((not hasattr(shared.sd_refiner.scheduler, 'name')) or (p.latent_sampler == 'DPM SDE') or (shared.sd_refiner.scheduler.name != p.latent_sampler)) and (p.latent_sampler != 'Default'):
             sampler = sd_samplers.all_samplers_map.get(p.latent_sampler, None)
             if sampler is None:
                 sampler = sd_samplers.all_samplers_map.get("UniPC")
diff --git a/modules/shared.py b/modules/shared.py
index 7ee44b668..bed228abf 100644
--- a/modules/shared.py
+++ b/modules/shared.py
@@ -434,7 +434,7 @@ options_templates.update(options_section(('diffusers', "Diffusers Settings"), {
     "diffusers_vae_upcast": OptionInfo("default", "VAE upcasting", gr.Radio, lambda: {"choices": ['default', 'true', 'false']}),
     "diffusers_vae_slicing": OptionInfo(True, "Enable VAE slicing"),
     "diffusers_vae_tiling": OptionInfo(False if cmd_opts.use_openvino else True, "Enable VAE tiling"),
-    "diffusers_attention_slicing": OptionInfo(False, "Enable attention slicing"),
+    "diffusers_attention_slicing": OptionInfo(True if devices.backend == "ipex" else False, "Enable attention slicing"),
     "diffusers_model_load_variant": OptionInfo("default", "Diffusers model loading variant", gr.Radio, lambda: {"choices": ['default', 'fp32', 'fp16']}),
     "diffusers_vae_load_variant": OptionInfo("default", "Diffusers VAE loading variant", gr.Radio, lambda: {"choices": ['default', 'fp32', 'fp16']}),
     "diffusers_lora_loader": OptionInfo("diffusers", "Diffusers LoRA loading variant", gr.Radio, lambda: {"choices": ['diffusers', 'sequential apply', 'merge and apply']}),