Decide attention and ANE on hardware capabilities (huggingface#29)

For Macs with >= 8 performance cores, we select CPU+GPU (original
attention). Otherwise we select CPU+ANE (split einsum).

Some computers (M1 Pro, 16 core GPU) might yield slightly better
performance using CPU+GPU+ANE with SPLIT_EINSUM.

Author: pcuenca

Diffusion-macOS/Capabilities.swift

@@ -0,0 +1,36 @@
+//
+//  Capabilities.swift
+//  Diffusion-macOS
+//
+//  Created by Pedro Cuenca on 20/2/23.
+//  See LICENSE at https://github.com/huggingface/swift-coreml-diffusers/LICENSE
+//
+
+import Foundation
+
+let runningOnMac = true
+
+#if canImport(MLCompute)
+import MLCompute
+let _hasANE = MLCDevice.ane() != nil
+#else
+let _hasANE = false
+#endif
+
+final class Capabilities {
+    static let hasANE = _hasANE
+    
+    // According to my tests this is a good proxy to estimate whether CPU+GPU
+    // or CPU+NE works better. Things may become more complicated if we
+    // choose all compute units.
+    static var performanceCores: Int = {
+        var ncores: Int32 = 0
+        var bytes = MemoryLayout<Int32>.size
+        
+        // In M1/M2 perflevel0 refers to the performance cores and perflevel1 are the efficiency cores
+        // In Intel there's only one performance level
+        let result = sysctlbyname("hw.perflevel0.physicalcpu", &ncores, &bytes, nil, 0)
+        guard result == 0 else { return 0 }
+        return Int(ncores)
+    }()
+}

Diffusion-macOS/ControlsView.swift

@@ -284,7 +284,7 @@ struct ControlsView: View {
                         }.foregroundColor(.secondary)
                     }
 
-                    if hasANE {
+                    if Capabilities.hasANE {
                         Divider()
                         DisclosureGroup(isExpanded: $disclosedAdvanced) {
                             HStack {

Diffusion-macOS/Diffusion_macOSApp.swift

@@ -16,12 +16,3 @@ struct Diffusion_macOSApp: App {
         }
     }
 }
-
-let runningOnMac = true
-
-#if canImport(MLCompute)
-import MLCompute
-let hasANE = MLCDevice.ane() != nil
-#else
-let hasANE = false
-#endif

Diffusion.xcodeproj/project.pbxproj

@@ -9,6 +9,7 @@
 /* Begin PBXBuildFile section */
 		EB067F872992E561004D1AD9 /* HelpContent.swift in Sources */ = {isa = PBXBuildFile; fileRef = EB067F862992E561004D1AD9 /* HelpContent.swift */; };
 		EB33A51D2954D89F00B16357 /* StableDiffusion in Frameworks */ = {isa = PBXBuildFile; productRef = EB33A51C2954D89F00B16357 /* StableDiffusion */; };
+		EB560F0429A3C20800C0F8B8 /* Capabilities.swift in Sources */ = {isa = PBXBuildFile; fileRef = EB560F0329A3C20800C0F8B8 /* Capabilities.swift */; };
 		EBB5BA5329425BEE003A2A5B /* PipelineLoader.swift in Sources */ = {isa = PBXBuildFile; fileRef = EBB5BA5229425BEE003A2A5B /* PipelineLoader.swift */; };
 		EBB5BA5829425E17003A2A5B /* Path in Frameworks */ = {isa = PBXBuildFile; productRef = EBB5BA5729425E17003A2A5B /* Path */; };
 		EBB5BA5A29426E06003A2A5B /* Downloader.swift in Sources */ = {isa = PBXBuildFile; fileRef = EBB5BA5929426E06003A2A5B /* Downloader.swift */; };
@@ -64,6 +65,7 @@
 /* Begin PBXFileReference section */
 		EB067F862992E561004D1AD9 /* HelpContent.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = HelpContent.swift; sourceTree = "<group>"; };
 		EB33A51E2954E1BC00B16357 /* Info.plist */ = {isa = PBXFileReference; lastKnownFileType = text.plist; path = Info.plist; sourceTree = "<group>"; };
+		EB560F0329A3C20800C0F8B8 /* Capabilities.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = Capabilities.swift; sourceTree = "<group>"; };
 		EBB5BA5229425BEE003A2A5B /* PipelineLoader.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = PipelineLoader.swift; sourceTree = "<group>"; };
 		EBB5BA5929426E06003A2A5B /* Downloader.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = Downloader.swift; sourceTree = "<group>"; };
 		EBDD7DB22973200200C1C4B2 /* Utils.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = Utils.swift; sourceTree = "<group>"; };
@@ -247,6 +249,7 @@
 				F1552030297109C300DC009B /* ControlsView.swift */,
 				F155203329710B3600DC009B /* StatusView.swift */,
 				EB067F862992E561004D1AD9 /* HelpContent.swift */,
+				EB560F0329A3C20800C0F8B8 /* Capabilities.swift */,
 				F155202C2971093400DC009B /* Diffusion_macOS.entitlements */,
 				F15520292971093400DC009B /* Preview Content */,
 			);
@@ -502,6 +505,7 @@
 				F1552031297109C300DC009B /* ControlsView.swift in Sources */,
 				EBDD7DB62973206600C1C4B2 /* Downloader.swift in Sources */,
 				F155203429710B3600DC009B /* StatusView.swift in Sources */,
+				EB560F0429A3C20800C0F8B8 /* Capabilities.swift in Sources */,
 				F15520242971093300DC009B /* Diffusion_macOSApp.swift in Sources */,
 				EBDD7DB52973201800C1C4B2 /* ModelInfo.swift in Sources */,
 				EBDD7DBD2977FFB300C1C4B2 /* GeneratedImageView.swift in Sources */,

Diffusion/ModelInfo.swift

@@ -40,10 +40,11 @@ struct ModelInfo {
 
 extension ModelInfo {
     static var defaultAttention: AttentionVariant {
-        return runningOnMac ? .original : .splitEinsum
+        guard runningOnMac else { return .splitEinsum }
+        guard Capabilities.hasANE else { return .original }
+        return Capabilities.performanceCores >= 8 ? .original : .splitEinsum
     }
 
-    // TODO: heuristics per {model, device}
     var bestAttention: AttentionVariant {
         return ModelInfo.defaultAttention
     }
@@ -60,7 +61,7 @@ extension ModelInfo {
     }
 
     /// Best variant for the current platform.
-    /// Currently using `split_einsum` for iOS and `original` for macOS, but could vary depending on model.
+    /// Currently using `split_einsum` for iOS and simple performance heuristics for macOS.
     var bestURL: URL { modelURL(for: bestAttention) }
 
     var reduceMemory: Bool {