Changes from PR comments. Removal of FastTreeNative flag.

Directory.Build.props

@@ -29,8 +29,13 @@
     <StrongNameKeyId>Open</StrongNameKeyId>
   </PropertyGroup>
 
+  <ItemGroup>
+    <TargetsArm Include="LdaNative" />
+    <TargetsArm Include="MatrixFactorizationNative" />
+  </ItemGroup>
+
   <PropertyGroup>
-    <DefineConstants Condition="'$(TargetPlatform)' == 'arm64'">$(DefineConstants);ARM64</DefineConstants>
+    <DefineConstants Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'">$(DefineConstants);ARM64</DefineConstants>
   </PropertyGroup>
 
   <PropertyGroup>

Directory.Build.targets

@@ -27,7 +27,15 @@
       </NativeAssemblyReference>
 		</ItemGroup>
 
-		<Copy SourceFiles = "@(NativeAssemblyReference->'%(FullAssemblyPath)')"
+    <PropertyGroup>
+      <ShouldCopyx64>false</ShouldCopyx64>
+      <ShouldCopyx64 Condition="'$(TargetArchitecture)' == 'x86' Or '$(TargetArchitecture)' == 'x64'">true</ShouldCopyx64>
+      <ShouldCopyArm>false</ShouldCopyArm>
+      <ShouldCopyArm Condition="('$(TargetArchitecture)' == 'arm64' Or '$(TargetArchitecture)' == 'arm') And %(NativeAssemblyReference.Item) == %(TargetsArm.Identity)">true</ShouldCopyArm>
+    </PropertyGroup>
+
+		<Copy Condition="$(ShouldCopyx64) or $(ShouldCopyArm)"
+      SourceFiles = "@(NativeAssemblyReference->'%(FullAssemblyPath)')"
 			DestinationFolder="$(OutDir)"
 			OverwriteReadOnlyFiles="$(OverwriteReadOnlyFiles)"
 			Retries="$(CopyRetryCount)"

docs/samples/Microsoft.ML.AutoML.Samples/Microsoft.ML.AutoML.Samples.csproj

@@ -16,7 +16,7 @@
     </ProjectReference>
 
     <NativeAssemblyReference Include="MatrixFactorizationNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="FastTreeNative" />
+    <NativeAssemblyReference Include="FastTreeNative" />
   </ItemGroup>
   <ItemGroup>
     <PackageReference Include="SciSharp.TensorFlow.Redist" Version="$(TensorFlowVersion)" />

docs/samples/Microsoft.ML.Samples.GPU/Microsoft.ML.Samples.GPU.csproj

@@ -28,12 +28,12 @@
     <ProjectReference Include="..\..\..\src\Microsoft.ML.StandardTrainers\Microsoft.ML.StandardTrainers.csproj" />
     <ProjectReference Include="..\..\..\src\Microsoft.ML.Transforms\Microsoft.ML.Transforms.csproj" />
 
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="CpuMathNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="FastTreeNative" />
+    <NativeAssemblyReference Include="CpuMathNative" />
+    <NativeAssemblyReference Include="FastTreeNative" />
     <NativeAssemblyReference Include="MatrixFactorizationNative" />
     <NativeAssemblyReference Include="LdaNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="SymSgdNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="MklProxyNative" />
+    <NativeAssemblyReference Include="SymSgdNative" />
+    <NativeAssemblyReference Include="MklProxyNative" />
 
   </ItemGroup>
 

docs/samples/Microsoft.ML.Samples/Microsoft.ML.Samples.csproj

@@ -34,12 +34,12 @@
     <ProjectReference Include="..\..\..\src\Microsoft.ML.Core\Microsoft.ML.Core.csproj" />
     <ProjectReference Include="..\..\..\src\Microsoft.ML.Data\Microsoft.ML.Data.csproj" />
 
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="CpuMathNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="FastTreeNative" />
+    <NativeAssemblyReference Include="CpuMathNative" />
+    <NativeAssemblyReference Include="FastTreeNative" />
     <NativeAssemblyReference Include="MatrixFactorizationNative" />
     <NativeAssemblyReference Include="LdaNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="SymSgdNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="MklProxyNative" />
+    <NativeAssemblyReference Include="SymSgdNative" />
+    <NativeAssemblyReference Include="MklProxyNative" />
     <None Include="Dynamic\Trainers\BinaryClassification\FastForest.cs">
       <DesignTime>True</DesignTime>
       <AutoGen>True</AutoGen>

src/Microsoft.ML.Console/Microsoft.ML.Console.csproj

@@ -28,13 +28,13 @@
     <ProjectReference Include="..\Microsoft.ML.TensorFlow\Microsoft.ML.TensorFlow.csproj" />
     <ProjectReference Include="..\Microsoft.ML.Transforms\Microsoft.ML.Transforms.csproj" />
 
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="FastTreeNative" />
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="CpuMathNative" />
+    <NativeAssemblyReference Include="FastTreeNative" />
+    <NativeAssemblyReference Include="CpuMathNative" />
     <NativeAssemblyReference Include="MatrixFactorizationNative" />
     <NativeAssemblyReference Include="LdaNative" />
     <!-- TODO: once we fix the 4 intel MKL methods, SymSgdNative will need to go back in. -->
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="SymSgdNative"/>
-    <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm'" Include="MklImports"/>
+    <NativeAssemblyReference Include="SymSgdNative"/>
+    <NativeAssemblyReference Include="MklImports"/>
     <NativeAssemblyReference Condition="'$(TargetArchitecture)' != 'arm64' And '$(TargetArchitecture)' != 'arm' And '$(OS)' == 'Windows_NT'" Include="libiomp5md"/>
   </ItemGroup>
 

src/Microsoft.ML.FastTree/Dataset/DenseIntArray.cs

@@ -24,10 +24,11 @@ internal abstract class DenseIntArray : IntArray, IIntArrayForwardIndexer
     {
         public override IntArrayType Type { get { return IntArrayType.Dense; } }
 
-        protected DenseIntArray(int length)
+        protected DenseIntArray(int length, PerformSumup sumupHandler = null)
         {
             Contracts.Assert(length >= 0);
             Length = length;
+            SumupHandler = sumupHandler;
         }
 
         public override int Length { get; }
@@ -70,7 +71,6 @@ public override IntArray[] Split(int[][] assignment)
             return newArrays;
         }
 
-#if USE_FASTTREENATIVE
         internal const string NativePath = "FastTreeNative";
         [DllImport(NativePath), SuppressUnmanagedCodeSecurity]
         private static extern unsafe int C_Sumup_float(
@@ -111,8 +111,6 @@ protected static unsafe void SumupCPlusPlusDense(SumupInputData input, FeatureHi
             }
         }
 
-#endif
-
         public override IIntArrayForwardIndexer GetIndexer()
         {
             return this;
@@ -391,18 +389,21 @@ public Dense8BitIntArray(int len)
             : base(len)
         {
             _data = new byte[len];
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public Dense8BitIntArray(byte[] buffer, ref int position)
             : base(buffer.ToInt(ref position))
         {
             _data = buffer.ToByteArray(ref position);
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public Dense8BitIntArray(int len, IEnumerable<int> values)
             : base(len)
         {
             _data = values.Select(i => (byte)i).ToArray(len);
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         /// <summary>
@@ -445,8 +446,7 @@ public override unsafe int this[int index]
             }
         }
 
-#if USE_FASTTREENATIVE
-        public override void Sumup(SumupInputData input, FeatureHistogram histogram)
+        private void SumupNative(SumupInputData input, FeatureHistogram histogram)
         {
             unsafe
             {
@@ -456,7 +456,8 @@ public override void Sumup(SumupInputData input, FeatureHistogram histogram)
                 }
             }
         }
-#endif
+
+        public override void Sumup(SumupInputData input, FeatureHistogram histogram) => SumupHandler(input, histogram);
     }
 
     /// <summary>
@@ -475,12 +476,14 @@ public Dense4BitIntArray(int len)
             : base(len)
         {
             _data = new byte[(len + 1) / 2]; // Even length = half the bytes. Odd length = half the bytes+0.5.
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public Dense4BitIntArray(int len, IEnumerable<int> values)
             : base(len)
         {
             _data = new byte[(len + 1) / 2];
+            SetupSumupHandler(SumupNative, base.Sumup);
 
             int currentIndex = 0;
             bool upper = true;
@@ -505,6 +508,7 @@ public Dense4BitIntArray(byte[] buffer, ref int position)
             : base(buffer.ToInt(ref position))
         {
             _data = buffer.ToByteArray(ref position);
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         /// <summary>
@@ -565,8 +569,7 @@ public override unsafe int this[int index]
             }
         }
 
-#if USE_FASTTREENATIVE
-        public override void Sumup(SumupInputData input, FeatureHistogram histogram)
+        public void SumupNative(SumupInputData input, FeatureHistogram histogram)
         {
             unsafe
             {
@@ -576,7 +579,8 @@ public override void Sumup(SumupInputData input, FeatureHistogram histogram)
                 }
             }
         }
-#endif
+
+        public override void Sumup(SumupInputData input, FeatureHistogram histogram) => SumupHandler(input, histogram);
     }
 
     /// <summary>
@@ -592,18 +596,21 @@ public Dense16BitIntArray(int len)
             : base(len)
         {
             _data = new ushort[len];
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public Dense16BitIntArray(int len, IEnumerable<int> values)
             : base(len)
         {
             _data = values.Select(i => (ushort)i).ToArray(len);
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public Dense16BitIntArray(byte[] buffer, ref int position)
             : base(buffer.ToInt(ref position))
         {
             _data = buffer.ToUShortArray(ref position);
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public override unsafe void Callback(Action<IntPtr> callback)
@@ -648,8 +655,8 @@ public override unsafe int this[int index]
                 _data[index] = (ushort)value;
             }
         }
-#if USE_FASTTREENATIVE
-        public override void Sumup(SumupInputData input, FeatureHistogram histogram)
+
+        public void SumupNative(SumupInputData input, FeatureHistogram histogram)
         {
             unsafe
             {
@@ -660,7 +667,9 @@ public override void Sumup(SumupInputData input, FeatureHistogram histogram)
                 }
             }
         }
-#endif
+
+        public override void Sumup(SumupInputData input, FeatureHistogram histogram) => SumupHandler(input, histogram);
+
     }
 
     /// <summary>
@@ -676,18 +685,21 @@ public Dense32BitIntArray(int len)
             : base(len)
         {
             _data = new int[len];
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public Dense32BitIntArray(int len, IEnumerable<int> values)
             : base(len)
         {
             _data = values.ToArray(len);
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public Dense32BitIntArray(byte[] buffer, ref int position)
             : base(buffer.ToInt(ref position))
         {
             _data = buffer.ToIntArray(ref position);
+            SetupSumupHandler(SumupNative, base.Sumup);
         }
 
         public override unsafe void Callback(Action<IntPtr> callback)
@@ -733,8 +745,7 @@ public override int this[int index]
             }
         }
 
-#if USE_FASTTREENATIVE
-        public override void Sumup(SumupInputData input, FeatureHistogram histogram)
+        public void SumupNative(SumupInputData input, FeatureHistogram histogram)
         {
             unsafe
             {
@@ -745,6 +756,7 @@ public override void Sumup(SumupInputData input, FeatureHistogram histogram)
                 }
             }
         }
-#endif
+
+        public override void Sumup(SumupInputData input, FeatureHistogram histogram) => SumupHandler(input, histogram);
     }
 }

src/Microsoft.ML.FastTree/Dataset/IntArray.cs

@@ -5,6 +5,8 @@
 using System;
 using System.Collections.Generic;
 using System.Linq;
+using System.Runtime.InteropServices;
+using System.Security;
 using Microsoft.ML.Runtime;
 
 namespace Microsoft.ML.Trainers.FastTree
@@ -32,6 +34,38 @@ internal abstract class IntArray : IEnumerable<int>
         /// </summary>
         public abstract int Length { get; }
 
+        // We can test if the dll is present by calling into a non-existant method.
+        // If we get an EntryPointNotFoundException then the dll exists and we can use the native code.
+        // If we get a DllNotFoundException then the dll doesn't exist and we should use the managed fallbacks.
+        private const string NativePath = "FastTreeNative";
+        [DllImport(NativePath), SuppressUnmanagedCodeSecurity]
+        private static extern void NonExistantMethod();
+
+        /// <summary>
+        /// Lazy bool that checks if the FastTreeNative dll exists so we know if we should use the native code
+        /// or the managed fallbacks.
+        /// </summary>
+        public static Lazy<bool> UseFastTreeNative = new(() =>
+        {
+            try
+            {
+                NonExistantMethod();
+            }
+            // If the entry point isn't found then the dll exists and we should use the managed code.
+            catch(EntryPointNotFoundException)
+            {
+                return true;
+            }
+            // If any other error is returned we should use the managed fallbacks.
+            catch
+            {
+                return false;
+            }
+
+            // We will never hit this but we need to have it so every path has a return value. Defaulting to managed fallbacks.
+            return false;
+        });
+
         /// <summary>
         /// Returns the number of bytes written by the member ToByteArray()
         /// </summary>
@@ -198,6 +232,17 @@ public static IntArray New(byte[] buffer, ref int position)
         /// <returns>An indexer into the array</returns>
         public abstract IIntArrayForwardIndexer GetIndexer();
 
+        // Used in the child classes so we can set either the native or managed Sumup method one time and then
+        // never have to check again.
+        protected delegate void PerformSumup(SumupInputData input, FeatureHistogram histogram);
+
+        // Handler so the child classes don't have to redefine it. If they don't have different logic for native vs managed
+        // code then they don't need to use this.
+        protected PerformSumup SumupHandler { get; set; }
+
+        // Helper to setup the SumupHandler for the derived classes that need it.
+        protected void SetupSumupHandler(PerformSumup native, PerformSumup managed) => SumupHandler = UseFastTreeNative.Value ? native : managed;
+
         public virtual void Sumup(SumupInputData input, FeatureHistogram histogram)
         {
             Contracts.Assert((input.Weights == null) == (histogram.SumWeightsByBin == null));
@@ -313,7 +358,7 @@ public IntArray Compress(uint[] workarray = null)
                 int bits = SegmentIntArray.BitsForValue((uint)maxval);
                 if (bits <= 21)
                 {
-                    SegmentIntArray.SegmentFindOptimalPath(workarray, Length,
+                    SegmentIntArray.SegmentFindOptimalPath.Value(workarray, Length,
                         bits, out segBits, out segTransitions);
                 }
             }