JIT: Fix removal of intermediate double cast in long->double->float chains

src/coreclr/jit/morph.cpp

@@ -359,6 +359,22 @@ GenTree* Compiler::fgMorphExpandCast(GenTreeCast* tree)
             }
         }
     }
+
+    // If we have a double->float cast, and the double node is itself a cast,
+    // look through it and see if we can cast directly to float. This is valid
+    // only if the cast to double would have been lossless.
+    //
+    // This pattern most often appears as CAST(float <- CAST(double <- float)),
+    // which is reduced to CAST(float <- float) and handled in codegen as an optional mov.
+    else if ((srcType == TYP_DOUBLE) && (dstType == TYP_FLOAT) && oper->OperIs(GT_CAST) &&
+             !varTypeIsLong(oper->AsCast()->CastOp()))
+    {
+        oper->gtType       = TYP_FLOAT;
+        oper->CastToType() = TYP_FLOAT;
+
+        return fgMorphTree(oper);
+    }
+
 #ifndef TARGET_64BIT
     // The code generation phase (for x86 & ARM32) does not handle casts
     // directly from [u]long to anything other than [u]int. Insert an
@@ -372,36 +388,6 @@ GenTree* Compiler::fgMorphExpandCast(GenTreeCast* tree)
     }
 #endif //! TARGET_64BIT
 
-#if defined(TARGET_ARMARCH) || defined(TARGET_XARCH)
-    // Because there is no IL instruction conv.r4.un, uint/ulong -> float
-    // casts are always imported as CAST(float <- CAST(double <- uint/ulong)).
-    // We can usually eliminate the redundant intermediate cast as an optimization.
-    //
-    // AArch and xarch+EVEX have instructions that can cast directly from
-    // all integers (except for longs on ARM32) to floats.
-    // On x64, we also have the option of widening uint -> long and
-    // using the signed conversion instructions, and ulong -> float/double
-    // is handled directly in codegen, so we can allow all casts.
-    //
-    // This logic will also catch CAST(float <- CAST(double <- float))
-    // and reduce it to CAST(float <- float), which is handled in codegen as
-    // an optional mov.
-    else if ((dstType == TYP_FLOAT) && (srcType == TYP_DOUBLE) && oper->OperIs(GT_CAST)
-#ifndef TARGET_64BIT
-             && !varTypeIsLong(oper->AsCast()->CastOp())
-#endif // !TARGET_64BIT
-#ifdef TARGET_X86
-             && canUseEvexEncoding()
-#endif // TARGET_X86
-    )
-    {
-        oper->gtType       = TYP_FLOAT;
-        oper->CastToType() = TYP_FLOAT;
-
-        return fgMorphTree(oper);
-    }
-#endif // TARGET_ARMARCH || TARGET_XARCH
-
 #ifdef TARGET_ARM
     // converts long/ulong --> float/double casts into helper calls.
     else if (varTypeIsFloating(dstType) && varTypeIsLong(srcType))

src/tests/JIT/Regression/JitBlue/Runtime_106338/Runtime_106338.cs

@@ -8,21 +8,74 @@
 // Debug: Outputs 1600094603
 // Release: Outputs 1600094604
 using System;
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics.X86;
 using Xunit;
 
 public class Runtime_106338
 {
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    static float CastToFloatDirect(long val) => (float)val;
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    static float CastToFloatDirect(ulong val) => (float)val;
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    static float CastToFloatThroughDouble(long val) => (float)(double)val;
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    static float CastToFloatThroughDouble(ulong val) => (float)(double)val;
+
     [Fact]
     [SkipOnMono("https://github.com/dotnet/runtime/issues/100368", TestPlatforms.Any)]
     public static void TestEntryPoint()
     {
-        ulong vr10 = 16105307123914158031UL;
-        float vr11 = 4294967295U | vr10;
-        uint result = BitConverter.SingleToUInt32Bits(vr11);
+        ulong vr10 = 0xDF818B7F_FFFFFFFF;
+        float vr11 = vr10;
+        float vr12 = (float)(double)vr10;
+
+        // These results will be const folded
+        uint resultDirect = BitConverter.SingleToUInt32Bits(vr11);
+        uint resultIntermediate = BitConverter.SingleToUInt32Bits(vr12);
 
         // Expected to cast ulong -> float directly
-        Assert.Equal(1600094603U, result);
+        Assert.Equal(1600094603U, resultDirect);
+
+        // Expected to preserve ulong -> double intermediate cast
+        Assert.Equal(1600094604U, resultIntermediate);
+
+        // Check that run-time computed values match
+        resultDirect = BitConverter.SingleToUInt32Bits(CastToFloatDirect(vr10));
+        resultIntermediate = BitConverter.SingleToUInt32Bits(CastToFloatThroughDouble(vr10));
+
+        Assert.Equal(1600094603U, resultDirect);
+        Assert.Equal(1600094604U, resultIntermediate);
+    }
+
+    [Fact]
+    [SkipOnMono("https://github.com/dotnet/runtime/issues/100368", TestPlatforms.Any)]
+    public static void TestEntryPointSigned()
+    {
+        long vr10 = 0x002FFFFF_DFFFFFFF;
+        float vr11 = vr10;
+        float vr12 = (float)(double)vr10;
+
+        // These results will be const folded
+        uint resultDirect = BitConverter.SingleToUInt32Bits(vr11);
+        uint resultIntermediate = BitConverter.SingleToUInt32Bits(vr12);
+
+        // Expected to cast long -> float directly
+        Assert.Equal(1514143743U, resultDirect);
+
+        // Expected to preserve long -> double intermediate cast
+        Assert.Equal(1514143744U, resultIntermediate);
+
+        // Check that run-time computed values match
+        resultDirect = BitConverter.SingleToUInt32Bits(CastToFloatDirect(vr10));
+        resultIntermediate = BitConverter.SingleToUInt32Bits(CastToFloatThroughDouble(vr10));
+
+        Assert.Equal(1514143743U, resultDirect);
+        Assert.Equal(1514143744U, resultIntermediate);
     }
 }