A64: Add fast path for Fcvtas_Gp/S/V, Fcvtau_Gp/S/V and Frinta_S/V in… (#3712)

* A64: Add fast path for Fcvtas_Gp/S/V, Fcvtau_Gp/S/V and Frinta_S/V instructions;

they use "Round to Nearest with Ties to Away" rounding mode not supported in x86.

All instructions involved have been tested locally in both release and debug modes, in both lowcq and highcq.

The titles Mario Strikers and Super Smash Bros. U. use these instructions intensively.

* Update Ptc.cs

* A32: Add fast path for Vcvta_RM, Vrinta_RM and Vrinta_V instructions aswell.

1 files changed, 44 insertions, 11 deletions

diff --git a/ARMeilleure/Instructions/InstEmitSimdHelper.cs b/ARMeilleure/Instructions/InstEmitSimdHelper.cs
index 49c17560..0e7af794 100644
--- a/ARMeilleure/Instructions/InstEmitSimdHelper.cs
+++ b/ARMeilleure/Instructions/InstEmitSimdHelper.cs
@@ -33,6 +33,14 @@ namespace ARMeilleure.Instructions
         };
 
         public static readonly long ZeroMask = 128L << 56 | 128L << 48 | 128L << 40 | 128L << 32 | 128L << 24 | 128L << 16 | 128L << 8 | 128L << 0;
+
+        public static ulong X86GetGf2p8LogicalShiftLeft(int shift)
+        {
+            ulong identity = (0b00000001UL << 56) | (0b00000010UL << 48) | (0b00000100UL << 40) | (0b00001000UL << 32) |
+                             (0b00010000UL << 24) | (0b00100000UL << 16) | (0b01000000UL <<  8) | (0b10000000UL <<  0);
+
+            return shift >= 0 ? identity >> (shift * 8) : identity << (-shift * 8);
+        }
 #endregion
 
 #region "X86 SSE Intrinsics"
@@ -243,19 +251,44 @@ namespace ARMeilleure.Instructions
             throw new ArgumentException($"Invalid rounding mode \"{roundMode}\".");
         }
 
-        public static ulong X86GetGf2p8LogicalShiftLeft(int shift)
+        public static Operand EmitSse41RoundToNearestWithTiesToAwayOpF(ArmEmitterContext context, Operand n, bool scalar)
         {
-            ulong identity = 
-                (0b00000001UL << 56) |
-                (0b00000010UL << 48) |
-                (0b00000100UL << 40) |
-                (0b00001000UL << 32) |
-                (0b00010000UL << 24) |
-                (0b00100000UL << 16) |
-                (0b01000000UL <<  8) |
-                (0b10000000UL <<  0);
+            Debug.Assert(n.Type == OperandType.V128);
 
-            return shift >= 0 ? identity >> (shift * 8) : identity << (-shift * 8);
+            Operand nCopy = context.Copy(n);
+
+            Operand rC = Const(X86GetRoundControl(FPRoundingMode.TowardsZero));
+
+            IOpCodeSimd op = (IOpCodeSimd)context.CurrOp;
+
+            if ((op.Size & 1) == 0)
+            {
+                Operand signMask = scalar ? X86GetScalar(context, int.MinValue) : X86GetAllElements(context, int.MinValue);
+                        signMask = context.AddIntrinsic(Intrinsic.X86Pand, signMask, nCopy);
+
+                // 0x3EFFFFFF == BitConverter.SingleToInt32Bits(0.5f) - 1
+                Operand valueMask = scalar ? X86GetScalar(context, 0x3EFFFFFF) : X86GetAllElements(context, 0x3EFFFFFF);
+                        valueMask = context.AddIntrinsic(Intrinsic.X86Por, valueMask, signMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Addss : Intrinsic.X86Addps, nCopy, valueMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Roundss : Intrinsic.X86Roundps, nCopy, rC);
+            }
+            else
+            {
+                Operand signMask = scalar ? X86GetScalar(context, long.MinValue) : X86GetAllElements(context, long.MinValue);
+                        signMask = context.AddIntrinsic(Intrinsic.X86Pand, signMask, nCopy);
+
+                // 0x3FDFFFFFFFFFFFFFL == BitConverter.DoubleToInt64Bits(0.5d) - 1L
+                Operand valueMask = scalar ? X86GetScalar(context, 0x3FDFFFFFFFFFFFFFL) : X86GetAllElements(context, 0x3FDFFFFFFFFFFFFFL);
+                        valueMask = context.AddIntrinsic(Intrinsic.X86Por, valueMask, signMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Addsd : Intrinsic.X86Addpd, nCopy, valueMask);
+
+                nCopy = context.AddIntrinsic(scalar ? Intrinsic.X86Roundsd : Intrinsic.X86Roundpd, nCopy, rC);
+            }
+
+            return nCopy;
         }
 
         public static Operand EmitCountSetBits8(ArmEmitterContext context, Operand op) // "size" is 8 (SIMD&FP Inst.).