Add Saddlv_V Inst. Improve Cnt_V, Dup_Gp & Ins_Gp Tests. Tuneup Cls_V & Clz_V Tests. (#720)

* Update PackageReferences.

* Improve Cnt_V Test. Tuneup Cls_V & Clz_V Tests.

Nit.

* Nit.

* Improve Dup_Gp & Ins_Gp Tests.

* Update for Saddlv_V Inst.

* Update for Saddlv_V Inst.

* Update for Saddlv_V Inst.

ChocolArm64/Instructions/InstEmitSimdArithmetic.cs

@@ -2206,6 +2206,11 @@ namespace ChocolArm64.Instructions
             EmitAddLongPairwise(context, signed: true, accumulate: false);
         }
 
+        public static void Saddlv_V(ILEmitterCtx context)
+        {
+            EmitVectorLongAcrossVectorOpSx(context, () => context.Emit(OpCodes.Add));
+        }
+
         public static void Saddw_V(ILEmitterCtx context)
         {
             if (Optimizations.UseSse41)
@@ -3041,21 +3046,7 @@ namespace ChocolArm64.Instructions
 
         public static void Uaddlv_V(ILEmitterCtx context)
         {
-            OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
+            EmitVectorLongAcrossVectorOpZx(context, () => context.Emit(OpCodes.Add));
-
-            int bytes = op.GetBitsCount() >> 3;
-            int elems = bytes >> op.Size;
-
-            EmitVectorExtractZx(context, op.Rn, 0, op.Size);
-
-            for (int index = 1; index < elems; index++)
-            {
-                EmitVectorExtractZx(context, op.Rn, index, op.Size);
-
-                context.Emit(OpCodes.Add);
-            }
-
-            EmitScalarSet(context, op.Rd, op.Size + 1);
         }
 
         public static void Uaddw_V(ILEmitterCtx context)

ChocolArm64/Instructions/InstEmitSimdHelper.cs

@@ -823,15 +823,29 @@ namespace ChocolArm64.Instructions
 
         public static void EmitVectorAcrossVectorOpSx(ILEmitterCtx context, Action emit)
         {
-            EmitVectorAcrossVectorOp(context, emit, true);
+            EmitVectorAcrossVectorOp(context, emit, signed: true, isLong: false);
         }
 
         public static void EmitVectorAcrossVectorOpZx(ILEmitterCtx context, Action emit)
         {
-            EmitVectorAcrossVectorOp(context, emit, false);
+            EmitVectorAcrossVectorOp(context, emit, signed: false, isLong: false);
         }
 
-        public static void EmitVectorAcrossVectorOp(ILEmitterCtx context, Action emit, bool signed)
+        public static void EmitVectorLongAcrossVectorOpSx(ILEmitterCtx context, Action emit)
+        {
+            EmitVectorAcrossVectorOp(context, emit, signed: true, isLong: true);
+        }
+
+        public static void EmitVectorLongAcrossVectorOpZx(ILEmitterCtx context, Action emit)
+        {
+            EmitVectorAcrossVectorOp(context, emit, signed: false, isLong: true);
+        }
+
+        public static void EmitVectorAcrossVectorOp(
+            ILEmitterCtx context,
+            Action emit,
+            bool signed,
+            bool isLong)
         {
             OpCodeSimd64 op = (OpCodeSimd64)context.CurrOp;
 
@@ -847,7 +861,7 @@ namespace ChocolArm64.Instructions
                 emit();
             }
 
-            EmitScalarSet(context, op.Rd, op.Size);
+            EmitScalarSet(context, op.Rd, isLong ? op.Size + 1 : op.Size);
         }
 
         public static void EmitVectorPairwiseOpF(ILEmitterCtx context, Action emit)

ChocolArm64/OpCodeTable.cs

@@ -439,6 +439,8 @@ namespace ChocolArm64
             SetA64("0x001110<<100000011010xxxxxxxxxx", InstEmit.Sadalp_V,        typeof(OpCodeSimd64));
             SetA64("0x001110<<1xxxxx000000xxxxxxxxxx", InstEmit.Saddl_V,         typeof(OpCodeSimdReg64));
             SetA64("0x001110<<100000001010xxxxxxxxxx", InstEmit.Saddlp_V,        typeof(OpCodeSimd64));
+            SetA64("000011100x110000001110xxxxxxxxxx", InstEmit.Saddlv_V,        typeof(OpCodeSimd64));
+            SetA64("01001110<<110000001110xxxxxxxxxx", InstEmit.Saddlv_V,        typeof(OpCodeSimd64));
             SetA64("0x001110<<1xxxxx000100xxxxxxxxxx", InstEmit.Saddw_V,         typeof(OpCodeSimdReg64));
             SetA64("x00111100x100010000000xxxxxxxxxx", InstEmit.Scvtf_Gp,        typeof(OpCodeSimdCvt64));
             SetA64(">00111100x000010>xxxxxxxxxxxxxxx", InstEmit.Scvtf_Gp_Fixed,  typeof(OpCodeSimdCvt64));

Ryujinx.Tests/Cpu/CpuTestSimd.cs

@@ -117,6 +117,12 @@ namespace Ryujinx.Tests.Cpu
                                  0x8000000080000000ul, 0xFFFFFFFFFFFFFFFFul };
         }
 
+        private static ulong[] _4H_()
+        {
+            return new ulong[] { 0x0000000000000000ul, 0x7FFF7FFF7FFF7FFFul,
+                                 0x8000800080008000ul, 0xFFFFFFFFFFFFFFFFul };
+        }
+
         private static ulong[] _4H2S1D_()
         {
             return new ulong[] { 0x0000000000000000ul, 0x7FFF7FFF7FFF7FFFul,
@@ -155,73 +161,16 @@ namespace Ryujinx.Tests.Cpu
                                  0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul };
         }
 
-        private static IEnumerable<ulong> _GenLeadingSigns8B_()
+        private static uint[] _W_()
         {
-            for (int cnt = 0; cnt <= 7; cnt++)
+            return new uint[] { 0x00000000u, 0x7FFFFFFFu,
-            {
+                                0x80000000u, 0xFFFFFFFFu };
-                ulong rnd1 = GenLeadingSignsMinus8(cnt);
-                ulong rnd2 = GenLeadingSignsPlus8(cnt);
-
-                yield return (rnd1 << 56) | (rnd1 << 48) | (rnd1 << 40) | (rnd1 << 32) |
-                             (rnd1 << 24) | (rnd1 << 16) | (rnd1 << 08) | rnd1;
-                yield return (rnd2 << 56) | (rnd2 << 48) | (rnd2 << 40) | (rnd2 << 32) |
-                             (rnd2 << 24) | (rnd2 << 16) | (rnd2 << 08) | rnd2;
-            }
         }
 
-        private static IEnumerable<ulong> _GenLeadingSigns4H_()
+        private static ulong[] _X_()
         {
-            for (int cnt = 0; cnt <= 15; cnt++)
+            return new ulong[] { 0x0000000000000000ul, 0x7FFFFFFFFFFFFFFFul,
-            {
+                                 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul };
-                ulong rnd1 = GenLeadingSignsMinus16(cnt);
-                ulong rnd2 = GenLeadingSignsPlus16(cnt);
-
-                yield return (rnd1 << 48) | (rnd1 << 32) | (rnd1 << 16) | rnd1;
-                yield return (rnd2 << 48) | (rnd2 << 32) | (rnd2 << 16) | rnd2;
-            }
-        }
-
-        private static IEnumerable<ulong> _GenLeadingSigns2S_()
-        {
-            for (int cnt = 0; cnt <= 31; cnt++)
-            {
-                ulong rnd1 = GenLeadingSignsMinus32(cnt);
-                ulong rnd2 = GenLeadingSignsPlus32(cnt);
-
-                yield return (rnd1 << 32) | rnd1;
-                yield return (rnd2 << 32) | rnd2;
-            }
-        }
-
-        private static IEnumerable<ulong> _GenLeadingZeros8B_()
-        {
-            for (int cnt = 0; cnt <= 8; cnt++)
-            {
-                ulong rnd = GenLeadingZeros8(cnt);
-
-                yield return (rnd << 56) | (rnd << 48) | (rnd << 40) | (rnd << 32) |
-                             (rnd << 24) | (rnd << 16) | (rnd << 08) | rnd;
-            }
-        }
-
-        private static IEnumerable<ulong> _GenLeadingZeros4H_()
-        {
-            for (int cnt = 0; cnt <= 16; cnt++)
-            {
-                ulong rnd = GenLeadingZeros16(cnt);
-
-                yield return (rnd << 48) | (rnd << 32) | (rnd << 16) | rnd;
-            }
-        }
-
-        private static IEnumerable<ulong> _GenLeadingZeros2S_()
-        {
-            for (int cnt = 0; cnt <= 32; cnt++)
-            {
-                ulong rnd = GenLeadingZeros32(cnt);
-
-                yield return (rnd << 32) | rnd;
-            }
         }
 
         private static IEnumerable<ulong> _1H_F_()
@@ -619,16 +568,82 @@ namespace Ryujinx.Tests.Cpu
             }
         }
 
-        private static uint[] _W_()
+        private static IEnumerable<ulong> _GenLeadingSigns8B_()
         {
-            return new uint[] { 0x00000000u, 0x7FFFFFFFu,
+            for (int cnt = 0; cnt <= 7; cnt++)
-                                0x80000000u, 0xFFFFFFFFu };
+            {
+                ulong rnd1 = GenLeadingSignsMinus8(cnt);
+                ulong rnd2 = GenLeadingSignsPlus8(cnt);
+
+                yield return (rnd1 << 56) | (rnd1 << 48) | (rnd1 << 40) | (rnd1 << 32) |
+                             (rnd1 << 24) | (rnd1 << 16) | (rnd1 << 08) | rnd1;
+                yield return (rnd2 << 56) | (rnd2 << 48) | (rnd2 << 40) | (rnd2 << 32) |
+                             (rnd2 << 24) | (rnd2 << 16) | (rnd2 << 08) | rnd2;
+            }
         }
 
-        private static ulong[] _X_()
+        private static IEnumerable<ulong> _GenLeadingSigns4H_()
         {
-            return new ulong[] { 0x0000000000000000ul, 0x7FFFFFFFFFFFFFFFul,
+            for (int cnt = 0; cnt <= 15; cnt++)
-                                 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul };
+            {
+                ulong rnd1 = GenLeadingSignsMinus16(cnt);
+                ulong rnd2 = GenLeadingSignsPlus16(cnt);
+
+                yield return (rnd1 << 48) | (rnd1 << 32) | (rnd1 << 16) | rnd1;
+                yield return (rnd2 << 48) | (rnd2 << 32) | (rnd2 << 16) | rnd2;
+            }
+        }
+
+        private static IEnumerable<ulong> _GenLeadingSigns2S_()
+        {
+            for (int cnt = 0; cnt <= 31; cnt++)
+            {
+                ulong rnd1 = GenLeadingSignsMinus32(cnt);
+                ulong rnd2 = GenLeadingSignsPlus32(cnt);
+
+                yield return (rnd1 << 32) | rnd1;
+                yield return (rnd2 << 32) | rnd2;
+            }
+        }
+
+        private static IEnumerable<ulong> _GenLeadingZeros8B_()
+        {
+            for (int cnt = 0; cnt <= 8; cnt++)
+            {
+                ulong rnd = GenLeadingZeros8(cnt);
+
+                yield return (rnd << 56) | (rnd << 48) | (rnd << 40) | (rnd << 32) |
+                             (rnd << 24) | (rnd << 16) | (rnd << 08) | rnd;
+            }
+        }
+
+        private static IEnumerable<ulong> _GenLeadingZeros4H_()
+        {
+            for (int cnt = 0; cnt <= 16; cnt++)
+            {
+                ulong rnd = GenLeadingZeros16(cnt);
+
+                yield return (rnd << 48) | (rnd << 32) | (rnd << 16) | rnd;
+            }
+        }
+
+        private static IEnumerable<ulong> _GenLeadingZeros2S_()
+        {
+            for (int cnt = 0; cnt <= 32; cnt++)
+            {
+                ulong rnd = GenLeadingZeros32(cnt);
+
+                yield return (rnd << 32) | rnd;
+            }
+        }
+
+        private static IEnumerable<ulong> _GenPopCnt8B_()
+        {
+            for (ulong cnt = 0ul; cnt <= 255ul; cnt++)
+            {
+                yield return (cnt << 56) | (cnt << 48) | (cnt << 40) | (cnt << 32) |
+                             (cnt << 24) | (cnt << 16) | (cnt << 08) | cnt;
+            }
         }
 #endregion
 
@@ -1264,7 +1279,7 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CLS <Vd>.<T>, <Vn>.<T>")]
         public void Cls_V_8B_16B([Values(0u)]     uint rd,
                                  [Values(1u, 0u)] uint rn,
-                                 [ValueSource("_GenLeadingSigns8B_")] [Random(RndCnt)] ulong z,
+                                 [ValueSource("_8B_")]                [Random(RndCnt)] ulong z,
                                  [ValueSource("_GenLeadingSigns8B_")] [Random(RndCnt)] ulong a,
                                  [Values(0b0u, 0b1u)] uint q) // <8B, 16B>
         {
@@ -1283,7 +1298,7 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CLS <Vd>.<T>, <Vn>.<T>")]
         public void Cls_V_4H_8H([Values(0u)]     uint rd,
                                 [Values(1u, 0u)] uint rn,
-                                [ValueSource("_GenLeadingSigns4H_")] [Random(RndCnt)] ulong z,
+                                [ValueSource("_4H_")]                [Random(RndCnt)] ulong z,
                                 [ValueSource("_GenLeadingSigns4H_")] [Random(RndCnt)] ulong a,
                                 [Values(0b0u, 0b1u)] uint q) // <4H, 8H>
         {
@@ -1302,7 +1317,7 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CLS <Vd>.<T>, <Vn>.<T>")]
         public void Cls_V_2S_4S([Values(0u)]     uint rd,
                                 [Values(1u, 0u)] uint rn,
-                                [ValueSource("_GenLeadingSigns2S_")] [Random(RndCnt)] ulong z,
+                                [ValueSource("_2S_")]                [Random(RndCnt)] ulong z,
                                 [ValueSource("_GenLeadingSigns2S_")] [Random(RndCnt)] ulong a,
                                 [Values(0b0u, 0b1u)] uint q) // <2S, 4S>
         {
@@ -1321,7 +1336,7 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CLZ <Vd>.<T>, <Vn>.<T>")]
         public void Clz_V_8B_16B([Values(0u)]     uint rd,
                                  [Values(1u, 0u)] uint rn,
-                                 [ValueSource("_GenLeadingZeros8B_")] [Random(RndCnt)] ulong z,
+                                 [ValueSource("_8B_")]                [Random(RndCnt)] ulong z,
                                  [ValueSource("_GenLeadingZeros8B_")] [Random(RndCnt)] ulong a,
                                  [Values(0b0u, 0b1u)] uint q) // <8B, 16B>
         {
@@ -1340,7 +1355,7 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CLZ <Vd>.<T>, <Vn>.<T>")]
         public void Clz_V_4H_8H([Values(0u)]     uint rd,
                                 [Values(1u, 0u)] uint rn,
-                                [ValueSource("_GenLeadingZeros4H_")] [Random(RndCnt)] ulong z,
+                                [ValueSource("_4H_")]                [Random(RndCnt)] ulong z,
                                 [ValueSource("_GenLeadingZeros4H_")] [Random(RndCnt)] ulong a,
                                 [Values(0b0u, 0b1u)] uint q) // <4H, 8H>
         {
@@ -1359,7 +1374,7 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CLZ <Vd>.<T>, <Vn>.<T>")]
         public void Clz_V_2S_4S([Values(0u)]     uint rd,
                                 [Values(1u, 0u)] uint rn,
-                                [ValueSource("_GenLeadingZeros2S_")] [Random(RndCnt)] ulong z,
+                                [ValueSource("_2S_")]                [Random(RndCnt)] ulong z,
                                 [ValueSource("_GenLeadingZeros2S_")] [Random(RndCnt)] ulong a,
                                 [Values(0b0u, 0b1u)] uint q) // <2S, 4S>
         {
@@ -1653,8 +1668,8 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CNT <Vd>.<T>, <Vn>.<T>")]
         public void Cnt_V_8B([Values(0u)]     uint rd,
                              [Values(1u, 0u)] uint rn,
-                             [ValueSource("_8B_")] [Random(RndCnt)] ulong z,
+                             [ValueSource("_8B_")]          [Random(RndCnt)] ulong z,
-                             [ValueSource("_8B_")] [Random(RndCnt)] ulong a)
+                             [ValueSource("_GenPopCnt8B_")] [Random(RndCnt)] ulong a)
         {
             uint opcode = 0x0E205800; // CNT V0.8B, V0.8B
             opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);
@@ -1670,8 +1685,8 @@ namespace Ryujinx.Tests.Cpu
         [Test, Pairwise, Description("CNT <Vd>.<T>, <Vn>.<T>")]
         public void Cnt_V_16B([Values(0u)]     uint rd,
                               [Values(1u, 0u)] uint rn,
-                              [ValueSource("_8B_")] [Random(RndCnt)] ulong z,
+                              [ValueSource("_8B_")]          [Random(RndCnt)] ulong z,
-                              [ValueSource("_8B_")] [Random(RndCnt)] ulong a)
+                              [ValueSource("_GenPopCnt8B_")] [Random(RndCnt)] ulong a)
         {
             uint opcode = 0x4E205800; // CNT V0.16B, V0.16B
             opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);

Ryujinx.Tests/Cpu/CpuTestSimdCvt.cs

@@ -14,6 +14,18 @@ namespace Ryujinx.Tests.Cpu
 #if SimdCvt
 
 #region "ValueSource (Types)"
+        private static uint[] _W_()
+        {
+            return new uint[] { 0x00000000u, 0x7FFFFFFFu,
+                                0x80000000u, 0xFFFFFFFFu };
+        }
+
+        private static ulong[] _X_()
+        {
+            return new ulong[] { 0x0000000000000000ul, 0x7FFFFFFFFFFFFFFFul,
+                                 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul };
+        }
+
         private static IEnumerable<ulong> _1S_F_WX_()
         {
             // int
@@ -177,18 +189,6 @@ namespace Ryujinx.Tests.Cpu
                 yield return rnd6;
             }
         }
-
-        private static uint[] _W_()
-        {
-            return new uint[] { 0x00000000u, 0x7FFFFFFFu,
-                                0x80000000u, 0xFFFFFFFFu };
-        }
-
-        private static ulong[] _X_()
-        {
-            return new ulong[] { 0x0000000000000000ul, 0x7FFFFFFFFFFFFFFFul,
-                                 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul };
-        }
 #endregion
 
 #region "ValueSource (Opcodes)"

Ryujinx.Tests/Cpu/CpuTestSimdIns.cs

@@ -84,10 +84,11 @@ namespace Ryujinx.Tests.Cpu
             opcode |= (imm5 << 16);
             opcode |= ((q & 1) << 30);
 
-            ulong z = TestContext.CurrentContext.Random.NextULong();
+            uint  w31 = TestContext.CurrentContext.Random.NextUInt();
+            ulong z   = TestContext.CurrentContext.Random.NextULong();
             Vector128<float> v0 = MakeVectorE0E1(z, z);
 
-            SingleOpcode(opcode, x1: wn, v0: v0);
+            SingleOpcode(opcode, x1: wn, x31: w31, v0: v0);
 
             CompareAgainstUnicorn();
         }
@@ -100,10 +101,11 @@ namespace Ryujinx.Tests.Cpu
             uint opcode = 0x4E080C00; // DUP V0.2D, X0
             opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);
 
-            ulong z = TestContext.CurrentContext.Random.NextULong();
+            ulong x31 = TestContext.CurrentContext.Random.NextULong();
+            ulong z   = TestContext.CurrentContext.Random.NextULong();
             Vector128<float> v0 = MakeVectorE0E1(z, z);
 
-            SingleOpcode(opcode, x1: xn, v0: v0);
+            SingleOpcode(opcode, x1: xn, x31: x31, v0: v0);
 
             CompareAgainstUnicorn();
         }
@@ -303,9 +305,10 @@ namespace Ryujinx.Tests.Cpu
             opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);
             opcode |= (imm5 << 16);
 
+            uint w31 = TestContext.CurrentContext.Random.NextUInt();
             Vector128<float> v0 = MakeVectorE0E1(z, z);
 
-            SingleOpcode(opcode, x1: wn, v0: v0);
+            SingleOpcode(opcode, x1: wn, x31: w31, v0: v0);
 
             CompareAgainstUnicorn();
         }
@@ -325,9 +328,10 @@ namespace Ryujinx.Tests.Cpu
             opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);
             opcode |= (imm5 << 16);
 
+            uint w31 = TestContext.CurrentContext.Random.NextUInt();
             Vector128<float> v0 = MakeVectorE0E1(z, z);
 
-            SingleOpcode(opcode, x1: wn, v0: v0);
+            SingleOpcode(opcode, x1: wn, x31: w31, v0: v0);
 
             CompareAgainstUnicorn();
         }
@@ -347,9 +351,10 @@ namespace Ryujinx.Tests.Cpu
             opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);
             opcode |= (imm5 << 16);
 
+            uint w31 = TestContext.CurrentContext.Random.NextUInt();
             Vector128<float> v0 = MakeVectorE0E1(z, z);
 
-            SingleOpcode(opcode, x1: wn, v0: v0);
+            SingleOpcode(opcode, x1: wn, x31: w31, v0: v0);
 
             CompareAgainstUnicorn();
         }
@@ -369,9 +374,10 @@ namespace Ryujinx.Tests.Cpu
             opcode |= ((rn & 31) << 5) | ((rd & 31) << 0);
             opcode |= (imm5 << 16);
 
+            ulong x31 = TestContext.CurrentContext.Random.NextULong();
             Vector128<float> v0 = MakeVectorE0E1(z, z);
 
-            SingleOpcode(opcode, x1: xn, v0: v0);
+            SingleOpcode(opcode, x1: xn, x31: x31, v0: v0);
 
             CompareAgainstUnicorn();
         }

Ryujinx.Tests/Ryujinx.Tests.csproj

@@ -27,8 +27,8 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="Microsoft.NET.Test.Sdk" Version="16.0.1" />
+    <PackageReference Include="Microsoft.NET.Test.Sdk" Version="16.2.0" />
-    <PackageReference Include="NUnit" Version="3.11.0" />
+    <PackageReference Include="NUnit" Version="3.12.0" />
     <PackageReference Include="NUnit3TestAdapter" Version="3.13.0" />
     <PackageReference Include="System.Runtime.Intrinsics.Experimental" Version="4.5.0-rc1" />
   </ItemGroup>