From 2ccd995cb27d95d056f8b6f270cb74012bfb7a33 Mon Sep 17 00:00:00 2001 From: LDj3SNuD <35856442+LDj3SNuD@users.noreply.github.com> Date: Fri, 20 Apr 2018 17:40:15 +0200 Subject: [PATCH] Add ADDHN{2}, RADDHN{2}, SUBHN{2}, RSUBHN{2} (vector) instructions. Add 8 Tests. (#92) * Update AOpCodeTable.cs * Update AInstEmitSimdArithmetic.cs * Update Pseudocode.cs * Update Instructions.cs * Update Bits.cs * Create CpuTestSimd.cs * Create CpuTestSimdReg.cs * Update CpuTestSimd.cs Provide a better supply of input values for the 20 Simd Tests. * Update CpuTestSimdReg.cs Provide a better supply of input values for the 20 Simd Tests. * Update AOpCodeTable.cs * Update AInstEmitSimdArithmetic.cs * Update CpuTestSimd.cs * Update CpuTestSimdReg.cs --- ChocolArm64/AOpCodeTable.cs | 4 + .../Instruction/AInstEmitSimdArithmetic.cs | 55 +- Ryujinx.Tests/Cpu/CpuTestSimd.cs | 166 +++++ Ryujinx.Tests/Cpu/CpuTestSimdReg.cs | 432 +++++++++++++ Ryujinx.Tests/Cpu/Tester/Instructions.cs | 577 +++++++++++++++++- Ryujinx.Tests/Cpu/Tester/Pseudocode.cs | 112 +++- Ryujinx.Tests/Cpu/Tester/Types/Bits.cs | 81 ++- 7 files changed, 1410 insertions(+), 17 deletions(-) create mode 100644 Ryujinx.Tests/Cpu/CpuTestSimd.cs create mode 100644 Ryujinx.Tests/Cpu/CpuTestSimdReg.cs diff --git a/ChocolArm64/AOpCodeTable.cs b/ChocolArm64/AOpCodeTable.cs index fb74f8925..154affec9 100644 --- a/ChocolArm64/AOpCodeTable.cs +++ b/ChocolArm64/AOpCodeTable.cs @@ -137,6 +137,7 @@ namespace ChocolArm64 Set("0>001110<<100000101110xxxxxxxxxx", AInstEmit.Abs_V, typeof(AOpCodeSimd)); Set("01011110111xxxxx100001xxxxxxxxxx", AInstEmit.Add_S, typeof(AOpCodeSimdReg)); Set("0>001110<<1xxxxx100001xxxxxxxxxx", AInstEmit.Add_V, typeof(AOpCodeSimdReg)); + Set("0x001110<<1xxxxx010000xxxxxxxxxx", AInstEmit.Addhn_V, typeof(AOpCodeSimdReg)); Set("01011110xx110001101110xxxxxxxxxx", AInstEmit.Addp_S, typeof(AOpCodeSimd)); Set("0>001110<<1xxxxx101111xxxxxxxxxx", AInstEmit.Addp_V, typeof(AOpCodeSimdReg)); Set("000011100x110001101110xxxxxxxxxx", AInstEmit.Addv_V, typeof(AOpCodeSimd)); @@ -290,7 +291,9 @@ namespace ChocolArm64 Set("0x10111000100000010110xxxxxxxxxx", AInstEmit.Not_V, typeof(AOpCodeSimd)); Set("0x001110101xxxxx000111xxxxxxxxxx", AInstEmit.Orr_V, typeof(AOpCodeSimdReg)); Set("0x00111100000xxx<101110<<1xxxxx100001xxxxxxxxxx", AInstEmit.Sub_V, typeof(AOpCodeSimdReg)); + Set("0x001110<<1xxxxx011000xxxxxxxxxx", AInstEmit.Subhn_V, typeof(AOpCodeSimdReg)); Set("0x001110000xxxxx0xx000xxxxxxxxxx", AInstEmit.Tbl_V, typeof(AOpCodeSimdTbl)); Set("0>001110<<0xxxxx001010xxxxxxxxxx", AInstEmit.Trn1_V, typeof(AOpCodeSimdReg)); Set("0>001110<<0xxxxx011010xxxxxxxxxx", AInstEmit.Trn2_V, typeof(AOpCodeSimdReg)); diff --git a/ChocolArm64/Instruction/AInstEmitSimdArithmetic.cs b/ChocolArm64/Instruction/AInstEmitSimdArithmetic.cs index 6b65c1561..2dce74107 100644 --- a/ChocolArm64/Instruction/AInstEmitSimdArithmetic.cs +++ b/ChocolArm64/Instruction/AInstEmitSimdArithmetic.cs @@ -26,7 +26,6 @@ namespace ChocolArm64.Instruction AILLabel LblTrue = new AILLabel(); Context.Emit(OpCodes.Dup); - Context.Emit(OpCodes.Ldc_I4_0); Context.Emit(OpCodes.Bge_S, LblTrue); @@ -45,6 +44,11 @@ namespace ChocolArm64.Instruction EmitVectorBinaryOpZx(Context, () => Context.Emit(OpCodes.Add)); } + public static void Addhn_V(AILEmitterCtx Context) + { + EmitHighNarrow(Context, () => Context.Emit(OpCodes.Add), Round: false); + } + public static void Addp_S(AILEmitterCtx Context) { AOpCodeSimd Op = (AOpCodeSimd)Context.CurrOp; @@ -130,6 +134,40 @@ namespace ChocolArm64.Instruction } } + private static void EmitHighNarrow(AILEmitterCtx Context, Action Emit, bool Round) + { + AOpCodeSimdReg Op = (AOpCodeSimdReg)Context.CurrOp; + + int Elems = 8 >> Op.Size; + int ESize = 8 << Op.Size; + + int Part = Op.RegisterSize == ARegisterSize.SIMD128 ? Elems : 0; + + for (int Index = 0; Index < Elems; Index++) + { + EmitVectorExtractZx(Context, Op.Rn, Index, Op.Size + 1); + EmitVectorExtractZx(Context, Op.Rm, Index, Op.Size + 1); + + Emit(); + + if (Round) + { + Context.EmitLdc_I8(1L << (ESize - 1)); + + Context.Emit(OpCodes.Add); + } + + Context.EmitLsr(ESize); + + EmitVectorInsert(Context, Op.Rd, Part + Index, Op.Size); + } + + if (Part == 0) + { + EmitVectorZeroUpper(Context, Op.Rd); + } + } + public static void Fabd_S(AILEmitterCtx Context) { EmitScalarBinaryOpF(Context, () => @@ -849,6 +887,16 @@ namespace ChocolArm64.Instruction EmitVectorUnaryOpSx(Context, () => Context.Emit(OpCodes.Neg)); } + public static void Raddhn_V(AILEmitterCtx Context) + { + EmitHighNarrow(Context, () => Context.Emit(OpCodes.Add), Round: true); + } + + public static void Rsubhn_V(AILEmitterCtx Context) + { + EmitHighNarrow(Context, () => Context.Emit(OpCodes.Sub), Round: true); + } + public static void Saddw_V(AILEmitterCtx Context) { EmitVectorWidenRmBinaryOpSx(Context, () => Context.Emit(OpCodes.Add)); @@ -896,6 +944,11 @@ namespace ChocolArm64.Instruction EmitVectorBinaryOpZx(Context, () => Context.Emit(OpCodes.Sub)); } + public static void Subhn_V(AILEmitterCtx Context) + { + EmitHighNarrow(Context, () => Context.Emit(OpCodes.Sub), Round: false); + } + public static void Uabd_V(AILEmitterCtx Context) { EmitVectorBinaryOpZx(Context, () => EmitAbd(Context)); diff --git a/Ryujinx.Tests/Cpu/CpuTestSimd.cs b/Ryujinx.Tests/Cpu/CpuTestSimd.cs new file mode 100644 index 000000000..9107a8f89 --- /dev/null +++ b/Ryujinx.Tests/Cpu/CpuTestSimd.cs @@ -0,0 +1,166 @@ +#define Simd + +using ChocolArm64.State; + +using NUnit.Framework; + +namespace Ryujinx.Tests.Cpu +{ + using Tester; + using Tester.Types; + + [Category("Simd")] + public sealed class CpuTestSimd : CpuTest + { +#if Simd + [SetUp] + public void SetupTester() + { + AArch64.TakeReset(false); + } + +#region "ValueSource" + private static ulong[] _D_() + { + return new ulong[] { 0x0000000000000000ul, 0x7FFFFFFFFFFFFFFFul, + 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul }; + } + + private static ulong[] _8B4H2S_() + { + return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, + 0x8080808080808080ul, 0x7FFF7FFF7FFF7FFFul, + 0x8000800080008000ul, 0x7FFFFFFF7FFFFFFFul, + 0x8000000080000000ul, 0xFFFFFFFFFFFFFFFFul }; + } + + private static ulong[] _16B8H4S2D_() + { + return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, + 0x8080808080808080ul, 0x7FFF7FFF7FFF7FFFul, + 0x8000800080008000ul, 0x7FFFFFFF7FFFFFFFul, + 0x8000000080000000ul, 0x7FFFFFFFFFFFFFFFul, + 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul }; + } +#endregion + + [Test, Description("ABS , ")] + public void Abs_S_D([ValueSource("_D_")] [Random(1)] ulong A) + { + uint Opcode = 0x5EE0B820; // ABS D0, D1 + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1); + + AArch64.V(1, new Bits(A)); + SimdFp.Abs_S(Op[23, 22], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Description("ABS ., .")] + public void Abs_V_8B_4H_2S([ValueSource("_8B4H2S_")] [Random(1)] ulong A, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8B, 4H, 2S> + { + uint Opcode = 0x0E20B820; // ABS V0.8B, V1.8B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1); + + AArch64.V(1, new Bits(A)); + SimdFp.Abs_V(Op[30], Op[23, 22], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Pairwise, Description("ABS ., .")] + public void Abs_V_16B_8H_4S_2D([ValueSource("_16B8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong A1, + [Values(0b00u, 0b01u, 0b10u, 0b11u)] uint size) // <16B, 8H, 4S, 2D> + { + uint Opcode = 0x4E20B820; // ABS V0.16B, V1.16B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V1: V1); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + SimdFp.Abs_V(Op[30], Op[23, 22], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.Vpart(64, 0, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } + + [Test, Description("NEG , ")] + public void Neg_S_D([ValueSource("_D_")] [Random(1)] ulong A) + { + uint Opcode = 0x7EE0B820; // NEG D0, D1 + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1); + + AArch64.V(1, new Bits(A)); + SimdFp.Neg_S(Op[23, 22], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Description("NEG ., .")] + public void Neg_V_8B_4H_2S([ValueSource("_8B4H2S_")] [Random(1)] ulong A, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8B, 4H, 2S> + { + uint Opcode = 0x2E20B820; // NEG V0.8B, V1.8B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1); + + AArch64.V(1, new Bits(A)); + SimdFp.Neg_V(Op[30], Op[23, 22], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Pairwise, Description("NEG ., .")] + public void Neg_V_16B_8H_4S_2D([ValueSource("_16B8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong A1, + [Values(0b00u, 0b01u, 0b10u, 0b11u)] uint size) // <16B, 8H, 4S, 2D> + { + uint Opcode = 0x6E20B820; // NEG V0.16B, V1.16B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V1: V1); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + SimdFp.Neg_V(Op[30], Op[23, 22], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.Vpart(64, 0, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } +#endif + } +} diff --git a/Ryujinx.Tests/Cpu/CpuTestSimdReg.cs b/Ryujinx.Tests/Cpu/CpuTestSimdReg.cs new file mode 100644 index 000000000..04cef60fb --- /dev/null +++ b/Ryujinx.Tests/Cpu/CpuTestSimdReg.cs @@ -0,0 +1,432 @@ +#define SimdReg + +using ChocolArm64.State; + +using NUnit.Framework; + +namespace Ryujinx.Tests.Cpu +{ + using Tester; + using Tester.Types; + + [Category("SimdReg")] + public sealed class CpuTestSimdReg : CpuTest + { +#if SimdReg + [SetUp] + public void SetupTester() + { + AArch64.TakeReset(false); + } + +#region "ValueSource" + private static ulong[] _D_() + { + return new ulong[] { 0x0000000000000000ul, 0x7FFFFFFFFFFFFFFFul, + 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul }; + } + + private static ulong[] _8B4H2S_() + { + return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, + 0x8080808080808080ul, 0x7FFF7FFF7FFF7FFFul, + 0x8000800080008000ul, 0x7FFFFFFF7FFFFFFFul, + 0x8000000080000000ul, 0xFFFFFFFFFFFFFFFFul }; + } + + private static ulong[] _16B8H4S2D_() + { + return new ulong[] { 0x0000000000000000ul, 0x7F7F7F7F7F7F7F7Ful, + 0x8080808080808080ul, 0x7FFF7FFF7FFF7FFFul, + 0x8000800080008000ul, 0x7FFFFFFF7FFFFFFFul, + 0x8000000080000000ul, 0x7FFFFFFFFFFFFFFFul, + 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul }; + } + + private static ulong[] _8H4S2D_() + { + return new ulong[] { 0x0000000000000000ul, 0x7FFF7FFF7FFF7FFFul, + 0x8000800080008000ul, 0x7FFFFFFF7FFFFFFFul, + 0x8000000080000000ul, 0x7FFFFFFFFFFFFFFFul, + 0x8000000000000000ul, 0xFFFFFFFFFFFFFFFFul }; + } +#endregion + + [Test, Description("ADD , , ")] + public void Add_S_D([ValueSource("_D_")] [Random(1)] ulong A, + [ValueSource("_D_")] [Random(1)] ulong B) + { + uint Opcode = 0x5EE28420; // ADD D0, D1, D2 + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AVec V2 = new AVec { X0 = B }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.V(1, new Bits(A)); + AArch64.V(2, new Bits(B)); + SimdFp.Add_S(Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Description("ADD ., ., .")] + public void Add_V_8B_4H_2S([ValueSource("_8B4H2S_")] [Random(1)] ulong A, + [ValueSource("_8B4H2S_")] [Random(1)] ulong B, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8B, 4H, 2S> + { + uint Opcode = 0x0E228420; // ADD V0.8B, V1.8B, V2.8B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AVec V2 = new AVec { X0 = B }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.V(1, new Bits(A)); + AArch64.V(2, new Bits(B)); + SimdFp.Add_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Pairwise, Description("ADD ., ., .")] + public void Add_V_16B_8H_4S_2D([ValueSource("_16B8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u, 0b11u)] uint size) // <16B, 8H, 4S, 2D> + { + uint Opcode = 0x4E228420; // ADD V0.16B, V1.16B, V2.16B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Add_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.Vpart(64, 0, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } + + [Test, Pairwise, Description("ADDHN{2} ., ., .")] + public void Addhn_V_8H8B_4S4H_2D2S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H8B, 4S4H, 2D2S> + { + uint Opcode = 0x0E224020; // ADDHN V0.8B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Addhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + }); + } + + [Test, Pairwise, Description("ADDHN{2} ., ., .")] + public void Addhn_V_8H16B_4S8H_2D4S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H16B, 4S8H, 2D4S> + { + uint Opcode = 0x4E224020; // ADDHN2 V0.16B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + ulong _X0 = TestContext.CurrentContext.Random.NextULong(); + AVec V0 = new AVec { X0 = _X0 }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Addhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(_X0)); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } + + [Test, Pairwise, Description("RADDHN{2} ., ., .")] + public void Raddhn_V_8H8B_4S4H_2D2S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H8B, 4S4H, 2D2S> + { + uint Opcode = 0x2E224020; // RADDHN V0.8B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Raddhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + }); + } + + [Test, Pairwise, Description("RADDHN{2} ., ., .")] + public void Raddhn_V_8H16B_4S8H_2D4S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H16B, 4S8H, 2D4S> + { + uint Opcode = 0x6E224020; // RADDHN2 V0.16B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + ulong _X0 = TestContext.CurrentContext.Random.NextULong(); + AVec V0 = new AVec { X0 = _X0 }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Raddhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(_X0)); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } + + [Test, Pairwise, Description("RSUBHN{2} ., ., .")] + public void Rsubhn_V_8H8B_4S4H_2D2S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H8B, 4S4H, 2D2S> + { + uint Opcode = 0x2E226020; // RSUBHN V0.8B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Rsubhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + }); + } + + [Test, Pairwise, Description("RSUBHN{2} ., ., .")] + public void Rsubhn_V_8H16B_4S8H_2D4S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H16B, 4S8H, 2D4S> + { + uint Opcode = 0x6E226020; // RSUBHN2 V0.16B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + ulong _X0 = TestContext.CurrentContext.Random.NextULong(); + AVec V0 = new AVec { X0 = _X0 }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Rsubhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(_X0)); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } + + [Test, Description("SUB , , ")] + public void Sub_S_D([ValueSource("_D_")] [Random(1)] ulong A, + [ValueSource("_D_")] [Random(1)] ulong B) + { + uint Opcode = 0x7EE28420; // SUB D0, D1, D2 + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AVec V2 = new AVec { X0 = B }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.V(1, new Bits(A)); + AArch64.V(2, new Bits(B)); + SimdFp.Sub_S(Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Description("SUB ., ., .")] + public void Sub_V_8B_4H_2S([ValueSource("_8B4H2S_")] [Random(1)] ulong A, + [ValueSource("_8B4H2S_")] [Random(1)] ulong B, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8B, 4H, 2S> + { + uint Opcode = 0x2E228420; // SUB V0.8B, V1.8B, V2.8B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A }; + AVec V2 = new AVec { X0 = B }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.V(1, new Bits(A)); + AArch64.V(2, new Bits(B)); + SimdFp.Sub_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + } + + [Test, Pairwise, Description("SUB ., ., .")] + public void Sub_V_16B_8H_4S_2D([ValueSource("_16B8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_16B8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u, 0b11u)] uint size) // <16B, 8H, 4S, 2D> + { + uint Opcode = 0x6E228420; // SUB V0.16B, V1.16B, V2.16B + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Sub_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.Vpart(64, 0, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } + + [Test, Pairwise, Description("SUBHN{2} ., ., .")] + public void Subhn_V_8H8B_4S4H_2D2S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H8B, 4S4H, 2D2S> + { + uint Opcode = 0x0E226020; // SUBHN V0.8B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + AVec V0 = new AVec { X1 = TestContext.CurrentContext.Random.NextULong() }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Subhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(AArch64.V(64, 0).ToUInt64())); + Assert.That(ThreadState.V0.X1, Is.Zero); + }); + } + + [Test, Pairwise, Description("SUBHN{2} ., ., .")] + public void Subhn_V_8H16B_4S8H_2D4S([ValueSource("_8H4S2D_")] [Random(1)] ulong A0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong A1, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B0, + [ValueSource("_8H4S2D_")] [Random(1)] ulong B1, + [Values(0b00u, 0b01u, 0b10u)] uint size) // <8H16B, 4S8H, 2D4S> + { + uint Opcode = 0x4E226020; // SUBHN2 V0.16B, V1.8H, V2.8H + Opcode |= ((size & 3) << 22); + Bits Op = new Bits(Opcode); + + ulong _X0 = TestContext.CurrentContext.Random.NextULong(); + AVec V0 = new AVec { X0 = _X0 }; + AVec V1 = new AVec { X0 = A0, X1 = A1 }; + AVec V2 = new AVec { X0 = B0, X1 = B1 }; + AThreadState ThreadState = SingleOpcode(Opcode, V0: V0, V1: V1, V2: V2); + + AArch64.Vpart(1, 0, new Bits(A0)); + AArch64.Vpart(1, 1, new Bits(A1)); + AArch64.Vpart(2, 0, new Bits(B0)); + AArch64.Vpart(2, 1, new Bits(B1)); + SimdFp.Subhn_V(Op[30], Op[23, 22], Op[20, 16], Op[9, 5], Op[4, 0]); + + Assert.Multiple(() => + { + Assert.That(ThreadState.V0.X0, Is.EqualTo(_X0)); + Assert.That(ThreadState.V0.X1, Is.EqualTo(AArch64.Vpart(64, 0, 1).ToUInt64())); + }); + } +#endif + } +} diff --git a/Ryujinx.Tests/Cpu/Tester/Instructions.cs b/Ryujinx.Tests/Cpu/Tester/Instructions.cs index 7a51923f8..2ca76ffdc 100644 --- a/Ryujinx.Tests/Cpu/Tester/Instructions.cs +++ b/Ryujinx.Tests/Cpu/Tester/Instructions.cs @@ -1,7 +1,7 @@ // https://github.com/LDj3SNuD/ARM_v8-A_AArch64_Instructions_Tester/blob/master/Tester/Instructions.cs // https://meriac.github.io/archex/A64_v83A_ISA/index.xml -/* https://meriac.github.io/archex/A64_v83A_ISA/fpsimdindex.xml */ +// https://meriac.github.io/archex/A64_v83A_ISA/fpsimdindex.xml using System.Numerics; @@ -1674,9 +1674,578 @@ namespace Ryujinx.Tests.Cpu.Tester #endregion } - /* - internal static class Advanced + internal static class SimdFp { +#region "Simd" + // https://meriac.github.io/archex/A64_v83A_ISA/abs_advsimd.xml#ABS_asisdmisc_R + public static void Abs_S(Bits size, Bits Rn, Bits Rd) + { + bool U = false; + + /* Decode Scalar */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + + /* if size != '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = esize; + int elements = 1; + + bool neg = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand = V(datasize, n); + BigInteger element; + + for (int e = 0; e <= elements - 1; e++) + { + element = SInt(Elem(operand, e, esize)); + + if (neg) + { + element = -element; + } + else + { + element = Abs(element); + } + + Elem(result, e, esize, element.SubBigInteger(esize - 1, 0)); + } + + V(d, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/abs_advsimd.xml#ABS_asimdmisc_R + public static void Abs_V(bool Q, Bits size, Bits Rn, Bits Rd) + { + bool U = false; + + /* Decode Vector */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + + /* if size:Q == '110' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = (Q ? 128 : 64); + int elements = datasize / esize; + + bool neg = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand = V(datasize, n); + BigInteger element; + + for (int e = 0; e <= elements - 1; e++) + { + element = SInt(Elem(operand, e, esize)); + + if (neg) + { + element = -element; + } + else + { + element = Abs(element); + } + + Elem(result, e, esize, element.SubBigInteger(esize - 1, 0)); + } + + V(d, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/neg_advsimd.xml#NEG_asisdmisc_R + public static void Neg_S(Bits size, Bits Rn, Bits Rd) + { + bool U = true; + + /* Decode Scalar */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + + /* if size != '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = esize; + int elements = 1; + + bool neg = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand = V(datasize, n); + BigInteger element; + + for (int e = 0; e <= elements - 1; e++) + { + element = SInt(Elem(operand, e, esize)); + + if (neg) + { + element = -element; + } + else + { + element = Abs(element); + } + + Elem(result, e, esize, element.SubBigInteger(esize - 1, 0)); + } + + V(d, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/neg_advsimd.xml#NEG_asimdmisc_R + public static void Neg_V(bool Q, Bits size, Bits Rn, Bits Rd) + { + bool U = true; + + /* Decode Vector */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + + /* if size:Q == '110' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = (Q ? 128 : 64); + int elements = datasize / esize; + + bool neg = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand = V(datasize, n); + BigInteger element; + + for (int e = 0; e <= elements - 1; e++) + { + element = SInt(Elem(operand, e, esize)); + + if (neg) + { + element = -element; + } + else + { + element = Abs(element); + } + + Elem(result, e, esize, element.SubBigInteger(esize - 1, 0)); + } + + V(d, result); + } +#endregion + +#region "SimdReg" + // https://meriac.github.io/archex/A64_v83A_ISA/add_advsimd.xml#ADD_asisdsame_only + public static void Add_S(Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = false; + + /* Decode Scalar */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size != '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = esize; + int elements = 1; + + bool sub_op = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(datasize, n); + Bits operand2 = V(datasize, m); + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, esize); + element2 = Elem(operand2, e, esize); + + if (sub_op) + { + Elem(result, e, esize, element1 - element2); + } + else + { + Elem(result, e, esize, element1 + element2); + } + } + + V(d, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/add_advsimd.xml#ADD_asimdsame_only + public static void Add_V(bool Q, Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = false; + + /* Decode Vector */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size:Q == '110' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = (Q ? 128 : 64); + int elements = datasize / esize; + + bool sub_op = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(datasize, n); + Bits operand2 = V(datasize, m); + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, esize); + element2 = Elem(operand2, e, esize); + + if (sub_op) + { + Elem(result, e, esize, element1 - element2); + } + else + { + Elem(result, e, esize, element1 + element2); + } + } + + V(d, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/addhn_advsimd.xml + public static void Addhn_V(bool Q, Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = false; + bool o1 = false; + + /* Decode */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size == '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = 64; + int part = (int)UInt(Q); + int elements = datasize / esize; + + bool sub_op = (o1 == true); + bool round = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(2 * datasize, n); + Bits operand2 = V(2 * datasize, m); + BigInteger round_const = (round ? (BigInteger)1 << (esize - 1) : 0); + Bits sum; + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, 2 * esize); + element2 = Elem(operand2, e, 2 * esize); + + if (sub_op) + { + sum = element1 - element2; + } + else + { + sum = element1 + element2; + } + + sum = sum + round_const; + + Elem(result, e, esize, sum[2 * esize - 1, esize]); + } + + Vpart(d, part, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/raddhn_advsimd.xml + public static void Raddhn_V(bool Q, Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = true; + bool o1 = false; + + /* Decode */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size == '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = 64; + int part = (int)UInt(Q); + int elements = datasize / esize; + + bool sub_op = (o1 == true); + bool round = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(2 * datasize, n); + Bits operand2 = V(2 * datasize, m); + BigInteger round_const = (round ? (BigInteger)1 << (esize - 1) : 0); + Bits sum; + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, 2 * esize); + element2 = Elem(operand2, e, 2 * esize); + + if (sub_op) + { + sum = element1 - element2; + } + else + { + sum = element1 + element2; + } + + sum = sum + round_const; + + Elem(result, e, esize, sum[2 * esize - 1, esize]); + } + + Vpart(d, part, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/rsubhn_advsimd.xml + public static void Rsubhn_V(bool Q, Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = true; + bool o1 = true; + + /* Decode */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size == '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = 64; + int part = (int)UInt(Q); + int elements = datasize / esize; + + bool sub_op = (o1 == true); + bool round = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(2 * datasize, n); + Bits operand2 = V(2 * datasize, m); + BigInteger round_const = (round ? (BigInteger)1 << (esize - 1) : 0); + Bits sum; + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, 2 * esize); + element2 = Elem(operand2, e, 2 * esize); + + if (sub_op) + { + sum = element1 - element2; + } + else + { + sum = element1 + element2; + } + + sum = sum + round_const; + + Elem(result, e, esize, sum[2 * esize - 1, esize]); + } + + Vpart(d, part, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/sub_advsimd.xml#SUB_asisdsame_only + public static void Sub_S(Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = true; + + /* Decode Scalar */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size != '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = esize; + int elements = 1; + + bool sub_op = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(datasize, n); + Bits operand2 = V(datasize, m); + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, esize); + element2 = Elem(operand2, e, esize); + + if (sub_op) + { + Elem(result, e, esize, element1 - element2); + } + else + { + Elem(result, e, esize, element1 + element2); + } + } + + V(d, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/sub_advsimd.xml#SUB_asimdsame_only + public static void Sub_V(bool Q, Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = true; + + /* Decode Vector */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size:Q == '110' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = (Q ? 128 : 64); + int elements = datasize / esize; + + bool sub_op = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(datasize, n); + Bits operand2 = V(datasize, m); + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, esize); + element2 = Elem(operand2, e, esize); + + if (sub_op) + { + Elem(result, e, esize, element1 - element2); + } + else + { + Elem(result, e, esize, element1 + element2); + } + } + + V(d, result); + } + + // https://meriac.github.io/archex/A64_v83A_ISA/subhn_advsimd.xml + public static void Subhn_V(bool Q, Bits size, Bits Rm, Bits Rn, Bits Rd) + { + bool U = false; + bool o1 = true; + + /* Decode */ + int d = (int)UInt(Rd); + int n = (int)UInt(Rn); + int m = (int)UInt(Rm); + + /* if size == '11' then ReservedValue(); */ + + int esize = 8 << (int)UInt(size); + int datasize = 64; + int part = (int)UInt(Q); + int elements = datasize / esize; + + bool sub_op = (o1 == true); + bool round = (U == true); + + /* Operation */ + /* CheckFPAdvSIMDEnabled64(); */ + + Bits result = new Bits(datasize); + Bits operand1 = V(2 * datasize, n); + Bits operand2 = V(2 * datasize, m); + BigInteger round_const = (round ? (BigInteger)1 << (esize - 1) : 0); + Bits sum; + Bits element1; + Bits element2; + + for (int e = 0; e <= elements - 1; e++) + { + element1 = Elem(operand1, e, 2 * esize); + element2 = Elem(operand2, e, 2 * esize); + + if (sub_op) + { + sum = element1 - element2; + } + else + { + sum = element1 + element2; + } + + sum = sum + round_const; + + Elem(result, e, esize, sum[2 * esize - 1, esize]); + } + + Vpart(d, part, result); + } +#endregion } - */ } diff --git a/Ryujinx.Tests/Cpu/Tester/Pseudocode.cs b/Ryujinx.Tests/Cpu/Tester/Pseudocode.cs index 1da17c497..21c3935ca 100644 --- a/Ryujinx.Tests/Cpu/Tester/Pseudocode.cs +++ b/Ryujinx.Tests/Cpu/Tester/Pseudocode.cs @@ -70,6 +70,7 @@ namespace Ryujinx.Tests.Cpu.Tester // ELR_ELx and SPSR_ELx have UNKNOWN values, so that it // is impossible to return from a reset in an architecturally defined way. AArch64.ResetGeneralRegisters(); + AArch64.ResetSIMDFPRegisters(); AArch64.ResetSpecialRegisters(); } #endregion @@ -85,6 +86,16 @@ namespace Ryujinx.Tests.Cpu.Tester } } + /* #AArch64.ResetSIMDFPRegisters.0 */ + public static void ResetSIMDFPRegisters() + { + for (int i = 0; i <= 31; i++) + { + /* V[i] = bits(128) UNKNOWN; */ + _V[i].SetAll(false); + } + } + /* #AArch64.ResetSpecialRegisters.0 */ public static void ResetSpecialRegisters() { @@ -116,8 +127,8 @@ namespace Ryujinx.Tests.Cpu.Tester default: case Bits bits when bits == EL1: SP_EL1 = ZeroExtend(64, value); - break; - /*case Bits bits when bits == EL2: + break;/* + case Bits bits when bits == EL2: SP_EL2 = ZeroExtend(64, value); break; case Bits bits when bits == EL3: @@ -144,8 +155,8 @@ namespace Ryujinx.Tests.Cpu.Tester return SP_EL0[width - 1, 0]; default: case Bits bits when bits == EL1: - return SP_EL1[width - 1, 0]; - /*case Bits bits when bits == EL2: + return SP_EL1[width - 1, 0];/* + case Bits bits when bits == EL2: return SP_EL2[width - 1, 0]; case Bits bits when bits == EL3: return SP_EL3[width - 1, 0];*/ @@ -153,6 +164,64 @@ namespace Ryujinx.Tests.Cpu.Tester } } + // #impl-aarch64.V.write.1 + public static void V(int n, Bits value) + { + /* int width = value.Count; */ + + /* assert n >= 0 && n <= 31; */ + /* assert width IN {8,16,32,64,128}; */ + + _V[n] = ZeroExtend(128, value); + } + + /* #impl-aarch64.V.read.1 */ + public static Bits V(int width, int n) + { + /* assert n >= 0 && n <= 31; */ + /* assert width IN {8,16,32,64,128}; */ + + return _V[n][width - 1, 0]; + } + + /* #impl-aarch64.Vpart.read.2 */ + public static Bits Vpart(int width, int n, int part) + { + /* assert n >= 0 && n <= 31; */ + /* assert part IN {0, 1}; */ + + if (part == 0) + { + /* assert width IN {8,16,32,64}; */ + return _V[n][width - 1, 0]; + } + else + { + /* assert width == 64; */ + return _V[n][(width * 2) - 1, width]; + } + } + + // #impl-aarch64.Vpart.write.2 + public static void Vpart(int n, int part, Bits value) + { + int width = value.Count; + + /* assert n >= 0 && n <= 31; */ + /* assert part IN {0, 1}; */ + + if (part == 0) + { + /* assert width IN {8,16,32,64}; */ + _V[n] = ZeroExtend(128, value); + } + else + { + /* assert width == 64; */ + _V[n][(width * 2) - 1, width] = value[width - 1, 0]; + } + } + // #impl-aarch64.X.write.1 public static void X(int n, Bits value) { @@ -214,6 +283,7 @@ namespace Ryujinx.Tests.Cpu.Tester public static Bits ExtendReg(int N, int reg, ExtendType type, int shift) { /* assert shift >= 0 && shift <= 4; */ + Bits val = X(N, reg); bool unsigned; int len; @@ -384,6 +454,12 @@ namespace Ryujinx.Tests.Cpu.Tester _R[i] = new Bits(64, false); } + _V = new Bits[32]; + for (int i = 0; i <= 31; i++) + { + _V[i] = new Bits(128, false); + } + SP_EL0 = new Bits(64, false); SP_EL1 = new Bits(64, false); @@ -437,6 +513,12 @@ namespace Ryujinx.Tests.Cpu.Tester return (result, carry_out); } + // #impl-shared.Abs.1 + public static BigInteger Abs(BigInteger x) + { + return (x >= 0 ? x : -x); + } + // #impl-shared.CountLeadingSignBits.1 public static int CountLeadingSignBits(Bits x) { @@ -453,6 +535,26 @@ namespace Ryujinx.Tests.Cpu.Tester return (N - 1 - HighestSetBit(x)); } + // #impl-shared.Elem.read.3 + public static Bits Elem(/*in */Bits vector, int e, int size) + { + /* int N = vector.Count; */ + + /* assert e >= 0 && (e+1)*size <= N; */ + + return vector[e * size + size - 1, e * size]; + } + + // #impl-shared.Elem.write.3 + public static void Elem(/*out */Bits vector, int e, int size, Bits value) + { + /* int N = vector.Count; */ + + /* assert e >= 0 && (e+1)*size <= N; */ + + vector[(e + 1) * size - 1, e * size] = value; + } + /* */ public static Bits EOR(Bits x, Bits y) { @@ -855,6 +957,8 @@ namespace Ryujinx.Tests.Cpu.Tester #region "functions/registers/" public static readonly Bits[] _R; + public static readonly Bits[] _V; + public static Bits SP_EL0; public static Bits SP_EL1; #endregion diff --git a/Ryujinx.Tests/Cpu/Tester/Types/Bits.cs b/Ryujinx.Tests/Cpu/Tester/Types/Bits.cs index f4ee966cf..30d632640 100644 --- a/Ryujinx.Tests/Cpu/Tester/Types/Bits.cs +++ b/Ryujinx.Tests/Cpu/Tester/Types/Bits.cs @@ -15,14 +15,15 @@ namespace Ryujinx.Tests.Cpu.Tester.Types public Bits(bool[] values) => bits = new BitArray(values); public Bits(byte[] bytes) => bits = new BitArray(bytes); public Bits(Bits bits) => this.bits = new BitArray(bits.bits); - private Bits(BitArray bitArray) => bits = new BitArray(bitArray); public Bits(int length) => bits = new BitArray(length); public Bits(int length, bool defaultValue) => bits = new BitArray(length, defaultValue); + private Bits(BitArray bitArray) => bits = new BitArray(bitArray); public Bits(ulong value) => bits = new BitArray(BitConverter.GetBytes(value)); public Bits(uint value) => bits = new BitArray(BitConverter.GetBytes(value)); public Bits(ushort value) => bits = new BitArray(BitConverter.GetBytes(value)); public Bits(byte value) => bits = new BitArray(new byte[1] {value}); + private BitArray ToBitArray() => new BitArray(bits); public ulong ToUInt64() { byte[] dst = new byte[8]; @@ -39,7 +40,22 @@ namespace Ryujinx.Tests.Cpu.Tester.Types return BitConverter.ToUInt32(dst, 0); } - private BitArray ToBitArray() => new BitArray(bits); + public ushort ToUInt16() + { + byte[] dst = new byte[2]; + + bits.CopyTo(dst, 0); + + return BitConverter.ToUInt16(dst, 0); + } + public byte ToByte() + { + byte[] dst = new byte[1]; + + bits.CopyTo(dst, 0); + + return dst[0]; + } public bool this[int index] // ASL: "<>". { @@ -166,17 +182,66 @@ namespace Ryujinx.Tests.Cpu.Tester.Types BigInteger dst; - if (left.Count <= 32) + switch (left.Count) { - dst = left.ToUInt32() + right; + case 8: dst = left.ToByte() + right; break; + case 16: dst = left.ToUInt16() + right; break; + case 32: dst = left.ToUInt32() + right; break; + case 64: dst = left.ToUInt64() + right; break; + + default: throw new ArgumentOutOfRangeException(); } - else if (left.Count <= 64) + + return dst.SubBigInteger(left.Count - 1, 0); + } + public static Bits operator +(Bits left, Bits right) // ASL: "+". + { + if (((object)left == null) || ((object)right == null)) { - dst = left.ToUInt64() + right; + throw new ArgumentNullException(); } - else + + if (left.Count != right.Count) { - throw new ArgumentOutOfRangeException(); + throw new ArgumentException(); + } + + BigInteger dst; + + switch (left.Count) + { + case 8: dst = left.ToByte() + (BigInteger)right.ToByte(); break; + case 16: dst = left.ToUInt16() + (BigInteger)right.ToUInt16(); break; + case 32: dst = left.ToUInt32() + (BigInteger)right.ToUInt32(); break; + case 64: dst = left.ToUInt64() + (BigInteger)right.ToUInt64(); break; + + default: throw new ArgumentOutOfRangeException(); + } + + return dst.SubBigInteger(left.Count - 1, 0); + } + public static Bits operator -(Bits left, Bits right) // ASL: "-". + { + if (((object)left == null) || ((object)right == null)) + { + throw new ArgumentNullException(); + } + + if (left.Count != right.Count) + { + throw new ArgumentException(); + } + + BigInteger dst; + + switch (left.Count) + { + case 8: dst = left.ToByte() - (BigInteger)right.ToByte(); break; + case 16: dst = left.ToUInt16() - (BigInteger)right.ToUInt16(); break; + case 32: dst = left.ToUInt32() - (BigInteger)right.ToUInt32(); break; + case 64: dst = left.ToUInt64() - (BigInteger)right.ToUInt64(); break; + + default: throw new ArgumentOutOfRangeException(); } return dst.SubBigInteger(left.Count - 1, 0);