diff --git a/Ryujinx.Graphics.Shader/Decoders/Block.cs b/Ryujinx.Graphics.Shader/Decoders/Block.cs index b5e610d71..238b0bd6e 100644 --- a/Ryujinx.Graphics.Shader/Decoders/Block.cs +++ b/Ryujinx.Graphics.Shader/Decoders/Block.cs @@ -11,15 +11,17 @@ namespace Ryujinx.Graphics.Shader.Decoders public Block Next { get; set; } public Block Branch { get; set; } - public List OpCodes { get; } - public List SsyOpCodes { get; } + public OpCodeBranchIndir BrIndir { get; set; } + + public List OpCodes { get; } + public List PushOpCodes { get; } public Block(ulong address) { Address = address; - OpCodes = new List(); - SsyOpCodes = new List(); + OpCodes = new List(); + PushOpCodes = new List(); } public void Split(Block rightBlock) @@ -45,7 +47,7 @@ namespace Ryujinx.Graphics.Shader.Decoders rightBlock.OpCodes.AddRange(OpCodes.GetRange(splitIndex, splitCount)); - rightBlock.UpdateSsyOpCodes(); + rightBlock.UpdatePushOps(); EndAddress = rightBlock.Address; @@ -54,7 +56,7 @@ namespace Ryujinx.Graphics.Shader.Decoders OpCodes.RemoveRange(splitIndex, splitCount); - UpdateSsyOpCodes(); + UpdatePushOps(); } private static int BinarySearch(List opCodes, ulong address) @@ -99,18 +101,18 @@ namespace Ryujinx.Graphics.Shader.Decoders return null; } - public void UpdateSsyOpCodes() + public void UpdatePushOps() { - SsyOpCodes.Clear(); + PushOpCodes.Clear(); for (int index = 0; index < OpCodes.Count; index++) { - if (!(OpCodes[index] is OpCodeSsy op)) + if (!(OpCodes[index] is OpCodePush op)) { continue; } - SsyOpCodes.Add(op); + PushOpCodes.Add(op); } } } diff --git a/Ryujinx.Graphics.Shader/Decoders/Decoder.cs b/Ryujinx.Graphics.Shader/Decoders/Decoder.cs index 4078440b1..6841c98d3 100644 --- a/Ryujinx.Graphics.Shader/Decoders/Decoder.cs +++ b/Ryujinx.Graphics.Shader/Decoders/Decoder.cs @@ -43,9 +43,7 @@ namespace Ryujinx.Graphics.Shader.Decoders return block; } - ulong startAddress = headerSize; - - GetBlock(startAddress); + GetBlock(0); while (workQueue.TryDequeue(out Block currBlock)) { @@ -67,7 +65,7 @@ namespace Ryujinx.Graphics.Shader.Decoders } // If we have a block after the current one, set the limit address. - ulong limitAddress = (ulong)code.Length; + ulong limitAddress = (ulong)code.Length - headerSize; if (nBlkIndex != blocks.Count) { @@ -85,13 +83,15 @@ namespace Ryujinx.Graphics.Shader.Decoders } } - FillBlock(code, currBlock, limitAddress, startAddress); + FillBlock(code, currBlock, limitAddress, headerSize); if (currBlock.OpCodes.Count != 0) { - foreach (OpCodeSsy ssyOp in currBlock.SsyOpCodes) + // We should have blocks for all possible branch targets, + // including those from SSY/PBK instructions. + foreach (OpCodePush pushOp in currBlock.PushOpCodes) { - GetBlock(ssyOp.GetAbsoluteAddress()); + GetBlock(pushOp.GetAbsoluteAddress()); } // Set child blocks. "Branch" is the block the branch instruction @@ -100,9 +100,25 @@ namespace Ryujinx.Graphics.Shader.Decoders // or end of program, Next is null. OpCode lastOp = currBlock.GetLastOp(); - if (lastOp is OpCodeBranch op) + if (lastOp is OpCodeBranch opBr) { - currBlock.Branch = GetBlock(op.GetAbsoluteAddress()); + currBlock.Branch = GetBlock(opBr.GetAbsoluteAddress()); + } + else if (lastOp is OpCodeBranchIndir opBrIndir) + { + // An indirect branch could go anywhere, we don't know the target. + // Those instructions are usually used on a switch to jump table + // compiler optimization, and in those cases the possible targets + // seems to be always right after the BRX itself. We can assume + // that the possible targets are all the blocks in-between the + // instruction right after the BRX, and the common target that + // all the "cases" should eventually jump to, acting as the + // switch break. + Block firstTarget = GetBlock(currBlock.EndAddress); + + firstTarget.BrIndir = opBrIndir; + + opBrIndir.PossibleTargets.Add(firstTarget); } if (!IsUnconditionalBranch(lastOp)) @@ -122,13 +138,28 @@ namespace Ryujinx.Graphics.Shader.Decoders { blocks.Add(currBlock); } + + // Do we have a block after the current one? + if (!IsExit(currBlock.GetLastOp()) && currBlock.BrIndir != null) + { + bool targetVisited = visited.ContainsKey(currBlock.EndAddress); + + Block possibleTarget = GetBlock(currBlock.EndAddress); + + currBlock.BrIndir.PossibleTargets.Add(possibleTarget); + + if (!targetVisited) + { + possibleTarget.BrIndir = currBlock.BrIndir; + } + } } - foreach (Block ssyBlock in blocks.Where(x => x.SsyOpCodes.Count != 0)) + foreach (Block block in blocks.Where(x => x.PushOpCodes.Count != 0)) { - for (int ssyIndex = 0; ssyIndex < ssyBlock.SsyOpCodes.Count; ssyIndex++) + for (int pushOpIndex = 0; pushOpIndex < block.PushOpCodes.Count; pushOpIndex++) { - PropagateSsy(visited, ssyBlock, ssyIndex); + PropagatePushOp(visited, block, pushOpIndex); } } @@ -180,21 +211,21 @@ namespace Ryujinx.Graphics.Shader.Decoders do { - if (address >= limitAddress) + if (address + 7 >= limitAddress) { break; } // Ignore scheduling instructions, which are written every 32 bytes. - if (((address - startAddress) & 0x1f) == 0) + if ((address & 0x1f) == 0) { address += 8; continue; } - uint word0 = BinaryPrimitives.ReadUInt32LittleEndian(code.Slice((int)address)); - uint word1 = BinaryPrimitives.ReadUInt32LittleEndian(code.Slice((int)address + 4)); + uint word0 = BinaryPrimitives.ReadUInt32LittleEndian(code.Slice((int)(startAddress + address))); + uint word1 = BinaryPrimitives.ReadUInt32LittleEndian(code.Slice((int)(startAddress + address + 4))); ulong opAddress = address; @@ -221,7 +252,7 @@ namespace Ryujinx.Graphics.Shader.Decoders block.EndAddress = address; - block.UpdateSsyOpCodes(); + block.UpdatePushOps(); } private static bool IsUnconditionalBranch(OpCode opCode) @@ -242,10 +273,16 @@ namespace Ryujinx.Graphics.Shader.Decoders private static bool IsBranch(OpCode opCode) { return (opCode is OpCodeBranch opBranch && !opBranch.PushTarget) || - opCode is OpCodeSync || + opCode is OpCodeBranchIndir || + opCode is OpCodeBranchPop || opCode is OpCodeExit; } + private static bool IsExit(OpCode opCode) + { + return opCode is OpCodeExit; + } + private static OpCode MakeOpCode(Type type, InstEmitter emitter, ulong address, long opCode) { if (type == null) @@ -282,8 +319,8 @@ namespace Ryujinx.Graphics.Shader.Decoders private enum RestoreType { None, - PopSsy, - PushSync + PopPushOp, + PushBranchOp } private RestoreType _restoreType; @@ -299,45 +336,45 @@ namespace Ryujinx.Graphics.Shader.Decoders _restoreValue = 0; } - public PathBlockState(int oldSsyStackSize) + public PathBlockState(int oldStackSize) { Block = null; - _restoreType = RestoreType.PopSsy; - _restoreValue = (ulong)oldSsyStackSize; + _restoreType = RestoreType.PopPushOp; + _restoreValue = (ulong)oldStackSize; } public PathBlockState(ulong syncAddress) { Block = null; - _restoreType = RestoreType.PushSync; + _restoreType = RestoreType.PushBranchOp; _restoreValue = syncAddress; } - public void RestoreStackState(Stack ssyStack) + public void RestoreStackState(Stack branchStack) { - if (_restoreType == RestoreType.PushSync) + if (_restoreType == RestoreType.PushBranchOp) { - ssyStack.Push(_restoreValue); + branchStack.Push(_restoreValue); } - else if (_restoreType == RestoreType.PopSsy) + else if (_restoreType == RestoreType.PopPushOp) { - while (ssyStack.Count > (uint)_restoreValue) + while (branchStack.Count > (uint)_restoreValue) { - ssyStack.Pop(); + branchStack.Pop(); } } } } - private static void PropagateSsy(Dictionary blocks, Block ssyBlock, int ssyIndex) + private static void PropagatePushOp(Dictionary blocks, Block currBlock, int pushOpIndex) { - OpCodeSsy ssyOp = ssyBlock.SsyOpCodes[ssyIndex]; + OpCodePush pushOp = currBlock.PushOpCodes[pushOpIndex]; Stack workQueue = new Stack(); HashSet visited = new HashSet(); - Stack ssyStack = new Stack(); + Stack branchStack = new Stack(); void Push(PathBlockState pbs) { @@ -347,32 +384,32 @@ namespace Ryujinx.Graphics.Shader.Decoders } } - Push(new PathBlockState(ssyBlock)); + Push(new PathBlockState(currBlock)); while (workQueue.TryPop(out PathBlockState pbs)) { if (pbs.ReturningFromVisit) { - pbs.RestoreStackState(ssyStack); + pbs.RestoreStackState(branchStack); continue; } Block current = pbs.Block; - int ssyOpCodesCount = current.SsyOpCodes.Count; + int pushOpsCount = current.PushOpCodes.Count; - if (ssyOpCodesCount != 0) + if (pushOpsCount != 0) { - Push(new PathBlockState(ssyStack.Count)); + Push(new PathBlockState(branchStack.Count)); - for (int index = ssyIndex; index < ssyOpCodesCount; index++) + for (int index = pushOpIndex; index < pushOpsCount; index++) { - ssyStack.Push(current.SsyOpCodes[index].GetAbsoluteAddress()); + branchStack.Push(current.PushOpCodes[index].GetAbsoluteAddress()); } } - ssyIndex = 0; + pushOpIndex = 0; if (current.Next != null) { @@ -383,17 +420,24 @@ namespace Ryujinx.Graphics.Shader.Decoders { Push(new PathBlockState(current.Branch)); } - else if (current.GetLastOp() is OpCodeSync op) + else if (current.GetLastOp() is OpCodeBranchIndir brIndir) { - ulong syncAddress = ssyStack.Pop(); - - if (ssyStack.Count == 0) + foreach (Block possibleTarget in brIndir.PossibleTargets) { - ssyStack.Push(syncAddress); + Push(new PathBlockState(possibleTarget)); + } + } + else if (current.GetLastOp() is OpCodeBranchPop op) + { + ulong syncAddress = branchStack.Pop(); - op.Targets.Add(ssyOp, op.Targets.Count); + if (branchStack.Count == 0) + { + branchStack.Push(syncAddress); - ssyOp.Syncs.TryAdd(op, Local()); + op.Targets.Add(pushOp, op.Targets.Count); + + pushOp.PopOps.TryAdd(op, Local()); } else { diff --git a/Ryujinx.Graphics.Shader/Decoders/FmulScale.cs b/Ryujinx.Graphics.Shader/Decoders/FPMultiplyScale.cs similarity index 89% rename from Ryujinx.Graphics.Shader/Decoders/FmulScale.cs rename to Ryujinx.Graphics.Shader/Decoders/FPMultiplyScale.cs index c35c6e489..398c0e66f 100644 --- a/Ryujinx.Graphics.Shader/Decoders/FmulScale.cs +++ b/Ryujinx.Graphics.Shader/Decoders/FPMultiplyScale.cs @@ -1,6 +1,6 @@ namespace Ryujinx.Graphics.Shader.Decoders { - enum FmulScale + enum FPMultiplyScale { None = 0, Divide2 = 1, diff --git a/Ryujinx.Graphics.Shader/Decoders/IOpCodeFArith.cs b/Ryujinx.Graphics.Shader/Decoders/IOpCodeFArith.cs index d68ccf593..3d06eae0d 100644 --- a/Ryujinx.Graphics.Shader/Decoders/IOpCodeFArith.cs +++ b/Ryujinx.Graphics.Shader/Decoders/IOpCodeFArith.cs @@ -4,7 +4,7 @@ namespace Ryujinx.Graphics.Shader.Decoders { RoundingMode RoundingMode { get; } - FmulScale Scale { get; } + FPMultiplyScale Scale { get; } bool FlushToZero { get; } bool AbsoluteA { get; } diff --git a/Ryujinx.Graphics.Shader/Decoders/OpCodeBranchIndir.cs b/Ryujinx.Graphics.Shader/Decoders/OpCodeBranchIndir.cs new file mode 100644 index 000000000..3e694e61c --- /dev/null +++ b/Ryujinx.Graphics.Shader/Decoders/OpCodeBranchIndir.cs @@ -0,0 +1,23 @@ +using Ryujinx.Graphics.Shader.Instructions; +using System.Collections.Generic; + +namespace Ryujinx.Graphics.Shader.Decoders +{ + class OpCodeBranchIndir : OpCode + { + public HashSet PossibleTargets { get; } + + public Register Ra { get; } + + public int Offset { get; } + + public OpCodeBranchIndir(InstEmitter emitter, ulong address, long opCode) : base(emitter, address, opCode) + { + PossibleTargets = new HashSet(); + + Ra = new Register(opCode.Extract(8, 8), RegisterType.Gpr); + + Offset = ((int)(opCode >> 20) << 8) >> 8; + } + } +} \ No newline at end of file diff --git a/Ryujinx.Graphics.Shader/Decoders/OpCodeBranchPop.cs b/Ryujinx.Graphics.Shader/Decoders/OpCodeBranchPop.cs new file mode 100644 index 000000000..7ea66fe45 --- /dev/null +++ b/Ryujinx.Graphics.Shader/Decoders/OpCodeBranchPop.cs @@ -0,0 +1,15 @@ +using Ryujinx.Graphics.Shader.Instructions; +using System.Collections.Generic; + +namespace Ryujinx.Graphics.Shader.Decoders +{ + class OpCodeBranchPop : OpCode + { + public Dictionary Targets { get; } + + public OpCodeBranchPop(InstEmitter emitter, ulong address, long opCode) : base(emitter, address, opCode) + { + Targets = new Dictionary(); + } + } +} \ No newline at end of file diff --git a/Ryujinx.Graphics.Shader/Decoders/OpCodeFArith.cs b/Ryujinx.Graphics.Shader/Decoders/OpCodeFArith.cs index c88f7f0ee..cfbf65c3d 100644 --- a/Ryujinx.Graphics.Shader/Decoders/OpCodeFArith.cs +++ b/Ryujinx.Graphics.Shader/Decoders/OpCodeFArith.cs @@ -6,7 +6,7 @@ namespace Ryujinx.Graphics.Shader.Decoders { public RoundingMode RoundingMode { get; } - public FmulScale Scale { get; } + public FPMultiplyScale Scale { get; } public bool FlushToZero { get; } public bool AbsoluteA { get; } @@ -15,7 +15,7 @@ namespace Ryujinx.Graphics.Shader.Decoders { RoundingMode = (RoundingMode)opCode.Extract(39, 2); - Scale = (FmulScale)opCode.Extract(41, 3); + Scale = (FPMultiplyScale)opCode.Extract(41, 3); FlushToZero = opCode.Extract(44); AbsoluteA = opCode.Extract(46); diff --git a/Ryujinx.Graphics.Shader/Decoders/OpCodeFArithImm32.cs b/Ryujinx.Graphics.Shader/Decoders/OpCodeFArithImm32.cs index ec9da6f30..aecc5143c 100644 --- a/Ryujinx.Graphics.Shader/Decoders/OpCodeFArithImm32.cs +++ b/Ryujinx.Graphics.Shader/Decoders/OpCodeFArithImm32.cs @@ -7,7 +7,7 @@ namespace Ryujinx.Graphics.Shader.Decoders { public RoundingMode RoundingMode => RoundingMode.ToNearest; - public FmulScale Scale => FmulScale.None; + public FPMultiplyScale Scale => FPMultiplyScale.None; public bool FlushToZero { get; } public bool AbsoluteA { get; } diff --git a/Ryujinx.Graphics.Shader/Decoders/OpCodeSsy.cs b/Ryujinx.Graphics.Shader/Decoders/OpCodePush.cs similarity index 58% rename from Ryujinx.Graphics.Shader/Decoders/OpCodeSsy.cs rename to Ryujinx.Graphics.Shader/Decoders/OpCodePush.cs index d3831e22d..a7657bcf8 100644 --- a/Ryujinx.Graphics.Shader/Decoders/OpCodeSsy.cs +++ b/Ryujinx.Graphics.Shader/Decoders/OpCodePush.cs @@ -4,13 +4,13 @@ using System.Collections.Generic; namespace Ryujinx.Graphics.Shader.Decoders { - class OpCodeSsy : OpCodeBranch + class OpCodePush : OpCodeBranch { - public Dictionary Syncs { get; } + public Dictionary PopOps { get; } - public OpCodeSsy(InstEmitter emitter, ulong address, long opCode) : base(emitter, address, opCode) + public OpCodePush(InstEmitter emitter, ulong address, long opCode) : base(emitter, address, opCode) { - Syncs = new Dictionary(); + PopOps = new Dictionary(); Predicate = new Register(RegisterConsts.PredicateTrueIndex, RegisterType.Predicate); diff --git a/Ryujinx.Graphics.Shader/Decoders/OpCodeSync.cs b/Ryujinx.Graphics.Shader/Decoders/OpCodeSync.cs deleted file mode 100644 index 081d08a0d..000000000 --- a/Ryujinx.Graphics.Shader/Decoders/OpCodeSync.cs +++ /dev/null @@ -1,15 +0,0 @@ -using Ryujinx.Graphics.Shader.Instructions; -using System.Collections.Generic; - -namespace Ryujinx.Graphics.Shader.Decoders -{ - class OpCodeSync : OpCode - { - public Dictionary Targets { get; } - - public OpCodeSync(InstEmitter emitter, ulong address, long opCode) : base(emitter, address, opCode) - { - Targets = new Dictionary(); - } - } -} \ No newline at end of file diff --git a/Ryujinx.Graphics.Shader/Decoders/OpCodeTable.cs b/Ryujinx.Graphics.Shader/Decoders/OpCodeTable.cs index 58bd2b88e..bc30940d8 100644 --- a/Ryujinx.Graphics.Shader/Decoders/OpCodeTable.cs +++ b/Ryujinx.Graphics.Shader/Decoders/OpCodeTable.cs @@ -41,7 +41,8 @@ namespace Ryujinx.Graphics.Shader.Decoders Set("0101001111110x", InstEmit.Bfi, typeof(OpCodeAluRegCbuf)); Set("0101101111110x", InstEmit.Bfi, typeof(OpCodeAluReg)); Set("111000100100xx", InstEmit.Bra, typeof(OpCodeBranch)); - Set("111000110100xx", InstEmit.Brk, typeof(OpCodeSync)); + Set("111000110100xx", InstEmit.Brk, typeof(OpCodeBranchPop)); + Set("111000100101xx", InstEmit.Brx, typeof(OpCodeBranchIndir)); Set("0101000010100x", InstEmit.Csetp, typeof(OpCodePsetp)); Set("111000110000xx", InstEmit.Exit, typeof(OpCodeExit)); Set("0100110010101x", InstEmit.F2F, typeof(OpCodeFArithCbuf)); @@ -137,7 +138,7 @@ namespace Ryujinx.Graphics.Shader.Decoders Set("0101110010011x", InstEmit.Mov, typeof(OpCodeAluReg)); Set("0101000010000x", InstEmit.Mufu, typeof(OpCodeFArith)); Set("1111101111100x", InstEmit.Out, typeof(OpCode)); - Set("111000101010xx", InstEmit.Pbk, typeof(OpCodeSsy)); + Set("111000101010xx", InstEmit.Pbk, typeof(OpCodePush)); Set("0100110000001x", InstEmit.Popc, typeof(OpCodeAluCbuf)); Set("0011100x00001x", InstEmit.Popc, typeof(OpCodeAluImm)); Set("0101110000001x", InstEmit.Popc, typeof(OpCodeAluReg)); @@ -157,12 +158,12 @@ namespace Ryujinx.Graphics.Shader.Decoders Set("0100110000101x", InstEmit.Shr, typeof(OpCodeAluCbuf)); Set("0011100x00101x", InstEmit.Shr, typeof(OpCodeAluImm)); Set("0101110000101x", InstEmit.Shr, typeof(OpCodeAluReg)); - Set("111000101001xx", InstEmit.Ssy, typeof(OpCodeSsy)); + Set("111000101001xx", InstEmit.Ssy, typeof(OpCodePush)); Set("1110111101010x", InstEmit.St, typeof(OpCodeMemory)); Set("1110111011011x", InstEmit.Stg, typeof(OpCodeMemory)); Set("1110111101011x", InstEmit.Sts, typeof(OpCodeMemory)); Set("11101011001xxx", InstEmit.Sust, typeof(OpCodeImage)); - Set("1111000011111x", InstEmit.Sync, typeof(OpCodeSync)); + Set("1111000011111x", InstEmit.Sync, typeof(OpCodeBranchPop)); Set("110000xxxx111x", InstEmit.Tex, typeof(OpCodeTex)); Set("1101111010111x", InstEmit.TexB, typeof(OpCodeTexB)); Set("1101x00xxxxxxx", InstEmit.Texs, typeof(OpCodeTexs)); diff --git a/Ryujinx.Graphics.Shader/Instructions/InstEmitFArith.cs b/Ryujinx.Graphics.Shader/Instructions/InstEmitFArith.cs index 4f7072eb0..79d92c2d7 100644 --- a/Ryujinx.Graphics.Shader/Instructions/InstEmitFArith.cs +++ b/Ryujinx.Graphics.Shader/Instructions/InstEmitFArith.cs @@ -97,14 +97,14 @@ namespace Ryujinx.Graphics.Shader.Instructions switch (op.Scale) { - case FmulScale.None: break; + case FPMultiplyScale.None: break; - case FmulScale.Divide2: srcA = context.FPDivide (srcA, ConstF(2)); break; - case FmulScale.Divide4: srcA = context.FPDivide (srcA, ConstF(4)); break; - case FmulScale.Divide8: srcA = context.FPDivide (srcA, ConstF(8)); break; - case FmulScale.Multiply2: srcA = context.FPMultiply(srcA, ConstF(2)); break; - case FmulScale.Multiply4: srcA = context.FPMultiply(srcA, ConstF(4)); break; - case FmulScale.Multiply8: srcA = context.FPMultiply(srcA, ConstF(8)); break; + case FPMultiplyScale.Divide2: srcA = context.FPDivide (srcA, ConstF(2)); break; + case FPMultiplyScale.Divide4: srcA = context.FPDivide (srcA, ConstF(4)); break; + case FPMultiplyScale.Divide8: srcA = context.FPDivide (srcA, ConstF(8)); break; + case FPMultiplyScale.Multiply2: srcA = context.FPMultiply(srcA, ConstF(2)); break; + case FPMultiplyScale.Multiply4: srcA = context.FPMultiply(srcA, ConstF(4)); break; + case FPMultiplyScale.Multiply8: srcA = context.FPMultiply(srcA, ConstF(8)); break; default: break; //TODO: Warning. } diff --git a/Ryujinx.Graphics.Shader/Instructions/InstEmitFlow.cs b/Ryujinx.Graphics.Shader/Instructions/InstEmitFlow.cs index e17c9d6c6..4a9f5f7fc 100644 --- a/Ryujinx.Graphics.Shader/Instructions/InstEmitFlow.cs +++ b/Ryujinx.Graphics.Shader/Instructions/InstEmitFlow.cs @@ -20,6 +20,36 @@ namespace Ryujinx.Graphics.Shader.Instructions EmitBrkOrSync(context); } + public static void Brx(EmitterContext context) + { + OpCodeBranchIndir op = (OpCodeBranchIndir)context.CurrOp; + + int offset = (int)op.Address + 8 + op.Offset; + + Operand address = context.IAdd(Register(op.Ra), Const(offset)); + + // Sorting the target addresses in descending order improves the code, + // since it will always check the most distant targets first, then the + // near ones. This can be easily transformed into if/else statements. + IOrderedEnumerable sortedTargets = op.PossibleTargets.OrderByDescending(x => x.Address); + + Block lastTarget = sortedTargets.LastOrDefault(); + + foreach (Block possibleTarget in sortedTargets) + { + Operand label = context.GetLabel(possibleTarget.Address); + + if (possibleTarget != lastTarget) + { + context.BranchIfTrue(label, context.ICompareEqual(address, Const((int)possibleTarget.Address))); + } + else + { + context.Branch(label); + } + } + } + public static void Exit(EmitterContext context) { OpCodeExit op = (OpCodeExit)context.CurrOp; @@ -54,45 +84,45 @@ namespace Ryujinx.Graphics.Shader.Instructions private static void EmitPbkOrSsy(EmitterContext context) { - OpCodeSsy op = (OpCodeSsy)context.CurrOp; + OpCodePush op = (OpCodePush)context.CurrOp; - foreach (KeyValuePair kv in op.Syncs) + foreach (KeyValuePair kv in op.PopOps) { - OpCodeSync opSync = kv.Key; + OpCodeBranchPop opSync = kv.Key; Operand local = kv.Value; - int ssyIndex = opSync.Targets[op]; + int pushOpIndex = opSync.Targets[op]; - context.Copy(local, Const(ssyIndex)); + context.Copy(local, Const(pushOpIndex)); } } private static void EmitBrkOrSync(EmitterContext context) { - OpCodeSync op = (OpCodeSync)context.CurrOp; + OpCodeBranchPop op = (OpCodeBranchPop)context.CurrOp; if (op.Targets.Count == 1) { - // If we have only one target, then the SSY is basically + // If we have only one target, then the SSY/PBK is basically // a branch, we can produce better codegen for this case. - OpCodeSsy opSsy = op.Targets.Keys.First(); + OpCodePush pushOp = op.Targets.Keys.First(); - EmitBranch(context, opSsy.GetAbsoluteAddress()); + EmitBranch(context, pushOp.GetAbsoluteAddress()); } else { - foreach (KeyValuePair kv in op.Targets) + foreach (KeyValuePair kv in op.Targets) { - OpCodeSsy opSsy = kv.Key; + OpCodePush pushOp = kv.Key; - Operand label = context.GetLabel(opSsy.GetAbsoluteAddress()); + Operand label = context.GetLabel(pushOp.GetAbsoluteAddress()); - Operand local = opSsy.Syncs[op]; + Operand local = pushOp.PopOps[op]; - int ssyIndex = kv.Value; + int pushOpIndex = kv.Value; - context.BranchIfTrue(label, context.ICompareEqual(local, Const(ssyIndex))); + context.BranchIfTrue(label, context.ICompareEqual(local, Const(pushOpIndex))); } } } diff --git a/Ryujinx.Graphics.Shader/ShaderProgram.cs b/Ryujinx.Graphics.Shader/ShaderProgram.cs index 5d04f2cf4..4d0c6e5bd 100644 --- a/Ryujinx.Graphics.Shader/ShaderProgram.cs +++ b/Ryujinx.Graphics.Shader/ShaderProgram.cs @@ -10,11 +10,14 @@ namespace Ryujinx.Graphics.Shader public string Code { get; private set; } - internal ShaderProgram(ShaderProgramInfo info, ShaderStage stage, string code) + public int Size { get; } + + internal ShaderProgram(ShaderProgramInfo info, ShaderStage stage, string code, int size) { Info = info; Stage = stage; Code = code; + Size = size; } public void Prepend(string line) diff --git a/Ryujinx.Graphics.Shader/Translation/Translator.cs b/Ryujinx.Graphics.Shader/Translation/Translator.cs index b7a5bffab..55617b24c 100644 --- a/Ryujinx.Graphics.Shader/Translation/Translator.cs +++ b/Ryujinx.Graphics.Shader/Translation/Translator.cs @@ -46,7 +46,12 @@ namespace Ryujinx.Graphics.Shader.Translation bool compute = (translationConfig.Flags & TranslationFlags.Compute) != 0; bool debugMode = (translationConfig.Flags & TranslationFlags.DebugMode) != 0; - Operation[] ops = DecodeShader(code, compute, debugMode, out ShaderHeader header); + Operation[] ops = DecodeShader( + code, + compute, + debugMode, + out ShaderHeader header, + out int size); ShaderStage stage; @@ -76,15 +81,15 @@ namespace Ryujinx.Graphics.Shader.Translation maxOutputVertexCount, outputTopology); - return Translate(ops, config); + return Translate(ops, config, size); } public static ShaderProgram Translate(Span vpACode, Span vpBCode, TranslationConfig translationConfig) { bool debugMode = (translationConfig.Flags & TranslationFlags.DebugMode) != 0; - Operation[] vpAOps = DecodeShader(vpACode, compute: false, debugMode, out _); - Operation[] vpBOps = DecodeShader(vpBCode, compute: false, debugMode, out ShaderHeader header); + Operation[] vpAOps = DecodeShader(vpACode, compute: false, debugMode, out _, out _); + Operation[] vpBOps = DecodeShader(vpBCode, compute: false, debugMode, out ShaderHeader header, out int sizeB); ShaderConfig config = new ShaderConfig( header.Stage, @@ -93,10 +98,10 @@ namespace Ryujinx.Graphics.Shader.Translation header.MaxOutputVertexCount, header.OutputTopology); - return Translate(Combine(vpAOps, vpBOps), config); + return Translate(Combine(vpAOps, vpBOps), config, sizeB); } - private static ShaderProgram Translate(Operation[] ops, ShaderConfig config) + private static ShaderProgram Translate(Operation[] ops, ShaderConfig config, int size) { BasicBlock[] irBlocks = ControlFlowGraph.MakeCfg(ops); @@ -122,17 +127,20 @@ namespace Ryujinx.Graphics.Shader.Translation string glslCode = program.Code; - return new ShaderProgram(spInfo, config.Stage, glslCode); + return new ShaderProgram(spInfo, config.Stage, glslCode, size); } - private static Operation[] DecodeShader(Span code, bool compute, bool debugMode, out ShaderHeader header) + private static Operation[] DecodeShader( + Span code, + bool compute, + bool debugMode, + out ShaderHeader header, + out int size) { Block[] cfg; EmitterContext context; - ulong headerSize; - if (compute) { header = null; @@ -140,8 +148,6 @@ namespace Ryujinx.Graphics.Shader.Translation cfg = Decoder.Decode(code, 0); context = new EmitterContext(ShaderStage.Compute, header); - - headerSize = 0; } else { @@ -150,14 +156,19 @@ namespace Ryujinx.Graphics.Shader.Translation cfg = Decoder.Decode(code, HeaderSize); context = new EmitterContext(header.Stage, header); - - headerSize = HeaderSize; } + ulong maxEndAddress = 0; + for (int blkIndex = 0; blkIndex < cfg.Length; blkIndex++) { Block block = cfg[blkIndex]; + if (maxEndAddress < block.EndAddress) + { + maxEndAddress = block.EndAddress; + } + context.CurrBlock = block; context.MarkLabel(context.GetLabel(block.Address)); @@ -179,7 +190,7 @@ namespace Ryujinx.Graphics.Shader.Translation instName = "???"; } - string dbgComment = $"0x{(op.Address - headerSize):X6}: 0x{op.RawOpCode:X16} {instName}"; + string dbgComment = $"0x{op.Address:X6}: 0x{op.RawOpCode:X16} {instName}"; context.Add(new CommentNode(dbgComment)); } @@ -193,13 +204,13 @@ namespace Ryujinx.Graphics.Shader.Translation bool skipPredicateCheck = op.Emitter == InstEmit.Bra; - if (op is OpCodeSync opSync) + if (op is OpCodeBranchPop opBranchPop) { // If the instruction is a SYNC instruction with only one // possible target address, then the instruction is basically // just a simple branch, we can generate code similar to branch // instructions, with the condition check on the branch itself. - skipPredicateCheck |= opSync.Targets.Count < 2; + skipPredicateCheck |= opBranchPop.Targets.Count < 2; } if (!(op.Predicate.IsPT || skipPredicateCheck)) @@ -243,6 +254,8 @@ namespace Ryujinx.Graphics.Shader.Translation } } + size = (int)maxEndAddress + (compute ? 0 : HeaderSize); + return context.GetOperations(); }