vleeuwenmenno/Ryujinx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Support texture rectangle targets (non-normalized coords)

Browse Source
This commit is contained in:
gdkchan
2019-12-16 01:59:46 -03:00
committed by Thog
parent 2eccc7023a
commit 9d7a142a48
23 changed files with 473 additions and 356 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
View File

@ -11,25 +11,15 @@ namespace Ryujinx.Graphics.Shader.Translation
public Block CurrBlock { get; set; }
public OpCode CurrOp { get; set; }
private ShaderStage _stage;
private ShaderHeader _header;
private ShaderCapabilities _capabilities;
private TranslationFlags _flags;
private ShaderConfig _config;
private List<Operation> _operations;
private Dictionary<ulong, Operand> _labels;
public EmitterContext(
ShaderStage stage,
ShaderHeader header,
ShaderCapabilities capabilities,
TranslationFlags flags)
public EmitterContext(ShaderConfig config)
{
_stage = stage;
_header = header;
_capabilities = capabilities;
_flags = flags;
_config = config;
_operations = new List<Operation>();
@ -69,24 +59,24 @@ namespace Ryujinx.Graphics.Shader.Translation
public void PrepareForReturn()
{
if (_stage == ShaderStage.Vertex)
if (_config.Stage == ShaderStage.Vertex)
{
if ((_flags & TranslationFlags.DividePosXY) != 0)
if (!_config.QueryInfoBool(QueryInfoName.ViewportTransformEnable))
{
Operand posX = Attribute(AttributeConsts.PositionX);
Operand posY = Attribute(AttributeConsts.PositionY);
this.Copy(posX, this.FPDivide(posX, ConstF(_capabilities.MaximumViewportDimensions / 2)));
this.Copy(posY, this.FPDivide(posY, ConstF(_capabilities.MaximumViewportDimensions / 2)));
this.Copy(posX, this.FPDivide(posX, ConstF(_config.QueryInfo(QueryInfoName.MaximumViewportDimensions) / 2)));
this.Copy(posY, this.FPDivide(posY, ConstF(_config.QueryInfo(QueryInfoName.MaximumViewportDimensions) / 2)));
}
}
else if (_stage == ShaderStage.Fragment)
else if (_config.Stage == ShaderStage.Fragment)
{
if (_header.OmapDepth)
if (_config.OmapDepth)
{
Operand dest = Attribute(AttributeConsts.FragmentOutputDepth);
Operand src = Register(_header.DepthRegister, RegisterType.Gpr);
Operand src = Register(_config.GetDepthRegister(), RegisterType.Gpr);
this.Copy(dest, src);
}
@ -95,7 +85,7 @@ namespace Ryujinx.Graphics.Shader.Translation
for (int attachment = 0; attachment < 8; attachment++)
{
OutputMapTarget target = _header.OmapTargets[attachment];
OutputMapTarget target = _config.OmapTargets[attachment];
for (int component = 0; component < 4; component++)
{

210
Ryujinx.Graphics.Shader/Translation/Lowering.cs
View File

@ -27,9 +27,9 @@ namespace Ryujinx.Graphics.Shader.Translation
node = RewriteGlobalAccess(node, config);
}
if (!config.Capabilities.SupportsNonConstantTextureOffset && operation.Inst == Instruction.TextureSample)
if (operation.Inst == Instruction.TextureSample)
{
node = RewriteTextureSample(node);
node = RewriteTextureSample(node, config);
}
}
}
@ -79,7 +79,7 @@ namespace Ryujinx.Graphics.Shader.Translation
sbSlot = PrependOperation(Instruction.ConditionalSelect, inRange, Const(slot), sbSlot);
}
Operand alignMask = Const(-config.Capabilities.StorageBufferOffsetAlignment);
Operand alignMask = Const(-config.QueryInfo(QueryInfoName.StorageBufferOffsetAlignment));
Operand baseAddrTrunc = PrependOperation(Instruction.BitwiseAnd, sbBaseAddrLow, Const(-64));
Operand byteOffset = PrependOperation(Instruction.Subtract, addrLow, baseAddrTrunc);
@ -124,23 +124,18 @@ namespace Ryujinx.Graphics.Shader.Translation
return node;
}
private static LinkedListNode<INode> RewriteTextureSample(LinkedListNode<INode> node)
private static LinkedListNode<INode> RewriteTextureSample(LinkedListNode<INode> node, ShaderConfig config)
{
// Technically, non-constant texture offsets are not allowed (according to the spec),
// however some GPUs does support that.
// For GPUs where it is not supported, we can replace the instruction with the following:
// For texture*Offset, we replace it by texture*, and add the offset to the P coords.
// The offset can be calculated as offset / textureSize(lod), where lod = textureQueryLod(coords).
// For texelFetchOffset, we replace it by texelFetch and add the offset to the P coords directly.
// For textureGatherOffset, we take advantage of the fact that the operation is already broken down
// to read the 4 pixels separately, and just replace it with 4 textureGather with a different offset
// for each pixel.
TextureOperation texOp = (TextureOperation)node.Value;
bool hasOffset = (texOp.Flags & TextureFlags.Offset) != 0;
bool hasOffsets = (texOp.Flags & TextureFlags.Offsets) != 0;
if (!(hasOffset || hasOffsets))
bool hasInvalidOffset = (hasOffset || hasOffsets) && !config.QueryInfoBool(QueryInfoName.SupportsNonConstantTextureOffset);
bool isRect = config.QueryInfoBool(QueryInfoName.IsTextureRectangle, texOp.Handle);
if (!(hasInvalidOffset || isRect))
{
return node;
}
@ -159,14 +154,24 @@ namespace Ryujinx.Graphics.Shader.Translation
int coordsCount = texOp.Type.GetDimensions();
int offsetsCount = coordsCount * (hasOffsets ? 4 : 1);
int offsetsCount;
if (hasOffsets)
{
offsetsCount = coordsCount * 4;
}
else if (hasOffset)
{
offsetsCount = coordsCount;
}
else
{
offsetsCount = 0;
}
Operand[] offsets = new Operand[offsetsCount];
Operand[] sources = new Operand[texOp.SourcesCount - offsetsCount];
int srcIndex = 0;
int dstIndex = 0;
int copyCount = 0;
if (isBindless || isIndexed)
@ -207,6 +212,9 @@ namespace Ryujinx.Graphics.Shader.Translation
copyCount++;
}
int srcIndex = 0;
int dstIndex = 0;
for (int index = 0; index < copyCount; index++)
{
sources[dstIndex++] = texOp.GetSource(srcIndex++);
@ -223,7 +231,9 @@ namespace Ryujinx.Graphics.Shader.Translation
offsets[index] = offset;
}
if (areAllOffsetsConstant)
hasInvalidOffset &= !areAllOffsetsConstant;
if (!(hasInvalidOffset || isRect))
{
return node;
}
@ -240,50 +250,32 @@ namespace Ryujinx.Graphics.Shader.Translation
int coordsIndex = isBindless || isIndexed ? 1 : 0;
if (intCoords)
int componentIndex = texOp.Index;
Operand Int(Operand value)
{
for (int index = 0; index < coordsCount; index++)
{
Operand source = sources[coordsIndex + index];
Operand res = Local();
Operand coordPlusOffset = Local();
node.List.AddBefore(node, new Operation(Instruction.ConvertFPToS32, res, value));
node.List.AddBefore(node, new Operation(Instruction.Add, coordPlusOffset, source, offsets[index]));
sources[coordsIndex + index] = coordPlusOffset;
}
return res;
}
else
Operand Float(Operand value)
{
Operand lod = Local();
Operand res = Local();
node.List.AddBefore(node, new TextureOperation(
Instruction.Lod,
texOp.Type,
texOp.Flags,
texOp.Handle,
1,
lod,
lodSources));
node.List.AddBefore(node, new Operation(Instruction.ConvertS32ToFP, res, value));
Operand Int(Operand value)
{
Operand res = Local();
node.List.AddBefore(node, new Operation(Instruction.ConvertFPToS32, res, value));
return res;
}
Operand Float(Operand value)
{
Operand res = Local();
node.List.AddBefore(node, new Operation(Instruction.ConvertS32ToFP, res, value));
return res;
}
return res;
}
// Emulate texture rectangle by normalizing the coordinates on the shader.
// When sampler*Rect is used, the coords are expected to the in the [0, W or H] range,
// and otherwise, it is expected to be in the [0, 1] range.
// We normalize by dividing the coords by the texture size.
if (isRect && !intCoords)
{
for (int index = 0; index < coordsCount; index++)
{
Operand coordSize = Local();
@ -292,11 +284,11 @@ namespace Ryujinx.Graphics.Shader.Translation
if (isBindless || isIndexed)
{
texSizeSources = new Operand[] { sources[0], Int(lod) };
texSizeSources = new Operand[] { sources[0], Const(0) };
}
else
{
texSizeSources = new Operand[] { Int(lod) };
texSizeSources = new Operand[] { Const(0) };
}
node.List.AddBefore(node, new TextureOperation(
@ -308,35 +300,101 @@ namespace Ryujinx.Graphics.Shader.Translation
coordSize,
texSizeSources));
Operand offset = Local();
Operand intOffset = offsets[index + (hasOffsets ? texOp.Index * coordsCount : 0)];
node.List.AddBefore(node, new Operation(Instruction.FP | Instruction.Divide, offset, Float(intOffset), Float(coordSize)));
Operand source = sources[coordsIndex + index];
Operand coordPlusOffset = Local();
Operand coordNormalized = Local();
node.List.AddBefore(node, new Operation(Instruction.FP | Instruction.Add, coordPlusOffset, source, offset));
node.List.AddBefore(node, new Operation(Instruction.FP | Instruction.Divide, coordNormalized, source, Float(coordSize)));
sources[coordsIndex + index] = coordPlusOffset;
sources[coordsIndex + index] = coordNormalized;
}
}
int componentIndex;
if (isGather && !isShadow)
// Technically, non-constant texture offsets are not allowed (according to the spec),
// however some GPUs does support that.
// For GPUs where it is not supported, we can replace the instruction with the following:
// For texture*Offset, we replace it by texture*, and add the offset to the P coords.
// The offset can be calculated as offset / textureSize(lod), where lod = textureQueryLod(coords).
// For texelFetchOffset, we replace it by texelFetch and add the offset to the P coords directly.
// For textureGatherOffset, we take advantage of the fact that the operation is already broken down
// to read the 4 pixels separately, and just replace it with 4 textureGather with a different offset
// for each pixel.
if (hasInvalidOffset)
{
Operand gatherComponent = sources[dstIndex - 1];
if (intCoords)
{
for (int index = 0; index < coordsCount; index++)
{
Operand source = sources[coordsIndex + index];
Debug.Assert(gatherComponent.Type == OperandType.Constant);
Operand coordPlusOffset = Local();
componentIndex = gatherComponent.Value;
}
else
{
componentIndex = texOp.Index;
node.List.AddBefore(node, new Operation(Instruction.Add, coordPlusOffset, source, offsets[index]));
sources[coordsIndex + index] = coordPlusOffset;
}
}
else
{
Operand lod = Local();
node.List.AddBefore(node, new TextureOperation(
Instruction.Lod,
texOp.Type,
texOp.Flags,
texOp.Handle,
1,
lod,
lodSources));
for (int index = 0; index < coordsCount; index++)
{
Operand coordSize = Local();
Operand[] texSizeSources;
if (isBindless || isIndexed)
{
texSizeSources = new Operand[] { sources[0], Int(lod) };
}
else
{
texSizeSources = new Operand[] { Int(lod) };
}
node.List.AddBefore(node, new TextureOperation(
Instruction.TextureSize,
texOp.Type,
texOp.Flags,
texOp.Handle,
index,
coordSize,
texSizeSources));
Operand offset = Local();
Operand intOffset = offsets[index + (hasOffsets ? texOp.Index * coordsCount : 0)];
node.List.AddBefore(node, new Operation(Instruction.FP | Instruction.Divide, offset, Float(intOffset), Float(coordSize)));
Operand source = sources[coordsIndex + index];
Operand coordPlusOffset = Local();
node.List.AddBefore(node, new Operation(Instruction.FP | Instruction.Add, coordPlusOffset, source, offset));
sources[coordsIndex + index] = coordPlusOffset;
}
}
if (isGather && !isShadow)
{
Operand gatherComponent = sources[dstIndex - 1];
Debug.Assert(gatherComponent.Type == OperandType.Constant);
componentIndex = gatherComponent.Value;
}
}
TextureOperation newTexOp = new TextureOperation(

2
Ryujinx.Graphics.Shader/Translation/Optimizations/GlobalToStorage.cs
View File

@ -52,7 +52,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
Operand baseAddrTrunc = Local();
Operand alignMask = Const(-config.Capabilities.StorageBufferOffsetAlignment);
Operand alignMask = Const(-config.QueryInfo(QueryInfoName.StorageBufferOffsetAlignment));
Operation andOp = new Operation(Instruction.BitwiseAnd, baseAddrTrunc, baseAddrLow, alignMask);

6
Ryujinx.Graphics.Shader/Translation/TranslationFlags.cs
View File

@ -4,9 +4,7 @@ namespace Ryujinx.Graphics.Shader.Translation
{
None = 0,
Compute = 1 << 0,
DebugMode = 1 << 1,
Unspecialized = 1 << 2,
DividePosXY = 1 << 3
Compute = 1 << 0,
DebugMode = 1 << 1
}
}

76
Ryujinx.Graphics.Shader/Translation/Translator.cs
View File

@ -16,14 +16,7 @@ namespace Ryujinx.Graphics.Shader.Translation
public static Span<byte> ExtractCode(Span<byte> code, bool compute, out int headerSize)
{
if (compute)
{
headerSize = 0;
}
else
{
headerSize = HeaderSize;
}
headerSize = compute ? 0 : HeaderSize;
Block[] cfg = Decoder.Decode(code, (ulong)headerSize);
@ -47,56 +40,21 @@ namespace Ryujinx.Graphics.Shader.Translation
return code.Slice(0, headerSize + (int)endAddress);
}
public static ShaderProgram Translate(Span<byte> code, ShaderCapabilities capabilities, TranslationFlags flags)
public static ShaderProgram Translate(Span<byte> code, QueryInfoCallback queryInfoCallback, TranslationFlags flags)
{
bool compute = (flags & TranslationFlags.Compute) != 0;
Operation[] ops = DecodeShader(code, capabilities, flags, out ShaderHeader header, out int size);
ShaderStage stage;
if (compute)
{
stage = ShaderStage.Compute;
}
else
{
stage = header.Stage;
}
int maxOutputVertexCount = 0;
OutputTopology outputTopology = OutputTopology.LineStrip;
if (!compute)
{
maxOutputVertexCount = header.MaxOutputVertexCount;
outputTopology = header.OutputTopology;
}
ShaderConfig config = new ShaderConfig(
stage,
capabilities,
flags,
maxOutputVertexCount,
outputTopology);
Operation[] ops = DecodeShader(code, queryInfoCallback, flags, out ShaderConfig config, out int size);
return Translate(ops, config, size);
}
public static ShaderProgram Translate(Span<byte> vpACode, Span<byte> vpBCode, ShaderCapabilities capabilities, TranslationFlags flags)
public static ShaderProgram Translate(Span<byte> vpACode, Span<byte> vpBCode, QueryInfoCallback queryInfoCallback, TranslationFlags flags)
{
bool debugMode = (flags & TranslationFlags.DebugMode) != 0;
Operation[] vpAOps = DecodeShader(vpACode, capabilities, flags, out _, out _);
Operation[] vpBOps = DecodeShader(vpBCode, capabilities, flags, out ShaderHeader header, out int sizeB);
ShaderConfig config = new ShaderConfig(
header.Stage,
capabilities,
flags,
header.MaxOutputVertexCount,
header.OutputTopology);
Operation[] vpAOps = DecodeShader(vpACode, queryInfoCallback, flags, out _, out _);
Operation[] vpBOps = DecodeShader(vpBCode, queryInfoCallback, flags, out ShaderConfig config, out int sizeB);
return Translate(Combine(vpAOps, vpBOps), config, sizeB);
}
@ -136,31 +94,25 @@ namespace Ryujinx.Graphics.Shader.Translation
}
private static Operation[] DecodeShader(
Span<byte> code,
ShaderCapabilities capabilities,
TranslationFlags flags,
out ShaderHeader header,
out int size)
Span<byte> code,
QueryInfoCallback queryInfoCallback,
TranslationFlags flags,
out ShaderConfig config,
out int size)
{
Block[] cfg;
EmitterContext context;
if ((flags & TranslationFlags.Compute) != 0)
{
header = null;
config = new ShaderConfig(flags, queryInfoCallback);
cfg = Decoder.Decode(code, 0);
context = new EmitterContext(ShaderStage.Compute, header, capabilities, flags);
}
else
{
header = new ShaderHeader(code);
config = new ShaderConfig(new ShaderHeader(code), flags, queryInfoCallback);
cfg = Decoder.Decode(code, HeaderSize);
context = new EmitterContext(header.Stage, header, capabilities, flags);
}
if (cfg == null)
@ -172,6 +124,8 @@ namespace Ryujinx.Graphics.Shader.Translation
return new Operation[0];
}
EmitterContext context = new EmitterContext(config);
ulong maxEndAddress = 0;
for (int blkIndex = 0; blkIndex < cfg.Length; blkIndex++)