Mirrors/Ryujinx
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Implement shader storage buffer operations using new Load/Store instructions (#4993)

Browse source 
* Implement storage buffer operations using new Load/Store instruction

* Extend GenerateMultiTargetStorageOp to also match access with constant offset, and log and comments

* Remove now unused code

* Catch more complex cases of global memory usage

* Shader cache version bump

* Extend global access elimination to work with more shared memory cases

* Change alignment requirement from 16 bytes to 8 bytes, handle cases where we need more than 16 storage buffers

* Tweak preferencing to catch more cases

* Enable CB0 elimination even when host storage buffer alignment is > 16 (for Intel)

* Fix storage buffer bindings

* Simplify some code

* Shader cache version bump

* Fix typo

* Extend global memory elimination to handle shared memory with multiple possible offsets and local memory
This commit is contained in:
gdkchan 2023-06-03 20:12:18 -03:00 • committed by GitHub
parent 81c9052847
commit 21c9ac6240
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
42 changed files with 1468 additions and 1259 deletions
Download patch file Download diff file Expand all files Collapse all files

56
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs
View file

@ -68,33 +68,45 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
string args = string.Empty;
for (int argIndex = 0; argIndex < arity; argIndex++)
if (atomic && operation.StorageKind == StorageKind.StorageBuffer)
{
args = GenerateLoadOrStore(context, operation, isStore: false);
AggregateType dstType = operation.Inst == Instruction.AtomicMaxS32 || operation.Inst == Instruction.AtomicMinS32
? AggregateType.S32
: AggregateType.U32;
for (int argIndex = operation.SourcesCount - arity + 2; argIndex < operation.SourcesCount; argIndex++)
{
args += ", " + GetSoureExpr(context, operation.GetSource(argIndex), dstType);
}
}
else if (atomic && operation.StorageKind == StorageKind.SharedMemory)
{
args = LoadShared(context, operation);
// For shared memory access, the second argument is unused and should be ignored.
// It is there to make both storage and shared access have the same number of arguments.
// For storage, both inputs are consumed when the argument index is 0, so we should skip it here.
if (argIndex == 1 && (atomic || operation.StorageKind == StorageKind.SharedMemory))
{
continue;
}
if (argIndex != 0)
for (int argIndex = 2; argIndex < arity; argIndex++)
{
args += ", ";
}
if (argIndex == 0 && atomic)
AggregateType dstType = GetSrcVarType(inst, argIndex);
args += GetSoureExpr(context, operation.GetSource(argIndex), dstType);
}
}
else
{
for (int argIndex = 0; argIndex < arity; argIndex++)
{
switch (operation.StorageKind)
if (argIndex != 0)
{
case StorageKind.SharedMemory: args += LoadShared(context, operation); break;
case StorageKind.StorageBuffer: args += LoadStorage(context, operation); break;
default: throw new InvalidOperationException($"Invalid storage kind \"{operation.StorageKind}\".");
args += ", ";
}
}
else
{
AggregateType dstType = GetSrcVarType(inst, argIndex);
args += GetSoureExpr(context, operation.GetSource(argIndex), dstType);
@ -173,9 +185,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
case Instruction.LoadShared:
return LoadShared(context, operation);
case Instruction.LoadStorage:
return LoadStorage(context, operation);
case Instruction.Lod:
return Lod(context, operation);
@ -203,15 +212,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
case Instruction.StoreShared8:
return StoreShared8(context, operation);
case Instruction.StoreStorage:
return StoreStorage(context, operation);
case Instruction.StoreStorage16:
return StoreStorage16(context, operation);
case Instruction.StoreStorage8:
return StoreStorage8(context, operation);
case Instruction.TextureSample:
return TextureSample(context, operation);