hugopeixoto/yuzu
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

shader/memory: Implement LDG.U8 and unaligned U8 loads

Browse Source 
LDG can load single bytes instead of full integers or packs of integers.
These have the advantage of loading bytes that are not aligned to 4
bytes.

To emulate these this commit gets the byte being referenced (by doing
"address & 3" and then using that to extract the byte from the loaded
integer:

result = bitfieldExtract(loaded_integer, (address % 4) * 8, 8)
This commit is contained in:
ReinUsesLisp 2019-12-18 00:36:21 -03:00
parent 8825b88a45
commit ae8d4b6c0c
No known key found for this signature in database
GPG Key ID: 2DFC508897B39CFE
1 changed files with 32 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
src/video_core/shader/decode/memory.cpp
View File

@ -22,7 +22,23 @@ using Tegra::Shader::Register;
namespace { namespace {
u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) { u32 GetLdgMemorySize(Tegra::Shader::UniformType uniform_type) {
switch (uniform_type) {
case Tegra::Shader::UniformType::UnsignedByte:
case Tegra::Shader::UniformType::Single:
return 1;
case Tegra::Shader::UniformType::Double:
return 2;
case Tegra::Shader::UniformType::Quad:
case Tegra::Shader::UniformType::UnsignedQuad:
return 4;
default:
UNIMPLEMENTED_MSG("Unimplemented size={}!", static_cast<u32>(uniform_type));
return 1;
}
}
u32 GetStgMemorySize(Tegra::Shader::UniformType uniform_type) {
switch (uniform_type) { switch (uniform_type) {
case Tegra::Shader::UniformType::Single: case Tegra::Shader::UniformType::Single:
return 1; return 1;
@ -170,7 +186,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
const auto [real_address_base, base_address, descriptor] = const auto [real_address_base, base_address, descriptor] =
TrackGlobalMemory(bb, instr, false); TrackGlobalMemory(bb, instr, false);
const u32 count = GetUniformTypeElementsCount(type); const u32 count = GetLdgMemorySize(type);
if (!real_address_base || !base_address) { if (!real_address_base || !base_address) {
// Tracking failed, load zeroes. // Tracking failed, load zeroes.
for (u32 i = 0; i < count; ++i) { for (u32 i = 0; i < count; ++i) {
@ -181,12 +197,22 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
for (u32 i = 0; i < count; ++i) { for (u32 i = 0; i < count; ++i) {
const Node it_offset = Immediate(i * 4); const Node it_offset = Immediate(i * 4);
const Node real_address = const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);
Operation(OperationCode::UAdd, NO_PRECISE, real_address_base, it_offset); Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
if (type == Tegra::Shader::UniformType::UnsignedByte) {
// To handle unaligned loads get the byte used to dereferenced global memory
// and extract that byte from the loaded uint32.
Node byte = Operation(OperationCode::UBitwiseAnd, real_address, Immediate(3));
byte = Operation(OperationCode::ULogicalShiftLeft, std::move(byte), Immediate(3));
gmem = Operation(OperationCode::UBitfieldExtract, std::move(gmem), std::move(byte),
Immediate(8));
}
SetTemporary(bb, i, gmem); SetTemporary(bb, i, gmem);
} }
for (u32 i = 0; i < count; ++i) { for (u32 i = 0; i < count; ++i) {
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
} }
@ -276,7 +302,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
break; break;
} }
const u32 count = GetUniformTypeElementsCount(type); const u32 count = GetStgMemorySize(type);
for (u32 i = 0; i < count; ++i) { for (u32 i = 0; i < count; ++i) {
const Node it_offset = Immediate(i * 4); const Node it_offset = Immediate(i * 4);
const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset); const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset);