raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: add distinct mesh shader type

Browse Source
This commit is contained in:
nmzik
2026-10-01 17:46:54 +04:00
parent 337f432016
commit 11001c97f7
5 changed files with 159 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/renderDraw.cpp
+1
View File
@@ -1032,6 +1032,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
uint32_t mesh_groups = 0;
if (mesh_active) {
const auto& mesh = state.vertex_info[0].mesh;
EXIT_NOT_IMPLEMENTED(mesh.fast_launch && (draw.IsIndexed() || primitive_restart_enable));
static std::atomic_bool restart_warned = false;
if (primitive_restart_enable && !restart_warned.exchange(true, std::memory_order_relaxed)) {
std::printf("Warning: primitive restart is not implemented for mesh shaders; "
src/graphics/shader/recompiler/frontend/translate/Translate.cpp
+62 -53
View File
@@ -1138,12 +1138,13 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
return entry_ir.ISub(lhs, minimum(lhs, rhs));
};
const auto local = builtin(IR::StageInputKind::LocalInvocationIndex);
const auto primitive_chunk = entry_ir.IMul(builtin(IR::StageInputKind::WorkgroupId, 0),
u32(mesh.primitives_per_group));
const auto group = builtin(IR::StageInputKind::WorkgroupId, 0);
const auto primitive_chunk = mesh.fast_launch ? group :
entry_ir.IMul(group, u32(mesh.primitives_per_group));
const auto step = u32(mesh.InputPrimitiveStep());
const auto size = u32(mesh.InputPrimitiveSize());
const auto chunk = entry_ir.IMul(primitive_chunk, step);
const auto vertices =
const auto chunk = mesh.fast_launch ? group : entry_ir.IMul(primitive_chunk, step);
const auto vertices = mesh.fast_launch ? u32(mesh.vertices_per_group) :
minimum(subtract_saturate(draw(0), chunk), u32(mesh.vertices_per_group));
const auto primitives = entry_ir.Select(
entry_ir.ULessThan(vertices, size), u32(0),
@@ -1164,56 +1165,64 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
entry_ir.BitwiseOr(wave_info, entry_ir.BitwiseOr(entry_ir.ShiftLeftLogical(
primitive_count, u32(8)),
vertex_count)));
// GS adjacency addresses local ES records in LDS. Fans retain the draw's
// center in every subgroup; strip winding follows the global primitive.
const auto vertex = entry_ir.IMul(local, step);
auto first = vertex;
auto second = u32(0);
auto third = u32(0);
if (mesh.InputPrimitiveSize() >= 2u) {
second = entry_ir.IAdd(vertex, u32(1));
if (mesh.fast_launch) {
// Fast launch broadcasts the subgroup's base vertex and instance to every lane.
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5), entry_ir.IAdd(draw(1), group));
entry_ir.SetVectorReg(
static_cast<IR::VectorReg>(6),
entry_ir.IAdd(draw(2), builtin(IR::StageInputKind::WorkgroupId, 1)));
} else {
// GS adjacency addresses local ES records in LDS. Fans retain the draw's
// center in every subgroup; strip winding follows the global primitive.
const auto vertex = entry_ir.IMul(local, step);
auto first = vertex;
auto second = u32(0);
auto third = u32(0);
if (mesh.InputPrimitiveSize() >= 2u) {
second = entry_ir.IAdd(vertex, u32(1));
}
if (mesh.InputPrimitiveSize() == 3u) {
third = entry_ir.IAdd(vertex, u32(2));
}
auto input_vertex = entry_ir.IAdd(chunk, local);
if (mesh.input_primitive == static_cast<uint32_t>(Prospero::PrimitiveType::kTriFan)) {
first = u32(0);
input_vertex = entry_ir.Select(entry_ir.IEqual(local, u32(0)), u32(0), input_vertex);
} else if (mesh.input_primitive == static_cast<uint32_t>(Prospero::PrimitiveType::kTriStrip)) {
const auto parity = entry_ir.BitwiseAnd(entry_ir.IAdd(primitive_chunk, local), u32(1));
first = entry_ir.IAdd(first, parity);
second = entry_ir.ISub(second, parity);
}
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(0),
entry_ir.BitwiseOr(entry_ir.ShiftLeftLogical(first, u32(2)),
entry_ir.ShiftLeftLogical(second, u32(18))));
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(1),
entry_ir.ShiftLeftLogical(third, u32(2)));
const auto index_bytes = draw(3);
const auto indexed = entry_ir.INotEqual(index_bytes, u32(0));
const auto index_low = draw(4);
const auto byte_offset = entry_ir.IAdd(entry_ir.BitwiseAnd(index_low, u32(3)),
entry_ir.IMul(input_vertex, index_bytes));
const auto index_resource = entry_ir.Emit(
IR::ValueOpcode::GetAddressResource,
{entry_ir.BitwiseAnd(index_low, u32(~3u)), draw(5)});
const auto memory_index = static_cast<uint32_t>(result.memory_info.size());
result.memory_info.push_back({.kind = IR::ResourceKind::Global});
const auto packed_index = entry_ir.Emit(
IR::ValueOpcode::LoadAddressU32,
{index_resource, entry_ir.BitwiseAnd(byte_offset, u32(~3u)), u32(0),
entry_ir.LogicalAnd(indexed, entry_ir.ULessThan(local, vertices))},
IR::MemoryFlags {.index = memory_index});
const auto index = IR::U32(entry_ir.Emit(
IR::ValueOpcode::BitFieldUExtract,
{packed_index, entry_ir.IMul(entry_ir.BitwiseAnd(byte_offset, u32(3)), u32(8)),
entry_ir.IMul(index_bytes, u32(8))}));
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5),
entry_ir.IAdd(draw(1), entry_ir.Select(indexed, index, input_vertex)));
entry_ir.SetVectorReg(
static_cast<IR::VectorReg>(8),
entry_ir.IAdd(draw(2), builtin(IR::StageInputKind::WorkgroupId, 1)));
}
if (mesh.InputPrimitiveSize() == 3u) {
third = entry_ir.IAdd(vertex, u32(2));
}
auto input_vertex = entry_ir.IAdd(chunk, local);
if (mesh.input_primitive == static_cast<uint32_t>(Prospero::PrimitiveType::kTriFan)) {
first = u32(0);
input_vertex = entry_ir.Select(entry_ir.IEqual(local, u32(0)), u32(0), input_vertex);
} else if (mesh.input_primitive == static_cast<uint32_t>(Prospero::PrimitiveType::kTriStrip)) {
const auto parity = entry_ir.BitwiseAnd(entry_ir.IAdd(primitive_chunk, local), u32(1));
first = entry_ir.IAdd(first, parity);
second = entry_ir.ISub(second, parity);
}
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(0),
entry_ir.BitwiseOr(entry_ir.ShiftLeftLogical(first, u32(2)),
entry_ir.ShiftLeftLogical(second, u32(18))));
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(1),
entry_ir.ShiftLeftLogical(third, u32(2)));
const auto index_bytes = draw(3);
const auto indexed = entry_ir.INotEqual(index_bytes, u32(0));
const auto index_low = draw(4);
const auto byte_offset = entry_ir.IAdd(entry_ir.BitwiseAnd(index_low, u32(3)),
entry_ir.IMul(input_vertex, index_bytes));
const auto index_resource = entry_ir.Emit(
IR::ValueOpcode::GetAddressResource,
{entry_ir.BitwiseAnd(index_low, u32(~3u)), draw(5)});
const auto memory_index = static_cast<uint32_t>(result.memory_info.size());
result.memory_info.push_back({.kind = IR::ResourceKind::Global});
const auto packed_index = entry_ir.Emit(
IR::ValueOpcode::LoadAddressU32,
{index_resource, entry_ir.BitwiseAnd(byte_offset, u32(~3u)), u32(0),
entry_ir.LogicalAnd(indexed, entry_ir.ULessThan(local, vertices))},
IR::MemoryFlags {.index = memory_index});
const auto index = IR::U32(entry_ir.Emit(
IR::ValueOpcode::BitFieldUExtract,
{packed_index, entry_ir.IMul(entry_ir.BitwiseAnd(byte_offset, u32(3)), u32(8)),
entry_ir.IMul(index_bytes, u32(8))}));
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5),
entry_ir.IAdd(draw(1), entry_ir.Select(indexed, index, input_vertex)));
entry_ir.SetVectorReg(
static_cast<IR::VectorReg>(8),
entry_ir.IAdd(draw(2), builtin(IR::StageInputKind::WorkgroupId, 1)));
} else if (options.stage == ShaderType::Local) {
entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3), IR::U32(IR::Value(64u)));
entry_ir.SetVectorReg(static_cast<IR::VectorReg>(2),
src/graphics/shader/shader.cpp
+20 -4
View File
@@ -617,7 +617,8 @@ void BuildStageStaticKey(const ShaderVertexInputInfo& info, std::vector<uint32_t
key.insert(key.end(), {mesh.wave_size, mesh.host_subgroup_size, mesh.lds_size_dwords,
mesh.scratch_size_dwords, mesh.input_primitive,
mesh.primitives_per_group, mesh.vertices_per_group,
mesh.max_vertices, mesh.max_primitives, mesh.provoking_vertex});
mesh.max_vertices, mesh.max_primitives, mesh.provoking_vertex,
static_cast<uint32_t>(mesh.fast_launch)});
}
key.push_back(info.tess.input_control_points);
if (info.tess.input_control_points != 0) {
@@ -719,6 +720,9 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
mesh.provoking_vertex = context.GetModeControl().provoking_vtx_last ? 2u : 0u;
mesh.lds_size_dwords = static_cast<uint32_t>(regs.gs_regs.rsrc2.lds_size) * 128u;
mesh.scratch_size_dwords = data.scratch_size_dwords;
const auto fast_launch = (context.GetShaderStages() >> 19u) & 3u;
EXIT_NOT_IMPLEMENTED(fast_launch > 1u);
mesh.fast_launch = fast_launch != 0;
if (data.type == Prospero::ShaderBinaryType::kGsFront) {
EXIT_IF(regs.gs_regs.data_addr == 0);
const auto [back, back_hash] = ShaderGetMappedData(regs.gs_regs.data_addr, "ShaderGetInputInfoGS():");
@@ -731,9 +735,20 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
params.hash = XXH3_64bits(hashes, sizeof(hashes));
mesh.scratch_size_dwords = std::max(mesh.scratch_size_dwords, back.scratch_size_dwords);
}
EXIT_NOT_IMPLEMENTED(regs.gs_regs.rsrc1.gs_vgpr_component_count != 3u ||
regs.gs_regs.rsrc2.es_vgpr_component_count != 3u);
const auto& group = user_config.GetGeControl();
if (mesh.fast_launch) {
const auto& user_vgpr = user_config.GetGeUserVgprEn();
EXIT_NOT_IMPLEMENTED(data.type != Prospero::ShaderBinaryType::kGs ||
regs.gs_regs.rsrc1.gs_vgpr_component_count != 0u ||
regs.gs_regs.rsrc2.es_vgpr_component_count != 1u ||
user_config.GetPrimType() != Prospero::PrimitiveType::kPointList ||
group.primitive_group_size != 1u || group.vertex_group_size != 1u ||
user_vgpr.vgpr1 || user_vgpr.vgpr2 || user_vgpr.vgpr3 ||
mesh.max_vertices != sh.m_vgtGsMaxVertOut);
} else {
EXIT_NOT_IMPLEMENTED(regs.gs_regs.rsrc1.gs_vgpr_component_count != 3u ||
regs.gs_regs.rsrc2.es_vgpr_component_count != 3u);
}
if ((user_config.GetPrimType() != Prospero::PrimitiveType::kPointList &&
user_config.GetPrimType() != Prospero::PrimitiveType::kLineList &&
user_config.GetPrimType() != Prospero::PrimitiveType::kTriFan &&
@@ -746,7 +761,8 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
mesh.input_primitive, sh.m_vgtGsOutPrimType, sh.m_vgtGsMaxVertOut,
group.primitive_group_size, group.vertex_group_size, mesh.max_vertices);
}
mesh.max_primitives = group.primitive_group_size * (sh.m_vgtGsMaxVertOut - 2u);
mesh.max_primitives = mesh.fast_launch ? mesh.max_vertices :
group.primitive_group_size * (sh.m_vgtGsMaxVertOut - 2u);
mesh.primitives_per_group = std::min({static_cast<uint32_t>(group.primitive_group_size),
mesh.InputPrimitiveCount(group.vertex_group_size),
mesh.max_vertices / sh.m_vgtGsMaxVertOut});
src/graphics/shader/shader.h
+1
View File
@@ -88,6 +88,7 @@ struct ShaderMeshInputInfo: ShaderWorkgroupInputInfo {
uint32_t max_vertices = 0;
uint32_t max_primitives = 0;
uint32_t provoking_vertex = 0;
bool fast_launch = false;
[[nodiscard]] constexpr uint32_t InputPrimitiveSize() const {
switch (static_cast<Prospero::PrimitiveType>(input_primitive)) {