raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-23 04:16:57 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: interpolate f16 parameter halves (V_INTERP_P1LL_F16, V_INTERP_P2_F16) (#1792)

Browse Source 
shader: interpolate packed f16 attributes for v_interp_p1ll_f16 and v_interp_p2_f16

Co-authored-by: nmzik <Nmzik@mail.ru>
This commit is contained in:
Nawid3333
2026-10-22 04:54:10 +02:00
committed by GitHub
parent 8502d8a77d
commit d7c1896647
10 changed files with 181 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterFlow.cpp
+81 -63
View File
@@ -4,6 +4,7 @@
#include <algorithm>
#include <atomic>
#include <optional>
namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
namespace {
@@ -140,6 +141,74 @@ uint32_t EmitDppWriteCondition(ValueEmitContext& ctx, const IR::DppMoveFlags& fl
return writable;
}
uint32_t LoadVertexParameter(EmitterState& state, const InputBinding& input, uint32_t vertex,
uint32_t chan) {
const auto pointer = state.builder.AllocateId();
const auto value = state.builder.AllocateId();
state.builder.AddFunction(spv::OpAccessChain,
TypePointer(state, spv::StorageClassInput, TypeF32(state)), pointer,
input.variable_id, ConstantU32(state, vertex),
ConstantU32(state, chan & 3u));
state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
return value;
}
// Interpolates a pixel input channel to f32. A half (0 low, 1 high) selects one f16 of a packed
// FP16_INTERP_MODE dword before blending; smooth inputs read by halves are always per-vertex.
uint32_t EmitPixelAttribute(EmitterState& state, const InputBinding& input, uint32_t attr,
uint32_t chan, std::optional<uint32_t> half) {
const auto select = [&](uint32_t value) {
if (!half) {
return value;
}
const auto bits = state.builder.AllocateId();
const auto unpacked = state.builder.AllocateId();
const auto word = state.builder.AllocateId();
state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
state.builder.AddFunction(spv::OpExtInst, TypeF32Vector(state, 2), unpacked,
GlslStd450(state), GLSLstd450UnpackHalf2x16, bits);
state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), word, unpacked, *half);
return word;
};
if (!input.per_vertex) {
const auto vector = state.builder.AllocateId();
const auto component = state.builder.AllocateId();
state.builder.AddFunction(spv::OpLoad, TypeF32Vector(state, 4), vector, input.variable_id);
state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), component, vector,
chan & 3u);
return select(component);
}
if (PixelParameterIsFlat(state, attr)) {
const auto vertex =
state.input_info.pixel->parameter_mode == ShaderPixelParameterMode::LastVertex ? 2u : 0u;
return select(LoadVertexParameter(state, input, vertex, chan));
}
const auto barycentric_kind = state.input_info.pixel->ps_no_perspective
? IR::StageInputKind::BaryCoordNoPerspective
: IR::StageInputKind::BaryCoordSmooth;
const auto barycentric = InputVariableForKind(state, barycentric_kind);
uint32_t sum = 0;
for (uint32_t vertex = 0; vertex < 3u; vertex++) {
const auto pointer = state.builder.AllocateId();
const auto weight = state.builder.AllocateId();
const auto product = state.builder.AllocateId();
state.builder.AddFunction(spv::OpAccessChain,
TypePointer(state, spv::StorageClassInput, TypeF32(state)),
pointer, barycentric, ConstantU32(state, vertex));
state.builder.AddFunction(spv::OpLoad, TypeF32(state), weight, pointer);
state.builder.AddFunction(spv::OpFMul, TypeF32(state), product,
select(LoadVertexParameter(state, input, vertex, chan)), weight);
if (vertex == 0u) {
sum = product;
} else {
const auto next = state.builder.AllocateId();
state.builder.AddFunction(spv::OpFAdd, TypeF32(state), next, sum, product);
sum = next;
}
}
return sum;
}
uint32_t EmitAttribute(EmitterState& state, uint32_t attr, uint32_t chan) {
const auto* input = InputBindingForParameter(state, attr);
if (input == nullptr || input->variable_id == 0) {
@@ -148,57 +217,9 @@ uint32_t EmitAttribute(EmitterState& state, uint32_t attr, uint32_t chan) {
if (state.program.stage == ShaderType::Vertex || state.program.stage == ShaderType::Local) {
return EmitVertexParameterComponentU32(state, *input, chan & 3u);
}
const auto load_per_vertex = [&](uint32_t vertex) {
const auto pointer = state.builder.AllocateId();
const auto value = state.builder.AllocateId();
state.builder.AddFunction(
spv::OpAccessChain, TypePointer(state, spv::StorageClassInput, TypeF32(state)), pointer,
input->variable_id, ConstantU32(state, vertex), ConstantU32(state, chan & 3u));
state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
return value;
};
if (input->per_vertex) {
if (PixelParameterIsFlat(state, attr)) {
const auto value = load_per_vertex(
state.input_info.pixel->parameter_mode == ShaderPixelParameterMode::LastVertex ? 2u : 0u);
const auto bits = state.builder.AllocateId();
state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
return bits;
}
const auto barycentric_kind = state.input_info.pixel->ps_no_perspective
? IR::StageInputKind::BaryCoordNoPerspective
: IR::StageInputKind::BaryCoordSmooth;
const auto barycentric = InputVariableForKind(state, barycentric_kind);
uint32_t sum = 0;
for (uint32_t vertex = 0; vertex < 3u; vertex++) {
const auto pointer = state.builder.AllocateId();
const auto weight = state.builder.AllocateId();
const auto product = state.builder.AllocateId();
state.builder.AddFunction(spv::OpAccessChain,
TypePointer(state, spv::StorageClassInput, TypeF32(state)),
pointer, barycentric, ConstantU32(state, vertex));
state.builder.AddFunction(spv::OpLoad, TypeF32(state), weight, pointer);
state.builder.AddFunction(spv::OpFMul, TypeF32(state), product, load_per_vertex(vertex),
weight);
if (vertex == 0u) {
sum = product;
} else {
const auto next = state.builder.AllocateId();
state.builder.AddFunction(spv::OpFAdd, TypeF32(state), next, sum, product);
sum = next;
}
}
const auto bits = state.builder.AllocateId();
state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, sum);
return bits;
}
const auto vector = state.builder.AllocateId();
const auto component = state.builder.AllocateId();
const auto bits = state.builder.AllocateId();
state.builder.AddFunction(spv::OpLoad, TypeF32Vector(state, 4), vector, input->variable_id);
state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), component, vector,
chan & 3u);
state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, component);
const auto value = EmitPixelAttribute(state, *input, attr, chan, std::nullopt);
const auto bits = state.builder.AllocateId();
state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
return bits;
}
@@ -209,25 +230,16 @@ uint32_t EmitInterpolationParameter(ValueEmitContext& ctx, uint32_t attr, uint32
if (!input->per_vertex) {
return EmitAttribute(ctx.state, attr, chan);
}
const auto load_vertex = [&](uint32_t vertex) {
const auto pointer = state.builder.AllocateId();
const auto value = state.builder.AllocateId();
state.builder.AddFunction(
spv::OpAccessChain, TypePointer(state, spv::StorageClassInput, TypeF32(state)), pointer,
input->variable_id, ConstantU32(state, vertex), ConstantU32(state, chan & 3u));
state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
return value;
};
uint32_t selected_vertex = (mode + 1u) % 3u;
if (mode == 2u && PixelParameterIsFlat(state, attr)) {
selected_vertex =
state.input_info.pixel->parameter_mode == ShaderPixelParameterMode::LastVertex ? 2u : 0u;
}
uint32_t value = load_vertex(selected_vertex);
uint32_t value = LoadVertexParameter(state, *input, selected_vertex, chan);
if (!PixelParameterIsCustom(state, attr) && mode < 2u) {
const auto delta = state.builder.AllocateId();
state.builder.AddFunction(spv::OpFSub, TypeF32(state), delta, value, load_vertex(0));
state.builder.AddFunction(spv::OpFSub, TypeF32(state), delta, value,
LoadVertexParameter(state, *input, 0, chan));
value = delta;
}
const auto bits = state.builder.AllocateId();
@@ -853,6 +865,12 @@ uint32_t EmitGetAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
return EmitAttribute(ctx.state, inst.Arg(0).U32(), inst.Arg(1).U32());
}
uint32_t EmitGetAttributeF16(ValueEmitContext& ctx, const IR::Inst& inst) {
const auto attr = inst.Arg(0).U32();
return EmitPixelAttribute(ctx.state, *InputBindingForParameter(ctx.state, attr), attr,
inst.Arg(1).U32(), inst.Arg(2).U32());
}
uint32_t EmitGetInterpolationParameter(ValueEmitContext& ctx, const IR::Inst& inst) {
return EmitInterpolationParameter(ctx, inst.Arg(0).U32(), inst.Arg(1).U32(), inst.Arg(2).U32());
}