summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/renderer_vulkan/vk_shader_decompiler.cpp83
-rw-r--r--src/video_core/renderer_vulkan/vk_shader_decompiler.h4
2 files changed, 49 insertions, 38 deletions
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index fae8c95c8..a8baf91de 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -543,7 +543,7 @@ private:
}
for (u32 rt = 0; rt < static_cast<u32>(frag_colors.size()); ++rt) {
- if (!IsRenderTargetUsed(rt)) {
+ if (!specialization.enabled_rendertargets[rt]) {
continue;
}
@@ -1555,40 +1555,48 @@ private:
Expression Texture(Operation operation) {
const auto& meta = std::get<MetaTexture>(operation.GetMeta());
- UNIMPLEMENTED_IF(!meta.aoffi.empty());
const bool can_implicit = stage == ShaderType::Fragment;
const Id sampler = GetTextureSampler(operation);
const Id coords = GetCoordinates(operation, Type::Float);
+ std::vector<Id> operands;
+ spv::ImageOperandsMask mask{};
+ if (meta.bias) {
+ mask = mask | spv::ImageOperandsMask::Bias;
+ operands.push_back(AsFloat(Visit(meta.bias)));
+ }
+
+ if (!can_implicit) {
+ mask = mask | spv::ImageOperandsMask::Lod;
+ operands.push_back(v_float_zero);
+ }
+
+ if (!meta.aoffi.empty()) {
+ mask = mask | spv::ImageOperandsMask::Offset;
+ operands.push_back(GetOffsetCoordinates(operation));
+ }
+
if (meta.depth_compare) {
// Depth sampling
UNIMPLEMENTED_IF(meta.bias);
const Id dref = AsFloat(Visit(meta.depth_compare));
if (can_implicit) {
- return {OpImageSampleDrefImplicitLod(t_float, sampler, coords, dref, {}),
- Type::Float};
+ return {
+ OpImageSampleDrefImplicitLod(t_float, sampler, coords, dref, mask, operands),
+ Type::Float};
} else {
- return {OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref,
- spv::ImageOperandsMask::Lod, v_float_zero),
- Type::Float};
+ return {
+ OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, operands),
+ Type::Float};
}
}
- std::vector<Id> operands;
- spv::ImageOperandsMask mask{};
- if (meta.bias) {
- mask = mask | spv::ImageOperandsMask::Bias;
- operands.push_back(AsFloat(Visit(meta.bias)));
- }
-
Id texture;
if (can_implicit) {
texture = OpImageSampleImplicitLod(t_float4, sampler, coords, mask, operands);
} else {
- texture = OpImageSampleExplicitLod(t_float4, sampler, coords,
- mask | spv::ImageOperandsMask::Lod, v_float_zero,
- operands);
+ texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, operands);
}
return GetTextureElement(operation, texture, Type::Float);
}
@@ -1601,7 +1609,8 @@ private:
const Id lod = AsFloat(Visit(meta.lod));
spv::ImageOperandsMask mask = spv::ImageOperandsMask::Lod;
- std::vector<Id> operands;
+ std::vector<Id> operands{lod};
+
if (!meta.aoffi.empty()) {
mask = mask | spv::ImageOperandsMask::Offset;
operands.push_back(GetOffsetCoordinates(operation));
@@ -1609,11 +1618,10 @@ private:
if (meta.sampler.IsShadow()) {
const Id dref = AsFloat(Visit(meta.depth_compare));
- return {
- OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, lod, operands),
- Type::Float};
+ return {OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, operands),
+ Type::Float};
}
- const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, lod, operands);
+ const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, operands);
return GetTextureElement(operation, texture, Type::Float);
}
@@ -1722,7 +1730,7 @@ private:
const std::vector grad = {dx, dy};
static constexpr auto mask = spv::ImageOperandsMask::Grad;
- const Id texture = OpImageSampleImplicitLod(t_float4, sampler, coords, mask, grad);
+ const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, grad);
return GetTextureElement(operation, texture, Type::Float);
}
@@ -1833,7 +1841,7 @@ private:
}
void PreExit() {
- if (stage == ShaderType::Vertex) {
+ if (stage == ShaderType::Vertex && specialization.ndc_minus_one_to_one) {
const u32 position_index = out_indices.position.value();
const Id z_pointer = AccessElement(t_out_float, out_vertex, position_index, 2U);
const Id w_pointer = AccessElement(t_out_float, out_vertex, position_index, 3U);
@@ -1860,12 +1868,18 @@ private:
// rendertargets/components are skipped in the register assignment.
u32 current_reg = 0;
for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
+ if (!specialization.enabled_rendertargets[rt]) {
+ // Skip rendertargets that are not enabled
+ continue;
+ }
// TODO(Subv): Figure out how dual-source blending is configured in the Switch.
for (u32 component = 0; component < 4; ++component) {
+ const Id pointer = AccessElement(t_out_float, frag_colors.at(rt), component);
if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
- OpStore(AccessElement(t_out_float, frag_colors.at(rt), component),
- SafeGetRegister(current_reg));
+ OpStore(pointer, SafeGetRegister(current_reg));
++current_reg;
+ } else {
+ OpStore(pointer, component == 3 ? v_float_one : v_float_zero);
}
}
}
@@ -1995,15 +2009,6 @@ private:
return DeclareBuiltIn(builtin, spv::StorageClass::Input, type, std::move(name));
}
- bool IsRenderTargetUsed(u32 rt) const {
- for (u32 component = 0; component < 4; ++component) {
- if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
- return true;
- }
- }
- return false;
- }
-
template <typename... Args>
Id AccessElement(Id pointer_type, Id composite, Args... elements_) {
std::vector<Id> members;
@@ -2567,7 +2572,7 @@ public:
const Id target = decomp.Constant(decomp.t_uint, expr.value);
Id gpr = decomp.OpLoad(decomp.t_float, decomp.registers.at(expr.gpr));
gpr = decomp.OpBitcast(decomp.t_uint, gpr);
- return decomp.OpLogicalEqual(decomp.t_uint, gpr, target);
+ return decomp.OpIEqual(decomp.t_bool, gpr, target);
}
Id Visit(const Expr& node) {
@@ -2637,11 +2642,11 @@ public:
const Id loop_label = decomp.OpLabel();
const Id endloop_label = decomp.OpLabel();
const Id loop_start_block = decomp.OpLabel();
- const Id loop_end_block = decomp.OpLabel();
+ const Id loop_continue_block = decomp.OpLabel();
current_loop_exit = endloop_label;
decomp.OpBranch(loop_label);
decomp.AddLabel(loop_label);
- decomp.OpLoopMerge(endloop_label, loop_end_block, spv::LoopControlMask::MaskNone);
+ decomp.OpLoopMerge(endloop_label, loop_continue_block, spv::LoopControlMask::MaskNone);
decomp.OpBranch(loop_start_block);
decomp.AddLabel(loop_start_block);
ASTNode current = ast.nodes.GetFirst();
@@ -2649,6 +2654,8 @@ public:
Visit(current);
current = current->GetNext();
}
+ decomp.OpBranch(loop_continue_block);
+ decomp.AddLabel(loop_continue_block);
ExprDecompiler expr_parser{decomp};
const Id condition = expr_parser.Visit(ast.condition);
decomp.OpBranchConditional(condition, loop_label, endloop_label);
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
index 2b01321b6..10794be1c 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
@@ -94,6 +94,7 @@ struct Specialization final {
Maxwell::PrimitiveTopology primitive_topology{};
std::optional<float> point_size{};
std::array<Maxwell::VertexAttribute::Type, Maxwell::NumVertexAttributes> attribute_types{};
+ bool ndc_minus_one_to_one{};
// Tessellation specific
struct {
@@ -101,6 +102,9 @@ struct Specialization final {
Maxwell::TessellationSpacing spacing{};
bool clockwise{};
} tessellation;
+
+ // Fragment specific
+ std::bitset<8> enabled_rendertargets;
};
// Old gcc versions don't consider this trivially copyable.
// static_assert(std::is_trivially_copyable_v<Specialization>);
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-10 13:42:02 +0100