aboutsummaryrefslogtreecommitdiff
path: root/src/video_core/renderer_vulkan
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/renderer_vulkan')
-rw-r--r--src/video_core/renderer_vulkan/vk_pipeline_cache.cpp3
-rw-r--r--src/video_core/renderer_vulkan/vk_shader_decompiler.cpp95
-rw-r--r--src/video_core/renderer_vulkan/vk_shader_decompiler.h3
3 files changed, 49 insertions, 52 deletions
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 48e23d4cd..7ddf7d3ee 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -325,9 +325,6 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
specialization.tessellation.primitive = fixed_state.tessellation.primitive;
specialization.tessellation.spacing = fixed_state.tessellation.spacing;
specialization.tessellation.clockwise = fixed_state.tessellation.clockwise;
- for (const auto& rt : key.renderpass_params.color_attachments) {
- specialization.enabled_rendertargets.set(rt.index);
- }
SPIRVProgram program;
std::vector<vk::DescriptorSetLayoutBinding> bindings;
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index dd6d2ef03..1ab22251e 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -542,11 +542,10 @@ private:
return;
}
- for (u32 rt = 0; rt < static_cast<u32>(frag_colors.size()); ++rt) {
- if (!specialization.enabled_rendertargets[rt]) {
+ for (u32 rt = 0; rt < static_cast<u32>(std::size(frag_colors)); ++rt) {
+ if (!IsRenderTargetEnabled(rt)) {
continue;
}
-
const Id id = AddGlobalVariable(OpVariable(t_out_float4, spv::StorageClass::Output));
Name(id, fmt::format("frag_color{}", rt));
Decorate(id, spv::Decoration::Location, rt);
@@ -852,6 +851,15 @@ private:
return binding;
}
+ bool IsRenderTargetEnabled(u32 rt) const {
+ for (u32 component = 0; component < 4; ++component) {
+ if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
+ return true;
+ }
+ }
+ return false;
+ }
+
bool IsInputAttributeArray() const {
return stage == ShaderType::TesselationControl || stage == ShaderType::TesselationEval ||
stage == ShaderType::Geometry;
@@ -1115,15 +1123,7 @@ private:
}
if (const auto gmem = std::get_if<GmemNode>(&*node)) {
- const Id gmem_buffer = global_buffers.at(gmem->GetDescriptor());
- const Id real = AsUint(Visit(gmem->GetRealAddress()));
- const Id base = AsUint(Visit(gmem->GetBaseAddress()));
-
- Id offset = OpISub(t_uint, real, base);
- offset = OpUDiv(t_uint, offset, Constant(t_uint, 4U));
- return {OpLoad(t_float,
- OpAccessChain(t_gmem_float, gmem_buffer, Constant(t_uint, 0U), offset)),
- Type::Float};
+ return {OpLoad(t_uint, GetGlobalMemoryPointer(*gmem)), Type::Uint};
}
if (const auto lmem = std::get_if<LmemNode>(&*node)) {
@@ -1134,10 +1134,7 @@ private:
}
if (const auto smem = std::get_if<SmemNode>(&*node)) {
- Id address = AsUint(Visit(smem->GetAddress()));
- address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
- const Id pointer = OpAccessChain(t_smem_uint, shared_memory, address);
- return {OpLoad(t_uint, pointer), Type::Uint};
+ return {OpLoad(t_uint, GetSharedMemoryPointer(*smem)), Type::Uint};
}
if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) {
@@ -1331,20 +1328,10 @@ private:
target = {OpAccessChain(t_prv_float, local_memory, address), Type::Float};
} else if (const auto smem = std::get_if<SmemNode>(&*dest)) {
- ASSERT(stage == ShaderType::Compute);
- Id address = AsUint(Visit(smem->GetAddress()));
- address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
- target = {OpAccessChain(t_smem_uint, shared_memory, address), Type::Uint};
+ target = {GetSharedMemoryPointer(*smem), Type::Uint};
} else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
- const Id real = AsUint(Visit(gmem->GetRealAddress()));
- const Id base = AsUint(Visit(gmem->GetBaseAddress()));
- const Id diff = OpISub(t_uint, real, base);
- const Id offset = OpShiftRightLogical(t_uint, diff, Constant(t_uint, 2));
-
- const Id gmem_buffer = global_buffers.at(gmem->GetDescriptor());
- target = {OpAccessChain(t_gmem_float, gmem_buffer, Constant(t_uint, 0), offset),
- Type::Float};
+ target = {GetGlobalMemoryPointer(*gmem), Type::Uint};
} else {
UNIMPLEMENTED();
@@ -1796,11 +1783,16 @@ private:
return {};
}
- Expression UAtomicAdd(Operation operation) {
- const auto& smem = std::get<SmemNode>(*operation[0]);
- Id address = AsUint(Visit(smem.GetAddress()));
- address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
- const Id pointer = OpAccessChain(t_smem_uint, shared_memory, address);
+ Expression AtomicAdd(Operation operation) {
+ Id pointer;
+ if (const auto smem = std::get_if<SmemNode>(&*operation[0])) {
+ pointer = GetSharedMemoryPointer(*smem);
+ } else if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) {
+ pointer = GetGlobalMemoryPointer(*gmem);
+ } else {
+ UNREACHABLE();
+ return {Constant(t_uint, 0), Type::Uint};
+ }
const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
const Id semantics = Constant(t_uint, 0U);
@@ -1889,19 +1881,14 @@ 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(pointer, SafeGetRegister(current_reg));
- ++current_reg;
- } else {
- OpStore(pointer, component == 3 ? v_float_one : v_float_zero);
+ if (!header.ps.IsColorComponentOutputEnabled(rt, component)) {
+ continue;
}
+ const Id pointer = AccessElement(t_out_float, frag_colors[rt], component);
+ OpStore(pointer, SafeGetRegister(current_reg));
+ ++current_reg;
}
}
if (header.ps.omap.depth) {
@@ -2240,6 +2227,22 @@ private:
return {};
}
+ Id GetGlobalMemoryPointer(const GmemNode& gmem) {
+ const Id real = AsUint(Visit(gmem.GetRealAddress()));
+ const Id base = AsUint(Visit(gmem.GetBaseAddress()));
+ const Id diff = OpISub(t_uint, real, base);
+ const Id offset = OpShiftRightLogical(t_uint, diff, Constant(t_uint, 2));
+ const Id buffer = global_buffers.at(gmem.GetDescriptor());
+ return OpAccessChain(t_gmem_uint, buffer, Constant(t_uint, 0), offset);
+ }
+
+ Id GetSharedMemoryPointer(const SmemNode& smem) {
+ ASSERT(stage == ShaderType::Compute);
+ Id address = AsUint(Visit(smem.GetAddress()));
+ address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
+ return OpAccessChain(t_smem_uint, shared_memory, address);
+ }
+
static constexpr std::array operation_decompilers = {
&SPIRVDecompiler::Assign,
@@ -2386,7 +2389,7 @@ private:
&SPIRVDecompiler::AtomicImageXor,
&SPIRVDecompiler::AtomicImageExchange,
- &SPIRVDecompiler::UAtomicAdd,
+ &SPIRVDecompiler::AtomicAdd,
&SPIRVDecompiler::Branch,
&SPIRVDecompiler::BranchIndirect,
@@ -2482,9 +2485,9 @@ private:
Id t_smem_uint{};
- const Id t_gmem_float = TypePointer(spv::StorageClass::StorageBuffer, t_float);
+ const Id t_gmem_uint = TypePointer(spv::StorageClass::StorageBuffer, t_uint);
const Id t_gmem_array =
- Name(Decorate(TypeRuntimeArray(t_float), spv::Decoration::ArrayStride, 4U), "GmemArray");
+ Name(Decorate(TypeRuntimeArray(t_uint), spv::Decoration::ArrayStride, 4U), "GmemArray");
const Id t_gmem_struct = MemberDecorate(
Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0);
const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct);
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
index 10794be1c..f5dc14d9e 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
@@ -102,9 +102,6 @@ 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 2025-12-18 00:37:35 +0000