summaryrefslogtreecommitdiffstats
path: root/src/video_core/shader
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/shader')
-rw-r--r--src/video_core/shader/decode/texture.cpp55
-rw-r--r--src/video_core/shader/node.h75
-rw-r--r--src/video_core/shader/node_helper.h2
-rw-r--r--src/video_core/shader/registry.cpp20
-rw-r--r--src/video_core/shader/registry.h35
-rw-r--r--src/video_core/shader/shader_ir.h14
-rw-r--r--src/video_core/shader/track.cpp78
7 files changed, 185 insertions, 94 deletions
diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp
index 8f0bb996e..29ebf65ba 100644
--- a/src/video_core/shader/decode/texture.cpp
+++ b/src/video_core/shader/decode/texture.cpp
@@ -357,13 +357,11 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
return pc;
}
-ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(SamplerInfo info, u32 offset,
- std::optional<u32> buffer) {
+ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(
+ SamplerInfo info, std::optional<Tegra::Engines::SamplerDescriptor> sampler) {
if (info.IsComplete()) {
return info;
}
- const auto sampler = buffer ? registry.ObtainBindlessSampler(*buffer, offset)
- : registry.ObtainBoundSampler(offset);
if (!sampler) {
LOG_WARNING(HW_GPU, "Unknown sampler info");
info.type = info.type.value_or(Tegra::Shader::TextureType::Texture2D);
@@ -381,8 +379,8 @@ ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(SamplerInfo info, u32 offset,
std::optional<Sampler> ShaderIR::GetSampler(Tegra::Shader::Sampler sampler,
SamplerInfo sampler_info) {
- const auto offset = static_cast<u32>(sampler.index.Value());
- const auto info = GetSamplerInfo(sampler_info, offset);
+ const u32 offset = static_cast<u32>(sampler.index.Value());
+ const auto info = GetSamplerInfo(sampler_info, registry.ObtainBoundSampler(offset));
// If this sampler has already been used, return the existing mapping.
const auto it = std::find_if(used_samplers.begin(), used_samplers.end(),
@@ -404,20 +402,19 @@ std::optional<Sampler> ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
const Node sampler_register = GetRegister(reg);
const auto [base_node, tracked_sampler_info] =
TrackBindlessSampler(sampler_register, global_code, static_cast<s64>(global_code.size()));
- ASSERT(base_node != nullptr);
- if (base_node == nullptr) {
+ if (!base_node) {
+ UNREACHABLE();
return std::nullopt;
}
- if (const auto bindless_sampler_info =
- std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) {
- const u32 buffer = bindless_sampler_info->GetIndex();
- const u32 offset = bindless_sampler_info->GetOffset();
- info = GetSamplerInfo(info, offset, buffer);
+ if (const auto sampler_info = std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) {
+ const u32 buffer = sampler_info->index;
+ const u32 offset = sampler_info->offset;
+ info = GetSamplerInfo(info, registry.ObtainBindlessSampler(buffer, offset));
// If this sampler has already been used, return the existing mapping.
const auto it = std::find_if(used_samplers.begin(), used_samplers.end(),
- [buffer = buffer, offset = offset](const Sampler& entry) {
+ [buffer, offset](const Sampler& entry) {
return entry.buffer == buffer && entry.offset == offset;
});
if (it != used_samplers.end()) {
@@ -431,10 +428,32 @@ std::optional<Sampler> ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
return used_samplers.emplace_back(next_index, offset, buffer, *info.type, *info.is_array,
*info.is_shadow, *info.is_buffer, false);
}
- if (const auto array_sampler_info = std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) {
- const u32 base_offset = array_sampler_info->GetBaseOffset() / 4;
- index_var = GetCustomVariable(array_sampler_info->GetIndexVar());
- info = GetSamplerInfo(info, base_offset);
+ if (const auto sampler_info = std::get_if<SeparateSamplerNode>(&*tracked_sampler_info)) {
+ const std::pair indices = sampler_info->indices;
+ const std::pair offsets = sampler_info->offsets;
+ info = GetSamplerInfo(info, registry.ObtainSeparateSampler(indices, offsets));
+
+ // Try to use an already created sampler if it exists
+ const auto it = std::find_if(
+ used_samplers.begin(), used_samplers.end(), [indices, offsets](const Sampler& entry) {
+ return offsets == std::pair{entry.offset, entry.secondary_offset} &&
+ indices == std::pair{entry.buffer, entry.secondary_buffer};
+ });
+ if (it != used_samplers.end()) {
+ ASSERT(it->is_separated && it->type == info.type && it->is_array == info.is_array &&
+ it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer);
+ return *it;
+ }
+
+ // Otherwise create a new mapping for this sampler
+ const u32 next_index = static_cast<u32>(used_samplers.size());
+ return used_samplers.emplace_back(next_index, offsets, indices, *info.type, *info.is_array,
+ *info.is_shadow, *info.is_buffer);
+ }
+ if (const auto sampler_info = std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) {
+ const u32 base_offset = sampler_info->base_offset / 4;
+ index_var = GetCustomVariable(sampler_info->bindless_var);
+ info = GetSamplerInfo(info, registry.ObtainBoundSampler(base_offset));
// If this sampler has already been used, return the existing mapping.
const auto it = std::find_if(
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index c5e5165ff..8f230d57a 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -275,10 +275,11 @@ using Node = std::shared_ptr<NodeData>;
using Node4 = std::array<Node, 4>;
using NodeBlock = std::vector<Node>;
-class BindlessSamplerNode;
-class ArraySamplerNode;
+struct ArraySamplerNode;
+struct BindlessSamplerNode;
+struct SeparateSamplerNode;
-using TrackSamplerData = std::variant<BindlessSamplerNode, ArraySamplerNode>;
+using TrackSamplerData = std::variant<BindlessSamplerNode, SeparateSamplerNode, ArraySamplerNode>;
using TrackSampler = std::shared_ptr<TrackSamplerData>;
struct Sampler {
@@ -288,63 +289,51 @@ struct Sampler {
: index{index}, offset{offset}, type{type}, is_array{is_array}, is_shadow{is_shadow},
is_buffer{is_buffer}, is_indexed{is_indexed} {}
+ /// Separate sampler constructor
+ constexpr explicit Sampler(u32 index, std::pair<u32, u32> offsets, std::pair<u32, u32> buffers,
+ Tegra::Shader::TextureType type, bool is_array, bool is_shadow,
+ bool is_buffer)
+ : index{index}, offset{offsets.first}, secondary_offset{offsets.second},
+ buffer{buffers.first}, secondary_buffer{buffers.second}, type{type}, is_array{is_array},
+ is_shadow{is_shadow}, is_buffer{is_buffer}, is_separated{true} {}
+
/// Bindless samplers constructor
constexpr explicit Sampler(u32 index, u32 offset, u32 buffer, Tegra::Shader::TextureType type,
bool is_array, bool is_shadow, bool is_buffer, bool is_indexed)
: index{index}, offset{offset}, buffer{buffer}, type{type}, is_array{is_array},
is_shadow{is_shadow}, is_buffer{is_buffer}, is_bindless{true}, is_indexed{is_indexed} {}
- u32 index = 0; ///< Emulated index given for the this sampler.
- u32 offset = 0; ///< Offset in the const buffer from where the sampler is being read.
- u32 buffer = 0; ///< Buffer where the bindless sampler is being read (unused on bound samplers).
- u32 size = 1; ///< Size of the sampler.
+ u32 index = 0; ///< Emulated index given for the this sampler.
+ u32 offset = 0; ///< Offset in the const buffer from where the sampler is being read.
+ u32 secondary_offset = 0; ///< Secondary offset in the const buffer.
+ u32 buffer = 0; ///< Buffer where the bindless sampler is read.
+ u32 secondary_buffer = 0; ///< Secondary buffer where the bindless sampler is read.
+ u32 size = 1; ///< Size of the sampler.
Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc)
- bool is_array = false; ///< Whether the texture is being sampled as an array texture or not.
- bool is_shadow = false; ///< Whether the texture is being sampled as a depth texture or not.
- bool is_buffer = false; ///< Whether the texture is a texture buffer without sampler.
- bool is_bindless = false; ///< Whether this sampler belongs to a bindless texture or not.
- bool is_indexed = false; ///< Whether this sampler is an indexed array of textures.
+ bool is_array = false; ///< Whether the texture is being sampled as an array texture or not.
+ bool is_shadow = false; ///< Whether the texture is being sampled as a depth texture or not.
+ bool is_buffer = false; ///< Whether the texture is a texture buffer without sampler.
+ bool is_bindless = false; ///< Whether this sampler belongs to a bindless texture or not.
+ bool is_indexed = false; ///< Whether this sampler is an indexed array of textures.
+ bool is_separated = false; ///< Whether the image and sampler is separated or not.
};
/// Represents a tracked bindless sampler into a direct const buffer
-class ArraySamplerNode final {
-public:
- explicit ArraySamplerNode(u32 index, u32 base_offset, u32 bindless_var)
- : index{index}, base_offset{base_offset}, bindless_var{bindless_var} {}
-
- constexpr u32 GetIndex() const {
- return index;
- }
-
- constexpr u32 GetBaseOffset() const {
- return base_offset;
- }
-
- constexpr u32 GetIndexVar() const {
- return bindless_var;
- }
-
-private:
+struct ArraySamplerNode {
u32 index;
u32 base_offset;
u32 bindless_var;
};
-/// Represents a tracked bindless sampler into a direct const buffer
-class BindlessSamplerNode final {
-public:
- explicit BindlessSamplerNode(u32 index, u32 offset) : index{index}, offset{offset} {}
-
- constexpr u32 GetIndex() const {
- return index;
- }
-
- constexpr u32 GetOffset() const {
- return offset;
- }
+/// Represents a tracked separate sampler image pair that was folded statically
+struct SeparateSamplerNode {
+ std::pair<u32, u32> indices;
+ std::pair<u32, u32> offsets;
+};
-private:
+/// Represents a tracked bindless sampler into a direct const buffer
+struct BindlessSamplerNode {
u32 index;
u32 offset;
};
diff --git a/src/video_core/shader/node_helper.h b/src/video_core/shader/node_helper.h
index 11231bbea..1e0886185 100644
--- a/src/video_core/shader/node_helper.h
+++ b/src/video_core/shader/node_helper.h
@@ -48,7 +48,7 @@ Node MakeNode(Args&&... args) {
template <typename T, typename... Args>
TrackSampler MakeTrackSampler(Args&&... args) {
static_assert(std::is_convertible_v<T, TrackSamplerData>);
- return std::make_shared<TrackSamplerData>(T(std::forward<Args>(args)...));
+ return std::make_shared<TrackSamplerData>(T{std::forward<Args>(args)...});
}
template <typename... Args>
diff --git a/src/video_core/shader/registry.cpp b/src/video_core/shader/registry.cpp
index af70b3f35..cdf274e54 100644
--- a/src/video_core/shader/registry.cpp
+++ b/src/video_core/shader/registry.cpp
@@ -93,6 +93,26 @@ std::optional<SamplerDescriptor> Registry::ObtainBoundSampler(u32 offset) {
return value;
}
+std::optional<Tegra::Engines::SamplerDescriptor> Registry::ObtainSeparateSampler(
+ std::pair<u32, u32> buffers, std::pair<u32, u32> offsets) {
+ SeparateSamplerKey key;
+ key.buffers = buffers;
+ key.offsets = offsets;
+ const auto iter = separate_samplers.find(key);
+ if (iter != separate_samplers.end()) {
+ return iter->second;
+ }
+ if (!engine) {
+ return std::nullopt;
+ }
+
+ const u32 handle_1 = engine->AccessConstBuffer32(stage, key.buffers.first, key.offsets.first);
+ const u32 handle_2 = engine->AccessConstBuffer32(stage, key.buffers.second, key.offsets.second);
+ const SamplerDescriptor value = engine->AccessSampler(handle_1 | handle_2);
+ separate_samplers.emplace(key, value);
+ return value;
+}
+
std::optional<Tegra::Engines::SamplerDescriptor> Registry::ObtainBindlessSampler(u32 buffer,
u32 offset) {
const std::pair key = {buffer, offset};
diff --git a/src/video_core/shader/registry.h b/src/video_core/shader/registry.h
index 0c80d35fd..231206765 100644
--- a/src/video_core/shader/registry.h
+++ b/src/video_core/shader/registry.h
@@ -19,8 +19,39 @@
namespace VideoCommon::Shader {
+struct SeparateSamplerKey {
+ std::pair<u32, u32> buffers;
+ std::pair<u32, u32> offsets;
+};
+
+} // namespace VideoCommon::Shader
+
+namespace std {
+
+template <>
+struct hash<VideoCommon::Shader::SeparateSamplerKey> {
+ std::size_t operator()(const VideoCommon::Shader::SeparateSamplerKey& key) const noexcept {
+ return std::hash<u32>{}(key.buffers.first ^ key.buffers.second ^ key.offsets.first ^
+ key.offsets.second);
+ }
+};
+
+template <>
+struct equal_to<VideoCommon::Shader::SeparateSamplerKey> {
+ bool operator()(const VideoCommon::Shader::SeparateSamplerKey& lhs,
+ const VideoCommon::Shader::SeparateSamplerKey& rhs) const noexcept {
+ return lhs.buffers == rhs.buffers && lhs.offsets == rhs.offsets;
+ }
+};
+
+} // namespace std
+
+namespace VideoCommon::Shader {
+
using KeyMap = std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>;
using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescriptor>;
+using SeparateSamplerMap =
+ std::unordered_map<SeparateSamplerKey, Tegra::Engines::SamplerDescriptor>;
using BindlessSamplerMap =
std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>;
@@ -73,6 +104,9 @@ public:
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBoundSampler(u32 offset);
+ std::optional<Tegra::Engines::SamplerDescriptor> ObtainSeparateSampler(
+ std::pair<u32, u32> buffers, std::pair<u32, u32> offsets);
+
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset);
/// Inserts a key.
@@ -128,6 +162,7 @@ private:
Tegra::Engines::ConstBufferEngineInterface* engine = nullptr;
KeyMap keys;
BoundSamplerMap bound_samplers;
+ SeparateSamplerMap separate_samplers;
BindlessSamplerMap bindless_samplers;
u32 bound_buffer;
GraphicsInfo graphics_info;
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index 15ae152f2..3a98b2104 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -330,8 +330,8 @@ private:
OperationCode GetPredicateCombiner(Tegra::Shader::PredOperation operation);
/// Queries the missing sampler info from the execution context.
- SamplerInfo GetSamplerInfo(SamplerInfo info, u32 offset,
- std::optional<u32> buffer = std::nullopt);
+ SamplerInfo GetSamplerInfo(SamplerInfo info,
+ std::optional<Tegra::Engines::SamplerDescriptor> sampler);
/// Accesses a texture sampler.
std::optional<Sampler> GetSampler(Tegra::Shader::Sampler sampler, SamplerInfo info);
@@ -409,8 +409,14 @@ private:
std::tuple<Node, u32, u32> TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const;
- std::tuple<Node, TrackSampler> TrackBindlessSampler(Node tracked, const NodeBlock& code,
- s64 cursor);
+ std::pair<Node, TrackSampler> TrackBindlessSampler(Node tracked, const NodeBlock& code,
+ s64 cursor);
+
+ std::pair<Node, TrackSampler> HandleBindlessIndirectRead(const CbufNode& cbuf,
+ const OperationNode& operation,
+ Node gpr, Node base_offset,
+ Node tracked, const NodeBlock& code,
+ s64 cursor);
std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const;
diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp
index eb97bfd41..d5ed81442 100644
--- a/src/video_core/shader/track.cpp
+++ b/src/video_core/shader/track.cpp
@@ -14,6 +14,7 @@
namespace VideoCommon::Shader {
namespace {
+
std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor,
OperationCode operation_code) {
for (; cursor >= 0; --cursor) {
@@ -63,7 +64,8 @@ bool AmendNodeCv(std::size_t amend_index, Node node) {
if (const auto operation = std::get_if<OperationNode>(&*node)) {
operation->SetAmendIndex(amend_index);
return true;
- } else if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
+ }
+ if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
conditional->SetAmendIndex(amend_index);
return true;
}
@@ -72,40 +74,27 @@ bool AmendNodeCv(std::size_t amend_index, Node node) {
} // Anonymous namespace
-std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, const NodeBlock& code,
- s64 cursor) {
+std::pair<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, const NodeBlock& code,
+ s64 cursor) {
if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
+ const u32 cbuf_index = cbuf->GetIndex();
+
// Constant buffer found, test if it's an immediate
const auto& offset = cbuf->GetOffset();
if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
- auto track =
- MakeTrackSampler<BindlessSamplerNode>(cbuf->GetIndex(), immediate->GetValue());
+ auto track = MakeTrackSampler<BindlessSamplerNode>(cbuf_index, immediate->GetValue());
return {tracked, track};
}
if (const auto operation = std::get_if<OperationNode>(&*offset)) {
const u32 bound_buffer = registry.GetBoundBuffer();
- if (bound_buffer != cbuf->GetIndex()) {
+ if (bound_buffer != cbuf_index) {
return {};
}
- const auto pair = DecoupleIndirectRead(*operation);
- if (!pair) {
- return {};
+ if (const std::optional pair = DecoupleIndirectRead(*operation)) {
+ auto [gpr, base_offset] = *pair;
+ return HandleBindlessIndirectRead(*cbuf, *operation, gpr, base_offset, tracked,
+ code, cursor);
}
- auto [gpr, base_offset] = *pair;
- const auto offset_inm = std::get_if<ImmediateNode>(&*base_offset);
- const auto& gpu_driver = registry.AccessGuestDriverProfile();
- const u32 bindless_cv = NewCustomVariable();
- Node op =
- Operation(OperationCode::UDiv, gpr, Immediate(gpu_driver.GetTextureHandlerSize()));
-
- const Node cv_node = GetCustomVariable(bindless_cv);
- Node amend_op = Operation(OperationCode::Assign, cv_node, std::move(op));
- const std::size_t amend_index = DeclareAmend(std::move(amend_op));
- AmendNodeCv(amend_index, code[cursor]);
- // TODO Implement Bindless Index custom variable
- auto track = MakeTrackSampler<ArraySamplerNode>(cbuf->GetIndex(),
- offset_inm->GetValue(), bindless_cv);
- return {tracked, track};
}
return {};
}
@@ -122,10 +111,23 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
return TrackBindlessSampler(source, code, new_cursor);
}
if (const auto operation = std::get_if<OperationNode>(&*tracked)) {
- for (std::size_t i = operation->GetOperandsCount(); i > 0; --i) {
- if (auto found = TrackBindlessSampler((*operation)[i - 1], code, cursor);
- std::get<0>(found)) {
- // Cbuf found in operand.
+ const OperationNode& op = *operation;
+
+ const OperationCode opcode = operation->GetCode();
+ if (opcode == OperationCode::IBitwiseOr || opcode == OperationCode::UBitwiseOr) {
+ ASSERT(op.GetOperandsCount() == 2);
+ auto [node_a, index_a, offset_a] = TrackCbuf(op[0], code, cursor);
+ auto [node_b, index_b, offset_b] = TrackCbuf(op[1], code, cursor);
+ if (node_a && node_b) {
+ auto track = MakeTrackSampler<SeparateSamplerNode>(std::pair{index_a, index_b},
+ std::pair{offset_a, offset_b});
+ return {tracked, std::move(track)};
+ }
+ }
+ std::size_t i = op.GetOperandsCount();
+ while (i--) {
+ if (auto found = TrackBindlessSampler(op[i - 1], code, cursor); std::get<0>(found)) {
+ // Constant buffer found in operand.
return found;
}
}
@@ -139,6 +141,26 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
return {};
}
+std::pair<Node, TrackSampler> ShaderIR::HandleBindlessIndirectRead(
+ const CbufNode& cbuf, const OperationNode& operation, Node gpr, Node base_offset, Node tracked,
+ const NodeBlock& code, s64 cursor) {
+ const auto offset_imm = std::get<ImmediateNode>(*base_offset);
+ const auto& gpu_driver = registry.AccessGuestDriverProfile();
+ const u32 bindless_cv = NewCustomVariable();
+ const u32 texture_handler_size = gpu_driver.GetTextureHandlerSize();
+ Node op = Operation(OperationCode::UDiv, gpr, Immediate(texture_handler_size));
+
+ Node cv_node = GetCustomVariable(bindless_cv);
+ Node amend_op = Operation(OperationCode::Assign, std::move(cv_node), std::move(op));
+ const std::size_t amend_index = DeclareAmend(std::move(amend_op));
+ AmendNodeCv(amend_index, code[cursor]);
+
+ // TODO: Implement bindless index custom variable
+ auto track =
+ MakeTrackSampler<ArraySamplerNode>(cbuf.GetIndex(), offset_imm.GetValue(), bindless_cv);
+ return {tracked, track};
+}
+
std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code,
s64 cursor) const {
if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-12 06:36:04 +0200