summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--src/video_core/CMakeLists.txt1
-rw-r--r--src/video_core/engines/shader_bytecode.h8
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.cpp44
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.h34
-rw-r--r--src/video_core/shader/decode/memory.cpp49
-rw-r--r--src/video_core/shader/shader_ir.h38
-rw-r--r--src/video_core/shader/track.cpp76
7 files changed, 240 insertions, 10 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 509ca117a..6113e17ff 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -87,6 +87,7 @@ add_library(video_core STATIC
shader/decode.cpp
shader/shader_ir.cpp
shader/shader_ir.h
+ shader/track.cpp
surface.cpp
surface.h
textures/astc.cpp
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index cdef97bc6..9989825f8 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -208,6 +208,8 @@ enum class UniformType : u64 {
SignedShort = 3,
Single = 4,
Double = 5,
+ Quad = 6,
+ UnsignedQuad = 7,
};
enum class StoreType : u64 {
@@ -785,6 +787,12 @@ union Instruction {
} st_l;
union {
+ BitField<48, 3, UniformType> type;
+ BitField<46, 2, u64> cache_mode;
+ BitField<20, 24, s64> immediate_offset;
+ } ldg;
+
+ union {
BitField<0, 3, u64> pred0;
BitField<3, 3, u64> pred3;
BitField<7, 1, u64> abs_a;
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 3411cf9e6..e072216f0 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -34,6 +34,8 @@ using Operation = const OperationNode&;
enum : u32 { POSITION_VARYING_LOCATION = 0, GENERIC_VARYING_START_LOCATION = 1 };
constexpr u32 MAX_CONSTBUFFER_ELEMENTS =
static_cast<u32>(RasterizerOpenGL::MaxConstbufferSize) / (4 * sizeof(float));
+constexpr u32 MAX_GLOBALMEMORY_ELEMENTS =
+ static_cast<u32>(RasterizerOpenGL::MaxGlobalMemorySize) / sizeof(float);
enum class Type { Bool, Bool2, Float, Int, Uint, HalfFloat };
@@ -143,6 +145,7 @@ public:
DeclareInputAttributes();
DeclareOutputAttributes();
DeclareConstantBuffers();
+ DeclareGlobalMemory();
DeclareSamplers();
code.AddLine("void execute_" + suffix + "() {");
@@ -190,12 +193,15 @@ public:
ShaderEntries GetShaderEntries() const {
ShaderEntries entries;
for (const auto& cbuf : ir.GetConstantBuffers()) {
- ConstBufferEntry desc(cbuf.second, stage, GetConstBufferBlock(cbuf.first), cbuf.first);
- entries.const_buffers.push_back(desc);
+ entries.const_buffers.emplace_back(cbuf.second, stage, GetConstBufferBlock(cbuf.first),
+ cbuf.first);
}
for (const auto& sampler : ir.GetSamplers()) {
- SamplerEntry desc(sampler, stage, GetSampler(sampler));
- entries.samplers.push_back(desc);
+ entries.samplers.emplace_back(sampler, stage, GetSampler(sampler));
+ }
+ for (const auto& gmem : ir.GetGlobalMemoryBases()) {
+ entries.global_memory_entries.emplace_back(gmem.cbuf_index, gmem.cbuf_offset, stage,
+ GetGlobalMemoryBlock(gmem));
}
entries.clip_distances = ir.GetClipDistances();
entries.shader_length = ir.GetLength();
@@ -375,6 +381,15 @@ private:
}
}
+ void DeclareGlobalMemory() {
+ for (const auto& entry : ir.GetGlobalMemoryBases()) {
+ code.AddLine("layout (std430) buffer " + GetGlobalMemoryBlock(entry) + " {");
+ code.AddLine(" float " + GetGlobalMemory(entry) + "[MAX_GLOBALMEMORY_ELEMENTS];");
+ code.AddLine("};");
+ code.AddNewLine();
+ }
+ }
+
void DeclareSamplers() {
const auto& samplers = ir.GetSamplers();
for (const auto& sampler : samplers) {
@@ -538,6 +553,12 @@ private:
UNREACHABLE_MSG("Unmanaged offset node type");
}
+ } else if (const auto gmem = std::get_if<GmemNode>(node)) {
+ const std::string real = Visit(gmem->GetRealAddress());
+ const std::string base = Visit(gmem->GetBaseAddress());
+ const std::string final_offset = "(ftou(" + real + ") - ftou(" + base + ")) / 4";
+ return fmt::format("{}[{}]", GetGlobalMemory(gmem->GetDescriptor()), final_offset);
+
} else if (const auto lmem = std::get_if<LmemNode>(node)) {
return fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
@@ -1471,6 +1492,15 @@ private:
return GetDeclarationWithSuffix(index, "cbuf");
}
+ std::string GetGlobalMemory(const GlobalMemoryBase& descriptor) const {
+ return fmt::format("gmem_{}_{}_{}", descriptor.cbuf_index, descriptor.cbuf_offset, suffix);
+ }
+
+ std::string GetGlobalMemoryBlock(const GlobalMemoryBase& descriptor) const {
+ return fmt::format("gmem_block_{}_{}_{}", descriptor.cbuf_index, descriptor.cbuf_offset,
+ suffix);
+ }
+
std::string GetConstBufferBlock(u32 index) const {
return GetDeclarationWithSuffix(index, "cbuf_block");
}
@@ -1505,8 +1535,10 @@ private:
};
std::string GetCommonDeclarations() {
- return "#define MAX_CONSTBUFFER_ELEMENTS " + std::to_string(MAX_CONSTBUFFER_ELEMENTS) +
- "\n"
+ const auto cbuf = std::to_string(MAX_CONSTBUFFER_ELEMENTS);
+ const auto gmem = std::to_string(MAX_GLOBALMEMORY_ELEMENTS);
+ return "#define MAX_CONSTBUFFER_ELEMENTS " + cbuf + "\n" +
+ "#define MAX_GLOBALMEMORY_ELEMENTS " + gmem + "\n" +
"#define ftoi floatBitsToInt\n"
"#define ftou floatBitsToUint\n"
"#define itof intBitsToFloat\n"
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h
index 396a560d8..e47bc3729 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.h
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h
@@ -71,9 +71,43 @@ private:
Maxwell::ShaderStage stage{};
};
+class GlobalMemoryEntry {
+public:
+ explicit GlobalMemoryEntry(u32 cbuf_index, u32 cbuf_offset, Maxwell::ShaderStage stage,
+ std::string name)
+ : cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset}, stage{stage}, name{std::move(name)} {}
+
+ u32 GetCbufIndex() const {
+ return cbuf_index;
+ }
+
+ u32 GetCbufOffset() const {
+ return cbuf_offset;
+ }
+
+ const std::string& GetName() const {
+ return name;
+ }
+
+ Maxwell::ShaderStage GetStage() const {
+ return stage;
+ }
+
+ u32 GetHash() const {
+ return (static_cast<u32>(stage) << 24) | (cbuf_index << 16) | cbuf_offset;
+ }
+
+private:
+ u32 cbuf_index{};
+ u32 cbuf_offset{};
+ Maxwell::ShaderStage stage{};
+ std::string name;
+};
+
struct ShaderEntries {
std::vector<ConstBufferEntry> const_buffers;
std::vector<SamplerEntry> samplers;
+ std::vector<GlobalMemoryEntry> global_memory_entries;
std::array<bool, Maxwell::NumClipDistances> clip_distances{};
std::size_t shader_length{};
};
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index ae71672d6..04cb386b7 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -4,6 +4,7 @@
#include <algorithm>
#include <vector>
+#include <fmt/format.h>
#include "common/assert.h"
#include "common/common_types.h"
@@ -119,6 +120,54 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, const BasicBlock& code, u32 pc) {
}
break;
}
+ case OpCode::Id::LDG: {
+ const u32 count = [&]() {
+ switch (instr.ldg.type) {
+ case Tegra::Shader::UniformType::Single:
+ return 1;
+ case Tegra::Shader::UniformType::Double:
+ return 2;
+ case Tegra::Shader::UniformType::Quad:
+ case Tegra::Shader::UniformType::UnsignedQuad:
+ return 4;
+ default:
+ UNIMPLEMENTED_MSG("Unimplemented LDG size!");
+ return 1;
+ }
+ }();
+
+ const Node addr_register = GetRegister(instr.gpr8);
+ const Node base_address = TrackCbuf(addr_register, code, static_cast<s64>(code.size()));
+ const auto cbuf = std::get_if<CbufNode>(base_address);
+ ASSERT(cbuf != nullptr);
+ const auto cbuf_offset_imm = std::get_if<ImmediateNode>(cbuf->GetOffset());
+ ASSERT(cbuf_offset_imm != nullptr);
+ const auto cbuf_offset = cbuf_offset_imm->GetValue() * 4;
+
+ bb.push_back(Comment(
+ fmt::format("Base address is c[0x{:x}][0x{:x}]", cbuf->GetIndex(), cbuf_offset)));
+
+ const GlobalMemoryBase descriptor{cbuf->GetIndex(), cbuf_offset};
+ used_global_memory_bases.insert(descriptor);
+
+ const Node immediate_offset =
+ Immediate(static_cast<u32>(instr.ldg.immediate_offset.Value()));
+ const Node base_real_address =
+ Operation(OperationCode::UAdd, NO_PRECISE, immediate_offset, addr_register);
+
+ for (u32 i = 0; i < count; ++i) {
+ const Node it_offset = Immediate(i * 4);
+ const Node real_address =
+ Operation(OperationCode::UAdd, NO_PRECISE, base_real_address, it_offset);
+ const Node gmem = StoreNode(GmemNode(real_address, base_address, descriptor));
+
+ SetTemporal(bb, i, gmem);
+ }
+ for (u32 i = 0; i < count; ++i) {
+ SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
+ }
+ break;
+ }
case OpCode::Id::ST_A: {
UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex,
"Indirect attribute loads are not supported");
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index ef8f94480..c4ecb2e3c 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -257,6 +257,15 @@ private:
bool is_indirect{};
};
+struct GlobalMemoryBase {
+ u32 cbuf_index{};
+ u32 cbuf_offset{};
+
+ bool operator<(const GlobalMemoryBase& rhs) const {
+ return std::tie(cbuf_index, cbuf_offset) < std::tie(rhs.cbuf_index, rhs.cbuf_offset);
+ }
+};
+
struct MetaArithmetic {
bool precise{};
};
@@ -478,14 +487,26 @@ private:
/// Global memory node
class GmemNode final {
public:
- explicit constexpr GmemNode(Node address) : address{address} {}
+ explicit constexpr GmemNode(Node real_address, Node base_address,
+ const GlobalMemoryBase& descriptor)
+ : real_address{real_address}, base_address{base_address}, descriptor{descriptor} {}
- Node GetAddress() const {
- return address;
+ Node GetRealAddress() const {
+ return real_address;
+ }
+
+ Node GetBaseAddress() const {
+ return base_address;
+ }
+
+ const GlobalMemoryBase& GetDescriptor() const {
+ return descriptor;
}
private:
- const Node address;
+ const Node real_address;
+ const Node base_address;
+ const GlobalMemoryBase descriptor;
};
/// Commentary, can be dropped
@@ -543,6 +564,10 @@ public:
return used_clip_distances;
}
+ const std::set<GlobalMemoryBase>& GetGlobalMemoryBases() const {
+ return used_global_memory_bases;
+ }
+
std::size_t GetLength() const {
return static_cast<std::size_t>(coverage_end * sizeof(u64));
}
@@ -734,6 +759,10 @@ private:
void WriteLop3Instruction(BasicBlock& bb, Tegra::Shader::Register dest, Node op_a, Node op_b,
Node op_c, Node imm_lut, bool sets_cc);
+ Node TrackCbuf(Node tracked, const BasicBlock& code, s64 cursor);
+
+ std::pair<Node, s64> TrackRegister(const GprNode* tracked, const BasicBlock& code, s64 cursor);
+
template <typename... T>
Node Operation(OperationCode code, const T*... operands) {
return StoreNode(OperationNode(code, operands...));
@@ -786,6 +815,7 @@ private:
std::map<u32, ConstBuffer> used_cbufs;
std::set<Sampler> used_samplers;
std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{};
+ std::set<GlobalMemoryBase> used_global_memory_bases;
Tegra::Shader::Header header;
};
diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp
new file mode 100644
index 000000000..d6d29ee9f
--- /dev/null
+++ b/src/video_core/shader/track.cpp
@@ -0,0 +1,76 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <utility>
+#include <variant>
+
+#include "video_core/shader/shader_ir.h"
+
+namespace VideoCommon::Shader {
+
+namespace {
+std::pair<Node, s64> FindOperation(const BasicBlock& code, s64 cursor,
+ OperationCode operation_code) {
+ for (; cursor >= 0; --cursor) {
+ const Node node = code[cursor];
+ if (const auto operation = std::get_if<OperationNode>(node)) {
+ if (operation->GetCode() == operation_code)
+ return {node, cursor};
+ }
+ }
+ return {};
+}
+} // namespace
+
+Node ShaderIR::TrackCbuf(Node tracked, const BasicBlock& code, s64 cursor) {
+ if (const auto cbuf = std::get_if<CbufNode>(tracked)) {
+ // Cbuf found, but it has to be immediate
+ return std::holds_alternative<ImmediateNode>(*cbuf->GetOffset()) ? tracked : nullptr;
+ }
+ if (const auto gpr = std::get_if<GprNode>(tracked)) {
+ if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) {
+ return nullptr;
+ }
+ // Reduce the cursor in one to avoid infinite loops when the instruction sets the same
+ // register that it uses as operand
+ const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1);
+ if (!source) {
+ return nullptr;
+ }
+ return TrackCbuf(source, code, new_cursor);
+ }
+ if (const auto operation = std::get_if<OperationNode>(tracked)) {
+ for (std::size_t i = 0; i < operation->GetOperandsCount(); ++i) {
+ if (const auto found = TrackCbuf((*operation)[i], code, cursor)) {
+ // Cbuf found in operand
+ return found;
+ }
+ }
+ return nullptr;
+ }
+ return nullptr;
+}
+
+std::pair<Node, s64> ShaderIR::TrackRegister(const GprNode* tracked, const BasicBlock& code,
+ s64 cursor) {
+ for (; cursor >= 0; --cursor) {
+ const auto [found_node, new_cursor] = FindOperation(code, cursor, OperationCode::Assign);
+ if (!found_node) {
+ return {};
+ }
+ const auto operation = std::get_if<OperationNode>(found_node);
+ ASSERT(operation);
+
+ const auto& target = (*operation)[0];
+ if (const auto gpr_target = std::get_if<GprNode>(target)) {
+ if (gpr_target->GetIndex() == tracked->GetIndex()) {
+ return {(*operation)[1], new_cursor};
+ }
+ }
+ }
+ return {};
+}
+
+} // namespace VideoCommon::Shader
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-14 18:34:28 +0100