Merge pull request #1533 from FernandoS27/lmem

Implemented Shader Local Memory
This commit is contained in:
bunnei 2018-10-26 00:16:25 -04:00 committed by GitHub
commit d278f25bda
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 138 additions and 1 deletions

View File

@ -208,6 +208,16 @@ enum class UniformType : u64 {
Double = 5, Double = 5,
}; };
enum class StoreType : u64 {
Unsigned8 = 0,
Signed8 = 1,
Unsigned16 = 2,
Signed16 = 3,
Bytes32 = 4,
Bytes64 = 5,
Bytes128 = 6,
};
enum class IMinMaxExchange : u64 { enum class IMinMaxExchange : u64 {
None = 0, None = 0,
XLo = 1, XLo = 1,
@ -747,6 +757,18 @@ union Instruction {
BitField<44, 2, u64> unknown; BitField<44, 2, u64> unknown;
} ld_c; } ld_c;
union {
BitField<48, 3, StoreType> type;
} ldst_sl;
union {
BitField<44, 2, u64> unknown;
} ld_l;
union {
BitField<44, 2, u64> unknown;
} st_l;
union { union {
BitField<0, 3, u64> pred0; BitField<0, 3, u64> pred0;
BitField<3, 3, u64> pred3; BitField<3, 3, u64> pred3;
@ -1209,6 +1231,7 @@ union Instruction {
BitField<61, 1, u64> is_b_imm; BitField<61, 1, u64> is_b_imm;
BitField<60, 1, u64> is_b_gpr; BitField<60, 1, u64> is_b_gpr;
BitField<59, 1, u64> is_c_gpr; BitField<59, 1, u64> is_c_gpr;
BitField<20, 24, s64> smem_imm;
Attribute attribute; Attribute attribute;
Sampler sampler; Sampler sampler;
@ -1232,8 +1255,12 @@ public:
BRA, BRA,
PBK, PBK,
LD_A, LD_A,
LD_L,
LD_S,
LD_C, LD_C,
ST_A, ST_A,
ST_L,
ST_S,
LDG, // Load from global memory LDG, // Load from global memory
STG, // Store in global memory STG, // Store in global memory
TEX, TEX,
@ -1490,8 +1517,12 @@ private:
INST("111000110100---", Id::BRK, Type::Flow, "BRK"), INST("111000110100---", Id::BRK, Type::Flow, "BRK"),
INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"), INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"), INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
INST("1110111101001---", Id::LD_S, Type::Memory, "LD_S"),
INST("1110111101000---", Id::LD_L, Type::Memory, "LD_L"),
INST("1110111110010---", Id::LD_C, Type::Memory, "LD_C"), INST("1110111110010---", Id::LD_C, Type::Memory, "LD_C"),
INST("1110111111110---", Id::ST_A, Type::Memory, "ST_A"), INST("1110111111110---", Id::ST_A, Type::Memory, "ST_A"),
INST("1110111101011---", Id::ST_S, Type::Memory, "ST_S"),
INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"),
INST("1110111011010---", Id::LDG, Type::Memory, "LDG"), INST("1110111011010---", Id::LDG, Type::Memory, "LDG"),
INST("1110111011011---", Id::STG, Type::Memory, "STG"), INST("1110111011011---", Id::STG, Type::Memory, "STG"),
INST("110000----111---", Id::TEX, Type::Memory, "TEX"), INST("110000----111---", Id::TEX, Type::Memory, "TEX"),

View File

@ -96,6 +96,11 @@ struct Header {
} }
} ps; } ps;
}; };
u64 GetLocalMemorySize() {
return (common1.shader_local_memory_low_size |
(common2.shader_local_memory_high_size << 24));
}
}; };
static_assert(sizeof(Header) == 0x50, "Incorrect structure size"); static_assert(sizeof(Header) == 0x50, "Incorrect structure size");

View File

@ -278,7 +278,7 @@ public:
const Maxwell3D::Regs::ShaderStage& stage, const std::string& suffix, const Maxwell3D::Regs::ShaderStage& stage, const std::string& suffix,
const Tegra::Shader::Header& header) const Tegra::Shader::Header& header)
: shader{shader}, declarations{declarations}, stage{stage}, suffix{suffix}, header{header}, : shader{shader}, declarations{declarations}, stage{stage}, suffix{suffix}, header{header},
fixed_pipeline_output_attributes_used{} { fixed_pipeline_output_attributes_used{}, local_memory_size{0} {
BuildRegisterList(); BuildRegisterList();
BuildInputList(); BuildInputList();
} }
@ -436,6 +436,25 @@ public:
shader.AddLine(dest + " = " + src + ';'); shader.AddLine(dest + " = " + src + ';');
} }
std::string GetLocalMemoryAsFloat(const std::string& index) {
return "lmem[" + index + ']';
}
std::string GetLocalMemoryAsInteger(const std::string& index, bool is_signed = false) {
const std::string func{is_signed ? "floatToIntBits" : "floatBitsToUint"};
return func + "(lmem[" + index + "])";
}
void SetLocalMemoryAsFloat(const std::string& index, const std::string& value) {
shader.AddLine("lmem[" + index + "] = " + value + ';');
}
void SetLocalMemoryAsInteger(const std::string& index, const std::string& value,
bool is_signed = false) {
const std::string func{is_signed ? "intBitsToFloat" : "uintBitsToFloat"};
shader.AddLine("lmem[" + index + "] = " + func + '(' + value + ");");
}
std::string GetControlCode(const Tegra::Shader::ControlCode cc) const { std::string GetControlCode(const Tegra::Shader::ControlCode cc) const {
switch (cc) { switch (cc) {
case Tegra::Shader::ControlCode::NEU: case Tegra::Shader::ControlCode::NEU:
@ -533,6 +552,7 @@ public:
void GenerateDeclarations(const std::string& suffix) { void GenerateDeclarations(const std::string& suffix) {
GenerateVertex(); GenerateVertex();
GenerateRegisters(suffix); GenerateRegisters(suffix);
GenerateLocalMemory();
GenerateInternalFlags(); GenerateInternalFlags();
GenerateInputAttrs(); GenerateInputAttrs();
GenerateOutputAttrs(); GenerateOutputAttrs();
@ -578,6 +598,10 @@ public:
return entry.GetName(); return entry.GetName();
} }
void SetLocalMemory(u64 lmem) {
local_memory_size = lmem;
}
private: private:
/// Generates declarations for registers. /// Generates declarations for registers.
void GenerateRegisters(const std::string& suffix) { void GenerateRegisters(const std::string& suffix) {
@ -588,6 +612,15 @@ private:
declarations.AddNewLine(); declarations.AddNewLine();
} }
/// Generates declarations for local memory.
void GenerateLocalMemory() {
if (local_memory_size > 0) {
declarations.AddLine("float lmem[" + std::to_string((local_memory_size - 1 + 4) / 4) +
"];");
declarations.AddNewLine();
}
}
/// Generates declarations for internal flags. /// Generates declarations for internal flags.
void GenerateInternalFlags() { void GenerateInternalFlags() {
for (u32 ii = 0; ii < static_cast<u64>(InternalFlag::Amount); ii++) { for (u32 ii = 0; ii < static_cast<u64>(InternalFlag::Amount); ii++) {
@ -895,6 +928,7 @@ private:
const std::string& suffix; const std::string& suffix;
const Tegra::Shader::Header& header; const Tegra::Shader::Header& header;
std::unordered_set<Attribute::Index> fixed_pipeline_output_attributes_used; std::unordered_set<Attribute::Index> fixed_pipeline_output_attributes_used;
u64 local_memory_size;
}; };
class GLSLGenerator { class GLSLGenerator {
@ -904,6 +938,8 @@ public:
: subroutines(subroutines), program_code(program_code), main_offset(main_offset), : subroutines(subroutines), program_code(program_code), main_offset(main_offset),
stage(stage), suffix(suffix) { stage(stage), suffix(suffix) {
std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header)); std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
local_memory_size = header.GetLocalMemorySize();
regs.SetLocalMemory(local_memory_size);
Generate(suffix); Generate(suffix);
} }
@ -2324,6 +2360,39 @@ private:
shader.AddLine("}"); shader.AddLine("}");
break; break;
} }
case OpCode::Id::LD_L: {
// Add an extra scope and declare the index register inside to prevent
// overwriting it in case it is used as an output of the LD instruction.
shader.AddLine('{');
++shader.scope;
std::string op = '(' + regs.GetRegisterAsInteger(instr.gpr8, 0, false) + " + " +
std::to_string(instr.smem_imm.Value()) + ')';
shader.AddLine("uint index = (" + op + " / 4);");
const std::string op_a = regs.GetLocalMemoryAsFloat("index");
if (instr.ld_l.unknown != 1) {
LOG_CRITICAL(HW_GPU, "LD_L Unhandled mode: {}",
static_cast<unsigned>(instr.ld_l.unknown.Value()));
UNREACHABLE();
}
switch (instr.ldst_sl.type.Value()) {
case Tegra::Shader::StoreType::Bytes32:
regs.SetRegisterToFloat(instr.gpr0, 0, op_a, 1, 1);
break;
default:
LOG_CRITICAL(HW_GPU, "LD_L Unhandled type: {}",
static_cast<unsigned>(instr.ldst_sl.type.Value()));
UNREACHABLE();
}
--shader.scope;
shader.AddLine('}');
break;
}
case OpCode::Id::ST_A: { case OpCode::Id::ST_A: {
ASSERT_MSG(instr.gpr8.Value() == Register::ZeroIndex, ASSERT_MSG(instr.gpr8.Value() == Register::ZeroIndex,
"Indirect attribute loads are not supported"); "Indirect attribute loads are not supported");
@ -2352,6 +2421,37 @@ private:
break; break;
} }
case OpCode::Id::ST_L: {
// Add an extra scope and declare the index register inside to prevent
// overwriting it in case it is used as an output of the LD instruction.
shader.AddLine('{');
++shader.scope;
std::string op = '(' + regs.GetRegisterAsInteger(instr.gpr8, 0, false) + " + " +
std::to_string(instr.smem_imm.Value()) + ')';
shader.AddLine("uint index = (" + op + " / 4);");
if (instr.st_l.unknown != 0) {
LOG_CRITICAL(HW_GPU, "ST_L Unhandled mode: {}",
static_cast<unsigned>(instr.st_l.unknown.Value()));
UNREACHABLE();
}
switch (instr.ldst_sl.type.Value()) {
case Tegra::Shader::StoreType::Bytes32:
regs.SetLocalMemoryAsFloat("index", regs.GetRegisterAsFloat(instr.gpr0));
break;
default:
LOG_CRITICAL(HW_GPU, "ST_L Unhandled type: {}",
static_cast<unsigned>(instr.ldst_sl.type.Value()));
UNREACHABLE();
}
--shader.scope;
shader.AddLine('}');
break;
}
case OpCode::Id::TEX: { case OpCode::Id::TEX: {
Tegra::Shader::TextureType texture_type{instr.tex.texture_type}; Tegra::Shader::TextureType texture_type{instr.tex.texture_type};
std::string coord; std::string coord;
@ -3575,6 +3675,7 @@ private:
const u32 main_offset; const u32 main_offset;
Maxwell3D::Regs::ShaderStage stage; Maxwell3D::Regs::ShaderStage stage;
const std::string& suffix; const std::string& suffix;
u64 local_memory_size;
ShaderWriter shader; ShaderWriter shader;
ShaderWriter declarations; ShaderWriter declarations;