diff options
Diffstat (limited to 'ptx/src/ptx.lalrpop')
| -rw-r--r-- | ptx/src/ptx.lalrpop | 169 |
1 files changed, 110 insertions, 59 deletions
diff --git a/ptx/src/ptx.lalrpop b/ptx/src/ptx.lalrpop index d2d5be8..2c0e365 100644 --- a/ptx/src/ptx.lalrpop +++ b/ptx/src/ptx.lalrpop @@ -70,6 +70,7 @@ match { ".ltu", ".lu", ".nan", + ".NaN", ".ne", ".neu", ".num", @@ -124,6 +125,8 @@ match { "ld", "mad", "map_f64_to_f32", + "max", + "min", "mov", "mul", "not", @@ -134,6 +137,7 @@ match { "shr", r"sm_[0-9]+" => ShaderModel, "st", + "sub", "texmode_independent", "texmode_unified", } else { @@ -153,6 +157,8 @@ ExtendedID : &'input str = { "ld", "mad", "map_f64_to_f32", + "max", + "min", "mov", "mul", "not", @@ -163,6 +169,7 @@ ExtendedID : &'input str = { "shr", ShaderModel, "st", + "sub", "texmode_independent", "texmode_unified", ID @@ -448,7 +455,10 @@ Instruction: ast::Instruction<ast::ParsedArgParams<'input>> = { InstCall, InstAbs, InstMad, - InstOr + InstOr, + InstSub, + InstMin, + InstMax, }; // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#data-movement-and-conversion-instructions-ld @@ -570,38 +580,19 @@ MovVectorType: ast::ScalarType = { // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#floating-point-instructions-mul // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#half-precision-floating-point-instructions-mul InstMul: ast::Instruction<ast::ParsedArgParams<'input>> = { - "mul" <d:InstMulMode> <a:Arg3> => ast::Instruction::Mul(d, a) + "mul" <d:MulDetails> <a:Arg3> => ast::Instruction::Mul(d, a) }; -InstMulMode: ast::MulDetails = { - <ctr:MulIntControl> <t:IntType> => ast::MulDetails::Int(ast::MulIntDesc { +MulDetails: ast::MulDetails = { + <ctr:MulIntControl> <t:UIntType> => ast::MulDetails::Unsigned(ast::MulUInt{ typ: t, control: ctr }), - <r:RoundingModeFloat?> <ftz:".ftz"?> <s:".sat"?> ".f32" => ast::MulDetails::Float(ast::MulFloatDesc { - typ: ast::FloatType::F32, - rounding: r, - flush_to_zero: ftz.is_some(), - saturate: s.is_some() - }), - <r:RoundingModeFloat?> ".f64" => ast::MulDetails::Float(ast::MulFloatDesc { - typ: ast::FloatType::F64, - rounding: r, - flush_to_zero: false, - saturate: false - }), - <r:".rn"?> <ftz:".ftz"?> <s:".sat"?> ".f16" => ast::MulDetails::Float(ast::MulFloatDesc { - typ: ast::FloatType::F16, - rounding: r.map(|_| ast::RoundingMode::NearestEven), - flush_to_zero: ftz.is_some(), - saturate: s.is_some() + <ctr:MulIntControl> <t:SIntType> => ast::MulDetails::Signed(ast::MulSInt{ + typ: t, + control: ctr }), - <r:".rn"?> <ftz:".ftz"?> <s:".sat"?> ".f16x2" => ast::MulDetails::Float(ast::MulFloatDesc { - typ: ast::FloatType::F16x2, - rounding: r.map(|_| ast::RoundingMode::NearestEven), - flush_to_zero: ftz.is_some(), - saturate: s.is_some() - }) + <f:ArithFloat> => ast::MulDetails::Float(f) }; MulIntControl: ast::MulIntControl = { @@ -634,41 +625,23 @@ IntType : ast::IntType = { ".s64" => ast::IntType::S64, }; +UIntType: ast::UIntType = { + ".u16" => ast::UIntType::U16, + ".u32" => ast::UIntType::U32, + ".u64" => ast::UIntType::U64, +}; + +SIntType: ast::SIntType = { + ".s16" => ast::SIntType::S16, + ".s32" => ast::SIntType::S32, + ".s64" => ast::SIntType::S64, +}; + // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#integer-arithmetic-instructions-add // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#floating-point-instructions-add // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#half-precision-floating-point-instructions-add InstAdd: ast::Instruction<ast::ParsedArgParams<'input>> = { - "add" <d:InstAddMode> <a:Arg3> => ast::Instruction::Add(d, a) -}; - -InstAddMode: ast::AddDetails = { - <t:IntType> => ast::AddDetails::Int(ast::AddIntDesc { - typ: t, - saturate: false, - }), - ".sat" ".s32" => ast::AddDetails::Int(ast::AddIntDesc { - typ: ast::IntType::S32, - saturate: true, - }), - <rn:RoundingModeFloat?> <ftz:".ftz"?> <sat:".sat"?> ".f32" => ast::AddDetails::Float(ast::AddFloatDesc { - typ: ast::FloatType::F32, - rounding: rn, - flush_to_zero: ftz.is_some(), - saturate: sat.is_some(), - }), - <rn:RoundingModeFloat?> ".f64" => ast::AddDetails::Float(ast::AddFloatDesc { - typ: ast::FloatType::F64, - rounding: rn, - flush_to_zero: false, - saturate: false, - }), - <rn:".rn"?> <ftz:".ftz"?> <sat:".sat"?>".f16" => ast::AddDetails::Float(ast::AddFloatDesc { - typ: ast::FloatType::F16, - rounding: rn.map(|_| ast::RoundingMode::NearestEven), - flush_to_zero: ftz.is_some(), - saturate: sat.is_some(), - }), - ".rn"? ".ftz"? ".sat"? ".f16x2" => todo!() + "add" <d:ArithDetails> <a:Arg3> => ast::Instruction::Add(d, a) }; // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#comparison-and-selection-instructions-setp @@ -1041,7 +1014,7 @@ InstAbs: ast::Instruction<ast::ParsedArgParams<'input>> = { // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#integer-arithmetic-instructions-mad // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#floating-point-instructions-mad InstMad: ast::Instruction<ast::ParsedArgParams<'input>> = { - "mad" <d:InstMulMode> <a:Arg4> => ast::Instruction::Mad(d, a), + "mad" <d:MulDetails> <a:Arg4> => ast::Instruction::Mad(d, a), "mad" ".hi" ".sat" ".s32" => todo!() }; @@ -1063,6 +1036,84 @@ OrType: ast::OrType = { ".b64" => ast::OrType::B64, } +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#integer-arithmetic-instructions-sub +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#floating-point-instructions-sub +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#half-precision-floating-point-instructions-sub +InstSub: ast::Instruction<ast::ParsedArgParams<'input>> = { + "sub" <d:ArithDetails> <a:Arg3> => ast::Instruction::Sub(d, a), +}; + +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#integer-arithmetic-instructions-min +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#floating-point-instructions-min +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#half-precision-floating-point-instructions-min +InstMin: ast::Instruction<ast::ParsedArgParams<'input>> = { + "min" <d:MinMaxDetails> <a:Arg3> => ast::Instruction::Min(d, a), +}; + +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#integer-arithmetic-instructions-max +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#floating-point-instructions-max +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#half-precision-floating-point-instructions-max +InstMax: ast::Instruction<ast::ParsedArgParams<'input>> = { + "max" <d:MinMaxDetails> <a:Arg3> => ast::Instruction::Max(d, a), +}; + +MinMaxDetails: ast::MinMaxDetails = { + <t:UIntType> => ast::MinMaxDetails::Unsigned(t), + <t:SIntType> => ast::MinMaxDetails::Signed(t), + <ftz:".ftz"?> <nan:".NaN"?> ".f32" => ast::MinMaxDetails::Float( + ast::MinMaxFloat{ ftz: ftz.is_some(), nan: nan.is_some(), typ: ast::FloatType::F32 } + ), + ".f64" => ast::MinMaxDetails::Float( + ast::MinMaxFloat{ ftz: false, nan: false, typ: ast::FloatType::F64 } + ), + <ftz:".ftz"?> <nan:".NaN"?> ".f16" => ast::MinMaxDetails::Float( + ast::MinMaxFloat{ ftz: ftz.is_some(), nan: nan.is_some(), typ: ast::FloatType::F16 } + ), + <ftz:".ftz"?> <nan:".NaN"?> ".f16x2" => ast::MinMaxDetails::Float( + ast::MinMaxFloat{ ftz: ftz.is_some(), nan: nan.is_some(), typ: ast::FloatType::F16x2 } + ) +} + +ArithDetails: ast::ArithDetails = { + <t:UIntType> => ast::ArithDetails::Unsigned(t), + <t:SIntType> => ast::ArithDetails::Signed(ast::ArithSInt { + typ: t, + saturate: false, + }), + ".sat" ".s32" => ast::ArithDetails::Signed(ast::ArithSInt { + typ: ast::SIntType::S32, + saturate: true, + }), + <f:ArithFloat> => ast::ArithDetails::Float(f) +} + +ArithFloat: ast::ArithFloat = { + <rn:RoundingModeFloat?> <ftz:".ftz"?> <sat:".sat"?> ".f32" => ast::ArithFloat { + typ: ast::FloatType::F32, + rounding: rn, + flush_to_zero: ftz.is_some(), + saturate: sat.is_some(), + }, + <rn:RoundingModeFloat?> ".f64" => ast::ArithFloat { + typ: ast::FloatType::F64, + rounding: rn, + flush_to_zero: false, + saturate: false, + }, + <rn:".rn"?> <ftz:".ftz"?> <sat:".sat"?> ".f16" => ast::ArithFloat { + typ: ast::FloatType::F16, + rounding: rn.map(|_| ast::RoundingMode::NearestEven), + flush_to_zero: ftz.is_some(), + saturate: sat.is_some(), + }, + <rn:".rn"?> <ftz:".ftz"?> <sat:".sat"?> ".f16x2" => ast::ArithFloat { + typ: ast::FloatType::F16x2, + rounding: rn.map(|_| ast::RoundingMode::NearestEven), + flush_to_zero: ftz.is_some(), + saturate: sat.is_some(), + }, +} + Operand: ast::Operand<&'input str> = { <r:ExtendedID> => ast::Operand::Reg(r), <r:ExtendedID> "+" <o:Num> => { |