Stowage/ladybird
2
0
Fork 0
mirror of https://github.com/LadybirdBrowser/ladybird.git synced 2025-11-09 01:31:02 +00:00
Code Issues Projects Releases Packages Wiki Activity

LibWasm: Implement proposal 'relaxed-simd'

Browse source
This commit is contained in:
Ali Mohammad Pur 2025-09-23 04:37:20 +02:00 • committed by Ali Mohammad Pur
parent 77237af33f
commit d6f3f5fd51
Notes: github-actions[bot] 2025-10-14 23:29:22 +00:00
7 changed files with 536 additions and 261 deletions
Download patch file Download diff file Expand all files Collapse all files

88
Libraries/LibWasm/AbstractMachine/BytecodeInterpreter.cpp
View file

@ -122,6 +122,16 @@ struct InstructionHandler { };
}; \
template<bool HasDynamicInsnLimit, typename Continue> \
Outcome InstructionHandler<Instructions::name.value()>::operator()(HANDLER_PARAMS(DECOMPOSE_PARAMS))
#define ALIAS_INSTRUCTION(new_name, existing_name) \
template<> \
struct InstructionHandler<Instructions::new_name.value()> { \
template<bool HasDynamicInsnLimit, typename Continue> \
static Outcome operator()(HANDLER_PARAMS(DECOMPOSE_PARAMS)) \
{ \
TAILCALL return InstructionHandler<Instructions::existing_name.value()>::operator()<HasDynamicInsnLimit, Continue>( \
HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY)); \
} \
};
struct Continue {
static Outcome operator()(BytecodeInterpreter& interpreter, Configuration& configuration, Instruction const*, SourcesAndDestination addresses, u64 current_ip_value, Dispatch const* cc)
@ -3670,6 +3680,80 @@ HANDLE_INSTRUCTION(i32x4_extmul_high_i16x8_s)
TAILCALL return continue_(HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY));
}
ALIAS_INSTRUCTION(i8x16_relaxed_swizzle, i8x16_swizzle)
ALIAS_INSTRUCTION(i32x4_relaxed_trunc_f32x4_s, i32x4_trunc_sat_f32x4_s)
ALIAS_INSTRUCTION(i32x4_relaxed_trunc_f32x4_u, i32x4_trunc_sat_f32x4_u)
ALIAS_INSTRUCTION(i32x4_relaxed_trunc_f64x2_s_zero, i32x4_trunc_sat_f64x2_s_zero)
ALIAS_INSTRUCTION(i32x4_relaxed_trunc_f64x2_u_zero, i32x4_trunc_sat_f64x2_u_zero)
HANDLE_INSTRUCTION(f32x4_relaxed_madd)
{
auto a = configuration.take_source(0, addresses.sources).to<u128>();
auto b = configuration.take_source(1, addresses.sources).to<u128>();
auto& c_slot = configuration.source_value(2, addresses.sources);
auto c = c_slot.to<u128>();
c_slot = Value { Operators::VectorMultiplyAdd<4> {}(a, b, c) };
TAILCALL return continue_(HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY));
}
HANDLE_INSTRUCTION(f32x4_relaxed_nmadd)
{
auto a = configuration.take_source(0, addresses.sources).to<u128>();
auto b = configuration.take_source(1, addresses.sources).to<u128>();
auto& c_slot = configuration.source_value(2, addresses.sources);
auto c = c_slot.to<u128>();
c_slot = Value { Operators::VectorMultiplySub<4> {}(a, b, c) };
TAILCALL return continue_(HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY));
}
HANDLE_INSTRUCTION(f64x2_relaxed_madd)
{
auto a = configuration.take_source(0, addresses.sources).to<u128>();
auto b = configuration.take_source(1, addresses.sources).to<u128>();
auto& c_slot = configuration.source_value(2, addresses.sources);
auto c = c_slot.to<u128>();
c_slot = Value { Operators::VectorMultiplyAdd<2> {}(a, b, c) };
TAILCALL return continue_(HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY));
}
HANDLE_INSTRUCTION(f64x2_relaxed_nmadd)
{
auto a = configuration.take_source(0, addresses.sources).to<u128>();
auto b = configuration.take_source(1, addresses.sources).to<u128>();
auto& c_slot = configuration.source_value(2, addresses.sources);
auto c = c_slot.to<u128>();
c_slot = Value { Operators::VectorMultiplySub<2> {}(a, b, c) };
TAILCALL return continue_(HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY));
}
ALIAS_INSTRUCTION(i8x16_relaxed_laneselect, v128_bitselect)
ALIAS_INSTRUCTION(i16x8_relaxed_laneselect, v128_bitselect)
ALIAS_INSTRUCTION(i32x4_relaxed_laneselect, v128_bitselect)
ALIAS_INSTRUCTION(i64x2_relaxed_laneselect, v128_bitselect)
ALIAS_INSTRUCTION(f32x4_relaxed_min, f32x4_min)
ALIAS_INSTRUCTION(f32x4_relaxed_max, f32x4_max)
ALIAS_INSTRUCTION(f64x2_relaxed_min, f64x2_min)
ALIAS_INSTRUCTION(f64x2_relaxed_max, f64x2_max)
ALIAS_INSTRUCTION(i16x8_relaxed_q15mulr_s, i16x8_q15mulr_sat_s)
HANDLE_INSTRUCTION(i16x8_relaxed_dot_i8x16_i7x16_s)
{
if (interpreter.binary_numeric_operation<u128, u128, Operators::VectorDotProduct<8>>(configuration, addresses))
return Outcome::Return;
TAILCALL return continue_(HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY));
}
HANDLE_INSTRUCTION(i32x4_relaxed_dot_i8x16_i7x16_add_s)
{
// do i16x8 dot first, then fold back down to i32, then do the final component add.
auto rhs = configuration.take_source(0, addresses.sources).to<u128>();
auto lhs = configuration.take_source(1, addresses.sources).to<u128>(); // bounds checked by verifier.
auto result = Operators::VectorDotProduct<4, Operators::VectorIntegerExtOpPairwise<4, Operators::Add>> {}(lhs, rhs);
auto& c_slot = configuration.source_value(2, addresses.sources);
c_slot = Value { Operators::VectorIntegerBinaryOp<4, Operators::Add, MakeSigned> {}(result, c_slot.to<u128>()) };
TAILCALL return continue_(HANDLER_PARAMS(DECOMPOSE_PARAMS_NAME_ONLY));
}
template<u64 opcode, bool HasDynamicInsnLimit, typename Continue, typename... Args>
constexpr static auto handle_instruction(Args&&... a)
{
@ -4000,10 +4084,10 @@ bool BytecodeInterpreter::binary_numeric_operation(Configuration& configuration,
return false;
}
template<typename PopType, typename PushType, typename Operator, typename... Args>
template<typename PopType, typename PushType, typename Operator, size_t input_arg, typename... Args>
bool BytecodeInterpreter::unary_operation(Configuration& configuration, SourcesAndDestination const& addresses, Args&&... args)
{
auto& entry = configuration.source_value(0, addresses.sources); // bounds checked by verifier.
auto& entry = configuration.source_value(input_arg, addresses.sources); // bounds checked by verifier.
auto value = entry.to<PopType>();
auto call_result = Operator { forward<Args>(args)... }(value);
PushType result;