mirror of
https://github.com/SerenityOS/serenity
synced 2024-10-02 22:24:26 +00:00
LibWasm: Implement f32x4 and f64x2 arithmetic SIMD ops
Adds all the arithmetic ops for f32x4 and f64x2 SIMD instructions. With this, we pass 8375 additional tests :) Quite a few of the spec tests for this are still failing. I confirmed with the wasmer runtime manually for a number of them, and we seem to match their and results. I'm not really sure what's happening here, a spec bug or wasmer is broken in the same way. 18476 failed before. 10101 failed after.
This commit is contained in:
parent
71cc8aeaef
commit
b43d2307cd
|
@ -1392,6 +1392,10 @@ void BytecodeInterpreter::interpret(Configuration& configuration, InstructionPoi
|
|||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<4, Operators::LessThanOrEquals>>(configuration);
|
||||
case Instructions::f32x4_ge.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<4, Operators::GreaterThanOrEquals>>(configuration);
|
||||
case Instructions::f32x4_min.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<4, Operators::Minimum>>(configuration);
|
||||
case Instructions::f32x4_max.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<4, Operators::Maximum>>(configuration);
|
||||
case Instructions::f64x2_eq.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<2, Operators::Equals>>(configuration);
|
||||
case Instructions::f64x2_ne.value():
|
||||
|
@ -1404,6 +1408,62 @@ void BytecodeInterpreter::interpret(Configuration& configuration, InstructionPoi
|
|||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<2, Operators::LessThanOrEquals>>(configuration);
|
||||
case Instructions::f64x2_ge.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<2, Operators::GreaterThanOrEquals>>(configuration);
|
||||
case Instructions::f64x2_min.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<2, Operators::Minimum>>(configuration);
|
||||
case Instructions::f64x2_max.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatCmpOp<2, Operators::Maximum>>(configuration);
|
||||
case Instructions::f32x4_div.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<4, Operators::Divide>>(configuration);
|
||||
case Instructions::f32x4_mul.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<4, Operators::Multiply>>(configuration);
|
||||
case Instructions::f32x4_sub.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<4, Operators::Subtract>>(configuration);
|
||||
case Instructions::f32x4_add.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<4, Operators::Add>>(configuration);
|
||||
case Instructions::f32x4_pmin.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<4, Operators::PseudoMinimum>>(configuration);
|
||||
case Instructions::f32x4_pmax.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<4, Operators::PseudoMaximum>>(configuration);
|
||||
case Instructions::f64x2_div.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<2, Operators::Divide>>(configuration);
|
||||
case Instructions::f64x2_mul.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<2, Operators::Multiply>>(configuration);
|
||||
case Instructions::f64x2_sub.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<2, Operators::Subtract>>(configuration);
|
||||
case Instructions::f64x2_add.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<2, Operators::Add>>(configuration);
|
||||
case Instructions::f64x2_pmin.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<2, Operators::PseudoMinimum>>(configuration);
|
||||
case Instructions::f64x2_pmax.value():
|
||||
return binary_numeric_operation<u128, u128, Operators::VectorFloatBinaryOp<2, Operators::PseudoMaximum>>(configuration);
|
||||
case Instructions::f32x4_ceil.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<4, Operators::Ceil>>(configuration);
|
||||
case Instructions::f32x4_floor.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<4, Operators::Floor>>(configuration);
|
||||
case Instructions::f32x4_trunc.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<4, Operators::Truncate>>(configuration);
|
||||
case Instructions::f32x4_nearest.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<4, Operators::NearbyIntegral>>(configuration);
|
||||
case Instructions::f32x4_sqrt.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<4, Operators::SquareRoot>>(configuration);
|
||||
case Instructions::f32x4_neg.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<4, Operators::Negate>>(configuration);
|
||||
case Instructions::f32x4_abs.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<4, Operators::Absolute>>(configuration);
|
||||
case Instructions::f64x2_ceil.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<2, Operators::Ceil>>(configuration);
|
||||
case Instructions::f64x2_floor.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<2, Operators::Floor>>(configuration);
|
||||
case Instructions::f64x2_trunc.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<2, Operators::Truncate>>(configuration);
|
||||
case Instructions::f64x2_nearest.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<2, Operators::NearbyIntegral>>(configuration);
|
||||
case Instructions::f64x2_sqrt.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<2, Operators::SquareRoot>>(configuration);
|
||||
case Instructions::f64x2_neg.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<2, Operators::Negate>>(configuration);
|
||||
case Instructions::f64x2_abs.value():
|
||||
return unary_operation<u128, u128, Operators::VectorFloatUnaryOp<2, Operators::Absolute>>(configuration);
|
||||
case Instructions::v128_not.value():
|
||||
case Instructions::v128_and.value():
|
||||
case Instructions::v128_andnot.value():
|
||||
|
@ -1430,23 +1490,16 @@ void BytecodeInterpreter::interpret(Configuration& configuration, InstructionPoi
|
|||
case Instructions::i8x16_bitmask.value():
|
||||
case Instructions::i8x16_narrow_i16x8_s.value():
|
||||
case Instructions::i8x16_narrow_i16x8_u.value():
|
||||
case Instructions::f32x4_ceil.value():
|
||||
case Instructions::f32x4_floor.value():
|
||||
case Instructions::f32x4_trunc.value():
|
||||
case Instructions::f32x4_nearest.value():
|
||||
case Instructions::i8x16_add.value():
|
||||
case Instructions::i8x16_add_sat_s.value():
|
||||
case Instructions::i8x16_add_sat_u.value():
|
||||
case Instructions::i8x16_sub.value():
|
||||
case Instructions::i8x16_sub_sat_s.value():
|
||||
case Instructions::i8x16_sub_sat_u.value():
|
||||
case Instructions::f64x2_ceil.value():
|
||||
case Instructions::f64x2_floor.value():
|
||||
case Instructions::i8x16_min_s.value():
|
||||
case Instructions::i8x16_min_u.value():
|
||||
case Instructions::i8x16_max_s.value():
|
||||
case Instructions::i8x16_max_u.value():
|
||||
case Instructions::f64x2_trunc.value():
|
||||
case Instructions::i8x16_avgr_u.value():
|
||||
case Instructions::i16x8_extadd_pairwise_i8x16_s.value():
|
||||
case Instructions::i16x8_extadd_pairwise_i8x16_u.value():
|
||||
|
@ -1469,7 +1522,6 @@ void BytecodeInterpreter::interpret(Configuration& configuration, InstructionPoi
|
|||
case Instructions::i16x8_sub.value():
|
||||
case Instructions::i16x8_sub_sat_s.value():
|
||||
case Instructions::i16x8_sub_sat_u.value():
|
||||
case Instructions::f64x2_nearest.value():
|
||||
case Instructions::i16x8_mul.value():
|
||||
case Instructions::i16x8_min_s.value():
|
||||
case Instructions::i16x8_min_u.value():
|
||||
|
@ -1508,7 +1560,6 @@ void BytecodeInterpreter::interpret(Configuration& configuration, InstructionPoi
|
|||
case Instructions::i64x2_extend_high_i32x4_s.value():
|
||||
case Instructions::i64x2_extend_low_i32x4_u.value():
|
||||
case Instructions::i64x2_extend_high_i32x4_u.value():
|
||||
case Instructions::i64x2_add.value():
|
||||
case Instructions::i64x2_sub.value():
|
||||
case Instructions::i64x2_mul.value():
|
||||
case Instructions::i64x2_eq.value():
|
||||
|
@ -1521,28 +1572,6 @@ void BytecodeInterpreter::interpret(Configuration& configuration, InstructionPoi
|
|||
case Instructions::i64x2_extmul_high_i32x4_s.value():
|
||||
case Instructions::i64x2_extmul_low_i32x4_u.value():
|
||||
case Instructions::i64x2_extmul_high_i32x4_u.value():
|
||||
case Instructions::f32x4_abs.value():
|
||||
case Instructions::f32x4_neg.value():
|
||||
case Instructions::f32x4_sqrt.value():
|
||||
case Instructions::f32x4_add.value():
|
||||
case Instructions::f32x4_sub.value():
|
||||
case Instructions::f32x4_mul.value():
|
||||
case Instructions::f32x4_div.value():
|
||||
case Instructions::f32x4_min.value():
|
||||
case Instructions::f32x4_max.value():
|
||||
case Instructions::f32x4_pmin.value():
|
||||
case Instructions::f32x4_pmax.value():
|
||||
case Instructions::f64x2_abs.value():
|
||||
case Instructions::f64x2_neg.value():
|
||||
case Instructions::f64x2_sqrt.value():
|
||||
case Instructions::f64x2_add.value():
|
||||
case Instructions::f64x2_sub.value():
|
||||
case Instructions::f64x2_mul.value():
|
||||
case Instructions::f64x2_div.value():
|
||||
case Instructions::f64x2_min.value():
|
||||
case Instructions::f64x2_max.value():
|
||||
case Instructions::f64x2_pmin.value():
|
||||
case Instructions::f64x2_pmax.value():
|
||||
case Instructions::i32x4_trunc_sat_f32x4_s.value():
|
||||
case Instructions::i32x4_trunc_sat_f32x4_u.value():
|
||||
case Instructions::f32x4_convert_i32x4_s.value():
|
||||
|
|
|
@ -364,6 +364,26 @@ struct Maximum {
|
|||
static StringView name() { return "maximum"sv; }
|
||||
};
|
||||
|
||||
struct PseudoMinimum {
|
||||
template<typename Lhs, typename Rhs>
|
||||
auto operator()(Lhs lhs, Rhs rhs) const
|
||||
{
|
||||
return rhs < lhs ? rhs : lhs;
|
||||
}
|
||||
|
||||
static StringView name() { return "pseudo_minimum"sv; }
|
||||
};
|
||||
|
||||
struct PseudoMaximum {
|
||||
template<typename Lhs, typename Rhs>
|
||||
auto operator()(Lhs lhs, Rhs rhs) const
|
||||
{
|
||||
return lhs < rhs ? rhs : lhs;
|
||||
}
|
||||
|
||||
static StringView name() { return "pseudo_maximum"sv; }
|
||||
};
|
||||
|
||||
struct CopySign {
|
||||
template<typename Lhs, typename Rhs>
|
||||
auto operator()(Lhs lhs, Rhs rhs) const
|
||||
|
@ -462,6 +482,61 @@ struct Ceil {
|
|||
static StringView name() { return "ceil"sv; }
|
||||
};
|
||||
|
||||
template<size_t VectorSize, typename Op>
|
||||
struct VectorFloatBinaryOp {
|
||||
auto operator()(u128 lhs, u128 rhs) const
|
||||
{
|
||||
using VectorType = NativeFloatingVectorType<128, VectorSize, NativeFloatingType<128 / VectorSize>>;
|
||||
auto first = bit_cast<VectorType>(lhs);
|
||||
auto second = bit_cast<VectorType>(rhs);
|
||||
VectorType result;
|
||||
Op op;
|
||||
for (size_t i = 0; i < VectorSize; ++i) {
|
||||
result[i] = op(first[i], second[i]);
|
||||
}
|
||||
return bit_cast<u128>(result);
|
||||
}
|
||||
|
||||
static StringView name()
|
||||
{
|
||||
switch (VectorSize) {
|
||||
case 4:
|
||||
return "vecf(32x4).binary_op"sv;
|
||||
case 2:
|
||||
return "vecf(64x2).binary_op"sv;
|
||||
default:
|
||||
VERIFY_NOT_REACHED();
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template<size_t VectorSize, typename Op>
|
||||
struct VectorFloatUnaryOp {
|
||||
auto operator()(u128 lhs) const
|
||||
{
|
||||
using VectorType = NativeFloatingVectorType<128, VectorSize, NativeFloatingType<128 / VectorSize>>;
|
||||
auto first = bit_cast<VectorType>(lhs);
|
||||
VectorType result;
|
||||
Op op;
|
||||
for (size_t i = 0; i < VectorSize; ++i) {
|
||||
result[i] = op(first[i]);
|
||||
}
|
||||
return bit_cast<u128>(result);
|
||||
}
|
||||
|
||||
static StringView name()
|
||||
{
|
||||
switch (VectorSize) {
|
||||
case 4:
|
||||
return "vecf(32x4).unary_op"sv;
|
||||
case 2:
|
||||
return "vecf(64x2).unary_op"sv;
|
||||
default:
|
||||
VERIFY_NOT_REACHED();
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
struct Floor {
|
||||
template<typename Lhs>
|
||||
auto operator()(Lhs lhs) const
|
||||
|
@ -479,7 +554,7 @@ struct Floor {
|
|||
|
||||
struct Truncate {
|
||||
template<typename Lhs>
|
||||
AK::ErrorOr<Lhs, StringView> operator()(Lhs lhs) const
|
||||
auto operator()(Lhs lhs) const
|
||||
{
|
||||
if constexpr (IsSame<Lhs, float>)
|
||||
return truncf(lhs);
|
||||
|
|
Loading…
Reference in a new issue