Implement Extractlane, UaddSat, and UsubSat for Cranelift interpreter (#3188)
* Implement `Extractlane`, `UaddSat`, and `UsubSat` for Cranelift interpreter Implemented the `Extractlane`, `UaddSat`, and `UsubSat` opcodes for the interpreter, and added helper functions for working with SIMD vectors (`extractlanes`, `vectorizelanes`, and `binary_arith`). Copyright (c) 2021, Arm Limited * Re-use tests + constrict Vector assert - Re-use interpreter tests as runtests where supported. - Constrict Vector assertion. - Code style adjustments following feedback. Copyright (c) 2021, Arm Limited * Runtest `i32x4` vectors on AArch64; add `i64x2` tests Copyright (c) 2021, Arm Limited * Add `simd-` prefix to test filenames Copyright (c) 2021, Arm Limited * Return aliased `SmallVec` from `extractlanes` Using a `SmallVec<[i128; 4]>` allows larger-width 128-bit vectors (`i32x4`, `i64x2`, ...) to not cause heap allocations. Copyright (c) 2021, Arm Limited * Accept slice to `vectorizelanes` rather than `Vec` Copyright (c) 2021, Arm Limited
This commit is contained in:
@@ -496,10 +496,22 @@ where
|
||||
binary(Value::div, inc, two)?
|
||||
}
|
||||
Opcode::Iadd => binary(Value::add, arg(0)?, arg(1)?)?,
|
||||
Opcode::UaddSat => unimplemented!("UaddSat"),
|
||||
Opcode::UaddSat => assign(binary_arith(
|
||||
arg(0)?,
|
||||
arg(1)?,
|
||||
ctrl_ty,
|
||||
Value::add_sat,
|
||||
true,
|
||||
)?),
|
||||
Opcode::SaddSat => unimplemented!("SaddSat"),
|
||||
Opcode::Isub => binary(Value::sub, arg(0)?, arg(1)?)?,
|
||||
Opcode::UsubSat => unimplemented!("UsubSat"),
|
||||
Opcode::UsubSat => assign(binary_arith(
|
||||
arg(0)?,
|
||||
arg(1)?,
|
||||
ctrl_ty,
|
||||
Value::sub_sat,
|
||||
true,
|
||||
)?),
|
||||
Opcode::SsubSat => unimplemented!("SsubSat"),
|
||||
Opcode::Ineg => binary(Value::sub, Value::int(0, ctrl_ty)?, arg(0)?)?,
|
||||
Opcode::Iabs => unimplemented!("Iabs"),
|
||||
@@ -661,7 +673,11 @@ where
|
||||
Opcode::Swizzle => unimplemented!("Swizzle"),
|
||||
Opcode::Splat => unimplemented!("Splat"),
|
||||
Opcode::Insertlane => unimplemented!("Insertlane"),
|
||||
Opcode::Extractlane => unimplemented!("Extractlane"),
|
||||
Opcode::Extractlane => {
|
||||
let value =
|
||||
extractlanes(&arg(0)?, ctrl_ty.lane_type())?[Value::into_int(imm())? as usize];
|
||||
assign(Value::int(value, ctrl_ty.lane_type())?)
|
||||
}
|
||||
Opcode::VhighBits => unimplemented!("VhighBits"),
|
||||
Opcode::Vsplit => unimplemented!("Vsplit"),
|
||||
Opcode::Vconcat => unimplemented!("Vconcat"),
|
||||
@@ -869,3 +885,80 @@ where
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
type SimdVec = SmallVec<[i128; 4]>;
|
||||
|
||||
/// Converts a SIMD vector value into a Rust vector of i128 for processing.
|
||||
fn extractlanes<V>(x: &V, lane_type: types::Type) -> ValueResult<SimdVec>
|
||||
where
|
||||
V: Value,
|
||||
{
|
||||
let iterations = match lane_type {
|
||||
types::I8 => 1,
|
||||
types::I16 => 2,
|
||||
types::I32 => 4,
|
||||
types::I64 => 8,
|
||||
_ => unimplemented!("Only 128-bit vectors are currently supported."),
|
||||
};
|
||||
|
||||
let x = x.into_array()?;
|
||||
let mut lanes = SimdVec::new();
|
||||
for (i, _) in x.iter().enumerate() {
|
||||
let mut lane: i128 = 0;
|
||||
if i % iterations != 0 {
|
||||
continue;
|
||||
}
|
||||
for j in 0..iterations {
|
||||
lane += (x[i + j] as i128) << (8 * j);
|
||||
}
|
||||
lanes.push(lane);
|
||||
}
|
||||
return Ok(lanes);
|
||||
}
|
||||
|
||||
/// Convert a Rust array of i128s back into a `Value::vector`.
|
||||
fn vectorizelanes<V>(x: &[i128], vector_type: types::Type) -> ValueResult<V>
|
||||
where
|
||||
V: Value,
|
||||
{
|
||||
let iterations = match vector_type.lane_type() {
|
||||
types::I8 => 1,
|
||||
types::I16 => 2,
|
||||
types::I32 => 4,
|
||||
types::I64 => 8,
|
||||
_ => unimplemented!("Only 128-bit vectors are currently supported."),
|
||||
};
|
||||
let mut result: [u8; 16] = [0; 16];
|
||||
for (i, val) in x.iter().enumerate() {
|
||||
let val = *val;
|
||||
for j in 0..iterations {
|
||||
result[(i * iterations) + j] = (val >> (8 * j)) as u8;
|
||||
}
|
||||
}
|
||||
Value::vector(result, vector_type)
|
||||
}
|
||||
|
||||
/// Performs the supplied binary arithmetic `op` on two SIMD vectors.
|
||||
fn binary_arith<V, F>(x: V, y: V, vector_type: types::Type, op: F, unsigned: bool) -> ValueResult<V>
|
||||
where
|
||||
V: Value,
|
||||
F: Fn(V, V) -> ValueResult<V>,
|
||||
{
|
||||
let arg0 = extractlanes(&x, vector_type.lane_type())?;
|
||||
let arg1 = extractlanes(&y, vector_type.lane_type())?;
|
||||
let mut result = Vec::new();
|
||||
for (lhs, rhs) in arg0.into_iter().zip(arg1) {
|
||||
// The initial Value::int needs to be on a separate line so the
|
||||
// compiler can determine concrete types.
|
||||
let mut lhs: V = Value::int(lhs, vector_type.lane_type())?;
|
||||
let mut rhs: V = Value::int(rhs, vector_type.lane_type())?;
|
||||
if unsigned {
|
||||
lhs = lhs.convert(ValueConversionKind::ToUnsigned)?;
|
||||
rhs = rhs.convert(ValueConversionKind::ToUnsigned)?;
|
||||
}
|
||||
let sum = op(lhs, rhs)?;
|
||||
let sum = sum.into_int()?;
|
||||
result.push(sum);
|
||||
}
|
||||
vectorizelanes(&result, vector_type)
|
||||
}
|
||||
|
||||
@@ -22,6 +22,7 @@ pub trait Value: Clone + From<DataValue> {
|
||||
fn bool(b: bool, ty: Type) -> ValueResult<Self>;
|
||||
fn into_bool(self) -> ValueResult<bool>;
|
||||
fn vector(v: [u8; 16], ty: Type) -> ValueResult<Self>;
|
||||
fn into_array(&self) -> ValueResult<[u8; 16]>;
|
||||
fn convert(self, kind: ValueConversionKind) -> ValueResult<Self>;
|
||||
fn concat(self, other: Self) -> ValueResult<Self>;
|
||||
|
||||
@@ -47,6 +48,10 @@ pub trait Value: Clone + From<DataValue> {
|
||||
fn div(self, other: Self) -> ValueResult<Self>;
|
||||
fn rem(self, other: Self) -> ValueResult<Self>;
|
||||
|
||||
// Saturating arithmetic.
|
||||
fn add_sat(self, other: Self) -> ValueResult<Self>;
|
||||
fn sub_sat(self, other: Self) -> ValueResult<Self>;
|
||||
|
||||
// Bitwise.
|
||||
fn shl(self, other: Self) -> ValueResult<Self>;
|
||||
fn ushr(self, other: Self) -> ValueResult<Self>;
|
||||
@@ -80,6 +85,7 @@ pub enum ValueTypeClass {
|
||||
Integer,
|
||||
Boolean,
|
||||
Float,
|
||||
Vector,
|
||||
}
|
||||
|
||||
impl Display for ValueTypeClass {
|
||||
@@ -88,6 +94,7 @@ impl Display for ValueTypeClass {
|
||||
ValueTypeClass::Integer => write!(f, "integer"),
|
||||
ValueTypeClass::Boolean => write!(f, "boolean"),
|
||||
ValueTypeClass::Float => write!(f, "float"),
|
||||
ValueTypeClass::Vector => write!(f, "vector"),
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -225,8 +232,16 @@ impl Value for DataValue {
|
||||
}
|
||||
}
|
||||
|
||||
fn vector(_v: [u8; 16], _ty: Type) -> ValueResult<Self> {
|
||||
unimplemented!()
|
||||
fn vector(v: [u8; 16], ty: Type) -> ValueResult<Self> {
|
||||
assert!(ty.is_vector() && ty.bytes() == 16);
|
||||
Ok(DataValue::V128(v))
|
||||
}
|
||||
|
||||
fn into_array(&self) -> ValueResult<[u8; 16]> {
|
||||
match *self {
|
||||
DataValue::V128(v) => Ok(v),
|
||||
_ => Err(ValueError::InvalidType(ValueTypeClass::Vector, self.ty())),
|
||||
}
|
||||
}
|
||||
|
||||
fn convert(self, kind: ValueConversionKind) -> ValueResult<Self> {
|
||||
@@ -380,6 +395,14 @@ impl Value for DataValue {
|
||||
binary_match!(%(&self, &other); [I8, I16, I32, I64])
|
||||
}
|
||||
|
||||
fn add_sat(self, other: Self) -> ValueResult<Self> {
|
||||
binary_match!(saturating_add(self, &other); [I8, I16, I32, I64, I128, U8, U16, U32, U64, U128])
|
||||
}
|
||||
|
||||
fn sub_sat(self, other: Self) -> ValueResult<Self> {
|
||||
binary_match!(saturating_sub(self, &other); [I8, I16, I32, I64, I128, U8, U16, U32, U64, U128])
|
||||
}
|
||||
|
||||
fn shl(self, other: Self) -> ValueResult<Self> {
|
||||
binary_match!(<<(&self, &other); [I8, I16, I32, I64])
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user