mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Auto merge of #2001 - RalfJung:simd, r=RalfJung

Browse Source 
add more simd_reduce intrinsics

This makes large parts of the portable-simd test suite work. :D
Cc https://github.com/rust-lang/portable-simd/pull/255
This commit is contained in:
bors 2022-03-06 00:25:19 +00:00
commit 54b14b7f01
3 changed files with 84 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/machine.rs
View File

@ -267,6 +267,7 @@ pub struct PrimitiveLayouts<'tcx> {
pub u8: TyAndLayout<'tcx>,
pub u32: TyAndLayout<'tcx>,
pub usize: TyAndLayout<'tcx>,
pub bool: TyAndLayout<'tcx>,
}
impl<'mir, 'tcx: 'mir> PrimitiveLayouts<'tcx> {
@ -279,6 +280,7 @@ fn new(layout_cx: LayoutCx<'tcx, TyCtxt<'tcx>>) -> Result<Self, LayoutError<'tcx
u8: layout_cx.layout_of(layout_cx.tcx.types.u8)?,
u32: layout_cx.layout_of(layout_cx.tcx.types.u32)?,
usize: layout_cx.layout_of(layout_cx.tcx.types.usize)?,
bool: layout_cx.layout_of(layout_cx.tcx.types.bool)?,
})
}
}

91
src/shims/intrinsics.rs
View File

@ -324,12 +324,22 @@ fn call_intrinsic(
assert_eq!(dest_len, op_len);
enum Op {
MirOp(mir::UnOp),
Abs,
}
let which = match intrinsic_name {
"simd_neg" => Op::MirOp(mir::UnOp::Neg),
"simd_fabs" => Op::Abs,
_ => unreachable!(),
};
for i in 0..dest_len {
let op = this.read_immediate(&this.mplace_index(&op, i)?.into())?;
let dest = this.mplace_index(&dest, i)?;
let val = match intrinsic_name {
"simd_neg" => this.unary_op(mir::UnOp::Neg, &op)?.to_scalar()?,
"simd_fabs" => {
let val = match which {
Op::MirOp(mir_op) => this.unary_op(mir_op, &op)?.to_scalar()?,
Op::Abs => {
// Works for f32 and f64.
let ty::Float(float_ty) = op.layout.ty.kind() else {
bug!("simd_fabs operand is not a float")
@ -341,7 +351,6 @@ fn call_intrinsic(
FloatTy::F64 => Scalar::from_f64(op.to_f64()?.abs()),
}
}
_ => bug!(),
};
this.write_scalar(val, &dest.into())?;
}
@ -373,7 +382,7 @@ fn call_intrinsic(
assert_eq!(dest_len, left_len);
assert_eq!(dest_len, right_len);
let op = match intrinsic_name {
let mir_op = match intrinsic_name {
"simd_add" => BinOp::Add,
"simd_sub" => BinOp::Sub,
"simd_mul" => BinOp::Mul,
@ -397,8 +406,8 @@ fn call_intrinsic(
let left = this.read_immediate(&this.mplace_index(&left, i)?.into())?;
let right = this.read_immediate(&this.mplace_index(&right, i)?.into())?;
let dest = this.mplace_index(&dest, i)?;
let (val, overflowed, ty) = this.overflowing_binary_op(op, &left, &right)?;
if matches!(op, BinOp::Shl | BinOp::Shr) {
let (val, overflowed, ty) = this.overflowing_binary_op(mir_op, &left, &right)?;
if matches!(mir_op, BinOp::Shl | BinOp::Shr) {
// Shifts have extra UB as SIMD operations that the MIR binop does not have.
// See <https://github.com/rust-lang/rust/issues/91237>.
if overflowed {
@ -406,7 +415,7 @@ fn call_intrinsic(
throw_ub_format!("overflowing shift by {} in `{}` in SIMD lane {}", r_val, intrinsic_name, i);
}
}
if matches!(op, BinOp::Eq | BinOp::Ne | BinOp::Lt | BinOp::Le | BinOp::Gt | BinOp::Ge) {
if matches!(mir_op, BinOp::Eq | BinOp::Ne | BinOp::Lt | BinOp::Le | BinOp::Gt | BinOp::Ge) {
// Special handling for boolean-returning operations
assert_eq!(ty, this.tcx.types.bool);
let val = val.to_bool().unwrap();
@ -419,28 +428,70 @@ fn call_intrinsic(
}
}
}
"simd_reduce_any" | "simd_reduce_all" => {
#[rustfmt::skip]
| "simd_reduce_and"
| "simd_reduce_or"
| "simd_reduce_xor"
| "simd_reduce_any"
| "simd_reduce_all" => {
use mir::BinOp;
let &[ref op] = check_arg_count(args)?;
let (op, op_len) = this.operand_to_simd(op)?;
// the neutral element
let mut res = match intrinsic_name {
"simd_reduce_any" => false,
"simd_reduce_all" => true,
_ => bug!(),
let imm_from_bool =
|b| ImmTy::from_scalar(Scalar::from_bool(b), this.machine.layouts.bool);
enum Op {
MirOp(BinOp),
MirOpBool(BinOp),
}
// The initial value is the neutral element.
let (which, init) = match intrinsic_name {
"simd_reduce_and" => (Op::MirOp(BinOp::BitAnd), ImmTy::from_int(-1, dest.layout)),
"simd_reduce_or" => (Op::MirOp(BinOp::BitOr), ImmTy::from_int(0, dest.layout)),
"simd_reduce_xor" => (Op::MirOp(BinOp::BitXor), ImmTy::from_int(0, dest.layout)),
"simd_reduce_any" => (Op::MirOpBool(BinOp::BitOr), imm_from_bool(false)),
"simd_reduce_all" => (Op::MirOpBool(BinOp::BitAnd), imm_from_bool(true)),
_ => unreachable!(),
};
let mut res = init;
for i in 0..op_len {
let op = this.read_immediate(&this.mplace_index(&op, i)?.into())?;
let val = simd_element_to_bool(op)?;
res = match intrinsic_name {
"simd_reduce_any" => res | val,
"simd_reduce_all" => res & val,
_ => bug!(),
res = match which {
Op::MirOp(mir_op) => {
this.binary_op(mir_op, &res, &op)?
}
Op::MirOpBool(mir_op) => {
let op = imm_from_bool(simd_element_to_bool(op)?);
this.binary_op(mir_op, &res, &op)?
}
};
}
this.write_immediate(*res, dest)?;
}
#[rustfmt::skip]
| "simd_reduce_add_ordered"
| "simd_reduce_mul_ordered" => {
use mir::BinOp;
this.write_scalar(Scalar::from_bool(res), dest)?;
let &[ref op, ref init] = check_arg_count(args)?;
let (op, op_len) = this.operand_to_simd(op)?;
let init = this.read_immediate(init)?;
let mir_op = match intrinsic_name {
"simd_reduce_add_ordered" => BinOp::Add,
"simd_reduce_mul_ordered" => BinOp::Mul,
_ => unreachable!(),
};
let mut res = init;
for i in 0..op_len {
let op = this.read_immediate(&this.mplace_index(&op, i)?.into())?;
res = this.binary_op(mir_op, &res, &op)?;
}
this.write_immediate(*res, dest)?;
}
"simd_select" => {
let &[ref mask, ref yes, ref no] = check_arg_count(args)?;

11
tests/run-pass/portable-simd.rs
View File

@ -68,6 +68,17 @@ fn simd_ops_i32() {
assert_eq!(a.lanes_lt(i32x4::splat(5)*b), Mask::from_int(i32x4::from_array([0, 0, -1, 0])));
assert_eq!(a.lanes_ge(i32x4::splat(5)*b), Mask::from_int(i32x4::from_array([-1, -1, 0, -1])));
assert_eq!(a.lanes_gt(i32x4::splat(5)*b), Mask::from_int(i32x4::from_array([-1, 0, 0, -1])));
assert_eq!(a.horizontal_and(), 10);
assert_eq!(b.horizontal_and(), 0);
assert_eq!(a.horizontal_or(), 10);
assert_eq!(b.horizontal_or(), -1);
assert_eq!(a.horizontal_xor(), 0);
assert_eq!(b.horizontal_xor(), -4);
assert_eq!(a.horizontal_sum(), 40);
assert_eq!(b.horizontal_sum(), 2);
assert_eq!(a.horizontal_product(), 100*100);
assert_eq!(b.horizontal_product(), 6*-4);
}
fn simd_intrinsics() {