compiler: add simd_ctpop intrinsic

This commit is contained in:
Jubilee Young 2024-05-18 17:56:49 -07:00
parent eb1a5c9bb3
commit 1914c722b5
7 changed files with 76 additions and 40 deletions

View file

@ -2336,7 +2336,10 @@ macro_rules! arith_unary {
}
// Unary integer intrinsics
if matches!(name, sym::simd_bswap | sym::simd_bitreverse | sym::simd_ctlz | sym::simd_cttz) {
if matches!(
name,
sym::simd_bswap | sym::simd_bitreverse | sym::simd_ctlz | sym::simd_ctpop | sym::simd_cttz
) {
let vec_ty = bx.cx.type_vector(
match *in_elem.kind() {
ty::Int(i) => bx.cx.type_int_from_ty(i),
@ -2354,31 +2357,38 @@ macro_rules! arith_unary {
sym::simd_bswap => "bswap",
sym::simd_bitreverse => "bitreverse",
sym::simd_ctlz => "ctlz",
sym::simd_ctpop => "ctpop",
sym::simd_cttz => "cttz",
_ => unreachable!(),
};
let int_size = in_elem.int_size_and_signed(bx.tcx()).0.bits();
let llvm_intrinsic = &format!("llvm.{}.v{}i{}", intrinsic_name, in_len, int_size,);
return if name == sym::simd_bswap && int_size == 8 {
return match name {
// byte swap is no-op for i8/u8
Ok(args[0].immediate())
} else if matches!(name, sym::simd_ctlz | sym::simd_cttz) {
let fn_ty = bx.type_func(&[vec_ty, bx.type_i1()], vec_ty);
let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
Ok(bx.call(
fn_ty,
None,
None,
f,
&[args[0].immediate(), bx.const_int(bx.type_i1(), 0)],
None,
None,
))
} else {
let fn_ty = bx.type_func(&[vec_ty], vec_ty);
let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
Ok(bx.call(fn_ty, None, None, f, &[args[0].immediate()], None, None))
sym::simd_bswap if int_size == 8 => Ok(args[0].immediate()),
sym::simd_ctlz | sym::simd_cttz => {
// this fun bonus i1 arg means "poison if the arg vector contains zero"
let fn_ty = bx.type_func(&[vec_ty, bx.type_i1()], vec_ty);
let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
Ok(bx.call(
fn_ty,
None,
None,
f,
// simd_ctlz and simd_cttz are exposed to safe code, so let's not poison anything
&[args[0].immediate(), bx.const_int(bx.type_i1(), 0)],
None,
None,
))
}
sym::simd_bswap | sym::simd_bitreverse | sym::simd_ctpop => {
// simple unary argument cases
let fn_ty = bx.type_func(&[vec_ty], vec_ty);
let f = bx.declare_cfn(llvm_intrinsic, llvm::UnnamedAddr::No, fn_ty);
Ok(bx.call(fn_ty, None, None, f, &[args[0].immediate()], None, None))
}
_ => unreachable!(),
};
}

View file

@ -607,6 +607,7 @@ pub fn check_intrinsic_type(
| sym::simd_bitreverse
| sym::simd_ctlz
| sym::simd_cttz
| sym::simd_ctpop
| sym::simd_fsqrt
| sym::simd_fsin
| sym::simd_fcos

View file

@ -1681,6 +1681,7 @@
simd_cast_ptr,
simd_ceil,
simd_ctlz,
simd_ctpop,
simd_cttz,
simd_div,
simd_eq,

View file

@ -569,6 +569,13 @@
#[rustc_nounwind]
pub fn simd_ctlz<T>(x: T) -> T;
/// Count the number of ones in each element.
///
/// `T` must be a vector of integers.
#[rustc_nounwind]
#[cfg(not(bootstrap))]
pub fn simd_ctpop<T>(x: T) -> T;
/// Count the trailing zeros of each element.
///
/// `T` must be a vector of integers.

View file

@ -30,6 +30,7 @@
fn simd_bswap<T>(x: T) -> T;
fn simd_bitreverse<T>(x: T) -> T;
fn simd_ctlz<T>(x: T) -> T;
fn simd_ctpop<T>(x: T) -> T;
fn simd_cttz<T>(x: T) -> T;
}
@ -77,7 +78,6 @@ fn main() {
simd_cttz(x);
simd_cttz(y);
simd_add(0, 0);
//~^ ERROR expected SIMD input type, found non-SIMD `i32`
simd_sub(0, 0);
@ -108,24 +108,25 @@ fn main() {
simd_cttz(0);
//~^ ERROR expected SIMD input type, found non-SIMD `i32`
simd_shl(z, z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_shr(z, z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_and(z, z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_or(z, z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_xor(z, z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_bswap(z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_bitreverse(z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_ctlz(z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_ctpop(z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
simd_cttz(z);
//~^ ERROR unsupported operation on `f32x4` with element `f32`
//~^ ERROR unsupported operation on `f32x4` with element `f32`
}
}

View file

@ -83,59 +83,65 @@ LL | simd_cttz(0);
| ^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_shl` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:112:9
--> $DIR/generic-arithmetic-2.rs:111:9
|
LL | simd_shl(z, z);
| ^^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_shr` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:114:9
--> $DIR/generic-arithmetic-2.rs:113:9
|
LL | simd_shr(z, z);
| ^^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_and` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:116:9
--> $DIR/generic-arithmetic-2.rs:115:9
|
LL | simd_and(z, z);
| ^^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_or` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:118:9
--> $DIR/generic-arithmetic-2.rs:117:9
|
LL | simd_or(z, z);
| ^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_xor` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:120:9
--> $DIR/generic-arithmetic-2.rs:119:9
|
LL | simd_xor(z, z);
| ^^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_bswap` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:122:9
--> $DIR/generic-arithmetic-2.rs:121:9
|
LL | simd_bswap(z);
| ^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_bitreverse` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:124:9
--> $DIR/generic-arithmetic-2.rs:123:9
|
LL | simd_bitreverse(z);
| ^^^^^^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_ctlz` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:126:9
--> $DIR/generic-arithmetic-2.rs:125:9
|
LL | simd_ctlz(z);
| ^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_ctpop` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:127:9
|
LL | simd_ctpop(z);
| ^^^^^^^^^^^^^
error[E0511]: invalid monomorphization of `simd_cttz` intrinsic: unsupported operation on `f32x4` with element `f32`
--> $DIR/generic-arithmetic-2.rs:128:9
--> $DIR/generic-arithmetic-2.rs:129:9
|
LL | simd_cttz(z);
| ^^^^^^^^^^^^
error: aborting due to 23 previous errors
error: aborting due to 24 previous errors
For more information about this error, try `rustc --explain E0511`.

View file

@ -47,6 +47,7 @@ macro_rules! all_eq_ {
fn simd_bswap<T>(x: T) -> T;
fn simd_bitreverse<T>(x: T) -> T;
fn simd_ctlz<T>(x: T) -> T;
fn simd_ctpop<T>(x: T) -> T;
fn simd_cttz<T>(x: T) -> T;
}
@ -57,6 +58,8 @@ fn main() {
let x2 = i32x4(2, 3, 4, 5);
let y2 = U32::<4>([2, 3, 4, 5]);
let z2 = f32x4(2.0, 3.0, 4.0, 5.0);
let x3 = i32x4(0, i32::MAX, i32::MIN, -1_i32);
let y3 = U32::<4>([0, i32::MAX as _, i32::MIN as _, -1_i32 as _]);
unsafe {
all_eq!(simd_add(x1, x2), i32x4(3, 5, 7, 9));
@ -147,6 +150,13 @@ fn main() {
all_eq!(simd_ctlz(x1), i32x4(31, 30, 30, 29));
all_eq_!(simd_ctlz(y1), U32::<4>([31, 30, 30, 29]));
all_eq!(simd_ctpop(x1), i32x4(1, 1, 2, 1));
all_eq_!(simd_ctpop(y1), U32::<4>([1, 1, 2, 1]));
all_eq!(simd_ctpop(x2), i32x4(1, 2, 1, 2));
all_eq_!(simd_ctpop(y2), U32::<4>([1, 2, 1, 2]));
all_eq!(simd_ctpop(x3), i32x4(0, 31, 1, 32));
all_eq_!(simd_ctpop(y3), U32::<4>([0, 31, 1, 32]));
all_eq!(simd_cttz(x1), i32x4(0, 1, 0, 2));
all_eq_!(simd_cttz(y1), U32::<4>([0, 1, 0, 2]));
}