Skip to content

Use platform intrinsics, not LLVM, for floor & ceil #47

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Dec 8, 2020
Merged

Use platform intrinsics, not LLVM, for floor & ceil #47

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
7538ff8
Revert "Disable riscv64gc"
workingjubilee Nov 30, 2020
e9cc306
Remove round, trunc tests
workingjubilee Nov 30, 2020
3d9bbf9
Use simd_{floor,ceil} intrinsics for round.rs
workingjubilee Nov 30, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 1 addition & 2 deletions .github/workflows/ci.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,8 +121,7 @@ jobs:
# for NaNs which makes it worth testing on despite that.
- mips-unknown-linux-gnu
- mips64-unknown-linux-gnuabi64
# TODO: reenable pending https://github.com/rust-lang/rust/issues/77866
# - riscv64gc-unknown-linux-gnu
- riscv64gc-unknown-linux-gnu

steps:
- uses: actions/checkout@v2
Expand Down
6 changes: 6 additions & 0 deletions crates/core_simd/src/intrinsics.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,4 +39,10 @@ extern "platform-intrinsic" {

/// fptoui/fptosi/uitofp/sitofp
pub(crate) fn simd_cast<T, U>(x: T) -> U;

// floor
pub(crate) fn simd_floor<T>(x: T) -> T;

// ceil
pub(crate) fn simd_ceil<T>(x: T) -> T;
}
85 changes: 10 additions & 75 deletions crates/core_simd/src/round.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,60 +1,23 @@
macro_rules! implement {
{
impl $type:ident {
int_type = $int_type:ident,
floor = $floor_intrinsic:literal,
ceil = $ceil_intrinsic:literal,
round = $round_intrinsic:literal,
trunc = $trunc_intrinsic:literal,
int_type = $int_type:ident
}
} => {
mod $type {
#[allow(improper_ctypes)]
extern "C" {
#[link_name = $floor_intrinsic]
fn floor_intrinsic(x: crate::$type) -> crate::$type;
#[link_name = $ceil_intrinsic]
fn ceil_intrinsic(x: crate::$type) -> crate::$type;
#[link_name = $round_intrinsic]
fn round_intrinsic(x: crate::$type) -> crate::$type;
#[link_name = $trunc_intrinsic]
fn trunc_intrinsic(x: crate::$type) -> crate::$type;
}

impl crate::$type {
/// Returns the largest integer less than or equal to each lane.
#[must_use = "method returns a new vector and does not mutate the original value"]
#[inline]
pub fn floor(self) -> Self {
unsafe { floor_intrinsic(self) }
unsafe { crate::intrinsics::simd_floor(self) }
}

/// Returns the smallest integer greater than or equal to each lane.
#[must_use = "method returns a new vector and does not mutate the original value"]
#[inline]
pub fn ceil(self) -> Self {
unsafe { ceil_intrinsic(self) }
}

/// Returns the nearest integer to each lane. Round half-way cases away from 0.0.
#[must_use = "method returns a new vector and does not mutate the original value"]
#[inline]
pub fn round(self) -> Self {
unsafe { round_intrinsic(self) }
}

/// Returns the integer part of each lane.
#[must_use = "method returns a new vector and does not mutate the original value"]
#[inline]
pub fn trunc(self) -> Self {
unsafe { trunc_intrinsic(self) }
}

/// Returns the fractional part of each lane.
#[must_use = "method returns a new vector and does not mutate the original value"]
#[inline]
pub fn fract(self) -> Self {
self - self.trunc()
unsafe { crate::intrinsics::simd_ceil(self) }
}

/// Rounds toward zero and converts to the same-width integer type, assuming that
Expand Down Expand Up @@ -84,70 +47,42 @@ macro_rules! implement {

implement! {
impl f32x2 {
int_type = i32x2,
floor = "llvm.floor.v2f32",
ceil = "llvm.ceil.v2f32",
round = "llvm.round.v2f32",
trunc = "llvm.trunc.v2f32",
int_type = i32x2
}
}

implement! {
impl f32x4 {
int_type = i32x4,
floor = "llvm.floor.v4f32",
ceil = "llvm.ceil.v4f32",
round = "llvm.round.v4f32",
trunc = "llvm.trunc.v4f32",
int_type = i32x4
}
}

implement! {
impl f32x8 {
int_type = i32x8,
floor = "llvm.floor.v8f32",
ceil = "llvm.ceil.v8f32",
round = "llvm.round.v8f32",
trunc = "llvm.trunc.v8f32",
int_type = i32x8
}
}

implement! {
impl f32x16 {
int_type = i32x16,
floor = "llvm.floor.v16f32",
ceil = "llvm.ceil.v16f32",
round = "llvm.round.v16f32",
trunc = "llvm.trunc.v16f32",
int_type = i32x16
}
}

implement! {
impl f64x2 {
int_type = i64x2,
floor = "llvm.floor.v2f64",
ceil = "llvm.ceil.v2f64",
round = "llvm.round.v2f64",
trunc = "llvm.trunc.v2f64",
int_type = i64x2
}
}

implement! {
impl f64x4 {
int_type = i64x4,
floor = "llvm.floor.v4f64",
ceil = "llvm.ceil.v4f64",
round = "llvm.round.v4f64",
trunc = "llvm.trunc.v4f64",
int_type = i64x4
}
}

implement! {
impl f64x8 {
int_type = i64x8,
floor = "llvm.floor.v8f64",
ceil = "llvm.ceil.v8f64",
round = "llvm.round.v8f64",
trunc = "llvm.trunc.v8f64",
int_type = i64x8
}
}
36 changes: 0 additions & 36 deletions crates/core_simd/tests/ops_impl/float_macros.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -353,42 +353,6 @@ macro_rules! float_tests {
}
}

#[test]
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)]
fn round_odd_floats() {
for v in slice_chunks(&C) {
let expected = apply_unary_lanewise(v, <$scalar>::round);
assert_biteq!(v.round(), expected);
}
}

#[test]
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)]
fn round_mode() {
assert_biteq!(core_simd::$vector::splat(1.5).round(), core_simd::$vector::splat(2.0));
assert_biteq!(core_simd::$vector::splat(2.5).round(), core_simd::$vector::splat(3.0));
assert_biteq!(core_simd::$vector::splat(-1.5).round(), core_simd::$vector::splat(-2.0));
assert_biteq!(core_simd::$vector::splat(-2.5).round(), core_simd::$vector::splat(-3.0));
}

#[test]
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)]
fn trunc_odd_floats() {
for v in slice_chunks(&C) {
let expected = apply_unary_lanewise(v, <$scalar>::trunc);
assert_biteq!(v.trunc(), expected);
}
}

#[test]
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)]
fn fract_odd_floats() {
for v in slice_chunks(&C) {
let expected = apply_unary_lanewise(v, <$scalar>::fract);
assert_biteq!(v.fract(), expected);
}
}

#[test]
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test)]
fn to_int_unchecked() {
Expand Down