Skip to content

Add f64<->f32 widening/narrowing conversions #127

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

Open
valadaptive wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add f64<->f32 widening/narrowing conversions #127

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
50 changes: 38 additions & 12 deletions fearless_simd/src/generated/avx2.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,6 +178,10 @@ impl Simd for Avx2 {
unsafe { _mm256_setr_m128(a.into(), b.into()).simd_into(self) }
}
#[inline(always)]
fn widen_f32x4(self, a: f32x4<Self>) -> f64x4<Self> {
unsafe { _mm256_cvtps_pd(a.into()).simd_into(self) }
}
#[inline(always)]
fn reinterpret_f64_f32x4(self, a: f32x4<Self>) -> f64x2<Self> {
f64x2 {
val: bytemuck::cast(a.val),
Expand Down Expand Up @@ -1449,6 +1453,15 @@ impl Simd for Avx2 {
}
}
#[inline(always)]
fn widen_f32x8(self, a: f32x8<Self>) -> f64x8<Self> {
unsafe {
let (a0, a1) = self.split_f32x8(a);
let high = _mm256_cvtps_pd(a0.into()).simd_into(self);
let low = _mm256_cvtps_pd(a1.into()).simd_into(self);
self.combine_f64x4(high, low)
}
}
#[inline(always)]
fn reinterpret_f64_f32x8(self, a: f32x8<Self>) -> f64x4<Self> {
f64x4 {
val: bytemuck::cast(a.val),
Expand Down Expand Up @@ -2818,6 +2831,10 @@ impl Simd for Avx2 {
}
}
#[inline(always)]
fn narrow_f64x4(self, a: f64x4<Self>) -> f32x4<Self> {
unsafe { _mm256_cvtpd_ps(a.into()).simd_into(self) }
}
#[inline(always)]
fn reinterpret_f32_f64x4(self, a: f64x4<Self>) -> f32x8<Self> {
f32x8 {
val: bytemuck::cast(a.val),
Expand Down Expand Up @@ -3052,6 +3069,18 @@ impl Simd for Avx2 {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn load_interleaved_128_f32x16(self, src: &[f32; 16usize]) -> f32x16<Self> {
crate::Fallback::new()
.load_interleaved_128_f32x16(src)
.val
.simd_into(self)
}
#[inline(always)]
fn store_interleaved_128_f32x16(self, a: f32x16<Self>, dest: &mut [f32; 16usize]) -> () {
let fb = crate::Fallback::new();
fb.store_interleaved_128_f32x16(a.val.simd_into(fb), dest);
}
#[inline(always)]
fn reinterpret_f64_f32x16(self, a: f32x16<Self>) -> f64x8<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_f64x4(
Expand All @@ -3068,18 +3097,6 @@ impl Simd for Avx2 {
)
}
#[inline(always)]
fn load_interleaved_128_f32x16(self, src: &[f32; 16usize]) -> f32x16<Self> {
crate::Fallback::new()
.load_interleaved_128_f32x16(src)
.val
.simd_into(self)
}
#[inline(always)]
fn store_interleaved_128_f32x16(self, a: f32x16<Self>, dest: &mut [f32; 16usize]) -> () {
let fb = crate::Fallback::new();
fb.store_interleaved_128_f32x16(a.val.simd_into(fb), dest);
}
#[inline(always)]
fn reinterpret_u8_f32x16(self, a: f32x16<Self>) -> u8x64<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_u8x32(self.reinterpret_u8_f32x8(a0), self.reinterpret_u8_f32x8(a1))
Expand Down Expand Up @@ -4484,6 +4501,15 @@ impl Simd for Avx2 {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn narrow_f64x8(self, a: f64x8<Self>) -> f32x8<Self> {
let (a, b) = self.split_f64x8(a);
unsafe {
let lo = _mm256_cvtpd_ps(a.into());
let hi = _mm256_cvtpd_ps(b.into());
_mm256_setr_m128(lo, hi).simd_into(self)
}
}
#[inline(always)]
fn reinterpret_f32_f64x8(self, a: f64x8<Self>) -> f32x16<Self> {
let (a0, a1) = self.split_f64x8(a);
self.combine_f32x8(
Expand Down
62 changes: 46 additions & 16 deletions fearless_simd/src/generated/fallback.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -338,6 +338,16 @@ impl Simd for Fallback {
result.simd_into(self)
}
#[inline(always)]
fn widen_f32x4(self, a: f32x4<Self>) -> f64x4<Self> {
[
a[0usize] as f64,
a[1usize] as f64,
a[2usize] as f64,
a[3usize] as f64,
]
.simd_into(self)
}
#[inline(always)]
fn reinterpret_f64_f32x4(self, a: f32x4<Self>) -> f64x2<Self> {
f64x2 {
val: bytemuck::cast(a.val),
Expand Down Expand Up @@ -3251,6 +3261,11 @@ impl Simd for Fallback {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn widen_f32x8(self, a: f32x8<Self>) -> f64x8<Self> {
let (a0, a1) = self.split_f32x8(a);
self.combine_f64x4(self.widen_f32x4(a0), self.widen_f32x4(a1))
}
#[inline(always)]
fn reinterpret_f64_f32x8(self, a: f32x8<Self>) -> f64x4<Self> {
let (a0, a1) = self.split_f32x8(a);
self.combine_f64x2(
Expand Down Expand Up @@ -4684,6 +4699,16 @@ impl Simd for Fallback {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn narrow_f64x4(self, a: f64x4<Self>) -> f32x4<Self> {
[
a[0usize] as f32,
a[1usize] as f32,
a[2usize] as f32,
a[3usize] as f32,
]
.simd_into(self)
}
#[inline(always)]
fn reinterpret_f32_f64x4(self, a: f64x4<Self>) -> f32x8<Self> {
let (a0, a1) = self.split_f64x4(a);
self.combine_f32x4(
Expand Down Expand Up @@ -4934,22 +4959,6 @@ impl Simd for Fallback {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn reinterpret_f64_f32x16(self, a: f32x16<Self>) -> f64x8<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_f64x4(
self.reinterpret_f64_f32x8(a0),
self.reinterpret_f64_f32x8(a1),
)
}
#[inline(always)]
fn reinterpret_i32_f32x16(self, a: f32x16<Self>) -> i32x16<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_i32x8(
self.reinterpret_i32_f32x8(a0),
self.reinterpret_i32_f32x8(a1),
)
}
#[inline(always)]
fn load_interleaved_128_f32x16(self, src: &[f32; 16usize]) -> f32x16<Self> {
[
src[0usize],
Expand Down Expand Up @@ -4980,6 +4989,22 @@ impl Simd for Fallback {
];
}
#[inline(always)]
fn reinterpret_f64_f32x16(self, a: f32x16<Self>) -> f64x8<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_f64x4(
self.reinterpret_f64_f32x8(a0),
self.reinterpret_f64_f32x8(a1),
)
}
#[inline(always)]
fn reinterpret_i32_f32x16(self, a: f32x16<Self>) -> i32x16<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_i32x8(
self.reinterpret_i32_f32x8(a0),
self.reinterpret_i32_f32x8(a1),
)
}
#[inline(always)]
fn reinterpret_u8_f32x16(self, a: f32x16<Self>) -> u8x64<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_u8x32(self.reinterpret_u8_f32x8(a0), self.reinterpret_u8_f32x8(a1))
Expand Down Expand Up @@ -6489,6 +6514,11 @@ impl Simd for Fallback {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn narrow_f64x8(self, a: f64x8<Self>) -> f32x8<Self> {
let (a0, a1) = self.split_f64x8(a);
self.combine_f32x4(self.narrow_f64x4(a0), self.narrow_f64x4(a1))
}
#[inline(always)]
fn reinterpret_f32_f64x8(self, a: f64x8<Self>) -> f32x16<Self> {
let (a0, a1) = self.split_f64x8(a);
self.combine_f32x8(
Expand Down
41 changes: 33 additions & 8 deletions fearless_simd/src/generated/neon.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -184,6 +184,14 @@ impl Simd for Neon {
result.simd_into(self)
}
#[inline(always)]
fn widen_f32x4(self, a: f32x4<Self>) -> f64x4<Self> {
unsafe {
let low = vcvt_f64_f32(vget_low_f32(a.into()));
let high = vcvt_high_f64_f32(a.into());
float64x2x2_t(low, high).simd_into(self)
}
}
#[inline(always)]
fn reinterpret_f64_f32x4(self, a: f32x4<Self>) -> f64x2<Self> {
unsafe { vreinterpretq_f64_f32(a.into()).simd_into(self) }
}
Expand Down Expand Up @@ -1401,6 +1409,11 @@ impl Simd for Neon {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn widen_f32x8(self, a: f32x8<Self>) -> f64x8<Self> {
let (a0, a1) = self.split_f32x8(a);
self.combine_f64x4(self.widen_f32x4(a0), self.widen_f32x4(a1))
}
#[inline(always)]
fn reinterpret_f64_f32x8(self, a: f32x8<Self>) -> f64x4<Self> {
let (a0, a1) = self.split_f32x8(a);
self.combine_f64x2(
Expand Down Expand Up @@ -2821,6 +2834,13 @@ impl Simd for Neon {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn narrow_f64x4(self, a: f64x4<Self>) -> f32x4<Self> {
unsafe {
let converted: float64x2x2_t = a.into();
vcvt_high_f32_f64(vcvt_f32_f64(converted.0), converted.1).simd_into(self)
}
}
#[inline(always)]
fn reinterpret_f32_f64x4(self, a: f64x4<Self>) -> f32x8<Self> {
let (a0, a1) = self.split_f64x4(a);
self.combine_f32x4(
Expand Down Expand Up @@ -3071,6 +3091,14 @@ impl Simd for Neon {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn load_interleaved_128_f32x16(self, src: &[f32; 16usize]) -> f32x16<Self> {
unsafe { vld4q_f32(src.as_ptr()).simd_into(self) }
}
#[inline(always)]
fn store_interleaved_128_f32x16(self, a: f32x16<Self>, dest: &mut [f32; 16usize]) -> () {
unsafe { vst4q_f32(dest.as_mut_ptr(), a.into()) }
}
#[inline(always)]
fn reinterpret_f64_f32x16(self, a: f32x16<Self>) -> f64x8<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_f64x4(
Expand All @@ -3087,14 +3115,6 @@ impl Simd for Neon {
)
}
#[inline(always)]
fn load_interleaved_128_f32x16(self, src: &[f32; 16usize]) -> f32x16<Self> {
unsafe { vld4q_f32(src.as_ptr()).simd_into(self) }
}
#[inline(always)]
fn store_interleaved_128_f32x16(self, a: f32x16<Self>, dest: &mut [f32; 16usize]) -> () {
unsafe { vst4q_f32(dest.as_mut_ptr(), a.into()) }
}
#[inline(always)]
fn reinterpret_u8_f32x16(self, a: f32x16<Self>) -> u8x64<Self> {
let (a0, a1) = self.split_f32x16(a);
self.combine_u8x32(self.reinterpret_u8_f32x8(a0), self.reinterpret_u8_f32x8(a1))
Expand Down Expand Up @@ -4465,6 +4485,11 @@ impl Simd for Neon {
(b0.simd_into(self), b1.simd_into(self))
}
#[inline(always)]
fn narrow_f64x8(self, a: f64x8<Self>) -> f32x8<Self> {
let (a0, a1) = self.split_f64x8(a);
self.combine_f32x4(self.narrow_f64x4(a0), self.narrow_f64x4(a1))
}
#[inline(always)]
fn reinterpret_f32_f64x8(self, a: f64x8<Self>) -> f32x16<Self> {
let (a0, a1) = self.split_f64x8(a);
self.combine_f32x8(
Expand Down
8 changes: 6 additions & 2 deletions fearless_simd/src/generated/simd_trait.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,6 +96,7 @@ pub trait Simd: Sized + Clone + Copy + Send + Sync + Seal + 'static {
fn trunc_f32x4(self, a: f32x4<Self>) -> f32x4<Self>;
fn select_f32x4(self, a: mask32x4<Self>, b: f32x4<Self>, c: f32x4<Self>) -> f32x4<Self>;
fn combine_f32x4(self, a: f32x4<Self>, b: f32x4<Self>) -> f32x8<Self>;
fn widen_f32x4(self, a: f32x4<Self>) -> f64x4<Self>;
fn reinterpret_f64_f32x4(self, a: f32x4<Self>) -> f64x2<Self>;
fn reinterpret_i32_f32x4(self, a: f32x4<Self>) -> i32x4<Self>;
fn reinterpret_u8_f32x4(self, a: f32x4<Self>) -> u8x16<Self>;
Expand Down Expand Up @@ -374,6 +375,7 @@ pub trait Simd: Sized + Clone + Copy + Send + Sync + Seal + 'static {
fn select_f32x8(self, a: mask32x8<Self>, b: f32x8<Self>, c: f32x8<Self>) -> f32x8<Self>;
fn combine_f32x8(self, a: f32x8<Self>, b: f32x8<Self>) -> f32x16<Self>;
fn split_f32x8(self, a: f32x8<Self>) -> (f32x4<Self>, f32x4<Self>);
fn widen_f32x8(self, a: f32x8<Self>) -> f64x8<Self>;
fn reinterpret_f64_f32x8(self, a: f32x8<Self>) -> f64x4<Self>;
fn reinterpret_i32_f32x8(self, a: f32x8<Self>) -> i32x8<Self>;
fn reinterpret_u8_f32x8(self, a: f32x8<Self>) -> u8x32<Self>;
Expand Down Expand Up @@ -619,6 +621,7 @@ pub trait Simd: Sized + Clone + Copy + Send + Sync + Seal + 'static {
fn select_f64x4(self, a: mask64x4<Self>, b: f64x4<Self>, c: f64x4<Self>) -> f64x4<Self>;
fn combine_f64x4(self, a: f64x4<Self>, b: f64x4<Self>) -> f64x8<Self>;
fn split_f64x4(self, a: f64x4<Self>) -> (f64x2<Self>, f64x2<Self>);
fn narrow_f64x4(self, a: f64x4<Self>) -> f32x4<Self>;
fn reinterpret_f32_f64x4(self, a: f64x4<Self>) -> f32x8<Self>;
fn splat_mask64x4(self, val: i64) -> mask64x4<Self>;
fn not_mask64x4(self, a: mask64x4<Self>) -> mask64x4<Self>;
Expand Down Expand Up @@ -663,10 +666,10 @@ pub trait Simd: Sized + Clone + Copy + Send + Sync + Seal + 'static {
fn trunc_f32x16(self, a: f32x16<Self>) -> f32x16<Self>;
fn select_f32x16(self, a: mask32x16<Self>, b: f32x16<Self>, c: f32x16<Self>) -> f32x16<Self>;
fn split_f32x16(self, a: f32x16<Self>) -> (f32x8<Self>, f32x8<Self>);
fn reinterpret_f64_f32x16(self, a: f32x16<Self>) -> f64x8<Self>;
fn reinterpret_i32_f32x16(self, a: f32x16<Self>) -> i32x16<Self>;
fn load_interleaved_128_f32x16(self, src: &[f32; 16usize]) -> f32x16<Self>;
fn store_interleaved_128_f32x16(self, a: f32x16<Self>, dest: &mut [f32; 16usize]) -> ();
fn reinterpret_f64_f32x16(self, a: f32x16<Self>) -> f64x8<Self>;
fn reinterpret_i32_f32x16(self, a: f32x16<Self>) -> i32x16<Self>;
fn reinterpret_u8_f32x16(self, a: f32x16<Self>) -> u8x64<Self>;
fn reinterpret_u32_f32x16(self, a: f32x16<Self>) -> u32x16<Self>;
fn cvt_u32_f32x16(self, a: f32x16<Self>) -> u32x16<Self>;
Expand Down Expand Up @@ -905,6 +908,7 @@ pub trait Simd: Sized + Clone + Copy + Send + Sync + Seal + 'static {
fn trunc_f64x8(self, a: f64x8<Self>) -> f64x8<Self>;
fn select_f64x8(self, a: mask64x8<Self>, b: f64x8<Self>, c: f64x8<Self>) -> f64x8<Self>;
fn split_f64x8(self, a: f64x8<Self>) -> (f64x4<Self>, f64x4<Self>);
fn narrow_f64x8(self, a: f64x8<Self>) -> f32x8<Self>;
fn reinterpret_f32_f64x8(self, a: f64x8<Self>) -> f32x16<Self>;
fn splat_mask64x8(self, val: i64) -> mask64x8<Self>;
fn not_mask64x8(self, a: mask64x8<Self>) -> mask64x8<Self>;
Expand Down
Loading