Rollup merge of #94452 - workingjubilee:sync-simd-bitmasks, r=workingjubilee

Sync portable-simd for bitmasks &c.

In the ideal case, where everything works easily and nothing has to be rearranged, it is as simple as:
- `git subtree pull -P library/portable-simd https://github.com/rust-lang/portable-simd - ${branch}`
- write the commit message
- `python x.py test --stage 1` to make sure it runs
- `git push` to your PR-to-rustc branch

If anything borks up this flow, you can fix it with sufficient git wizardry but you are usually better off going back to the source, fixing it, and starting over, before you open the PR.

r? `@calebzulawski`

Author: Dylan-DPC

library/portable-simd/crates/core_simd/examples/spectral_norm.rs

@@ -0,0 +1,77 @@
+#![feature(portable_simd)]
+
+use core_simd::simd::*;
+
+fn a(i: usize, j: usize) -> f64 {
+    ((i + j) * (i + j + 1) / 2 + i + 1) as f64
+}
+
+fn mult_av(v: &[f64], out: &mut [f64]) {
+    assert!(v.len() == out.len());
+    assert!(v.len() % 2 == 0);
+
+    for (i, out) in out.iter_mut().enumerate() {
+        let mut sum = f64x2::splat(0.0);
+
+        let mut j = 0;
+        while j < v.len() {
+            let b = f64x2::from_slice(&v[j..]);
+            let a = f64x2::from_array([a(i, j), a(i, j + 1)]);
+            sum += b / a;
+            j += 2
+        }
+        *out = sum.horizontal_sum();
+    }
+}
+
+fn mult_atv(v: &[f64], out: &mut [f64]) {
+    assert!(v.len() == out.len());
+    assert!(v.len() % 2 == 0);
+
+    for (i, out) in out.iter_mut().enumerate() {
+        let mut sum = f64x2::splat(0.0);
+
+        let mut j = 0;
+        while j < v.len() {
+            let b = f64x2::from_slice(&v[j..]);
+            let a = f64x2::from_array([a(j, i), a(j + 1, i)]);
+            sum += b / a;
+            j += 2
+        }
+        *out = sum.horizontal_sum();
+    }
+}
+
+fn mult_atav(v: &[f64], out: &mut [f64], tmp: &mut [f64]) {
+    mult_av(v, tmp);
+    mult_atv(tmp, out);
+}
+
+pub fn spectral_norm(n: usize) -> f64 {
+    assert!(n % 2 == 0, "only even lengths are accepted");
+
+    let mut u = vec![1.0; n];
+    let mut v = u.clone();
+    let mut tmp = u.clone();
+
+    for _ in 0..10 {
+        mult_atav(&u, &mut v, &mut tmp);
+        mult_atav(&v, &mut u, &mut tmp);
+    }
+    (dot(&u, &v) / dot(&v, &v)).sqrt()
+}
+
+fn dot(x: &[f64], y: &[f64]) -> f64 {
+    // This is auto-vectorized:
+    x.iter().zip(y).map(|(&x, &y)| x * y).sum()
+}
+
+#[cfg(test)]
+#[test]
+fn test() {
+    assert_eq!(&format!("{:.9}", spectral_norm(100)), "1.274219991");
+}
+
+fn main() {
+    // Empty main to make cargo happy
+}

library/portable-simd/crates/core_simd/src/comparisons.rs

@@ -10,13 +10,17 @@ where
     #[inline]
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub fn lanes_eq(self, other: Self) -> Mask<T::Mask, LANES> {
+        // Safety: `self` is a vector, and the result of the comparison
+        // is always a valid mask.
         unsafe { Mask::from_int_unchecked(intrinsics::simd_eq(self, other)) }
     }
 
     /// Test if each lane is not equal to the corresponding lane in `other`.
     #[inline]
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub fn lanes_ne(self, other: Self) -> Mask<T::Mask, LANES> {
+        // Safety: `self` is a vector, and the result of the comparison
+        // is always a valid mask.
         unsafe { Mask::from_int_unchecked(intrinsics::simd_ne(self, other)) }
     }
 }
@@ -30,27 +34,35 @@ where
     #[inline]
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub fn lanes_lt(self, other: Self) -> Mask<T::Mask, LANES> {
+        // Safety: `self` is a vector, and the result of the comparison
+        // is always a valid mask.
         unsafe { Mask::from_int_unchecked(intrinsics::simd_lt(self, other)) }
     }
 
     /// Test if each lane is greater than the corresponding lane in `other`.
     #[inline]
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub fn lanes_gt(self, other: Self) -> Mask<T::Mask, LANES> {
+        // Safety: `self` is a vector, and the result of the comparison
+        // is always a valid mask.
         unsafe { Mask::from_int_unchecked(intrinsics::simd_gt(self, other)) }
     }
 
     /// Test if each lane is less than or equal to the corresponding lane in `other`.
     #[inline]
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub fn lanes_le(self, other: Self) -> Mask<T::Mask, LANES> {
+        // Safety: `self` is a vector, and the result of the comparison
+        // is always a valid mask.
         unsafe { Mask::from_int_unchecked(intrinsics::simd_le(self, other)) }
     }
 
     /// Test if each lane is greater than or equal to the corresponding lane in `other`.
     #[inline]
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub fn lanes_ge(self, other: Self) -> Mask<T::Mask, LANES> {
+        // Safety: `self` is a vector, and the result of the comparison
+        // is always a valid mask.
         unsafe { Mask::from_int_unchecked(intrinsics::simd_ge(self, other)) }
     }
 }

library/portable-simd/crates/core_simd/src/intrinsics.rs

@@ -2,31 +2,55 @@
 //! crate.
 //!
 //! The LLVM assembly language is documented here: <https://llvm.org/docs/LangRef.html>
+//!
+//! A quick glossary of jargon that may appear in this module, mostly paraphrasing LLVM's LangRef:
+//! - poison: "undefined behavior as a value". specifically, it is like uninit memory (such as padding bytes). it is "safe" to create poison, BUT
+//!   poison MUST NOT be observed from safe code, as operations on poison return poison, like NaN. unlike NaN, which has defined comparisons,
+//!   poison is neither true nor false, and LLVM may also convert it to undef (at which point it is both). so, it can't be conditioned on, either.
+//! - undef: "a value that is every value". functionally like poison, insofar as Rust is concerned. poison may become this. note:
+//!   this means that division by poison or undef is like division by zero, which means it inflicts...
+//! - "UB": poison and undef cover most of what people call "UB". "UB" means this operation immediately invalidates the program:
+//!   LLVM is allowed to lower it to `ud2` or other opcodes that may cause an illegal instruction exception, and this is the "good end".
+//!   The "bad end" is that LLVM may reverse time to the moment control flow diverged on a path towards undefined behavior,
+//!   and destroy the other branch, potentially deleting safe code and violating Rust's `unsafe` contract.
+//!
+//! Note that according to LLVM, vectors are not arrays, but they are equivalent when stored to and loaded from memory.
+//!
+//! Unless stated otherwise, all intrinsics for binary operations require SIMD vectors of equal types and lengths.
 
 /// These intrinsics aren't linked directly from LLVM and are mostly undocumented, however they are
-/// simply lowered to the matching LLVM instructions by the compiler.  The associated instruction
-/// is documented alongside each intrinsic.
+/// mostly lowered to the matching LLVM instructions by the compiler in a fairly straightforward manner.
+/// The associated LLVM instruction or intrinsic is documented alongside each Rust intrinsic function.
 extern "platform-intrinsic" {
     /// add/fadd
     pub(crate) fn simd_add<T>(x: T, y: T) -> T;
 
     /// sub/fsub
-    pub(crate) fn simd_sub<T>(x: T, y: T) -> T;
+    pub(crate) fn simd_sub<T>(lhs: T, rhs: T) -> T;
 
     /// mul/fmul
     pub(crate) fn simd_mul<T>(x: T, y: T) -> T;
 
     /// udiv/sdiv/fdiv
-    pub(crate) fn simd_div<T>(x: T, y: T) -> T;
+    /// ints and uints: {s,u}div incur UB if division by zero occurs.
+    /// ints: sdiv is UB for int::MIN / -1.
+    /// floats: fdiv is never UB, but may create NaNs or infinities.
+    pub(crate) fn simd_div<T>(lhs: T, rhs: T) -> T;
 
     /// urem/srem/frem
-    pub(crate) fn simd_rem<T>(x: T, y: T) -> T;
+    /// ints and uints: {s,u}rem incur UB if division by zero occurs.
+    /// ints: srem is UB for int::MIN / -1.
+    /// floats: frem is equivalent to libm::fmod in the "default" floating point environment, sans errno.
+    pub(crate) fn simd_rem<T>(lhs: T, rhs: T) -> T;
 
     /// shl
-    pub(crate) fn simd_shl<T>(x: T, y: T) -> T;
+    /// for (u)ints. poison if rhs >= lhs::BITS
+    pub(crate) fn simd_shl<T>(lhs: T, rhs: T) -> T;
 
-    /// lshr/ashr
-    pub(crate) fn simd_shr<T>(x: T, y: T) -> T;
+    /// ints: ashr
+    /// uints: lshr
+    /// poison if rhs >= lhs::BITS
+    pub(crate) fn simd_shr<T>(lhs: T, rhs: T) -> T;
 
     /// and
     pub(crate) fn simd_and<T>(x: T, y: T) -> T;
@@ -38,12 +62,18 @@ extern "platform-intrinsic" {
     pub(crate) fn simd_xor<T>(x: T, y: T) -> T;
 
     /// fptoui/fptosi/uitofp/sitofp
+    /// casting floats to integers is truncating, so it is safe to convert values like e.g. 1.5
+    /// but the truncated value must fit in the target type or the result is poison.
+    /// use `simd_as` instead for a cast that performs a saturating conversion.
     pub(crate) fn simd_cast<T, U>(x: T) -> U;
     /// follows Rust's `T as U` semantics, including saturating float casts
     /// which amounts to the same as `simd_cast` for many cases
     pub(crate) fn simd_as<T, U>(x: T) -> U;
 
     /// neg/fneg
+    /// ints: ultimately becomes a call to cg_ssa's BuilderMethods::neg. cg_llvm equates this to `simd_sub(Simd::splat(0), x)`.
+    /// floats: LLVM's fneg, which changes the floating point sign bit. Some arches have instructions for it.
+    /// Rust panics for Neg::neg(int::MIN) due to overflow, but it is not UB in LLVM without `nsw`.
     pub(crate) fn simd_neg<T>(x: T) -> T;
 
     /// fabs
@@ -53,6 +83,7 @@ extern "platform-intrinsic" {
     pub(crate) fn simd_fmin<T>(x: T, y: T) -> T;
     pub(crate) fn simd_fmax<T>(x: T, y: T) -> T;
 
+    // these return Simd<int, N> with the same BITS size as the inputs
     pub(crate) fn simd_eq<T, U>(x: T, y: T) -> U;
     pub(crate) fn simd_ne<T, U>(x: T, y: T) -> U;
     pub(crate) fn simd_lt<T, U>(x: T, y: T) -> U;
@@ -61,19 +92,31 @@ extern "platform-intrinsic" {
     pub(crate) fn simd_ge<T, U>(x: T, y: T) -> U;
 
     // shufflevector
+    // idx: LLVM calls it a "shuffle mask vector constant", a vector of i32s
     pub(crate) fn simd_shuffle<T, U, V>(x: T, y: T, idx: U) -> V;
 
+    /// llvm.masked.gather
+    /// like a loop of pointer reads
+    /// val: vector of values to select if a lane is masked
+    /// ptr: vector of pointers to read from
+    /// mask: a "wide" mask of integers, selects as if simd_select(mask, read(ptr), val)
+    /// note, the LLVM intrinsic accepts a mask vector of <N x i1>
+    /// FIXME: review this if/when we fix up our mask story in general?
     pub(crate) fn simd_gather<T, U, V>(val: T, ptr: U, mask: V) -> T;
+    /// llvm.masked.scatter
+    /// like gather, but more spicy, as it writes instead of reads
     pub(crate) fn simd_scatter<T, U, V>(val: T, ptr: U, mask: V);
 
     // {s,u}add.sat
     pub(crate) fn simd_saturating_add<T>(x: T, y: T) -> T;
 
     // {s,u}sub.sat
-    pub(crate) fn simd_saturating_sub<T>(x: T, y: T) -> T;
+    pub(crate) fn simd_saturating_sub<T>(lhs: T, rhs: T) -> T;
 
     // reductions
+    // llvm.vector.reduce.{add,fadd}
     pub(crate) fn simd_reduce_add_ordered<T, U>(x: T, y: U) -> U;
+    // llvm.vector.reduce.{mul,fmul}
     pub(crate) fn simd_reduce_mul_ordered<T, U>(x: T, y: U) -> U;
     #[allow(unused)]
     pub(crate) fn simd_reduce_all<T>(x: T) -> bool;
@@ -90,7 +133,10 @@ extern "platform-intrinsic" {
     pub(crate) fn simd_bitmask<T, U>(x: T) -> U;
 
     // select
-    pub(crate) fn simd_select<M, T>(m: M, a: T, b: T) -> T;
+    // first argument is a vector of integers, -1 (all bits 1) is "true"
+    // logically equivalent to (yes & m) | (no & (m^-1),
+    // but you can use it on floats.
+    pub(crate) fn simd_select<M, T>(m: M, yes: T, no: T) -> T;
     #[allow(unused)]
-    pub(crate) fn simd_select_bitmask<M, T>(m: M, a: T, b: T) -> T;
+    pub(crate) fn simd_select_bitmask<M, T>(m: M, yes: T, no: T) -> T;
 }

library/portable-simd/crates/core_simd/src/lib.rs

@@ -1,7 +1,9 @@
 #![cfg_attr(not(feature = "std"), no_std)]
 #![feature(
     const_fn_trait_bound,
+    convert_float_to_int,
     decl_macro,
+    intra_doc_pointers,
     platform_intrinsics,
     repr_simd,
     simd_ffi,

library/portable-simd/crates/core_simd/src/masks.rs

@@ -12,8 +12,10 @@
 )]
 mod mask_impl;
 
-use crate::simd::intrinsics;
-use crate::simd::{LaneCount, Simd, SimdElement, SupportedLaneCount};
+mod to_bitmask;
+pub use to_bitmask::ToBitMask;
+
+use crate::simd::{intrinsics, LaneCount, Simd, SimdElement, SupportedLaneCount};
 use core::cmp::Ordering;
 use core::{fmt, mem};
 
@@ -42,6 +44,9 @@ mod sealed {
 use sealed::Sealed;
 
 /// Marker trait for types that may be used as SIMD mask elements.
+///
+/// # Safety
+/// Type must be a signed integer.
 pub unsafe trait MaskElement: SimdElement + Sealed {}
 
 macro_rules! impl_element {
@@ -149,6 +154,7 @@ where
     #[inline]
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub unsafe fn from_int_unchecked(value: Simd<T, LANES>) -> Self {
+        // Safety: the caller must confirm this invariant
         unsafe { Self(mask_impl::Mask::from_int_unchecked(value)) }
     }
 
@@ -161,6 +167,7 @@ where
     #[must_use = "method returns a new mask and does not mutate the original value"]
     pub fn from_int(value: Simd<T, LANES>) -> Self {
         assert!(T::valid(value), "all values must be either 0 or -1",);
+        // Safety: the validity has been checked
         unsafe { Self::from_int_unchecked(value) }
     }
 
@@ -179,6 +186,7 @@ where
     #[inline]
     #[must_use = "method returns a new bool and does not mutate the original value"]
     pub unsafe fn test_unchecked(&self, lane: usize) -> bool {
+        // Safety: the caller must confirm this invariant
         unsafe { self.0.test_unchecked(lane) }
     }
 
@@ -190,6 +198,7 @@ where
     #[must_use = "method returns a new bool and does not mutate the original value"]
     pub fn test(&self, lane: usize) -> bool {
         assert!(lane < LANES, "lane index out of range");
+        // Safety: the lane index has been checked
         unsafe { self.test_unchecked(lane) }
     }
 
@@ -199,6 +208,7 @@ where
     /// `lane` must be less than `LANES`.
     #[inline]
     pub unsafe fn set_unchecked(&mut self, lane: usize, value: bool) {
+        // Safety: the caller must confirm this invariant
         unsafe {
             self.0.set_unchecked(lane, value);
         }
@@ -211,27 +221,12 @@ where
     #[inline]
     pub fn set(&mut self, lane: usize, value: bool) {
         assert!(lane < LANES, "lane index out of range");
+        // Safety: the lane index has been checked
         unsafe {
             self.set_unchecked(lane, value);
         }
     }
 
-    /// Convert this mask to a bitmask, with one bit set per lane.
-    #[cfg(feature = "generic_const_exprs")]
-    #[inline]
-    #[must_use = "method returns a new array and does not mutate the original value"]
-    pub fn to_bitmask(self) -> [u8; LaneCount::<LANES>::BITMASK_LEN] {
-        self.0.to_bitmask()
-    }
-
-    /// Convert a bitmask to a mask.
-    #[cfg(feature = "generic_const_exprs")]
-    #[inline]
-    #[must_use = "method returns a new mask and does not mutate the original value"]
-    pub fn from_bitmask(bitmask: [u8; LaneCount::<LANES>::BITMASK_LEN]) -> Self {
-        Self(mask_impl::Mask::from_bitmask(bitmask))
-    }
-
     /// Returns true if any lane is set, or false otherwise.
     #[inline]
     #[must_use = "method returns a new bool and does not mutate the original value"]