pub struct Scalar256;Expand description
Portable pair of 128-bit lanes, useful for AVX2-equivalent conformance tests.
Implementations§
Trait Implementations§
impl Copy for Scalar256
Source§impl Simd for Scalar256
impl Simd for Scalar256
type V = [u8; 32]
fn zero(self) -> Self::V
fn splat8(self, v: u8) -> Self::V
fn splat16(self, v: u16) -> Self::V
fn splat32(self, v: u32) -> Self::V
fn splat64(self, v: u64) -> Self::V
fn splatf(self, v: f32) -> Self::V
fn splatd(self, v: f64) -> Self::V
fn store(self, b: &mut [u8], v: Self::V)
Source§fn load_raw(self, b: &[u8]) -> Self::V
fn load_raw(self, b: &[u8]) -> Self::V
4 * PX bytes as they are (“raw” pixels: SSE2 the low 8 bytes, AVX2 the low 16)
fn store_raw(self, b: &mut [u8], v: Self::V)
Source§fn store_px(self, b: &mut [u8], v: Self::V)
fn store_px(self, b: &mut [u8], v: Self::V)
pixel lanes saturated to bytes (packuswb), 4 * PX bytes
Source§fn load_lane32(self, b: &[u8]) -> Self::V
fn load_lane32(self, b: &[u8]) -> Self::V
4 bytes into the low dword of every 128-bit lane (BYTES / 4 bytes), the rest 0
fn to_i32(self, v: Self::V) -> i32
fn store_i16(self, b: &mut [i16], v: Self::V)
fn store_u16(self, b: &mut [u16], v: Self::V)
fn store_i32(self, b: &mut [i32], v: Self::V)
fn store_f32(self, b: &mut [f32], v: Self::V)
Source§fn load_u16_lo(self, b: &[u16]) -> Self::V
fn load_u16_lo(self, b: &[u16]) -> Self::V
movq: 4 u16 values into the low 64 bits (the rest 0) / the low 64 bits to 4 u16 values
fn store_u16_lo(self, b: &mut [u16], v: Self::V)
Source§fn store_i32_half(self, b: &mut [i32], v: Self::V)
fn store_i32_half(self, b: &mut [i32], v: Self::V)
the first BYTES / 8 i32 lanes
fn add8(self, a: Self::V, b: Self::V) -> Self::V
fn sub8(self, a: Self::V, b: Self::V) -> Self::V
fn cmpeq8(self, a: Self::V, b: Self::V) -> Self::V
fn add16(self, a: Self::V, b: Self::V) -> Self::V
fn sub16(self, a: Self::V, b: Self::V) -> Self::V
fn cmpeq16(self, a: Self::V, b: Self::V) -> Self::V
fn add32(self, a: Self::V, b: Self::V) -> Self::V
fn sub32(self, a: Self::V, b: Self::V) -> Self::V
fn cmpeq32(self, a: Self::V, b: Self::V) -> Self::V
fn subus8(self, a: Self::V, b: Self::V) -> Self::V
fn addus8(self, a: Self::V, b: Self::V) -> Self::V
fn maxu8(self, a: Self::V, b: Self::V) -> Self::V
fn minu8(self, a: Self::V, b: Self::V) -> Self::V
fn avgu8(self, a: Self::V, b: Self::V) -> Self::V
Source§fn adds8(self, a: Self::V, b: Self::V) -> Self::V
fn adds8(self, a: Self::V, b: Self::V) -> Self::V
paddsb: signed saturating add of the i8 lanes
Source§fn subs8(self, a: Self::V, b: Self::V) -> Self::V
fn subs8(self, a: Self::V, b: Self::V) -> Self::V
psubsb: signed saturating subtract of the i8 lanes
fn subus16(self, a: Self::V, b: Self::V) -> Self::V
fn mullo16(self, a: Self::V, b: Self::V) -> Self::V
fn mulhiu16(self, a: Self::V, b: Self::V) -> Self::V
Source§fn mulhis16(self, a: Self::V, b: Self::V) -> Self::V
fn mulhis16(self, a: Self::V, b: Self::V) -> Self::V
pmulhw: high 16 bits of the signed 16-bit products
fn mins16(self, a: Self::V, b: Self::V) -> Self::V
fn maxs16(self, a: Self::V, b: Self::V) -> Self::V
fn cmpgts16(self, a: Self::V, b: Self::V) -> Self::V
fn srl16<const N: i32>(self, a: Self::V) -> Self::V
fn sll16<const N: i32>(self, a: Self::V) -> Self::V
Source§fn sra16<const N: i32>(self, a: Self::V) -> Self::V
fn sra16<const N: i32>(self, a: Self::V) -> Self::V
psraw: arithmetic right shift of the i16 lanes
fn cmpgts32(self, a: Self::V, b: Self::V) -> Self::V
fn srl32<const N: i32>(self, a: Self::V) -> Self::V
fn sll32<const N: i32>(self, a: Self::V) -> Self::V
fn sra32<const N: i32>(self, a: Self::V) -> Self::V
Source§fn srl16x(self, a: Self::V, n: u32) -> Self::V
fn srl16x(self, a: Self::V, n: u32) -> Self::V
psrlw with a register count: logical right shift of the 16-bit lanes by n (n > 15: 0)
Source§fn movemask32(self, a: Self::V) -> u32
fn movemask32(self, a: Self::V) -> u32
bit i = top bit of dword i (BYTES / 4 bits)
fn unpacklo8(self, a: Self::V, b: Self::V) -> Self::V
fn unpackhi8(self, a: Self::V, b: Self::V) -> Self::V
fn unpacklo16(self, a: Self::V, b: Self::V) -> Self::V
fn unpackhi16(self, a: Self::V, b: Self::V) -> Self::V
fn packus16(self, a: Self::V, b: Self::V) -> Self::V
Source§fn packs16(self, a: Self::V, b: Self::V) -> Self::V
fn packs16(self, a: Self::V, b: Self::V) -> Self::V
packsswb: i16 lanes of a, then b, saturated to i8
fn packs32(self, a: Self::V, b: Self::V) -> Self::V
Source§fn packssdw(self, a: Self::V, b: Self::V) -> Self::V
fn packssdw(self, a: Self::V, b: Self::V) -> Self::V
packs32 (packssdw), on AVX2 without the clamps
Source§fn packuswb_px(self, v: Self::V) -> Self::V
fn packuswb_px(self, v: Self::V) -> Self::V
pixel lanes to raw pixels: packuswb (+ vpermq 0x08 on AVX2) as in store_px, on AVX2
without the clamps; the inverse of widen_px
Source§fn madd16(self, a: Self::V, b: Self::V) -> Self::V
fn madd16(self, a: Self::V, b: Self::V) -> Self::V
pmaddwd: i32 lanes = a0 * b0 + a1 * b1 of the signed 16-bit pairs
Source§fn shuf16<const IMM: i32>(self, a: Self::V) -> Self::V
fn shuf16<const IMM: i32>(self, a: Self::V) -> Self::V
pshuflw + pshufhw with the same immediate (per 64-bit half)
Source§fn shufps<const IMM: i32>(self, a: Self::V, b: Self::V) -> Self::V
fn shufps<const IMM: i32>(self, a: Self::V, b: Self::V) -> Self::V
shufps on integer registers: dwords 0, 1 from a, 2, 3 from b (per 128-bit lane)
Source§fn bsll<const N: i32>(self, a: Self::V) -> Self::V
fn bsll<const N: i32>(self, a: Self::V) -> Self::V
pslldq / psrldq: byte shifts per 128-bit lane
fn bsrl<const N: i32>(self, a: Self::V) -> Self::V
fn and(self, a: Self::V, b: Self::V) -> Self::V
fn or(self, a: Self::V, b: Self::V) -> Self::V
fn xor(self, a: Self::V, b: Self::V) -> Self::V
fn addf(self, a: Self::V, b: Self::V) -> Self::V
fn subf(self, a: Self::V, b: Self::V) -> Self::V
fn mulf(self, a: Self::V, b: Self::V) -> Self::V
fn divf(self, a: Self::V, b: Self::V) -> Self::V
fn minf(self, a: Self::V, b: Self::V) -> Self::V
Source§fn cmpltf(self, a: Self::V, b: Self::V) -> Self::V
fn cmpltf(self, a: Self::V, b: Self::V) -> Self::V
cmpltps: all ones where a < b (false when unordered)
fn addd(self, a: Self::V, b: Self::V) -> Self::V
fn subd(self, a: Self::V, b: Self::V) -> Self::V
fn muld(self, a: Self::V, b: Self::V) -> Self::V
fn divd(self, a: Self::V, b: Self::V) -> Self::V
fn mind(self, a: Self::V, b: Self::V) -> Self::V
Source§fn cmpltd(self, a: Self::V, b: Self::V) -> Self::V
fn cmpltd(self, a: Self::V, b: Self::V) -> Self::V
cmpltpd: all ones where a < b (false when unordered)
fn sqrtd(self, a: Self::V) -> Self::V
fn cvt_i32_f32(self, a: Self::V) -> Self::V
Source§fn cvtt_f32_i32(self, a: Self::V) -> Self::V
fn cvtt_f32_i32(self, a: Self::V) -> Self::V
truncating (0x80000000 for NaN / out of range)
Source§fn cvtr_f32_i32(self, a: Self::V) -> Self::V
fn cvtr_f32_i32(self, a: Self::V) -> Self::V
rounding to nearest even (cvtps2dq)
Source§fn cvt_i32_f64(self, a: Self::V) -> Self::V
fn cvt_i32_f64(self, a: Self::V) -> Self::V
cvtdq2pd: the first BYTES / 8 i32 lanes to f64
Source§fn cvtt_f64_i32(self, a: Self::V) -> Self::V
fn cvtt_f64_i32(self, a: Self::V) -> Self::V
cvttpd2dq: f64 lanes to the first BYTES / 8 i32 lanes (truncating, 0x80000000 when
out of range), the rest 0
Source§fn store_lanes2(self, b: &mut [u8], lo: Self::V, hi: Self::V)
fn store_lanes2(self, b: &mut [u8], lo: Self::V, hi: Self::V)
2 * BYTES bytes: per 128-bit lane k, lane k of lo then lane k of hi (SSE2: lo, hi;
AVX2: vperm2i128) – 2 x 4 pixels per lane in pixel order
Source§fn load_lanes2(self, b: &[u8]) -> (Self::V, Self::V)
fn load_lanes2(self, b: &[u8]) -> (Self::V, Self::V)
the inverse of store_lanes2: 2 * BYTES bytes -> (lo, hi)
Source§fn load_i16_lanes(self, a: &[i16], b: &[i16]) -> Self::V
fn load_i16_lanes(self, a: &[i16], b: &[i16]) -> Self::V
8 i16 per 128-bit lane from separate sources: lane 0 from a, lane 1 from b (SSE2: a)
Source§fn store_raw_lanes(self, a: &mut [u8], b: &mut [u8], v: Self::V)
fn store_raw_lanes(self, a: &mut [u8], b: &mut [u8], v: Self::V)
the low 8 bytes of every 128-bit lane to separate places: lane 0 to a, lane 1 to b
(SSE2: a)
Source§fn load_f32_lanes(self, a: &[f32], b: &[f32]) -> Self::V
fn load_f32_lanes(self, a: &[f32], b: &[f32]) -> Self::V
4 f32 per 128-bit lane from separate sources: lane 0 from a, lane 1 from b (SSE2: a)
Source§fn store_f32_lanes(self, a: &mut [f32], b: &mut [f32], v: Self::V)
fn store_f32_lanes(self, a: &mut [f32], b: &mut [f32], v: Self::V)
the 4 f32 of every 128-bit lane to separate places: lane 0 to a, lane 1 to b (SSE2: a)
Source§fn sadu8(self, a: Self::V, b: Self::V) -> Self::V
fn sadu8(self, a: Self::V, b: Self::V) -> Self::V
(main agent) psadbw: per 8-byte group the sum of |a - b| (u8) in its u64 lane
Source§fn ptest_raw(self, a: Self::V, b: Self::V) -> u32
fn ptest_raw(self, a: Self::V, b: Self::V) -> u32
ptest on the raw-pixel bytes (SSE2: the low 8, AVX2: the low 16): 1 when a & b == 0
(ZF), else 2 when !a & b == 0 (CF), else 0. AVX2: vptest + jz + jnc, the CF case
falls through (inline assembly: core::arch writes ptest as or-reductions, and LLVM
then often builds shuffle / reduce sequences or loses the CF test); branch on the
result, the branches thread through
Auto Trait Implementations§
impl Forget for Scalar256
impl Freeze for Scalar256
impl RefUnwindSafe for Scalar256
impl Send for Scalar256
impl Sync for Scalar256
impl Unpin for Scalar256
impl UnsafeUnpin for Scalar256
impl UnwindSafe for Scalar256
Blanket Implementations§
Source§impl<T> BorrowMut<T> for Twhere
T: ?Sized,
impl<T> BorrowMut<T> for Twhere
T: ?Sized,
Source§fn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
Mutably borrows from an owned value. Read more