vendor/crc-fast/src/traits.rs at switch-says · andymccall/oxygen

◆ switch-says oxygen / vendor / crc-fast / src / traits.rs

9.6 KiB · Rust · last changed Strata: the cloud stores send through ureq, so every release target links 1 month ago source blame history raw

// Copyright 2025 Don MacAskill. Licensed under MIT or Apache-2.0.

#![allow(dead_code)]

#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
use crate::enums::Reflector;

use crate::CrcParams;

#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
use crate::structs::CrcState;

use core::ops::BitXor;

/// Marker trait for CRC width
pub trait CrcWidth {
    /// The width in bits
    const WIDTH: u32;
    /// The natural value type for this width
    type Value: Copy + BitXor<Output = Self::Value>;
}

pub(crate) trait CrcCalculator {
    fn update(data: &[u8], state: u64, params: &CrcParams) -> u64 {
        Self::calculate(state, data, params)
    }

    fn checksum(data: &[u8], params: &CrcParams) -> u64 {
        Self::calculate(params.init, data, params) ^ params.xorout
    }

    fn calculate(state: u64, data: &[u8], params: &CrcParams) -> u64;
}

#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
/// Trait defining architecture-specific SIMD operations for CRC calculation
pub trait ArchOps: Sized + Copy + Clone {
    /// The SIMD vector type used by this architecture
    type Vector;

    /// Process aligned blocks using potentially accelerated SIMD operations
    ///
    /// Returns true if the operation was handled by the accelerated path (for example,
    /// using VPCLMULQDQ)
    unsafe fn process_enhanced_simd_blocks<W: EnhancedCrcWidth>(
        &self,
        _state: &mut CrcState<Self::Vector>,
        _first: &[Self::Vector; 8],
        _rest: &[[Self::Vector; 8]],
        _reflector: &Reflector<Self::Vector>,
        _keys: &[u64; 23],
    ) -> bool
    where
        Self::Vector: Copy,
    {
        // Default implementation just returns false
        // indicating the non-enhanced algorithm should be used
        false
    }

    /// Create a SIMD vector from a u64 pair
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn create_vector_from_u64_pair(
        &self,
        high: u64,
        low: u64,
        reflected: bool,
    ) -> Self::Vector;

    /// Create a SIMD vector from a u64 pair without reflection
    ///
    /// TODO: I have no idea (yet) why CRC-32 doesn't use reflection, but CRC-64 does.
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn create_vector_from_u64_pair_non_reflected(&self, high: u64, low: u64)
        -> Self::Vector;

    /// Create a SIMD vector with a single u64 value
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn create_vector_from_u64(&self, value: u64, high: bool) -> Self::Vector;

    /// Extract two u64 values from a SIMD vector
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn extract_u64s(&self, vector: Self::Vector) -> [u64; 2];

    /// Extract two polynomial values (for carryless multiplication)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn extract_poly64s(&self, vector: Self::Vector) -> [u64; 2];

    /// XOR two SIMD vectors
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn xor_vectors(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;

    /// Load bytes from memory into a SIMD vector
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn load_bytes(&self, ptr: *const u8) -> Self::Vector;

    /// Load aligned bytes from memory
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn load_aligned(&self, ptr: *const [u64; 2]) -> Self::Vector;

    //unsafe fn load_aligned(&self, ptr: &[u64]) -> Self::Vector;

    //unsafe fn load_aligned_const(&self, ptr: *const [u64; 2]) -> Self::Vector;

    /// Shuffle/permute bytes according to a mask
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn shuffle_bytes(&self, data: Self::Vector, mask: Self::Vector) -> Self::Vector;

    /// Blend two vectors using a mask (select from a or b based on mask bits)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn blend_vectors(
        &self,
        a: Self::Vector,
        b: Self::Vector,
        mask: Self::Vector,
    ) -> Self::Vector;

    /// Shift a vector left by 8 bytes
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn shift_left_8(&self, vector: Self::Vector) -> Self::Vector;

    /// Create a vector with all bytes set to the same value
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn set_all_bytes(&self, value: u8) -> Self::Vector;

    /// Create a comparison mask (for blending operations)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn create_compare_mask(&self, vector: Self::Vector) -> Self::Vector;

    /// AND two vectors
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn and_vectors(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;

    /// Shift a vector right by 32 bits (4 bytes)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn shift_right_32(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector left by 32 bits (4 bytes)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn shift_left_32(&self, vector: Self::Vector) -> Self::Vector;

    /// Create a SIMD vector with a single u32 value
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn create_vector_from_u32(&self, value: u32, high: bool) -> Self::Vector;

    /// Shift a vector left by 4 bytes (32 bits)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn shift_left_4(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector right by 4 bytes (32 bits)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn shift_right_4(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector right by 8 bytes (64 bits)
    ///
    /// # Safety
    /// May use native CPU features
    unsafe fn shift_right_8(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector right by 5 bytes
    unsafe fn shift_right_5(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector right by 6 bytes
    unsafe fn shift_right_6(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector right by 7 bytes
    unsafe fn shift_right_7(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector right by 12 bytes
    unsafe fn shift_right_12(&self, vector: Self::Vector) -> Self::Vector;

    /// Shift a vector left by 12 bytes
    unsafe fn shift_left_12(&self, vector: Self::Vector) -> Self::Vector;

    /// Perform carryless multiplication with immediate value 0x00 (low 64 bits of both vectors)
    unsafe fn carryless_mul_00(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;

    /// Perform carryless multiplication with immediate value 0x01 (low 64 bits of a, high 64 bits of b)
    unsafe fn carryless_mul_01(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;

    /// Perform carryless multiplication with immediate value 0x10 (high 64 bits of a, low 64 bits of b)
    unsafe fn carryless_mul_10(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;

    /// Perform carryless multiplication with immediate value 0x11 (high 64 bits of both vectors)
    unsafe fn carryless_mul_11(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;

    /// XOR three vectors together: a XOR b XOR c
    /// Uses native XOR3 instructions when available, falls back to two XOR operations otherwise
    unsafe fn xor3_vectors(
        &self,
        a: Self::Vector,
        b: Self::Vector,
        c: Self::Vector,
    ) -> Self::Vector;
}

#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
/// Enhanced CrcWidth trait with additional operations for generic CRC implementation
pub trait EnhancedCrcWidth: CrcWidth {
    /// Load constants specific to CRC width
    fn load_constants(reflected: bool) -> [[u64; 2]; 4];

    /// Create a CRC state with the initial value positioned correctly for the width
    unsafe fn create_state<T: ArchOps>(
        value: Self::Value,
        reflected: bool,
        ops: &T,
    ) -> CrcState<T::Vector>
    where
        T::Vector: Copy;

    /// Extract the final CRC result from a SIMD vector
    unsafe fn extract_result<T: ArchOps>(
        vector: T::Vector,
        reflected: bool,
        ops: &T,
    ) -> Self::Value
    where
        T::Vector: Copy;

    /// Perform width-specific folding operations using CLMUL and two XOR operations (or one XOR3)
    unsafe fn fold_16<T: ArchOps>(
        state: &mut CrcState<T::Vector>,
        coefficient: T::Vector,
        data_to_xor: T::Vector,
        ops: &T,
    ) where
        T::Vector: Copy;

    /// Fold width-specific number of bytes
    unsafe fn fold_width<T: ArchOps>(state: &mut CrcState<T::Vector>, high: u64, low: u64, ops: &T)
    where
        T::Vector: Copy;

    /// Width-specific Barrett reduction
    unsafe fn barrett_reduction<T: ArchOps>(
        state: &CrcState<T::Vector>,
        poly: u64,
        mu: u64,
        ops: &T,
    ) -> Self::Value
    where
        T::Vector: Copy;

    /// Create a coefficient vector for folding operations
    unsafe fn create_coefficient<T: ArchOps>(
        high: u64,
        low: u64,
        reflected: bool,
        ops: &T,
    ) -> T::Vector
    where
        T::Vector: Copy;

    /// Perform final reduction for the specific width
    unsafe fn perform_final_reduction<T: ArchOps>(
        state: T::Vector,
        reflected: bool,
        keys: &[u64; 23],
        ops: &T,
    ) -> Self::Value
    where
        T::Vector: Copy;

    /// Get the appropriate shuffle table pointer and offset for handling last bytes
    fn get_last_bytes_table_ptr(reflected: bool, remaining_len: usize) -> (*const u8, usize);
}

`/`	Search
`g` `c`	Code
`g` `h`	History
`g` `b`	Bookmarks
`g` `t`	Trust
`y`	Permalink to this state (files and trees)
`?`	This sheet

2cd40ea1 Andy McCall 1 month ago ↶	1	// Copyright 2025 Don MacAskill. Licensed under MIT or Apache-2.0.
	2
	3	#![allow(dead_code)]
	4
	5	#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
	6	use crate::enums::Reflector;
	7
	8	use crate::CrcParams;
	9
	10	#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
	11	use crate::structs::CrcState;
	12
	13	use core::ops::BitXor;
	14
	15	/// Marker trait for CRC width
	16	pub trait CrcWidth {
	17	/// The width in bits
	18	const WIDTH: u32;
	19	/// The natural value type for this width
	20	type Value: Copy + BitXor<Output = Self::Value>;
	21	}
	22
	23	pub(crate) trait CrcCalculator {
	24	fn update(data: &[u8], state: u64, params: &CrcParams) -> u64 {
	25	Self::calculate(state, data, params)
	26	}
	27
	28	fn checksum(data: &[u8], params: &CrcParams) -> u64 {
	29	Self::calculate(params.init, data, params) ^ params.xorout
	30	}
	31
	32	fn calculate(state: u64, data: &[u8], params: &CrcParams) -> u64;
	33	}
	34
	35	#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
	36	/// Trait defining architecture-specific SIMD operations for CRC calculation
	37	pub trait ArchOps: Sized + Copy + Clone {
	38	/// The SIMD vector type used by this architecture
	39	type Vector;
	40
	41	/// Process aligned blocks using potentially accelerated SIMD operations
	42	///
	43	/// Returns true if the operation was handled by the accelerated path (for example,
	44	/// using VPCLMULQDQ)
	45	unsafe fn process_enhanced_simd_blocks<W: EnhancedCrcWidth>(
	46	&self,
	47	_state: &mut CrcState<Self::Vector>,
	48	_first: &[Self::Vector; 8],
	49	_rest: &[[Self::Vector; 8]],
	50	_reflector: &Reflector<Self::Vector>,
	51	_keys: &[u64; 23],
	52	) -> bool
	53	where
	54	Self::Vector: Copy,
	55	{
	56	// Default implementation just returns false
	57	// indicating the non-enhanced algorithm should be used
	58	false
	59	}
	60
	61	/// Create a SIMD vector from a u64 pair
	62	///
	63	/// # Safety
	64	/// May use native CPU features
65	unsafe fn create_vector_from_u64_pair(
66	&self,
67	high: u64,
68	low: u64,
69	reflected: bool,
70	) -> Self::Vector;
71
72	/// Create a SIMD vector from a u64 pair without reflection
73	///
74	/// TODO: I have no idea (yet) why CRC-32 doesn't use reflection, but CRC-64 does.
75	///
76	/// # Safety
77	/// May use native CPU features
78	unsafe fn create_vector_from_u64_pair_non_reflected(&self, high: u64, low: u64)
79	-> Self::Vector;
80
81	/// Create a SIMD vector with a single u64 value
82	///
83	/// # Safety
84	/// May use native CPU features
85	unsafe fn create_vector_from_u64(&self, value: u64, high: bool) -> Self::Vector;
86
87	/// Extract two u64 values from a SIMD vector
88	///
89	/// # Safety
90	/// May use native CPU features
91	unsafe fn extract_u64s(&self, vector: Self::Vector) -> [u64; 2];
92
93	/// Extract two polynomial values (for carryless multiplication)
94	///
95	/// # Safety
96	/// May use native CPU features
97	unsafe fn extract_poly64s(&self, vector: Self::Vector) -> [u64; 2];
98
99	/// XOR two SIMD vectors
100	///
101	/// # Safety
102	/// May use native CPU features
103	unsafe fn xor_vectors(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;
104
105	/// Load bytes from memory into a SIMD vector
106	///
107	/// # Safety
108	/// May use native CPU features
109	unsafe fn load_bytes(&self, ptr: *const u8) -> Self::Vector;
110
111	/// Load aligned bytes from memory
112	///
113	/// # Safety
114	/// May use native CPU features
115	unsafe fn load_aligned(&self, ptr: *const [u64; 2]) -> Self::Vector;
116
117	//unsafe fn load_aligned(&self, ptr: &[u64]) -> Self::Vector;
118
119	//unsafe fn load_aligned_const(&self, ptr: *const [u64; 2]) -> Self::Vector;
120
121	/// Shuffle/permute bytes according to a mask
122	///
123	/// # Safety
124	/// May use native CPU features
125	unsafe fn shuffle_bytes(&self, data: Self::Vector, mask: Self::Vector) -> Self::Vector;
126
127	/// Blend two vectors using a mask (select from a or b based on mask bits)
128	///
129	/// # Safety
130	/// May use native CPU features
131	unsafe fn blend_vectors(
132	&self,
133	a: Self::Vector,
134	b: Self::Vector,
135	mask: Self::Vector,
136	) -> Self::Vector;
137
138	/// Shift a vector left by 8 bytes
139	///
140	/// # Safety
141	/// May use native CPU features
142	unsafe fn shift_left_8(&self, vector: Self::Vector) -> Self::Vector;
143
144	/// Create a vector with all bytes set to the same value
145	///
146	/// # Safety
147	/// May use native CPU features
148	unsafe fn set_all_bytes(&self, value: u8) -> Self::Vector;
149
150	/// Create a comparison mask (for blending operations)
151	///
152	/// # Safety
153	/// May use native CPU features
154	unsafe fn create_compare_mask(&self, vector: Self::Vector) -> Self::Vector;
155
156	/// AND two vectors
157	///
158	/// # Safety
159	/// May use native CPU features
160	unsafe fn and_vectors(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;
161
162	/// Shift a vector right by 32 bits (4 bytes)
163	///
164	/// # Safety
165	/// May use native CPU features
166	unsafe fn shift_right_32(&self, vector: Self::Vector) -> Self::Vector;
167
168	/// Shift a vector left by 32 bits (4 bytes)
169	///
170	/// # Safety
171	/// May use native CPU features
172	unsafe fn shift_left_32(&self, vector: Self::Vector) -> Self::Vector;
173
174	/// Create a SIMD vector with a single u32 value
175	///
176	/// # Safety
177	/// May use native CPU features
178	unsafe fn create_vector_from_u32(&self, value: u32, high: bool) -> Self::Vector;
179
180	/// Shift a vector left by 4 bytes (32 bits)
181	///
182	/// # Safety
183	/// May use native CPU features
184	unsafe fn shift_left_4(&self, vector: Self::Vector) -> Self::Vector;
185
186	/// Shift a vector right by 4 bytes (32 bits)
187	///
188	/// # Safety
189	/// May use native CPU features
190	unsafe fn shift_right_4(&self, vector: Self::Vector) -> Self::Vector;
191
192	/// Shift a vector right by 8 bytes (64 bits)
193	///
194	/// # Safety
195	/// May use native CPU features
196	unsafe fn shift_right_8(&self, vector: Self::Vector) -> Self::Vector;
197
198	/// Shift a vector right by 5 bytes
199	unsafe fn shift_right_5(&self, vector: Self::Vector) -> Self::Vector;
200
201	/// Shift a vector right by 6 bytes
202	unsafe fn shift_right_6(&self, vector: Self::Vector) -> Self::Vector;
203
204	/// Shift a vector right by 7 bytes
205	unsafe fn shift_right_7(&self, vector: Self::Vector) -> Self::Vector;
206
207	/// Shift a vector right by 12 bytes
208	unsafe fn shift_right_12(&self, vector: Self::Vector) -> Self::Vector;
209
210	/// Shift a vector left by 12 bytes
211	unsafe fn shift_left_12(&self, vector: Self::Vector) -> Self::Vector;
212
213	/// Perform carryless multiplication with immediate value 0x00 (low 64 bits of both vectors)
214	unsafe fn carryless_mul_00(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;
215
216	/// Perform carryless multiplication with immediate value 0x01 (low 64 bits of a, high 64 bits of b)
217	unsafe fn carryless_mul_01(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;
218
219	/// Perform carryless multiplication with immediate value 0x10 (high 64 bits of a, low 64 bits of b)
220	unsafe fn carryless_mul_10(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;
221
222	/// Perform carryless multiplication with immediate value 0x11 (high 64 bits of both vectors)
223	unsafe fn carryless_mul_11(&self, a: Self::Vector, b: Self::Vector) -> Self::Vector;
224
225	/// XOR three vectors together: a XOR b XOR c
226	/// Uses native XOR3 instructions when available, falls back to two XOR operations otherwise
227	unsafe fn xor3_vectors(
228	&self,
229	a: Self::Vector,
230	b: Self::Vector,
231	c: Self::Vector,
232	) -> Self::Vector;
233	}
234
235	#[cfg(any(target_arch = "x86", target_arch = "x86_64", target_arch = "aarch64"))]
236	/// Enhanced CrcWidth trait with additional operations for generic CRC implementation
237	pub trait EnhancedCrcWidth: CrcWidth {
238	/// Load constants specific to CRC width
239	fn load_constants(reflected: bool) -> [[u64; 2]; 4];
240
241	/// Create a CRC state with the initial value positioned correctly for the width
242	unsafe fn create_state<T: ArchOps>(
243	value: Self::Value,
244	reflected: bool,
245	ops: &T,
246	) -> CrcState<T::Vector>
247	where
248	T::Vector: Copy;
249
250	/// Extract the final CRC result from a SIMD vector
251	unsafe fn extract_result<T: ArchOps>(
252	vector: T::Vector,
253	reflected: bool,
254	ops: &T,
255	) -> Self::Value
256	where
257	T::Vector: Copy;
258
259	/// Perform width-specific folding operations using CLMUL and two XOR operations (or one XOR3)
260	unsafe fn fold_16<T: ArchOps>(
261	state: &mut CrcState<T::Vector>,
262	coefficient: T::Vector,
263	data_to_xor: T::Vector,
264	ops: &T,
265	) where
266	T::Vector: Copy;
267
268	/// Fold width-specific number of bytes
269	unsafe fn fold_width<T: ArchOps>(state: &mut CrcState<T::Vector>, high: u64, low: u64, ops: &T)
270	where
271	T::Vector: Copy;
272
273	/// Width-specific Barrett reduction
274	unsafe fn barrett_reduction<T: ArchOps>(
275	state: &CrcState<T::Vector>,
276	poly: u64,
277	mu: u64,
278	ops: &T,
279	) -> Self::Value
280	where
281	T::Vector: Copy;
282
283	/// Create a coefficient vector for folding operations
284	unsafe fn create_coefficient<T: ArchOps>(
285	high: u64,
286	low: u64,
287	reflected: bool,
288	ops: &T,
289	) -> T::Vector
290	where
291	T::Vector: Copy;
292
293	/// Perform final reduction for the specific width
294	unsafe fn perform_final_reduction<T: ArchOps>(
295	state: T::Vector,
296	reflected: bool,
297	keys: &[u64; 23],
298	ops: &T,
299	) -> Self::Value
300	where
301	T::Vector: Copy;
302
303	/// Get the appropriate shuffle table pointer and offset for handling last bytes
304	fn get_last_bytes_table_ptr(reflected: bool, remaining_len: usize) -> (*const u8, usize);
305	}

Keyboard