Pass large array parametes (keys) by reference.

Author: onethumb

src/algorithm.rs

@@ -80,6 +80,9 @@ where
     // Create initial CRC state
     let mut crc_state = W::create_state(state, params.refin, ops);
 
+    // Extract keys once and pass by reference to avoid repeated stack copies
+    let keys = extract_keys_array(params);
+
     // Process data differently based on length
     // On ARM M4 Max, ARM c8g, x86 c7a, and x86 c7i, using 128 bytes is a measurably faster
     // threshold than 256 bytes...
@@ -91,19 +94,13 @@ where
             bytes,
             &mut crc_state,
             reflector,
-            extract_keys_array(params),
+            &keys,
             ops,
         );
     }
 
     // Process large inputs with SIMD-optimized approach
-    process_large_aligned::<T, W>(
-        bytes,
-        &mut crc_state,
-        reflector,
-        extract_keys_array(params),
-        ops,
-    )
+    process_large_aligned::<T, W>(bytes, &mut crc_state, reflector, &keys, ops)
 }
 
 /// Process data with the selected strategy
@@ -118,7 +115,7 @@ unsafe fn process_by_strategy<T: ArchOps, W: EnhancedCrcWidth>(
     data: &[u8],
     state: &mut CrcState<T::Vector>,
     reflector: Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) -> W::Value
 where
@@ -153,7 +150,7 @@ unsafe fn process_large_aligned<T: ArchOps, W: EnhancedCrcWidth>(
     bytes: &[u8],
     state: &mut CrcState<T::Vector>,
     reflector: Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) -> W::Value
 where
@@ -174,7 +171,7 @@ where
 
         // try to use the enhanced SIMD implementation first, fall back to non-enhanced if necessary
         if rest.is_empty()
-            || !ops.process_enhanced_simd_blocks::<W>(state, first, rest, &reflector, keys)
+            || !ops.process_enhanced_simd_blocks::<W>(state, first, rest, &reflector, *keys)
         {
             process_simd_chunks::<T, W>(state, first, rest, &reflector, keys, ops);
         }
@@ -207,7 +204,7 @@ unsafe fn process_simd_chunks<T: ArchOps, W: EnhancedCrcWidth>(
     first: &[T::Vector; 8],
     rest: &[[T::Vector; 8]],
     reflector: &Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) where
     T::Vector: Copy,
@@ -271,7 +268,7 @@ unsafe fn process_simd_chunks<T: ArchOps, W: EnhancedCrcWidth>(
     }
 
     // Perform final reduction and update state
-    let final_value = W::perform_final_reduction(res, state.reflected, keys, ops);
+    let final_value = W::perform_final_reduction(res, state.reflected, *keys, ops);
     *state = W::create_state(final_value, state.reflected, ops);
 }
 
@@ -286,7 +283,7 @@ unsafe fn process_exactly_16<T: ArchOps, W: EnhancedCrcWidth>(
     data: &[u8],
     state: &mut CrcState<T::Vector>,
     reflector: &Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) -> W::Value
 where
@@ -296,7 +293,7 @@ where
     W::perform_final_reduction(
         process_16_byte_block(data.as_ptr(), state.value, reflector, ops),
         state.reflected,
-        keys,
+        *keys,
         ops,
     )
 }
@@ -386,7 +383,7 @@ unsafe fn process_17_to_31<T: ArchOps, W: EnhancedCrcWidth>(
     data: &[u8],
     state: &mut CrcState<T::Vector>,
     reflector: &Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) -> W::Value
 where
@@ -413,7 +410,7 @@ where
     );
 
     // Perform final reduction
-    W::perform_final_reduction(final_xmm7, state.reflected, keys, ops)
+    W::perform_final_reduction(final_xmm7, state.reflected, *keys, ops)
 }
 
 // Process inputs between 32 and 255 bytes
@@ -428,7 +425,7 @@ unsafe fn process_32_to_255<T: ArchOps, W: EnhancedCrcWidth>(
     data: &[u8],
     state: &mut CrcState<T::Vector>,
     reflector: &Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) -> W::Value
 where
@@ -478,7 +475,7 @@ where
     }
 
     // Perform final reduction
-    W::perform_final_reduction(xmm7, state.reflected, keys, ops)
+    W::perform_final_reduction(xmm7, state.reflected, *keys, ops)
 }
 
 /// Data region descriptor for overlapping SIMD reads in CRC processing
@@ -512,7 +509,7 @@ struct DataRegion<'a> {
 unsafe fn get_last_two_xmms<T: ArchOps, W: EnhancedCrcWidth>(
     region: DataRegion,
     current_state: T::Vector,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     reflector: &Reflector<T::Vector>,
     reflected: bool,
     ops: &T,

src/combine.rs

@@ -84,7 +84,7 @@ fn gf2_matrix_square(square: &mut [u64; 64], mat: &[u64; 64]) {
 first sequence of bytes, crc2 is the CRC of the immediately following
 sequence of bytes, and len2 is the length of the second sequence.  The CRC
 of the combined sequence is returned. */
-pub fn checksums(mut crc1: u64, crc2: u64, mut len2: u64, params: CrcParams) -> u64 {
+pub fn checksums(mut crc1: u64, crc2: u64, mut len2: u64, params: &CrcParams) -> u64 {
     let mut col: u64;
     let mut even = [0u64; 64]; /* even-power-of-two zeros operator */
     let mut odd = [0u64; 64]; /* odd-power-of-two zeros operator */

src/crc32/algorithm.rs

@@ -237,7 +237,7 @@ pub(crate) unsafe fn process_0_to_15<T: ArchOps, W: EnhancedCrcWidth>(
     data: &[u8],
     state: &mut CrcState<T::Vector>,
     reflector: &Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) -> W::Value
 where
@@ -321,7 +321,7 @@ where
     };
 
     if len >= 4 {
-        return W::perform_final_reduction(xmm7, state.reflected, keys, ops);
+        return W::perform_final_reduction(xmm7, state.reflected, *keys, ops);
     }
 
     let final_state = CrcState {

src/crc64/algorithm.rs

@@ -213,7 +213,7 @@ pub(crate) unsafe fn process_0_to_15<T: ArchOps, W: EnhancedCrcWidth>(
     data: &[u8],
     state: &mut CrcState<T::Vector>,
     reflector: &Reflector<T::Vector>,
-    keys: [u64; 23],
+    keys: &[u64; 23],
     ops: &T,
 ) -> W::Value
 where
@@ -285,7 +285,7 @@ where
 
     if len >= CRC_HALF_CHUNK_SIZE as i32 {
         // For 8-15 bytes, perform additional folding
-        return W::perform_final_reduction(xmm7, state.reflected, keys, ops);
+        return W::perform_final_reduction(xmm7, state.reflected, *keys, ops);
     }
 
     let final_state = CrcState {

src/lib.rs

@@ -592,7 +592,7 @@ impl Digest {
             self.state ^ self.params.xorout,
             other_crc,
             other.amount,
-            self.params,
+            &self.params,
         ) ^ self.params.xorout;
     }
 
@@ -893,7 +893,7 @@ pub fn checksum_combine(
 ) -> u64 {
     let params = get_calculator_params(algorithm).1;
 
-    combine::checksums(checksum1, checksum2, checksum2_len, params)
+    combine::checksums(checksum1, checksum2, checksum2_len, &params)
 }
 
 /// Combines two CRC checksums using custom CRC parameters.
@@ -926,7 +926,7 @@ pub fn checksum_combine_with_params(
     checksum2: u64,
     checksum2_len: u64,
 ) -> u64 {
-    combine::checksums(checksum1, checksum2, checksum2_len, params)
+    combine::checksums(checksum1, checksum2, checksum2_len, &params)
 }
 
 /// Returns the target used to calculate the CRC checksum for the specified algorithm.