fix: improve parallelization

fuel-vm/src/interpreter/memory.rs

@@ -1115,7 +1115,12 @@ fn slices_equal_avx2(a: &[u8], b: &[u8]) -> bool {
     }
 
     unsafe {
-        let mut aggregate_mask = -1i32;
+        let mut aggregate_mask_a = -1i32;
+        let mut aggregate_mask_b = -1i32;
+        let mut aggregate_mask_c = -1i32;
+        let mut aggregate_mask_d = -1i32;
+        let mut aggregate_mask_a_b = -1i32;
+        let mut aggregate_mask_c_d = -1i32;
 
         while i + CHUNK <= len {
             let simd_a1 = _mm256_loadu_si256(a.as_ptr().add(i) as *const _);
@@ -1151,9 +1156,15 @@ fn slices_equal_avx2(a: &[u8], b: &[u8]) -> bool {
             let cmp7 = _mm256_movemask_epi8(_mm256_cmpeq_epi8(simd_a7, simd_b7));
             let cmp8 = _mm256_movemask_epi8(_mm256_cmpeq_epi8(simd_a8, simd_b8));
 
-            aggregate_mask &= cmp1 & cmp2 & cmp3 & cmp4 & cmp5 & cmp6 & cmp7 & cmp8;
+            aggregate_mask_a &= cmp1 & cmp2;
+            aggregate_mask_b &= cmp3 & cmp4;
+            aggregate_mask_c &= cmp5 & cmp6;
+            aggregate_mask_d &= cmp7 & cmp8;
 
-            if aggregate_mask != -1i32 {
+            aggregate_mask_a_b &= aggregate_mask_a & aggregate_mask_b;
+            aggregate_mask_c_d &= aggregate_mask_c & aggregate_mask_d;
+
+            if aggregate_mask_a_b & aggregate_mask_c_d != -1i32 {
                 return false;
             }
 
@@ -1209,7 +1220,6 @@ fn slices_equal_avx512(a: &[u8], b: &[u8]) -> bool {
             let simd_a8 = _mm512_loadu_si512(a.as_ptr().add(i + 448) as *const _);
             let simd_b8 = _mm512_loadu_si512(b.as_ptr().add(i + 448) as *const _);
 
-            // Compare each pair of registers
             let cmp1 = _mm512_cmpeq_epi8_mask(simd_a1, simd_b1);
             let cmp2 = _mm512_cmpeq_epi8_mask(simd_a2, simd_b2);
             let cmp3 = _mm512_cmpeq_epi8_mask(simd_a3, simd_b3);
@@ -1219,11 +1229,17 @@ fn slices_equal_avx512(a: &[u8], b: &[u8]) -> bool {
             let cmp7 = _mm512_cmpeq_epi8_mask(simd_a7, simd_b7);
             let cmp8 = _mm512_cmpeq_epi8_mask(simd_a8, simd_b8);
 
-            // Combine all comparison masks
-            let combined_cmp = cmp1 & cmp2 & cmp3 & cmp4 & cmp5 & cmp6 & cmp7 & cmp8;
+            let cmp1_2 = cmp1 & cmp2;
+            let cmp3_4 = cmp3 & cmp4;
+            let cmp5_6 = cmp5 & cmp6;
+            let cmp7_8 = cmp7 & cmp8;
+
+            let cmp1_4 = cmp1_2 & cmp3_4;
+            let cmp5_8 = cmp5_6 & cmp7_8;
+
+            let full_cmp = cmp1_4 & cmp5_8;
 
-            // Check if all bytes are equal (mask should have all bits set)
-            if combined_cmp != u64::MAX {
+            if full_cmp != u64::MAX {
                 return false;
             }