test: different intrinsic for cmp

FuelLabs · Jan 1, 2025 · 179fd24 · 179fd24
1 parent 123aab9
commit 179fd24
Showing 1 changed file with 17 additions and 22 deletions.
diff --git a/fuel-vm/src/interpreter/memory.rs b/fuel-vm/src/interpreter/memory.rs
@@ -1117,13 +1117,6 @@ fn slices_equal_avx2(a: &[u8], b: &[u8]) -> bool {
     let mut i = 0;
 
     unsafe {
-        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 _);
             let simd_b1 = _mm256_loadu_si256(b.as_ptr().add(i) as *const _);
@@ -1149,24 +1142,26 @@ fn slices_equal_avx2(a: &[u8], b: &[u8]) -> bool {
             let simd_a8 = _mm256_loadu_si256(a.as_ptr().add(i + 224) as *const _);
             let simd_b8 = _mm256_loadu_si256(b.as_ptr().add(i + 224) as *const _);
 
-            let cmp1 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a1, simd_b1));
-            let cmp2 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a2, simd_b2));
-            let cmp3 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a3, simd_b3));
-            let cmp4 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a4, simd_b4));
-            let cmp5 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a5, simd_b5));
-            let cmp6 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a6, simd_b6));
-            let cmp7 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a7, simd_b7));
-            let cmp8 = _mm256_movemask_epi8(_mm256_cmpeq_epi32(simd_a8, simd_b8));
+            let cmp1 = _mm256_mpsadbw_epu8(simd_a1, simd_b1, 0);
+            let cmp2 = _mm256_mpsadbw_epu8(simd_a2, simd_b2, 0);
+            let cmp3 = _mm256_mpsadbw_epu8(simd_a3, simd_b3, 0);
+            let cmp4 = _mm256_mpsadbw_epu8(simd_a4, simd_b4, 0);
+            let cmp5 = _mm256_mpsadbw_epu8(simd_a5, simd_b5, 0);
+            let cmp6 = _mm256_mpsadbw_epu8(simd_a6, simd_b6, 0);
+            let cmp7 = _mm256_mpsadbw_epu8(simd_a7, simd_b7, 0);
+            let cmp8 = _mm256_mpsadbw_epu8(simd_a8, simd_b8, 0);
+
+            let cmp9 = _mm256_testz_si256(cmp1, cmp2);
+            let cmp10 = _mm256_testz_si256(cmp3, cmp4);
+            let cmp11 = _mm256_testz_si256(cmp5, cmp6);
+            let cmp12 = _mm256_testz_si256(cmp7, cmp8);
 
-            aggregate_mask_a &= cmp1 & cmp2;
-            aggregate_mask_b &= cmp3 & cmp4;
-            aggregate_mask_c &= cmp5 & cmp6;
-            aggregate_mask_d &= cmp7 & cmp8;
+            let cmp13 = _mm256_testz_si256(cmp9, cmp10);
+            let cmp14 = _mm256_testz_si256(cmp11, cmp12);
 
-            aggregate_mask_a_b &= aggregate_mask_a & aggregate_mask_b;
-            aggregate_mask_c_d &= aggregate_mask_c & aggregate_mask_d;
+            let cmp15 = _mm256_testz_si256(cmp13, cmp14);
 
-            if aggregate_mask_a_b & aggregate_mask_c_d != -1i32 {
+            if cmp15 != -1i32 {
                 return false;
             }