diff --git a/src/ops/i16.rs b/src/ops/i16.rs
index 2835271..97b32ee 100644
--- a/src/ops/i16.rs
+++ b/src/ops/i16.rs
@@ -231,19 +231,13 @@ impl_op! {
             _mm512_movm_epi16(_mm512_cmplt_epi16_mask(a, b))
         }
         for Avx2(a: __m256i, b: __m256i) -> __m256i {
-            let gt = _mm256_cmpgt_epi16(a, b);
-            let eq = _mm256_cmpeq_epi16(a, b);
-            _mm256_andnot_si256(_mm256_or_si256(gt, eq), _mm256_set1_epi16(u32::MAX as i16))
+            _mm256_cmpgt_epi16(b, a)
         }
         for Sse41(a: __m128i, b: __m128i) -> __m128i {
-            let gt = _mm_cmpgt_epi16(a, b);
-            let eq = _mm_cmpeq_epi16(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi16(u32::MAX as i16))
+            _mm_cmpgt_epi16(b, a)
         }
         for Sse2(a: __m128i, b: __m128i) -> __m128i {
-            let gt = _mm_cmpgt_epi16(a, b);
-            let eq = _mm_cmpeq_epi16(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi16(u32::MAX as i16))
+            _mm_cmpgt_epi16(b, a)
         }
         for Scalar(a: i16, b: i16) -> i16 {
             if a < b {
diff --git a/src/ops/i32.rs b/src/ops/i32.rs
index 1014726..8596516 100644
--- a/src/ops/i32.rs
+++ b/src/ops/i32.rs
@@ -244,19 +244,13 @@ impl_op! {
             _mm512_movm_epi32(k)
         }
         for Avx2(a: __m256i, b: __m256i) -> __m256i {
-            let gt = _mm256_cmpgt_epi32(a, b);
-            let eq = _mm256_cmpeq_epi32(a, b);
-            _mm256_andnot_si256(_mm256_or_si256(gt, eq), _mm256_set1_epi32(u32::MAX as i32))
+            _mm256_cmpgt_epi32(b, a)
         }
         for Sse41(a: __m128i, b: __m128i) -> __m128i {
-            let gt = _mm_cmpgt_epi32(a, b);
-            let eq = _mm_cmpeq_epi32(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi32(u32::MAX as i32))
+            _mm_cmpgt_epi32(b, a)
         }
         for Sse2(a: __m128i, b: __m128i) -> __m128i {
-            let gt = _mm_cmpgt_epi32(a, b);
-            let eq = _mm_cmpeq_epi32(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi32(u32::MAX as i32))
+            _mm_cmpgt_epi32(b, a)
         }
         for Scalar(a: i32, b: i32) -> i32 {
             if a < b {
diff --git a/src/ops/i64.rs b/src/ops/i64.rs
index 7b925f9..d64b7aa 100644
--- a/src/ops/i64.rs
+++ b/src/ops/i64.rs
@@ -310,19 +310,13 @@ impl_op! {
             _mm512_movm_epi64(k)
         }
         for Avx2(a: __m256i, b: __m256i) -> __m256i {
-            let gt = _mm256_cmpgt_epi64(a, b);
-            let eq = _mm256_cmpeq_epi64(a, b);
-            _mm256_andnot_si256(_mm256_or_si256(gt, eq), _mm256_set1_epi64x(u64::MAX as i64))
+            _mm256_cmpgt_epi64(b, a)
         }
         for Sse41(a: __m128i, b: __m128i) -> __m128i {
-            let gt = sse_cmpgt_epi64_compat(a, b);
-            let eq = sse_cmpeq_epi64_compat(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi64x(u64::MAX as i64))
+            _mm_cmpgt_epi64(b, a)
         }
         for Sse2(a: __m128i, b: __m128i) -> __m128i {
-            let gt = sse_cmpgt_epi64_compat(a, b);
-            let eq = sse_cmpeq_epi64_compat(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi64x(u64::MAX as i64))
+            _mm_cmpgt_epi64(b, a)
         }
         for Scalar(a: i64, b: i64) -> i64 {
             if a < b {
diff --git a/src/ops/i8.rs b/src/ops/i8.rs
index 072a928..856a0e9 100644
--- a/src/ops/i8.rs
+++ b/src/ops/i8.rs
@@ -276,19 +276,13 @@ impl_op! {
             _mm512_movm_epi8(k)
         }
         for Avx2(a: __m256i, b: __m256i) -> __m256i {
-            let gt = _mm256_cmpgt_epi8(a, b);
-            let eq = _mm256_cmpeq_epi8(a, b);
-            _mm256_andnot_si256(_mm256_or_si256(gt, eq), _mm256_set1_epi8(u32::MAX as i8))
+            _mm256_cmpgt_epi8(b, a)
         }
         for Sse41(a: __m128i, b: __m128i) -> __m128i {
-            let gt = _mm_cmpgt_epi8(a, b);
-            let eq = _mm_cmpeq_epi8(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi8(u32::MAX as i8))
+            _mm_cmpgt_epi8(b, a)
         }
         for Sse2(a: __m128i, b: __m128i) -> __m128i {
-            let gt = _mm_cmpgt_epi8(a, b);
-            let eq = _mm_cmpeq_epi8(a, b);
-            _mm_andnot_si128(_mm_or_si128(gt, eq), _mm_set1_epi8(u32::MAX as i8))
+            _mm_cmpgt_epi8(b, a)
         }
         for Scalar(a: i8, b: i8) -> i8 {
             if a < b {