diff --git a/raid/avx2.c b/raid/avx2.c
index 86cea670..3354afa0 100644
--- a/raid/avx2.c
+++ b/raid/avx2.c
@@ -317,7 +317,8 @@ static __always_inline void raid_gen3_avx2ext_gen(int nd, size_t size, void **vv
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 3; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -574,7 +575,8 @@ static __always_inline void raid_gen4_avx2ext_gen(int nd, size_t size, void **vv
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 4; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -736,7 +738,8 @@ static __always_inline void raid_genX_avx2ext(int nd, size_t size, void **vv, in
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
diff --git a/raid/avx512.c b/raid/avx512.c
index cd808353..d23f67d6 100644
--- a/raid/avx512.c
+++ b/raid/avx512.c
@@ -57,8 +57,10 @@ void raid_gen2_avx512bw(int nd, size_t size, void **vv)
 	int d;
 
 	if (nd == 1) {
-		memcpy(v[1], v[0], size);
-		memcpy(v[2], v[0], size);
+		if (v[1] != v[0])
+			memcpy(v[1], v[0], size);
+		if (v[2] != v[0])
+			memcpy(v[2], v[0], size);
 		return;
 	}
 
@@ -281,9 +283,12 @@ void raid_gen3_avx512bw(int nd, size_t size, void **vv)
 	int d;
 
 	if (nd == 1) {
-		memcpy(v[1], v[0], size);
-		memcpy(v[2], v[0], size);
-		memcpy(v[3], v[0], size);
+		if (v[1] != v[0])
+			memcpy(v[1], v[0], size);
+		if (v[2] != v[0])
+			memcpy(v[2], v[0], size);
+		if (v[3] != v[0])
+			memcpy(v[3], v[0], size);
 		return;
 	}
 
@@ -450,7 +455,8 @@ static __always_inline void raid_genX_avx512bw(int nd, size_t size, void **vv, i
 
 	if (nd == 1) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
diff --git a/raid/gfni.c b/raid/gfni.c
index 60d6a3be..9fd6fb99 100644
--- a/raid/gfni.c
+++ b/raid/gfni.c
@@ -18,7 +18,8 @@ static __always_inline void raid_genX_avx2gfni_raid(int nd, size_t size, void **
 
 	if (nd == 1) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
@@ -129,7 +130,8 @@ static __always_inline void raid_genX_avx512gfni_raid(int nd, size_t size, void
 
 	if (nd == 1) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
@@ -212,7 +214,8 @@ static __always_inline void raid_genX_avx2gfni_aes(int nd, size_t size, void **v
 
 	if (nd == 1) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
@@ -330,7 +333,8 @@ static __always_inline void raid_genX_avx512gfni_aes(int nd, size_t size, void *
 
 	if (nd == 1) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
@@ -779,7 +783,7 @@ static __always_inline void raid_recX_avx512gfni_raid(int nr, int *id, int *ip,
 
 	raid_avx_end();
 }
- 
+
 static __always_inline void raid_recX_avx2gfni_aes(int nr, int *id, int *ip, int nd, size_t size, void **vv)
 {
 	uint8_t **v = (uint8_t **)vv;
diff --git a/raid/neon.c b/raid/neon.c
index f7ac4740..7af6ae34 100644
--- a/raid/neon.c
+++ b/raid/neon.c
@@ -250,7 +250,8 @@ static __always_inline void raid_genX_neon(int nd, size_t size, void **vv, int n
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
diff --git a/raid/neon32.c b/raid/neon32.c
index c64e541a..1483a6ff 100644
--- a/raid/neon32.c
+++ b/raid/neon32.c
@@ -244,7 +244,8 @@ static __always_inline void raid_genX_neon32(int nd, size_t size,
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
diff --git a/raid/ssse3.c b/raid/ssse3.c
index 464257ca..c5165bf1 100644
--- a/raid/ssse3.c
+++ b/raid/ssse3.c
@@ -26,7 +26,8 @@ static __always_inline void raid_gen3_ssse3_gen(int nd, size_t size, void **vv,
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 3; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -131,7 +132,8 @@ static __always_inline void raid_gen4_ssse3_gen(int nd, size_t size, void **vv,
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 4; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -257,7 +259,8 @@ void raid_gen5_ssse3_raid(int nd, size_t size, void **vv)
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 5; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -394,7 +397,8 @@ void raid_gen6_ssse3_raid(int nd, size_t size, void **vv)
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 6; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -545,7 +549,8 @@ static __always_inline void raid_gen3_ssse3ext_gen(int nd, size_t size, void **v
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 3; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -805,7 +810,8 @@ static __always_inline void raid_gen4_ssse3ext_gen(int nd, size_t size, void **v
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (i = 0; i < 4; ++i)
-			memcpy(v[1 + i], v[0], size);
+			if (v[1 + i] != v[0])
+				memcpy(v[1 + i], v[0], size);
 		return;
 	}
 
@@ -995,7 +1001,8 @@ static __always_inline void raid_genX_ssse3ext(int nd, size_t size, void **vv, i
 	/* special case with only one data disk */
 	if (l == 0) {
 		for (d = 0; d < np; ++d)
-			memcpy(v[1 + d], v[0], size);
+			if (v[1 + d] != v[0])
+				memcpy(v[1 + d], v[0], size);
 		return;
 	}
 
@@ -1773,7 +1780,7 @@ static __always_inline void raid_recX_ssse3(int nr, int *id, int *ip, int nd, si
  *
  * Process two 16-byte blocks per iteration, grouping equal operations across
  * both lanes to expose instruction-level parallelism.
- * 
+ *
  * A four-way 64-byte unroll was tested but is slower than this two-way 32-byte
  * implementation.
  */