diff --git a/aes/hash.go b/aes/hash.go
index 654f9ff..6e7c3a8 100644
--- a/aes/hash.go
+++ b/aes/hash.go
@@ -57,21 +57,21 @@ func HashAes1Rx4(input []byte, output *[64]byte) {
 	for input_ptr := 0; input_ptr < len(input); input_ptr += 64 {
 		in := (*[4][4]uint32)(unsafe.Pointer(unsafe.SliceData(input[input_ptr:])))
 
-		soft_aesenc(&states[0], &in[0])
-		soft_aesdec(&states[1], &in[1])
-		soft_aesenc(&states[2], &in[2])
-		soft_aesdec(&states[3], &in[3])
+		aesenc(&states[0], &in[0])
+		aesdec(&states[1], &in[1])
+		aesenc(&states[2], &in[2])
+		aesdec(&states[3], &in[3])
 	}
 
-	soft_aesenc(&states[0], &keys.AesHash1R_XKeys[0])
-	soft_aesdec(&states[1], &keys.AesHash1R_XKeys[0])
-	soft_aesenc(&states[2], &keys.AesHash1R_XKeys[0])
-	soft_aesdec(&states[3], &keys.AesHash1R_XKeys[0])
+	aesenc(&states[0], &keys.AesHash1R_XKeys[0])
+	aesdec(&states[1], &keys.AesHash1R_XKeys[0])
+	aesenc(&states[2], &keys.AesHash1R_XKeys[0])
+	aesdec(&states[3], &keys.AesHash1R_XKeys[0])
 
-	soft_aesenc(&states[0], &keys.AesHash1R_XKeys[1])
-	soft_aesdec(&states[1], &keys.AesHash1R_XKeys[1])
-	soft_aesenc(&states[2], &keys.AesHash1R_XKeys[1])
-	soft_aesdec(&states[3], &keys.AesHash1R_XKeys[1])
+	aesenc(&states[0], &keys.AesHash1R_XKeys[1])
+	aesdec(&states[1], &keys.AesHash1R_XKeys[1])
+	aesenc(&states[2], &keys.AesHash1R_XKeys[1])
+	aesdec(&states[3], &keys.AesHash1R_XKeys[1])
 
 	copy(output[:], (*[64]byte)(unsafe.Pointer(&states))[:])
 }
@@ -95,10 +95,10 @@ func FillAes1Rx4(state *[64]byte, output []byte) {
 	states := (*[4][4]uint32)(unsafe.Pointer(state))
 
 	for outptr := 0; outptr < len(output); outptr += len(state) {
-		soft_aesdec(&states[0], &keys.AesGenerator1R_Keys[0])
-		soft_aesenc(&states[1], &keys.AesGenerator1R_Keys[1])
-		soft_aesdec(&states[2], &keys.AesGenerator1R_Keys[2])
-		soft_aesenc(&states[3], &keys.AesGenerator1R_Keys[3])
+		aesdec(&states[0], &keys.AesGenerator1R_Keys[0])
+		aesenc(&states[1], &keys.AesGenerator1R_Keys[1])
+		aesdec(&states[2], &keys.AesGenerator1R_Keys[2])
+		aesenc(&states[3], &keys.AesGenerator1R_Keys[3])
 
 		copy(output[outptr:], state[:])
 	}
@@ -116,25 +116,25 @@ func FillAes4Rx4(state [64]byte, output []byte) {
 	states := (*[4][4]uint32)(unsafe.Pointer(&state))
 
 	for outptr := 0; outptr < len(output); outptr += len(state) {
-		soft_aesdec(&states[0], &keys.AesGenerator4R_Keys[0])
-		soft_aesenc(&states[1], &keys.AesGenerator4R_Keys[0])
-		soft_aesdec(&states[2], &keys.AesGenerator4R_Keys[4])
-		soft_aesenc(&states[3], &keys.AesGenerator4R_Keys[4])
+		aesdec(&states[0], &keys.AesGenerator4R_Keys[0])
+		aesenc(&states[1], &keys.AesGenerator4R_Keys[0])
+		aesdec(&states[2], &keys.AesGenerator4R_Keys[4])
+		aesenc(&states[3], &keys.AesGenerator4R_Keys[4])
 
-		soft_aesdec(&states[0], &keys.AesGenerator4R_Keys[1])
-		soft_aesenc(&states[1], &keys.AesGenerator4R_Keys[1])
-		soft_aesdec(&states[2], &keys.AesGenerator4R_Keys[5])
-		soft_aesenc(&states[3], &keys.AesGenerator4R_Keys[5])
+		aesdec(&states[0], &keys.AesGenerator4R_Keys[1])
+		aesenc(&states[1], &keys.AesGenerator4R_Keys[1])
+		aesdec(&states[2], &keys.AesGenerator4R_Keys[5])
+		aesenc(&states[3], &keys.AesGenerator4R_Keys[5])
 
-		soft_aesdec(&states[0], &keys.AesGenerator4R_Keys[2])
-		soft_aesenc(&states[1], &keys.AesGenerator4R_Keys[2])
-		soft_aesdec(&states[2], &keys.AesGenerator4R_Keys[6])
-		soft_aesenc(&states[3], &keys.AesGenerator4R_Keys[6])
+		aesdec(&states[0], &keys.AesGenerator4R_Keys[2])
+		aesenc(&states[1], &keys.AesGenerator4R_Keys[2])
+		aesdec(&states[2], &keys.AesGenerator4R_Keys[6])
+		aesenc(&states[3], &keys.AesGenerator4R_Keys[6])
 
-		soft_aesdec(&states[0], &keys.AesGenerator4R_Keys[3])
-		soft_aesenc(&states[1], &keys.AesGenerator4R_Keys[3])
-		soft_aesdec(&states[2], &keys.AesGenerator4R_Keys[7])
-		soft_aesenc(&states[3], &keys.AesGenerator4R_Keys[7])
+		aesdec(&states[0], &keys.AesGenerator4R_Keys[3])
+		aesenc(&states[1], &keys.AesGenerator4R_Keys[3])
+		aesdec(&states[2], &keys.AesGenerator4R_Keys[7])
+		aesenc(&states[3], &keys.AesGenerator4R_Keys[7])
 
 		copy(output[outptr:], state[:])
 	}
diff --git a/aes/round_amd64.go b/aes/round_amd64.go
new file mode 100644
index 0000000..6be0afe
--- /dev/null
+++ b/aes/round_amd64.go
@@ -0,0 +1,33 @@
+//go:build amd64 && !purego
+
+package aes
+
+import (
+	_ "git.gammaspectra.live/P2Pool/go-randomx/v2/asm"
+	"golang.org/x/sys/cpu"
+	_ "unsafe"
+)
+
+//go:linkname hard_aesdec git.gammaspectra.live/P2Pool/go-randomx/v2/asm.aesdec
+func hard_aesdec(state *[4]uint32, key *[4]uint32)
+
+//go:linkname hard_aesenc git.gammaspectra.live/P2Pool/go-randomx/v2/asm.aesenc
+func hard_aesenc(state *[4]uint32, key *[4]uint32)
+
+var supportsAES = cpu.X86.HasAES
+
+func aesenc(state *[4]uint32, key *[4]uint32) {
+	if supportsAES {
+		hard_aesenc(state, key)
+	} else {
+		soft_aesenc(state, key)
+	}
+}
+
+func aesdec(state *[4]uint32, key *[4]uint32) {
+	if supportsAES {
+		hard_aesdec(state, key)
+	} else {
+		soft_aesdec(state, key)
+	}
+}
diff --git a/aes/round_generic.go b/aes/round_generic.go
new file mode 100644
index 0000000..396cbb1
--- /dev/null
+++ b/aes/round_generic.go
@@ -0,0 +1,11 @@
+//go:build !amd64 || purego
+
+package aes
+
+func aesenc(state *[4]uint32, key *[4]uint32) {
+	soft_aesenc(state, key)
+}
+
+func aesdec(state *[4]uint32, key *[4]uint32) {
+	soft_aesdec(state, key)
+}
diff --git a/asm/aes.go b/asm/aes.go
new file mode 100644
index 0000000..698bc4c
--- /dev/null
+++ b/asm/aes.go
@@ -0,0 +1,11 @@
+//go:build amd64 && !purego
+
+package asm
+
+func AESRoundEncrypt(state *[4]uint32, key *[4]uint32) {
+	aesenc(state, key)
+}
+
+func AESRoundDecrypt(state *[4]uint32, key *[4]uint32) {
+	aesdec(state, key)
+}
diff --git a/asm/aes_amd64.go b/asm/aes_amd64.go
new file mode 100644
index 0000000..8e7b982
--- /dev/null
+++ b/asm/aes_amd64.go
@@ -0,0 +1,9 @@
+//go:build amd64 && !purego
+
+package asm
+
+//go:noescape
+func aesenc(state *[4]uint32, key *[4]uint32)
+
+//go:noescape
+func aesdec(state *[4]uint32, key *[4]uint32)
diff --git a/asm/aes_amd64.s b/asm/aes_amd64.s
new file mode 100644
index 0000000..3a09f95
--- /dev/null
+++ b/asm/aes_amd64.s
@@ -0,0 +1,23 @@
+//go:build amd64 && !purego
+
+#include "textflag.h"
+
+TEXT ·aesenc(SB),NOSPLIT|NOFRAME,$0-16
+	MOVQ state+0(FP), AX
+	MOVQ key+8(FP), BX
+	MOVUPS 0(AX), X0
+	MOVUPS 0(BX), X1
+	AESENC X1, X0
+	MOVUPS X0, 0(AX)
+	MOVUPS X1, 0(BX)
+	RET
+
+TEXT ·aesdec(SB),NOSPLIT|NOFRAME,$0-16
+	MOVQ state+0(FP), AX
+	MOVQ key+8(FP), BX
+	MOVUPS 0(AX), X0
+	MOVUPS 0(BX), X1
+	AESDEC X1, X0
+	MOVUPS X0, 0(AX)
+	MOVUPS X1, 0(BX)
+	RET
diff --git a/asm/aes_noasm.go b/asm/aes_noasm.go
new file mode 100644
index 0000000..3f4006b
--- /dev/null
+++ b/asm/aes_noasm.go
@@ -0,0 +1,11 @@
+//go:build !amd64 || purego
+
+package asm
+
+func AESRoundEncrypt(state *[4]uint32, key *[4]uint32) {
+	panic("not implemented")
+}
+
+func AESRoundDecrypt(state *[4]uint32, key *[4]uint32) {
+	panic("not implemented")
+}