diff --git a/algo/cryptonight/cryptonight_av2.c b/algo/cryptonight/cryptonight_av2.c
index 7e5f4109f..654dd5bcb 100644
--- a/algo/cryptonight/cryptonight_av2.c
+++ b/algo/cryptonight/cryptonight_av2.c
@@ -204,5 +204,101 @@ void cryptonight_av2_v1(const uint8_t *restrict input, size_t size, uint8_t *res
 
 void cryptonight_av2_v2(const uint8_t *restrict input, size_t size, uint8_t *restrict output, struct cryptonight_ctx **restrict ctx)
 {
+    keccak(input,        size, ctx[0]->state, 200);
+    keccak(input + size, size, ctx[1]->state, 200);
 
+    const uint8_t* l0 = ctx[0]->memory;
+    const uint8_t* l1 = ctx[1]->memory;
+    uint64_t* h0 = (uint64_t*) ctx[0]->state;
+    uint64_t* h1 = (uint64_t*) ctx[1]->state;
+
+    VARIANT2_INIT(0);
+    VARIANT2_INIT(1);
+    VARIANT2_SET_ROUNDING_MODE();
+
+    cn_explode_scratchpad((__m128i*) h0, (__m128i*) l0);
+    cn_explode_scratchpad((__m128i*) h1, (__m128i*) l1);
+
+    uint64_t al0 = h0[0] ^ h0[4];
+    uint64_t al1 = h1[0] ^ h1[4];
+    uint64_t ah0 = h0[1] ^ h0[5];
+    uint64_t ah1 = h1[1] ^ h1[5];
+
+    __m128i bx00 = _mm_set_epi64x(h0[3] ^ h0[7], h0[2] ^ h0[6]);
+    __m128i bx01 = _mm_set_epi64x(h0[9] ^ h0[11], h0[8] ^ h0[10]);
+    __m128i bx10 = _mm_set_epi64x(h1[3] ^ h1[7], h1[2] ^ h1[6]);
+    __m128i bx11 = _mm_set_epi64x(h1[9] ^ h1[11], h1[8] ^ h1[10]);
+
+    uint64_t idx0 = al0;
+    uint64_t idx1 = al1;
+
+    for (size_t i = 0; __builtin_expect(i < 0x80000, 1); i++) {
+        __m128i cx0       = _mm_load_si128((__m128i *) &l0[idx0 & 0x1FFFF0]);
+        __m128i cx1       = _mm_load_si128((__m128i *) &l1[idx1 & 0x1FFFF0]);
+
+        const __m128i ax0 = _mm_set_epi64x(ah0, al0);
+        const __m128i ax1 = _mm_set_epi64x(ah1, al1);
+
+        cx0 = _mm_aesenc_si128(cx0, ax0);
+        cx1 = _mm_aesenc_si128(cx1, ax1);
+
+        VARIANT2_SHUFFLE(l0, idx0 & 0x1FFFF0, ax0, bx00, bx01);
+        _mm_store_si128((__m128i *) &l0[idx0 & 0x1FFFF0], _mm_xor_si128(bx00, cx0));
+
+        VARIANT2_SHUFFLE(l1, idx1 & 0x1FFFF0, ax1, bx10, bx11);
+        _mm_store_si128((__m128i *) &l1[idx1 & 0x1FFFF0], _mm_xor_si128(bx10, cx1));
+
+        idx0 = _mm_cvtsi128_si64(cx0);
+        idx1 = _mm_cvtsi128_si64(cx1);
+
+        uint64_t hi, lo, cl, ch;
+        cl = ((uint64_t*) &l0[idx0 & 0x1FFFF0])[0];
+        ch = ((uint64_t*) &l0[idx0 & 0x1FFFF0])[1];
+
+        VARIANT2_INTEGER_MATH(0, cl, cx0);
+        lo = _umul128(idx0, cl, &hi);
+        VARIANT2_SHUFFLE2(l0, idx0 & 0x1FFFF0, ax0, bx00, bx01, hi, lo);
+
+        al0 += hi;
+        ah0 += lo;
+
+        ((uint64_t*)&l0[idx0 & 0x1FFFF0])[0] = al0;
+        ((uint64_t*)&l0[idx0 & 0x1FFFF0])[1] = ah0;
+
+        al0 ^= cl;
+        ah0 ^= ch;
+        idx0 = al0;
+
+        cl = ((uint64_t*) &l1[idx1 & 0x1FFFF0])[0];
+        ch = ((uint64_t*) &l1[idx1 & 0x1FFFF0])[1];
+
+        VARIANT2_INTEGER_MATH(1, cl, cx1);
+        lo = _umul128(idx1, cl, &hi);
+        VARIANT2_SHUFFLE2(l1, idx1 & 0x1FFFF0, ax1, bx10, bx11, hi, lo);
+
+        al1 += hi;
+        ah1 += lo;
+
+        ((uint64_t*)&l1[idx1 & 0x1FFFF0])[0] = al1;
+        ((uint64_t*)&l1[idx1 & 0x1FFFF0])[1] = ah1;
+
+        al1 ^= cl;
+        ah1 ^= ch;
+        idx1 = al1;
+
+        bx01 = bx00;
+        bx11 = bx10;
+
+        bx00 = cx0;
+        bx10 = cx1;
+    }
+
+    cn_implode_scratchpad((__m128i*) l0, (__m128i*) h0);
+    cn_implode_scratchpad((__m128i*) l1, (__m128i*) h1);
+
+    keccakf(h0, 24);
+    keccakf(h1, 24);
+
+    extra_hashes[ctx[0]->state[0] & 3](ctx[0]->state, 200, output);
+    extra_hashes[ctx[1]->state[0] & 3](ctx[1]->state, 200, output + 32);
 }
diff --git a/algo/cryptonight/cryptonight_av4.c b/algo/cryptonight/cryptonight_av4.c
index bb4840952..5ff299c12 100644
--- a/algo/cryptonight/cryptonight_av4.c
+++ b/algo/cryptonight/cryptonight_av4.c
@@ -202,6 +202,103 @@ void cryptonight_av4_v1(const uint8_t *restrict input, size_t size, uint8_t *res
 }
 
 
-void cryptonight_av4_v2(const uint8_t *restrict input, size_t size, uint8_t *restrict output, struct cryptonight_ctx *restrict ctx)
+void cryptonight_av4_v2(const uint8_t *restrict input, size_t size, uint8_t *restrict output, struct cryptonight_ctx **restrict ctx)
 {
+    keccak(input,        size, ctx[0]->state, 200);
+    keccak(input + size, size, ctx[1]->state, 200);
+
+    const uint8_t* l0 = ctx[0]->memory;
+    const uint8_t* l1 = ctx[1]->memory;
+    uint64_t* h0 = (uint64_t*) ctx[0]->state;
+    uint64_t* h1 = (uint64_t*) ctx[1]->state;
+
+    VARIANT2_INIT(0);
+    VARIANT2_INIT(1);
+    VARIANT2_SET_ROUNDING_MODE();
+
+    cn_explode_scratchpad((__m128i*) h0, (__m128i*) l0);
+    cn_explode_scratchpad((__m128i*) h1, (__m128i*) l1);
+
+    uint64_t al0 = h0[0] ^ h0[4];
+    uint64_t al1 = h1[0] ^ h1[4];
+    uint64_t ah0 = h0[1] ^ h0[5];
+    uint64_t ah1 = h1[1] ^ h1[5];
+
+    __m128i bx00 = _mm_set_epi64x(h0[3] ^ h0[7], h0[2] ^ h0[6]);
+    __m128i bx01 = _mm_set_epi64x(h0[9] ^ h0[11], h0[8] ^ h0[10]);
+    __m128i bx10 = _mm_set_epi64x(h1[3] ^ h1[7], h1[2] ^ h1[6]);
+    __m128i bx11 = _mm_set_epi64x(h1[9] ^ h1[11], h1[8] ^ h1[10]);
+
+    uint64_t idx0 = al0;
+    uint64_t idx1 = al1;
+
+    for (size_t i = 0; __builtin_expect(i < 0x80000, 1); i++) {
+        __m128i cx0       = _mm_load_si128((__m128i *) &l0[idx0 & 0x1FFFF0]);
+        __m128i cx1       = _mm_load_si128((__m128i *) &l1[idx1 & 0x1FFFF0]);
+
+        const __m128i ax0 = _mm_set_epi64x(ah0, al0);
+        const __m128i ax1 = _mm_set_epi64x(ah1, al1);
+
+        cx0 = soft_aesenc(cx0, ax0);
+        cx1 = soft_aesenc(cx1, ax1);
+
+        VARIANT2_SHUFFLE(l0, idx0 & 0x1FFFF0, ax0, bx00, bx01);
+        _mm_store_si128((__m128i *) &l0[idx0 & 0x1FFFF0], _mm_xor_si128(bx00, cx0));
+
+        VARIANT2_SHUFFLE(l1, idx1 & 0x1FFFF0, ax1, bx10, bx11);
+        _mm_store_si128((__m128i *) &l1[idx1 & 0x1FFFF0], _mm_xor_si128(bx10, cx1));
+
+        idx0 = _mm_cvtsi128_si64(cx0);
+        idx1 = _mm_cvtsi128_si64(cx1);
+
+        uint64_t hi, lo, cl, ch;
+        cl = ((uint64_t*) &l0[idx0 & 0x1FFFF0])[0];
+        ch = ((uint64_t*) &l0[idx0 & 0x1FFFF0])[1];
+
+        VARIANT2_INTEGER_MATH(0, cl, cx0);
+        lo = _umul128(idx0, cl, &hi);
+        VARIANT2_SHUFFLE2(l0, idx0 & 0x1FFFF0, ax0, bx00, bx01, hi, lo);
+
+        al0 += hi;
+        ah0 += lo;
+
+        ((uint64_t*)&l0[idx0 & 0x1FFFF0])[0] = al0;
+        ((uint64_t*)&l0[idx0 & 0x1FFFF0])[1] = ah0;
+
+        al0 ^= cl;
+        ah0 ^= ch;
+        idx0 = al0;
+
+        cl = ((uint64_t*) &l1[idx1 & 0x1FFFF0])[0];
+        ch = ((uint64_t*) &l1[idx1 & 0x1FFFF0])[1];
+
+        VARIANT2_INTEGER_MATH(1, cl, cx1);
+        lo = _umul128(idx1, cl, &hi);
+        VARIANT2_SHUFFLE2(l1, idx1 & 0x1FFFF0, ax1, bx10, bx11, hi, lo);
+
+        al1 += hi;
+        ah1 += lo;
+
+        ((uint64_t*)&l1[idx1 & 0x1FFFF0])[0] = al1;
+        ((uint64_t*)&l1[idx1 & 0x1FFFF0])[1] = ah1;
+
+        al1 ^= cl;
+        ah1 ^= ch;
+        idx1 = al1;
+
+        bx01 = bx00;
+        bx11 = bx10;
+
+        bx00 = cx0;
+        bx10 = cx1;
+    }
+
+    cn_implode_scratchpad((__m128i*) l0, (__m128i*) h0);
+    cn_implode_scratchpad((__m128i*) l1, (__m128i*) h1);
+
+    keccakf(h0, 24);
+    keccakf(h1, 24);
+
+    extra_hashes[ctx[0]->state[0] & 3](ctx[0]->state, 200, output);
+    extra_hashes[ctx[1]->state[0] & 3](ctx[1]->state, 200, output + 32);
 }