rol32
(a) = _mm_add_epi32((a), (b)); (d) ^= a; (d) = rol32((d), 16); \
(c) = _mm_add_epi32((c), (d)); (b) ^= c; (b) = rol32((b), 12); \
(a) = _mm_add_epi32((a), (b)); (d) ^= a; (d) = rol32((d), 8); \
(c) = _mm_add_epi32((c), (d)); (b) ^= c; (b) = rol32((b), 7); \
r0 = _mm_add_epi32(r0, r1); r3 ^= r0; r3 = rol32(r3, 16);
r2 = _mm_add_epi32(r2, r3); r1 ^= r2; r1 = rol32(r1, 12);
r0 = _mm_add_epi32(r0, r1); r3 ^= r0; r3 = rol32(r3, 8);
r2 = _mm_add_epi32(r2, r3); r1 ^= r2; r1 = rol32(r1, 7);
c0 = _mm_add_epi32(c0, c1); c3 ^= c0; c3 = rol32(c3, 16);
c2 = _mm_add_epi32(c2, c3); c1 ^= c2; c1 = rol32(c1, 12);
c0 = _mm_add_epi32(c0, c1); c3 ^= c0; c3 = rol32(c3, 8);
c2 = _mm_add_epi32(c2, c3); c1 ^= c2; c1 = rol32(c1, 7);
(a) += (b); (d) ^= (a); (d) = rol32((d), 16); \
(c) += (d); (b) ^= (c); (b) = rol32((b), 12); \
(a) += (b); (d) ^= (a); (d) = rol32((d), 8); \
(c) += (d); (b) ^= (c); (b) = rol32((b), 7); \