Assuming I have SSE to SSE4.1, but not AVX(2), what is the fastest way to load a packed memory layout like this (all 32-bit integers):
a0 b0 c0 d0 a1 b1 c1 d1 a2 b2 c2 d2 a3 b3 c3 d3
Into four vectors a, b, c, d?
a: {a0, a1, a2, a3}
b: {b0, b1, b2, b3}
c: {c0, c1, c2, c3}
d: {d0, d1, d2, d3}
I'm not sure whether this is relevant or not, but in my actual application I have 16 vectors and as such a0 and a1 are 16*4 bytes apart in memory.
What you need here is 4 loads followed by a 4x4 transpose:
#include "emmintrin.h" // SSE2
v0 = _mm_load_si128((__m128i *)&a[0]); // v0 = a0 b0 c0 d0
v1 = _mm_load_si128((__m128i *)&a[16]); // v1 = a1 b1 c1 d1
v2 = _mm_load_si128((__m128i *)&a[32]); // v2 = a2 b2 c2 d2
v3 = _mm_load_si128((__m128i *)&a[48]); // v3 = a3 b3 c3 d3
// 4x4 transpose
w0 = _mm_unpacklo_epi32(v0, v1); // w0 = a0 a1 b0 b1
w1 = _mm_unpackhi_epi32(v0, v1); // w1 = c0 c1 d0 d1
w2 = _mm_unpacklo_epi32(v2, v3); // w2 = a2 a3 b2 b3
w3 = _mm_unpackhi_epi32(v2, v3); // w3 = c2 c3 d2 d3
v0 = _mm_unpacklo_epi64(w0, w2); // v0 = a0 a1 a2 a3
v1 = _mm_unpackhi_epi64(w0, w2); // v1 = b0 b1 b2 b3
v2 = _mm_unpacklo_epi64(w1, w3); // v2 = c0 c1 c2 c3
v3 = _mm_unpackhi_epi64(w1, w3); // v3 = d0 d1 d2 d3
Note: this is probably more efficient than using AVX2 gathered loads, since they generate a read cycle per element, which makes them really only useful when the access pattern is unknown or difficult to work with.
来源：https://stackoverflow.com/questions/19557746/what-is-the-fastest-way-to-do-a-simd-gather-without-avx2