Use the following SSSE3 intrinsics for horizontal addition.
extern __m128i _mm_hadd_epi16 (__m128i a, __m128i b);
Add horizontally packed signed words.
Interpreting a, b, and r as arrays of 16-bit signed integers:
for (i = 0; i < 4; i++) {
r[i] = a[2*i] + a[2i+1];
r[i+4] = b[2*i] + b[2*i+1];
}
extern __m128i _mm_hadd_epi32 (__m128i a, __m128i b);
Add horizontally packed signed dwords.
Interpreting a, b, and r as arrays of 32-bit signed integers:
for (i = 0; i < 2; i++) {
r[i] = a[2*i] + a[2i+1];
r[i+2] = b[2*i] + b[2*i+1];
}
extern __m128i _mm_hadds_epi16 (__m128i a, __m128i b);
Add horizontally packed signed words with signed saturation.
Interpreting a, b, and r as arrays of 16-bit signed integers:
for (i = 0; i < 4; i++) {
r[i] = signed_saturate_to_word(a[2*i] + a[2i+1]);
r[i+4] = signed_saturate_to_word(b[2*i] + b[2*i+1]);
}
extern __m64 _mm_hadd_pi16 (__m64 a, __m64 b);
Add horizontally packed signed words.
Interpreting a, b, and r as arrays of 16-bit signed integers:
for (i = 0; i < 2; i++) {
r[i] = a[2*i] + a[2i+1];
r[i+2] = b[2*i] + b[2*i+1];
}
extern __m64 _mm_hadd_pi32 (__m64 a, __m64 b);
Add horizontally packed signed dwords.
Interpreting a, b, and r as arrays of 32-bit signed integers:
r[0] = a[1] + a[0];
r[1] = b[1] + b[0];
extern __m64 _mm_hadds_pi16 (__m64 a, __m64 b);
Add horizontally packed signed words with signed saturation.
Interpreting a, b, and r as arrays of 16-bit signed integers:
for (i = 0; i < 2; i++) {
r[i] = signed_saturate_to_word(a[2*i] + a[2i+1]);
r[i+2] = signed_saturate_to_word(b[2*i] + b[2*i+1]);
}