How to compare two vectors using SIMD and get a single boolean result?

Question

I have two vectors of 4 integers each and I'd like to use a SIMD command to compare them (say generate a result vector where each entry is 0 or 1 according to the result of the comparison).

Then, I'd like to compare the result vector to a vector of 4 zeros and only if they're equal do something.

Do you know what SIMD commands I can use to do it?

Solution

To compare two SIMD vectors:

#include <stdint.h>
#include <xmmintrin.h>

int32_t __attribute__ ((aligned(16))) vector1[4] = { 1, 2, 3, 4 };
int32_t __attribute__ ((aligned(16))) vector2[4] = { 1, 2, 2, 2 };
int32_t __attribute__ ((aligned(16))) result[4];

__m128i v1 = _mm_load_si128((__m128i *)vector1);
__m128i v2 = _mm_load_si128((__m128i *)vector2);
__m128i vcmp = _mm_cmpeq_epi32(v1, v2);
_mm_store_si128((__m128i *)result, vcmp);

Notes:

• data is assumed to be 32 bit integers
• vector1, vector2, result all need to be 16 byte aligned
• result will be -1 for equal, 0 for not equal ({ -1, -1, 0, 0 } for above code example)

---

UPDATE

If you just want a single Boolean result for the case where all 4 elements match then you can do it like this:

#include <stdint.h>
#include <xmmintrin.h>

int32_t __attribute__ ((aligned(16))) vector1[4] = { 1, 2, 3, 4 };
int32_t __attribute__ ((aligned(16))) vector2[4] = { 1, 2, 2, 2 };

__m128i v1 = _mm_load_si128((__m128i *)vector1);
__m128i v2 = _mm_load_si128((__m128i *)vector2);
__m128i vcmp = _mm_cmpeq_epi32(v1, v2);
uint16_t mask = _mm_movemask_epi8(vcmp);
int result = (mask == 0xffff);

Comments

[N.M]

Thanks. When I get the result vector, is there a simple way to check it is all zeros (I'd like to have a Boolean representing if it is all zeros or not).

[N.M]

I could use _mm_cmpgt_epi32 if I want to get a result for greater then, right?

[Paul R]

@N.M: yes, _mm_cmpgt_epi32.

[Simon]

You need [] after vector1 & 2 no ?

[Paul R]

@Simon: good catch - it's taken 4 years for anyone to notice !

[Heygard Flisch]

Typo: _mm_movemask_epi8 returns and int and no __m128i.

[Paul R]

@HeygardFlisch: thanks - I normally test the code I post - I obviously didn't this time !

[user1794469]

mask should be a uint16_t otherwise 0xffff is out of range.