AVX intrinsics with uin8_t and uint32_t

[maeshiro_mi]

Hi, all. I'm new with intrinsics and your help would be very appreciated.

I want to vectorize some std::vectors processing and my question is, how can I add(or subtract)  std::vector<uint8_t> and std::vector<uint32_t>?

Thanks in advance!

[Om_S_Intel]

The data types need to be of the same type. Also data access must be done using single stride for vectorization. These condition will not be met in your case.

[maeshiro_mi]

Thanks om-sachan.

The Auto-Vectorizer is able to vectorize this for loop, right?

std::vector<uint8_t> vec1;
vec1.resize(8192);
std::vector<uint32_t> vec2;
vec2.resize(8192);

for (int i = 0; i < 8192; i++) {
  vec2 += vec1;
}

So I thought I could code it with intrinsics to start learning.

Thanks again om-sachan!

[Bernard]

@maeshiro_mi

Run your code and inspect disassembly you should see vmovaps(d) and vaddps(d) instructions or their integer counterparts.

[maeshiro_mi]

Thanks iliyapolak.

I found very interesting checking assembly and converting it to intrinsics.

[Bernard]

maeshiro_mi wrote:

Thanks iliyapolak.

I found very interesting checking assembly and converting it to intrinsics.

I must admit that I am doing the same:)