Storing 0 into a string

[HellOfMice]

Just a proposition, I don't know if it is possible, so I need your advices

_mm256_xor_si256

Compute the bitwise XOR of the 256-bit integer value of a and b.

char szString1[32] ;
_mm256_xor_si256(szString1,szString1) ;


That would fill a 32 bytes string with 0.

[HellOfMice]

                        __m256i a = _mm256_loadu_si256((__m256i*) _szTmp1);
                                 __m256i b = _mm256_loadu_si256((__m256i*) _szTmp1);


                        _mm256_xor_si256(a,b) ;

The compiler does not generate code for __mm_xor_si256 but it does for loading the two addresses

[HellOfMice]

The goo syntax for storing 32 Zeroes in a string is:

Code: [Select]


__m256i a = _mm256_loadu_si256((__m256i*) _szTmp1) ;
__m256i b = _mm256_loadu_si256((__m256i*) _szTmp1) ;


__m256i c = _mm256_xor_si256(a,b) ;
__mm256_storeu_si256((__m256i*) _szTmp1,c) ;


Verified under Pelle's C Compiler with pope

Code: [Select]


0x0000ED51  vmovdqu                         ymm0,ymmword ptr [rsp+188]  C5 FE 6F 84 24 88 01 00 00
0x0000ED5A  vmovdqu                         ymm1,ymmword ptr [rsp+188]  C5 FE 6F 8C 24 88 01 00 00
0x0000ED63  vpxor                              ymm0,ymm0,ymm1        C5 FD EF C1
0x0000ED67  vmovdqu                         ymmword ptr [rsp+188],ymm0  C5 FE 7F 84 24 88 01 00 00
Helped with this link: https://jcma.me/blog/2017/01/23/accelerate-c-code-with-avx2-instructions

[Vortex]

/arch option (POCC) [9.00] :
 

Syntax:
/arch:{SSE2 | AVX | AVX2}

Description:
The /arch option selects the processor architecture (for X64).

Architecture Description Default
SSE2 Enable use of Streaming SIMD Extensions 2 instructions. Yes
AVX Enable use of Advanced Vector Extensions instructions. No
AVX2 Enable use of Advanced Vector Extensions 2 instructions. No

[HellOfMice]

Hi Vortex


In the project file I select the SSE2 for the code generation


Could it be possible to compute the mean of RGB values with these instruction; I am searching.
I think that first I must have A0R0G0B int this order for the number and after it is possible to add them.
It's an idea, maybe I dream


Thank you for your comments