Whats the fastest way to convert RGB to ARGB and optionally xor at the same time? This is what I currently use:
bool DrawImage(const unsigned char* __restrict a, unsigned int* __restrict b, const int size, const bool xor){
if (xor){
#pragma simd
for (int i = size / 3; 0 < i; --i , a += 3){
*b++ = *reinterpret_cast<const unsigned int*>(a) ^ *b | 0xFF000000;
}
}else{
#pragma ivdep
#pragma simd
for (int i = size / 3; 0 < i; --i , a += 3){
*b++ = *reinterpret_cast<const unsigned int*>(a) | 0xFF000000;
}
}
return true;
}These functions are performance critical and I want to convert them to inline assembly can this be optimized further:
lea ecx,[esi+esi*2] movd xmm1,dword ptr [edi+ecx+9] movd xmm3,dword ptr [edi+ecx+3] movd xmm2,dword ptr [edi+ecx+6] movd xmm4,dword ptr [edi+ecx] punpckldq xmm3,xmm1 punpckldq xmm4,xmm2 punpckldq xmm4,xmm3 pxor xmm4,xmmword ptr [ebx+esi*4] por xmm4,xmm0 movdqa xmmword ptr [ebx+esi*4],xmm4 add esi,4 cmp esi,eax jb DrawImage+0A4h (0F8F96F4h)
lea ecx,[esi+esi*2] movd xmm1,dword ptr [edi+ecx+9] movd xmm3,dword ptr [edi+ecx+3] movd xmm2,dword ptr [edi+ecx+6] movd xmm4,dword ptr [edi+ecx] punpckldq xmm3,xmm1 punpckldq xmm4,xmm2 punpckldq xmm4,xmm3 pxor xmm4,xmmword ptr [ebx+esi*4] por xmm4,xmm0 movdqa xmmword ptr [ebx+esi*4],xmm4 add esi,4 cmp esi,eax jb DrawImage+0A4h (0F8F96F4h)