Reverse Engineering for Beginners

CHAPTER 25. SIMD CHAPTER 25. SIMD

mov eax, 13

jmp .L4

25.2 SIMDstrlen()implementation.

It has to be noted that theSIMDinstructions can be inserted in C/C++ code via special macros^7. For MSVC, some of them are
located in theintrin.hfile.

It is possible to implement thestrlen()function^8 using SIMD instructions that works 2-2.5 times faster than the common
implementation. This function loads 16 characters into a XMM-register and check each against zero^9.

size_t strlen_sse2(const char str)
{
register size_t len = 0;
const char s=str;
bool str_is_aligned=(((unsigned int)str)&0xFFFFFFF0) == (unsigned int)str;

if (str_is_aligned==false)

return strlen (str);

__m128i xmm0 = _mm_setzero_si128();

__m128i xmm1;

int mask = 0;

for (;;)

{

xmm1 = _mm_load_si128((__m128i *)s);

xmm1 = _mm_cmpeq_epi8(xmm1, xmm0);

if ((mask = _mm_movemask_epi8(xmm1)) != 0)

{

unsigned long pos;

_BitScanForward(&pos, mask);

len += (size_t)pos;

break;

}

s += sizeof(__m128i);

len += sizeof(__m128i);

};

return len;
}

Let’s compile it in MSVC 2010 with/Oxoption:

Listing 25.2: Optimizing MSVC 2010

_pos$75552 = -4 ; size = 4
_str$ = 8 ; size = 4
?strlen_sse2@@YAIPBD@Z PROC ; strlen_sse2

push ebp

mov ebp, esp

and esp, -16 ; fffffff0H

mov eax, DWORD PTR _str$[ebp]

sub esp, 12 ; 0000000cH

push esi

mov esi, eax

and esi, -16 ; fffffff0H

xor edx, edx

mov ecx, eax

cmp esi, eax

je SHORT $LN4@strlen_sse

lea edx, DWORD PTR [eax+1]

(^7) MSDN: MMX, SSE, and SSE2 Intrinsics
(^8) strlen() —standard C library function for calculating string length
(^9) The example is based on source code from:http://go.yurichev.com/17330.