Test Asm

gistfile1.asm

void Test( float* kernel, float* src, float* dst, int n )
{
 in          al,dx  
 and         esp,0FFFFFFF0h  
 sub         esp,2Ch  
	__m128 zero = _mm_setzero_ps();
	__m128 x0 = zero;
	__m128 x1 = zero;
	__m128 x2 = zero;
	__m128 x3 = zero;
	__m128 f0;
	__m128 f1;
	__m128 f2;
	__m128 f3;

	// init filter
	__m128 k0 = _mm_loadu_ps(kernel + 0);
	__m128 k1 = _mm_loadu_ps(kernel + 4);
 movups      xmm6,xmmword ptr [ecx+10h]  
 movups      xmm7,xmmword ptr [ecx]  

	f0 = _mm_shuffle_ps(k1, k1, _MM_SHUFFLE(0, 1, 2, 3));
 movaps      xmm1,xmm6  
 shufps      xmm1,xmm6,1Bh  
 xorps       xmm0,xmm0  
 movaps      xmmword ptr [esp+0Ch],xmm1  
	f1 = _mm_shuffle_ps(k0, k0, _MM_SHUFFLE(0, 1, 2, 3));
 movaps      xmm1,xmm7  
 shufps      xmm1,xmm7,1Bh  
	f2 = _mm_move_ss(k0, k1);
 movss       xmm7,xmm6  
	f2 = _mm_shuffle_ps(f2, f2, _MM_SHUFFLE(0, 3, 2, 1));
	f3 = _mm_move_ss(k1, zero);
 movss       xmm6,xmm0  
 push        esi  
 movaps      xmm5,xmm0  
 movaps      xmm2,xmm0  
 movaps      xmm3,xmm0  
 movaps      xmm4,xmm0  
 movaps      xmmword ptr [esp+20h],xmm1  
 shufps      xmm7,xmm7,39h  
	f3 = _mm_shuffle_ps(f3, f3, _MM_SHUFFLE(0, 3, 2, 1));
 shufps      xmm6,xmm6,39h  
 mov         ecx,0Eh  
 lea         ecx,[ecx]  
		x0 = _mm_move_ss(x0, x1);
		x0 = _mm_shuffle_ps(x0, x0, _MM_SHUFFLE(0, 3, 2, 1));
		x1 = _mm_move_ss(x1, x2);
		x1 = _mm_shuffle_ps(x1, x1, _MM_SHUFFLE(0, 3, 2, 1));
		x2 = _mm_move_ss(x2, x3);
		x2 = _mm_shuffle_ps(x2, x2, _MM_SHUFFLE(0, 3, 2, 1));
		x3 = _mm_move_ss(x3, zero);
		x3 = _mm_shuffle_ps(x3, x3, _MM_SHUFFLE(0, 3, 2, 1));


		__m128 s = _mm_load1_ps(src++);
 movss       xmm1,dword ptr [edx]  
 movss       xmm5,xmm2  
 movss       xmm2,xmm3  
 movss       xmm3,xmm4  
		x0 = _mm_move_ss(x0, x1);
		x0 = _mm_shuffle_ps(x0, x0, _MM_SHUFFLE(0, 3, 2, 1));
		x1 = _mm_move_ss(x1, x2);
		x1 = _mm_shuffle_ps(x1, x1, _MM_SHUFFLE(0, 3, 2, 1));
		x2 = _mm_move_ss(x2, x3);
		x2 = _mm_shuffle_ps(x2, x2, _MM_SHUFFLE(0, 3, 2, 1));
		x3 = _mm_move_ss(x3, zero);
		x3 = _mm_shuffle_ps(x3, x3, _MM_SHUFFLE(0, 3, 2, 1));


		__m128 s = _mm_load1_ps(src++);
 movss       xmm4,xmm0  
 shufps      xmm1,xmm1,0  
		x0 = _mm_add_ps(x0, _mm_mul_ps(f0, s));
 movaps      xmm0,xmm1  
 mulps       xmm0,xmmword ptr [esp+10h]  
 shufps      xmm5,xmm5,39h  
 shufps      xmm2,xmm2,39h  
 shufps      xmm3,xmm3,39h  
 addps       xmm5,xmm0  
		x1 = _mm_add_ps(x1, _mm_mul_ps(f1, s));
 movaps      xmm0,xmm1  
 mulps       xmm0,xmmword ptr [esp+20h]  
 add         edx,4  
 shufps      xmm4,xmm4,39h  
 addps       xmm2,xmm0  
		x2 = _mm_add_ps(x2, _mm_mul_ps(f2, s));
 movaps      xmm0,xmm1  
 mulps       xmm0,xmm7  
		x3 = _mm_add_ps(x3, _mm_mul_ps(f3, s));
 mulps       xmm1,xmm6  
 addps       xmm3,xmm0  
 xorps       xmm0,xmm0  
 addps       xmm4,xmm1  
 dec         ecx  
 jne         Test+50h (010B1050h)  

	// prime
	#pragma loop( no_vector )
	for(int i=0; i<14; ++i) {
 mov         esi,dword ptr [n]  
 mov         ecx,dword ptr [dst]  
 lea         esp,[esp]  
	}

	// pipeline
	do {
		x0 = _mm_move_ss(x0, x1);
		x0 = _mm_shuffle_ps(x0, x0, _MM_SHUFFLE(0, 3, 2, 1));
		x1 = _mm_move_ss(x1, x2);
		x1 = _mm_shuffle_ps(x1, x1, _MM_SHUFFLE(0, 3, 2, 1));
		x2 = _mm_move_ss(x2, x3);
		x2 = _mm_shuffle_ps(x2, x2, _MM_SHUFFLE(0, 3, 2, 1));
		x3 = _mm_move_ss(x3, zero);
		x3 = _mm_shuffle_ps(x3, x3, _MM_SHUFFLE(0, 3, 2, 1));

		__m128 s = _mm_load1_ps(src++);
 movss       xmm1,dword ptr [edx]  
 xorps       xmm0,xmm0  
 movss       xmm5,xmm2  
 movss       xmm2,xmm3  
 movss       xmm3,xmm4  
 movss       xmm4,xmm0  
 shufps      xmm1,xmm1,0  
		x0 = _mm_add_ps(x0, _mm_mul_ps(f0, s));
 movaps      xmm0,xmm1  
 mulps       xmm0,xmmword ptr [esp+10h]  
 shufps      xmm5,xmm5,39h  
 shufps      xmm2,xmm2,39h  
		x1 = _mm_add_ps(x1, _mm_mul_ps(f1, s));
		x2 = _mm_add_ps(x2, _mm_mul_ps(f2, s));
		x3 = _mm_add_ps(x3, _mm_mul_ps(f3, s));

		_mm_store_ss(dst++, x0);
 mov         eax,ecx  
 addps       xmm5,xmm0  
 movaps      xmm0,xmm1  
 mulps       xmm0,xmmword ptr [esp+20h]  
 shufps      xmm3,xmm3,39h  
 shufps      xmm4,xmm4,39h  
 add         ecx,4  
 addps       xmm2,xmm0  
 movaps      xmm0,xmm1  
 mulps       xmm0,xmm7  
 mulps       xmm1,xmm6  
 lea         edx,[edx+4]  
 addps       xmm3,xmm0  
 movss       dword ptr [eax],xmm5  
 addps       xmm4,xmm1  
	} while(--n);
 dec         esi  
 jne         Test+0B0h (010B10B0h)  
}
 pop         esi  
 mov         esp,ebp  
 pop         ebp  
 ret