Benchmark of SSE2 sinus calculation (based on intel's Approximate Math Library) vs FPU fsin instruction

SSE2_VS_FPU_SIN

format PE CONSOLE

include 'win32ax.inc'

entry main

pow = 20

section '.code' code readable executable
	proc main
		mov esi, 1 shl pow
		.loop:
			rdtsc
			mov dword[time], eax
			mov dword[time + 4], edx
			
			repeat 50
				movss xmm0, [src]
				;unedited
				movss	[esp - 4], xmm0
				movss	xmm1, [_ps_am_inv_sign_mask]
				mov		eax, [esp - 4]
				mulss	xmm0, [_ps_am_2_o_pi]
				andps	xmm0, xmm1
				and		eax, 0x80000000
				cvttss2si	ecx, xmm0
				movss	xmm1, [_ps_am_1]
				mov		edx, ecx
				shl		edx, (31 - 1)
				cvtsi2ss	xmm2, ecx
				and		ecx, 0x1
				and		edx, 0x80000000
				subss	xmm0, xmm2
				movss	xmm6, [_sincos_masks + ecx * 4]
				minss	xmm0, xmm1
				movss	xmm5, [_ps_sincos_p3]
				subss	xmm1, xmm0
				andps	xmm1, xmm6
				andnps	xmm6, xmm0
				orps	xmm1, xmm6
				movss	xmm4, [_ps_sincos_p2]
				movss	xmm0, xmm1
				mulss	xmm1, xmm1
				movss	xmm7, [_ps_sincos_p1]
				xor		eax, edx
				movss	xmm2, xmm1
				mulss	xmm1, xmm5
				movss	xmm5, [_ps_sincos_p0]
				mov		[esp - 4], eax
				addss	xmm1, xmm4
				mulss	xmm1, xmm2
				movss	xmm3, [esp - 4]
				addss	xmm1, xmm7
				mulss	xmm1, xmm2
				orps	xmm0, xmm3
				addss	xmm1, xmm5
				mulss	xmm0, xmm1
				
				cvtss2sd xmm0, xmm0
				movlpd [dest], xmm0
			end repeat
			
			rdtsc
			sub eax, dword[time]
			sbb edx, dword[time + 4]
			
			add dword[total_time], eax
			adc dword[total_time + 4], edx
			dec esi
		jnz .loop
				
		fild [total_time]
		fidiv [divv]
		fstp [total_time]
		
		cinvoke printf, fmt1, double[total_time], double[dest]
		
		xorps xmm0, xmm0
		movlpd [total_time], xmm0
		
		finit
		
		mov esi, 1 shl pow
		.loop2:
			rdtsc
			mov dword[time], eax
			mov dword[time + 4], edx
			
			repeat 50
				fld [src]
				fsin
				fstp [dest]
			end repeat
			
			rdtsc
			sub eax, dword[time]
			sbb edx, dword[time + 4]
			
			add dword[total_time], eax
			adc dword[total_time + 4], edx
			dec esi
		jnz .loop2
				
		fild [total_time]
		fidiv [divv]
		fstp [total_time]
		
		cinvoke printf, fmt2, double[total_time], double[dest]
		
		.below:
		cinvoke getch
		ret
	endp

section '.data' data readable writeable
	time dq 0
	total_time dq 0
	divv dd 1 shl pow
	fmt1 db 'SSE2 %lf sin(45 degree): %lf', 10, 0
	fmt2 db 'FPU  %lf sin(45 degree): %lf', 10, 0
	
	src dd 0.78539816339744830961566084581988
	dest dq ?
	
	align 4
	
    _ps_am_inv_sign_mask dd 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF
    _ps_am_pi_o_2 dd 1.57079632679, 1.57079632679, 1.57079632679, 1.57079632679
    _ps_am_2_o_pi dd 0.63661977236, 0.63661977236, 0.63661977236, 0.63661977236
    _epi32_1 dd 1, 1, 1, 1
    _ps_am_1 dd 1.0, 1.0, 1.0, 1.0
    _epi32_2 dd 2, 2, 2, 2

    _ps_sincos_p3 dd -0.00468175413, -0.00468175413, -0.00468175413, -0.00468175413
    _ps_sincos_p2 dd 0.0796926262, 0.0796926262, 0.0796926262, 0.0796926262
    _ps_sincos_p1 dd -0.64596409750621,-0.64596409750621,-0.64596409750621,-0.64596409750621 
    _ps_sincos_p0 dd 1.570796326794896, 1.570796326794896, 1.570796326794896, 1.570796326794896

	_sincos_masks dd 0x0, not 0x0
	
section '.idata' import data readable writeable
	library msvcrt, 'msvcrt.dll'
	import msvcrt,\
		printf, 'printf',\
		getch, '_getch'