format ELF64 executable 3 entry _start ;------------------------------------------------------------------------------- segment readable executable ;------------------------------------------------------------------------------- ; NAME: logss ; IN: xmm0.x function argument ; OUT: xmm0.x function result ;------------------------------------------------------------------------------- align 16 logss: maxss xmm0,[g_min_norm_pos] movss xmm1,[g_1_0] movd edx,xmm0 andps xmm0,dqword [g_inv_mant_mask] orps xmm0,xmm1 movaps xmm4,xmm0 subss xmm0,xmm1 addss xmm4,xmm1 shr edx,23 rcpss xmm4,xmm4 mulss xmm0,xmm4 addss xmm0,xmm0 movaps xmm2,xmm0 mulss xmm0,xmm0 sub edx,0x7f movss xmm4,[g_log_p0] movss xmm6,[g_log_q0] mulss xmm4,xmm0 movss xmm5,[g_log_p1] mulss xmm6,xmm0 movss xmm7,[g_log_q1] addss xmm4,xmm5 addss xmm6,xmm7 movss xmm5,[g_log_p2] mulss xmm4,xmm0 movss xmm7,[g_log_q2] mulss xmm6,xmm0 addss xmm4,xmm5 movss xmm5,[g_log_c0] addss xmm6,xmm7 cvtsi2ss xmm1,edx mulss xmm0,xmm4 rcpss xmm6,xmm6 mulss xmm0,xmm6 mulss xmm0,xmm2 mulss xmm1,xmm5 addss xmm0,xmm2 addss xmm0,xmm1 ret ;------------------------------------------------------------------------------- ; NAME: main ; DESC: Program main function. ;------------------------------------------------------------------------------- align 16 main: imgptr equ rbp-8 push rbp mov rbp,rsp sub rsp,128 ; alloc memory for the image mov eax,9 ; sys_mmap xor edi,edi ; addr mov esi,SIZE*SIZE*4 ; length mov edx,0x1+0x2 ; PROT_READ | PROT_WRITE mov r10d,0x02+0x20 ; MAP_PRIVATE | MAP_ANONYMOUS mov r8,-1 ; fd xor r9d,r9d ; offset syscall mov [imgptr],rax mov rbx,rax ; begin loops xor r13d,r13d ; .LoopY index .LoopY: xor r12d,r12d ; .LoopX index .LoopX: ; compute c cvtsi2ss xmm0,r12d cvtsi2ss xmm1,r13d shufps xmm0,xmm1,00000000b shufps xmm0,xmm0,00001000b divps xmm0,dqword [g_size] subps xmm0,dqword [g_0_5] addps xmm0,xmm0 movaps xmm13,xmm0 divps xmm13,dqword [g_zoom] addps xmm13,dqword [g_center] ; c = xmm13 ; z = (0.0,0.0) dz = (1.0,0.0) xorps xmm14,xmm14 ; z = xmm14 xorps xmm15,xmm15 movss xmm15,[g_1_0] ; dz = xmm15 mov ecx,[g_bailout] .LoopBailout: ; dz = 2.0 * z * dz + (1.0,0.0) movaps xmm0,xmm14 movaps xmm1,xmm15 shufps xmm0,xmm0,01000100b shufps xmm1,xmm1,00010100b mulps xmm0,xmm1 xorps xmm0,dqword [g_inv_y_sign] movaps xmm1,xmm0 shufps xmm0,xmm0,00001000b shufps xmm1,xmm1,00001101b addps xmm0,xmm1 addps xmm0,xmm0 addss xmm0,[g_1_0] movaps xmm15,xmm0 ; z = z * z + c movaps xmm0,xmm14 movaps xmm1,xmm0 shufps xmm0,xmm0,00000100b shufps xmm1,xmm1,01010100b mulps xmm0,xmm1 xorps xmm0,dqword [g_inv_y_sign] movaps xmm1,xmm0 shufps xmm0,xmm0,00001000b shufps xmm1,xmm1,00001101b addps xmm0,xmm1 addps xmm0,xmm13 movaps xmm14,xmm0 ; compute dot(z,z) mulps xmm0,xmm0 movaps xmm1,xmm0 shufps xmm1,xmm1,01010101b addps xmm0,xmm1 ; if dot(z,z) > g_z_max break .LoopBailout ucomiss xmm0,[g_z_max] ja .NotInSet sub ecx,1 jnz .LoopBailout xorps xmm0,xmm0 ; distance is zero jmp .InSet .NotInSet: movaps xmm8,xmm0 call logss movaps xmm9,xmm0 movaps xmm1,xmm15 mulps xmm1,xmm1 movaps xmm2,xmm1 shufps xmm2,xmm2,01010101b addps xmm1,xmm2 ; dot(dz,dz) divps xmm8,xmm1 ; dot(z,z) / dot(dz,dz) sqrtps xmm0,xmm8 mulps xmm0,dqword [g_0_5] mulps xmm0,xmm9 mulps xmm0,dqword [g_zoom] sqrtps xmm0,xmm0 sqrtps xmm0,xmm0 shufps xmm0,xmm0,00000000b mulps xmm0,dqword [g_brightness] .InSet: ; convert from [0.0,1.0] to [0,255] mulps xmm0,dqword [g_255_0] cvttps2dq xmm0,xmm0 movd eax,xmm0 mov [rbx+0],al pshufd xmm1,xmm0,00000001b movd eax,xmm1 mov [rbx+1],al pshufd xmm1,xmm0,00000010b movd eax,xmm1 mov [rbx+2],al mov byte [rbx+3],255 ; advance pixel pointer add rbx,4 ; continue .LoopX inc r12d cmp r12d,SIZE jne .LoopX ; continue .LoopY inc r13d cmp r13d,SIZE jne .LoopY ; create TGA file mov eax,85 mov rdi,g_tga_name mov esi,110000000b syscall mov rbx,rax ; write header mov eax,1 mov rdi,rbx mov rsi,g_tga_head mov edx,18 syscall ; write pixel data mov eax,1 mov rdi,rbx mov rsi,[imgptr] mov edx,SIZE*SIZE*4 syscall mov rsp,rbp pop rbp ret restore imgptr ;------------------------------------------------------------------------------- ; NAME: _start ; DESC: Program entry point. ;------------------------------------------------------------------------------- _start: call main ; terminate process mov eax,60 xor edi,edi syscall ;------------------------------------------------------------------------------- segment readable ;------------------------------------------------------------------------------- align 1 g_tga_name db 'mandelbrot.tga',0 g_tga_head db 0,0,2,9 dup 0 db (SIZE and 0x00ff),(SIZE and 0xff00) shr 8 db (SIZE and 0x00ff),(SIZE and 0xff00) shr 8,32,0 align 4 g_min_norm_pos dd 0x00800000 g_log_p0 dd -0.789580278884799154124 g_log_p1 dd 16.3866645699558079767 g_log_p2 dd -64.1409952958715622951 g_log_q0 dd -35.6722798256324312549 g_log_q1 dd 312.093766372244180303 g_log_q2 dd -769.691943550460008604 g_log_c0 dd 0.693147180559945 g_z_max dd 100.0 g_bailout dd 1000 align 16 SIZE=800 g_size dd 4 dup 800.0 g_center dd -0.761574,-0.0847596,0.0,0.0 g_zoom dd 4 dup 1000.0 g_brightness dd 4 dup 1.2 g_0_0 dd 4 dup 0.0 g_0_5 dd 4 dup 0.5 g_1_0 dd 4 dup 1.0 g_255_0 dd 4 dup 255.0 g_inv_mant_mask dd 4 dup (not 0x7f800000) g_inv_y_sign dd 0x0,0x80000000,0x0,0x0 ;-------------------------------------------------------------------------------