mirror of
https://github.com/WinampDesktop/winamp.git
synced 2024-09-24 15:54:12 +00:00
883 lines
No EOL
24 KiB
C++
883 lines
No EOL
24 KiB
C++
/*
|
|
LICENSE
|
|
-------
|
|
Copyright 2005 Nullsoft, Inc.
|
|
All rights reserved.
|
|
|
|
Redistribution and use in source and binary forms, with or without modification,
|
|
are permitted provided that the following conditions are met:
|
|
|
|
* Redistributions of source code must retain the above copyright notice,
|
|
this list of conditions and the following disclaimer.
|
|
|
|
* Redistributions in binary form must reproduce the above copyright notice,
|
|
this list of conditions and the following disclaimer in the documentation
|
|
and/or other materials provided with the distribution.
|
|
|
|
* Neither the name of Nullsoft nor the names of its contributors may be used to
|
|
endorse or promote products derived from this software without specific prior written permission.
|
|
|
|
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR
|
|
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
|
|
FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
|
CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
|
|
IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
|
|
OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
// alphachannel safe 11/21/99
|
|
|
|
#include <windows.h>
|
|
#include <commctrl.h>
|
|
#include "r_defs.h"
|
|
#include "resource.h"
|
|
|
|
#include "timing.h"
|
|
#include "../Agave/Language/api_language.h"
|
|
|
|
|
|
#ifndef LASER
|
|
|
|
#define C_THISCLASS C_BlurClass
|
|
#define MOD_NAME "Trans / Blur"
|
|
|
|
static const int zero=0;
|
|
|
|
class C_THISCLASS : public C_RBASE2 {
|
|
protected:
|
|
public:
|
|
C_THISCLASS();
|
|
virtual ~C_THISCLASS();
|
|
virtual int render(char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h);
|
|
virtual char *get_desc() { static char desc[128]; return (!desc[0]?WASABI_API_LNGSTRING_BUF(IDS_TRANS_BLUR,desc,128):desc); }
|
|
virtual HWND conf(HINSTANCE hInstance, HWND hwndParent);
|
|
virtual void load_config(unsigned char *data, int len);
|
|
virtual int save_config(unsigned char *data);
|
|
|
|
virtual int smp_getflags() { return 1; }
|
|
virtual int smp_begin(int max_threads, char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h);
|
|
virtual void smp_render(int this_thread, int max_threads, char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h);
|
|
virtual int smp_finish(char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h); // return value is that of render() for fbstuff etc
|
|
|
|
int enabled;
|
|
|
|
int roundmode;
|
|
};
|
|
|
|
#define PUT_INT(y) data[pos]=(y)&255; data[pos+1]=(y>>8)&255; data[pos+2]=(y>>16)&255; data[pos+3]=(y>>24)&255
|
|
#define GET_INT() (data[pos]|(data[pos+1]<<8)|(data[pos+2]<<16)|(data[pos+3]<<24))
|
|
void C_THISCLASS::load_config(unsigned char *data, int len)
|
|
{
|
|
int pos=0;
|
|
if (len-pos >= 4) { enabled=GET_INT(); pos+=4; }
|
|
if (len-pos >= 4) { roundmode=GET_INT(); pos+=4; }
|
|
else roundmode=0;
|
|
}
|
|
int C_THISCLASS::save_config(unsigned char *data)
|
|
{
|
|
int pos=0;
|
|
PUT_INT(enabled); pos+=4;
|
|
PUT_INT(roundmode); pos+=4;
|
|
return pos;
|
|
}
|
|
|
|
|
|
|
|
|
|
C_THISCLASS::C_THISCLASS()
|
|
{
|
|
roundmode=0;
|
|
enabled=1;
|
|
}
|
|
|
|
C_THISCLASS::~C_THISCLASS()
|
|
{
|
|
}
|
|
|
|
#define MASK_SH1 (~(((1<<7)|(1<<15)|(1<<23))<<1))
|
|
#define MASK_SH2 (~(((3<<6)|(3<<14)|(3<<22))<<2))
|
|
#define MASK_SH3 (~(((7<<5)|(7<<13)|(7<<21))<<3))
|
|
#define MASK_SH4 (~(((15<<4)|(15<<12)|(15<<20))<<4))
|
|
static unsigned int mmx_mask1[2]={MASK_SH1,MASK_SH1};
|
|
static unsigned int mmx_mask2[2]={MASK_SH2,MASK_SH2};
|
|
static unsigned int mmx_mask3[2]={MASK_SH3,MASK_SH3};
|
|
static unsigned int mmx_mask4[2]={MASK_SH4,MASK_SH4};
|
|
|
|
#define DIV_2(x) ((( x ) & MASK_SH1)>>1)
|
|
#define DIV_4(x) ((( x ) & MASK_SH2)>>2)
|
|
#define DIV_8(x) ((( x ) & MASK_SH3)>>3)
|
|
#define DIV_16(x) ((( x ) & MASK_SH4)>>4)
|
|
|
|
|
|
void C_THISCLASS::smp_render(int this_thread, int max_threads, char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h)
|
|
{
|
|
if (!enabled) return;
|
|
|
|
timingEnter(0);
|
|
|
|
unsigned int *f = (unsigned int *) framebuffer;
|
|
unsigned int *of = (unsigned int *) fbout;
|
|
|
|
if (max_threads < 1) max_threads=1;
|
|
|
|
int start_l = ( this_thread * h ) / max_threads;
|
|
int end_l;
|
|
|
|
if (this_thread >= max_threads - 1) end_l = h;
|
|
else end_l = ( (this_thread+1) * h ) / max_threads;
|
|
|
|
int outh=end_l-start_l;
|
|
if (outh<1) return;
|
|
|
|
int skip_pix=start_l*w;
|
|
|
|
f += skip_pix;
|
|
of+= skip_pix;
|
|
|
|
int at_top=0, at_bottom=0;
|
|
|
|
if (!this_thread) at_top=1;
|
|
if (this_thread >= max_threads - 1) at_bottom=1;
|
|
|
|
if (enabled == 2)
|
|
{
|
|
// top line
|
|
if (at_top)
|
|
{
|
|
unsigned int *f2=f+w;
|
|
int x;
|
|
int adj_tl=0, adj_tl2=0;
|
|
if (roundmode) { adj_tl = 0x03030303; adj_tl2 = 0x04040404; }
|
|
// top left
|
|
*of++=DIV_2(f[0])+DIV_4(f[0])+DIV_8(f[1])+DIV_8(f2[0]) + adj_tl; f++; f2++;
|
|
// top center
|
|
x=(w-2)/4;
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_2(f[0]) + DIV_8(f[0]) + DIV_8(f[1]) + DIV_8(f[-1]) + DIV_8(f2[0]) + adj_tl2;
|
|
of[1]=DIV_2(f[1]) + DIV_8(f[1]) + DIV_8(f[2]) + DIV_8(f[0]) + DIV_8(f2[1]) + adj_tl2;
|
|
of[2]=DIV_2(f[2]) + DIV_8(f[2]) + DIV_8(f[3]) + DIV_8(f[1]) + DIV_8(f2[2]) + adj_tl2;
|
|
of[3]=DIV_2(f[3]) + DIV_8(f[3]) + DIV_8(f[4]) + DIV_8(f[2]) + DIV_8(f2[3]) + adj_tl2;
|
|
f+=4;
|
|
f2+=4;
|
|
of+=4;
|
|
}
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_2(f[0]) + DIV_8(f[0]) + DIV_8(f[1]) + DIV_8(f[-1]) + DIV_8(f2[0]) + adj_tl2;
|
|
f++;
|
|
f2++;
|
|
}
|
|
// top right
|
|
*of++=DIV_2(f[0])+DIV_4(f[0]) + DIV_8(f[-1])+DIV_8(f2[0]) + adj_tl; f++; f2++;
|
|
}
|
|
|
|
|
|
// middle block
|
|
{
|
|
int y=outh-at_top-at_bottom;
|
|
unsigned int adj_tl1=0,adj_tl2=0;
|
|
unsigned __int64 adj2=0;
|
|
if (roundmode) { adj_tl1=0x04040404; adj_tl2=0x05050505; adj2=0x0505050505050505i64; }
|
|
while (y--)
|
|
{
|
|
int x;
|
|
unsigned int *f2=f+w;
|
|
unsigned int *f3=f-w;
|
|
|
|
// left edge
|
|
*of++=DIV_2(f[0])+DIV_8(f[0])+DIV_8(f[1])+DIV_8(f2[0])+DIV_8(f3[0])+adj_tl1; f++; f2++; f3++;
|
|
|
|
// middle of line
|
|
#ifdef NO_MMX
|
|
x=(w-2)/4;
|
|
if (roundmode)
|
|
{
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_2(f[0]) + DIV_4(f[0]) + DIV_16(f[1]) + DIV_16(f[-1]) + DIV_16(f2[0]) + DIV_16(f3[0]) + 0x05050505;
|
|
of[1]=DIV_2(f[1]) + DIV_4(f[1]) + DIV_16(f[2]) + DIV_16(f[0]) + DIV_16(f2[1]) + DIV_16(f3[1]) + 0x05050505;
|
|
of[2]=DIV_2(f[2]) + DIV_4(f[2]) + DIV_16(f[3]) + DIV_16(f[1]) + DIV_16(f2[2]) + DIV_16(f3[2]) + 0x05050505;
|
|
of[3]=DIV_2(f[3]) + DIV_4(f[3]) + DIV_16(f[4]) + DIV_16(f[2]) + DIV_16(f2[3]) + DIV_16(f3[3]) + 0x05050505;
|
|
f+=4;
|
|
f2+=4;
|
|
f3+=4;
|
|
of+=4;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_2(f[0]) + DIV_4(f[0]) + DIV_16(f[1]) + DIV_16(f[-1]) + DIV_16(f2[0]) + DIV_16(f3[0]);
|
|
of[1]=DIV_2(f[1]) + DIV_4(f[1]) + DIV_16(f[2]) + DIV_16(f[0]) + DIV_16(f2[1]) + DIV_16(f3[1]);
|
|
of[2]=DIV_2(f[2]) + DIV_4(f[2]) + DIV_16(f[3]) + DIV_16(f[1]) + DIV_16(f2[2]) + DIV_16(f3[2]);
|
|
of[3]=DIV_2(f[3]) + DIV_4(f[3]) + DIV_16(f[4]) + DIV_16(f[2]) + DIV_16(f2[3]) + DIV_16(f3[3]);
|
|
f+=4;
|
|
f2+=4;
|
|
f3+=4;
|
|
of+=4;
|
|
}
|
|
}
|
|
#else
|
|
{
|
|
__asm
|
|
{
|
|
mov ecx, w
|
|
mov edx, ecx
|
|
mov ebx, edx
|
|
neg ebx
|
|
mov esi, f
|
|
mov edi, of
|
|
sub ecx, 2
|
|
shr ecx, 2
|
|
movq mm1, [esi-4]
|
|
align 16
|
|
mmx_light_blur_loop:
|
|
movq mm0, [esi]
|
|
|
|
movq mm2, [esi+4]
|
|
pand mm0, mmx_mask1
|
|
|
|
movq mm5, mm2
|
|
psrl mm0, 1
|
|
|
|
movq mm7, [esi+8]
|
|
movq mm4, mm0
|
|
|
|
pand mm1, mmx_mask4
|
|
pand mm4, mmx_mask1
|
|
|
|
movq mm3, [esi+edx*4]
|
|
psrl mm4, 1
|
|
|
|
paddb mm0, mm4
|
|
pand mm2, mmx_mask4
|
|
|
|
movq mm4, [esi+ebx*4]
|
|
pand mm3, mmx_mask4
|
|
|
|
pand mm4, mmx_mask4
|
|
|
|
psrl mm1, 4
|
|
pand mm7, mmx_mask1
|
|
|
|
movq mm6, [esi+12]
|
|
|
|
psrl mm2, 4
|
|
add esi, 16
|
|
|
|
psrl mm3, 4
|
|
|
|
paddb mm0, mm1
|
|
psrl mm4, 4
|
|
|
|
movq mm1, mm6
|
|
psrl mm7, 1
|
|
|
|
paddb mm2, mm3
|
|
paddb mm0, mm4
|
|
|
|
movq mm3, [esi+edx*4-8]
|
|
paddb mm0, mm2
|
|
|
|
movq mm4, [esi+ebx*4-8]
|
|
paddb mm0, [adj2]
|
|
|
|
pand mm6, mmx_mask4
|
|
|
|
movq [edi],mm0
|
|
pand mm5, mmx_mask4
|
|
|
|
movq mm0, mm7
|
|
pand mm3, mmx_mask4
|
|
|
|
psrl mm6, 4
|
|
pand mm0, mmx_mask1
|
|
|
|
pand mm4, mmx_mask4
|
|
psrl mm5, 4
|
|
|
|
psrl mm0, 1
|
|
paddb mm7, mm6
|
|
|
|
paddb mm7, mm0
|
|
add edi, 16
|
|
|
|
psrl mm3, 4
|
|
|
|
psrl mm4, 4
|
|
paddb mm5, mm3
|
|
|
|
paddb mm7, mm4
|
|
dec ecx
|
|
|
|
paddb mm7, mm5
|
|
paddb mm7, [adj2]
|
|
|
|
movq [edi-8],mm7
|
|
|
|
jnz mmx_light_blur_loop
|
|
mov of, edi
|
|
mov f, esi
|
|
};
|
|
f2=f+w; // update these bitches
|
|
f3=f-w;
|
|
}
|
|
#endif
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_2(f[0]) + DIV_4(f[0]) + DIV_16(f[1]) + DIV_16(f[-1]) + DIV_16(f2[0]) + DIV_16(f3[0]) + adj_tl2;
|
|
f++;
|
|
f2++;
|
|
f3++;
|
|
}
|
|
|
|
// right block
|
|
*of++=DIV_2(f[0])+DIV_8(f[0])+DIV_8(f[-1])+DIV_8(f2[0])+DIV_8(f3[0])+adj_tl1; f++;
|
|
}
|
|
}
|
|
// bottom block
|
|
if (at_bottom)
|
|
{
|
|
unsigned int *f2=f-w;
|
|
int x;
|
|
int adj_tl=0, adj_tl2=0;
|
|
if (roundmode) { adj_tl = 0x03030303; adj_tl2 = 0x04040404; }
|
|
// bottom left
|
|
*of++=DIV_2(f[0])+DIV_4(f[0])+DIV_8(f[1])+DIV_8(f2[0]) + adj_tl; f++; f2++;
|
|
// bottom center
|
|
x=(w-2)/4;
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_2(f[0]) + DIV_8(f[0]) + DIV_8(f[1]) + DIV_8(f[-1]) + DIV_8(f2[0]) + adj_tl2;
|
|
of[1]=DIV_2(f[1]) + DIV_8(f[1]) + DIV_8(f[2]) + DIV_8(f[0]) + DIV_8(f2[1]) + adj_tl2;
|
|
of[2]=DIV_2(f[2]) + DIV_8(f[2]) + DIV_8(f[3]) + DIV_8(f[1]) + DIV_8(f2[2])+adj_tl2;
|
|
of[3]=DIV_2(f[3]) + DIV_8(f[3]) + DIV_8(f[4]) + DIV_8(f[2]) + DIV_8(f2[3])+adj_tl2;
|
|
f+=4;
|
|
f2+=4;
|
|
of+=4;
|
|
}
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_2(f[0]) + DIV_8(f[0]) + DIV_8(f[1]) + DIV_8(f[-1]) + DIV_8(f2[0])+adj_tl2;
|
|
f++;
|
|
f2++;
|
|
}
|
|
// bottom right
|
|
*of++=DIV_2(f[0])+DIV_4(f[0]) + DIV_8(f[-1])+DIV_8(f2[0])+adj_tl; f++; f2++;
|
|
}
|
|
}
|
|
else if (enabled == 3) // more blur
|
|
{
|
|
// top line
|
|
if (at_top) {
|
|
unsigned int *f2=f+w;
|
|
int x;
|
|
int adj_tl=0, adj_tl2=0;
|
|
if (roundmode) { adj_tl = 0x02020202; adj_tl2 = 0x01010101; }
|
|
// top left
|
|
*of++=DIV_2(f[1])+DIV_2(f2[0]) + adj_tl2; f++; f2++;
|
|
// top center
|
|
x=(w-2)/4;
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_4(f[1]) + DIV_4(f[-1]) + DIV_2(f2[0]) + adj_tl;
|
|
of[1]=DIV_4(f[2]) + DIV_4(f[0]) + DIV_2(f2[1]) +adj_tl;
|
|
of[2]=DIV_4(f[3]) + DIV_4(f[1]) + DIV_2(f2[2]) + adj_tl;
|
|
of[3]=DIV_4(f[4]) + DIV_4(f[2]) + DIV_2(f2[3]) + adj_tl;
|
|
f+=4;
|
|
f2+=4;
|
|
of+=4;
|
|
}
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_4(f[1]) + DIV_4(f[-1]) + DIV_2(f2[0])+adj_tl;
|
|
f++;
|
|
f2++;
|
|
}
|
|
// top right
|
|
*of++=DIV_2(f[-1])+DIV_2(f2[0])+adj_tl2; f++; f2++;
|
|
}
|
|
|
|
|
|
// middle block
|
|
{
|
|
int y=outh-at_top-at_bottom;
|
|
int adj_tl1=0,adj_tl2=0;
|
|
unsigned __int64 adj2=0;
|
|
if (roundmode) { adj_tl1=0x02020202; adj_tl2=0x03030303; adj2=0x0303030303030303i64; }
|
|
|
|
while (y--)
|
|
{
|
|
int x;
|
|
unsigned int *f2=f+w;
|
|
unsigned int *f3=f-w;
|
|
|
|
// left edge
|
|
*of++=DIV_2(f[1])+DIV_4(f2[0])+DIV_4(f3[0]) + adj_tl1; f++; f2++; f3++;
|
|
|
|
// middle of line
|
|
#ifdef NO_MMX
|
|
x=(w-2)/4;
|
|
if (roundmode)
|
|
{
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_4(f[1]) + DIV_4(f[-1]) + DIV_4(f2[0]) + DIV_4(f3[0]) + 0x03030303;
|
|
of[1]=DIV_4(f[2]) + DIV_4(f[0]) + DIV_4(f2[1]) + DIV_4(f3[1]) + 0x03030303;
|
|
of[2]=DIV_4(f[3]) + DIV_4(f[1]) + DIV_4(f2[2]) + DIV_4(f3[2]) + 0x03030303;
|
|
of[3]=DIV_4(f[4]) + DIV_4(f[2]) + DIV_4(f2[3]) + DIV_4(f3[3]) + 0x03030303;
|
|
f+=4; f2+=4; f3+=4; of+=4;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_4(f[1]) + DIV_4(f[-1]) + DIV_4(f2[0]) + DIV_4(f3[0]);
|
|
of[1]=DIV_4(f[2]) + DIV_4(f[0]) + DIV_4(f2[1]) + DIV_4(f3[1]);
|
|
of[2]=DIV_4(f[3]) + DIV_4(f[1]) + DIV_4(f2[2]) + DIV_4(f3[2]);
|
|
of[3]=DIV_4(f[4]) + DIV_4(f[2]) + DIV_4(f2[3]) + DIV_4(f3[3]);
|
|
f+=4; f2+=4; f3+=4; of+=4;
|
|
}
|
|
}
|
|
#else
|
|
{
|
|
__asm
|
|
{
|
|
mov ecx, w
|
|
mov edx, ecx
|
|
mov ebx, edx
|
|
neg ebx
|
|
mov esi, f
|
|
mov edi, of
|
|
sub ecx, 2
|
|
shr ecx, 2
|
|
movq mm1, [esi-4]
|
|
align 16
|
|
mmx_heavy_blur_loop:
|
|
movq mm2, [esi+4]
|
|
pxor mm0, mm0
|
|
|
|
movq mm5, mm2
|
|
pxor mm7, mm7
|
|
|
|
movq mm3, [esi+edx*4]
|
|
pand mm1, mmx_mask2
|
|
|
|
movq mm4, [esi+ebx*4]
|
|
pand mm2, mmx_mask2
|
|
|
|
pand mm3, mmx_mask2
|
|
pand mm4, mmx_mask2
|
|
|
|
psrl mm1, 2
|
|
|
|
movq mm6, [esi+12]
|
|
psrl mm2, 2
|
|
|
|
psrl mm3, 2
|
|
|
|
paddb mm0, mm1
|
|
psrl mm4, 2
|
|
|
|
movq mm1, mm6
|
|
|
|
paddb mm2, mm3
|
|
paddb mm0, mm4
|
|
|
|
movq mm3, [esi+edx*4+8]
|
|
paddb mm0, mm2
|
|
|
|
movq mm4, [esi+ebx*4+8]
|
|
paddb mm0, [adj2]
|
|
|
|
pand mm6, mmx_mask2
|
|
|
|
movq [edi],mm0
|
|
pand mm5, mmx_mask2
|
|
|
|
pand mm3, mmx_mask2
|
|
add esi, 16
|
|
|
|
psrl mm6, 2
|
|
pand mm4, mmx_mask2
|
|
|
|
psrl mm5, 2
|
|
paddb mm7, mm6
|
|
|
|
psrl mm3, 2
|
|
add edi, 16
|
|
|
|
psrl mm4, 2
|
|
paddb mm5, mm3
|
|
|
|
paddb mm7, mm4
|
|
|
|
paddb mm7, mm5
|
|
paddb mm7, [adj2]
|
|
|
|
movq [edi-8],mm7
|
|
|
|
dec ecx
|
|
jnz mmx_heavy_blur_loop
|
|
mov of, edi
|
|
mov f, esi
|
|
};
|
|
f2=f+w; // update these bitches
|
|
f3=f-w;
|
|
}
|
|
#endif
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_4(f[1]) + DIV_4(f[-1]) + DIV_4(f2[0]) + DIV_4(f3[0]) + adj_tl2;
|
|
f++;
|
|
f2++;
|
|
f3++;
|
|
}
|
|
|
|
// right block
|
|
*of++=DIV_2(f[-1])+DIV_4(f2[0])+DIV_4(f3[0]) + adj_tl1; f++;
|
|
}
|
|
}
|
|
|
|
// bottom block
|
|
if (at_bottom)
|
|
{
|
|
unsigned int *f2=f-w;
|
|
int x;
|
|
int adj_tl=0, adj_tl2=0;
|
|
if (roundmode) { adj_tl = 0x02020202; adj_tl2 = 0x01010101; }
|
|
// bottom left
|
|
*of++=DIV_2(f[1])+DIV_2(f2[0]) + adj_tl2; f++; f2++;
|
|
// bottom center
|
|
x=(w-2)/4;
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_4(f[1]) + DIV_4(f[-1]) + DIV_2(f2[0])+adj_tl;
|
|
of[1]=DIV_4(f[2]) + DIV_4(f[0]) + DIV_2(f2[1])+adj_tl;
|
|
of[2]=DIV_4(f[3]) + DIV_4(f[1]) + DIV_2(f2[2])+adj_tl;
|
|
of[3]=DIV_4(f[4]) + DIV_4(f[2]) + DIV_2(f2[3])+adj_tl;
|
|
f+=4;
|
|
f2+=4;
|
|
of+=4;
|
|
}
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_4(f[1]) + DIV_4(f[-1]) + DIV_2(f2[0])+adj_tl;
|
|
f++;
|
|
f2++;
|
|
}
|
|
// bottom right
|
|
*of++=DIV_2(f[-1])+DIV_2(f2[0])+adj_tl2; f++; f2++;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
// top line
|
|
if (at_top)
|
|
{
|
|
unsigned int *f2=f+w;
|
|
int x;
|
|
int adj_tl=0, adj_tl2=0;
|
|
if (roundmode) { adj_tl = 0x02020202; adj_tl2 = 0x03030303; }
|
|
// top left
|
|
*of++=DIV_2(f[0])+DIV_4(f[1])+DIV_4(f2[0]) + adj_tl; f++; f2++;
|
|
// top center
|
|
x=(w-2)/4;
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_4(f[0]) + DIV_4(f[1]) + DIV_4(f[-1]) + DIV_4(f2[0]) + adj_tl2;
|
|
of[1]=DIV_4(f[1]) + DIV_4(f[2]) + DIV_4(f[0]) + DIV_4(f2[1]) + adj_tl2;
|
|
of[2]=DIV_4(f[2]) + DIV_4(f[3]) + DIV_4(f[1]) + DIV_4(f2[2]) + adj_tl2;
|
|
of[3]=DIV_4(f[3]) + DIV_4(f[4]) + DIV_4(f[2]) + DIV_4(f2[3]) + adj_tl2;
|
|
f+=4;
|
|
f2+=4;
|
|
of+=4;
|
|
}
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_4(f[0]) + DIV_4(f[1]) + DIV_4(f[-1]) + DIV_4(f2[0]) + adj_tl2;
|
|
f++;
|
|
f2++;
|
|
}
|
|
// top right
|
|
*of++=DIV_2(f[0])+DIV_4(f[-1])+DIV_4(f2[0]) + adj_tl; f++; f2++;
|
|
}
|
|
|
|
|
|
// middle block
|
|
{
|
|
int y=outh-at_top-at_bottom;
|
|
int adj_tl1=0,adj_tl2=0;
|
|
unsigned __int64 adj2=0;
|
|
if (roundmode) { adj_tl1=0x03030303; adj_tl2=0x04040404; adj2=0x0404040404040404i64; }
|
|
while (y--)
|
|
{
|
|
int x;
|
|
unsigned int *f2=f+w;
|
|
unsigned int *f3=f-w;
|
|
|
|
// left edge
|
|
*of++=DIV_4(f[0])+DIV_4(f[1])+DIV_4(f2[0])+DIV_4(f3[0])+adj_tl1; f++; f2++; f3++;
|
|
|
|
// middle of line
|
|
#ifdef NO_MMX
|
|
x=(w-2)/4;
|
|
if (roundmode)
|
|
{
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_2(f[0]) + DIV_8(f[1]) + DIV_8(f[-1]) + DIV_8(f2[0]) + DIV_8(f3[0]) + 0x04040404;
|
|
of[1]=DIV_2(f[1]) + DIV_8(f[2]) + DIV_8(f[0]) + DIV_8(f2[1]) + DIV_8(f3[1]) + 0x04040404;
|
|
of[2]=DIV_2(f[2]) + DIV_8(f[3]) + DIV_8(f[1]) + DIV_8(f2[2]) + DIV_8(f3[2]) + 0x04040404;
|
|
of[3]=DIV_2(f[3]) + DIV_8(f[4]) + DIV_8(f[2]) + DIV_8(f2[3]) + DIV_8(f3[3]) + 0x04040404;
|
|
f+=4; f2+=4; f3+=4; of+=4;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_2(f[0]) + DIV_8(f[1]) + DIV_8(f[-1]) + DIV_8(f2[0]) + DIV_8(f3[0]);
|
|
of[1]=DIV_2(f[1]) + DIV_8(f[2]) + DIV_8(f[0]) + DIV_8(f2[1]) + DIV_8(f3[1]);
|
|
of[2]=DIV_2(f[2]) + DIV_8(f[3]) + DIV_8(f[1]) + DIV_8(f2[2]) + DIV_8(f3[2]);
|
|
of[3]=DIV_2(f[3]) + DIV_8(f[4]) + DIV_8(f[2]) + DIV_8(f2[3]) + DIV_8(f3[3]);
|
|
f+=4; f2+=4; f3+=4; of+=4;
|
|
}
|
|
}
|
|
#else
|
|
{
|
|
__asm
|
|
{
|
|
mov ecx, w
|
|
mov edx, ecx
|
|
mov ebx, edx
|
|
neg ebx
|
|
mov esi, f
|
|
mov edi, of
|
|
sub ecx, 2
|
|
shr ecx, 2
|
|
movq mm1, [esi-4]
|
|
align 16
|
|
mmx_normal_blur_loop:
|
|
movq mm0, [esi]
|
|
|
|
movq mm2, [esi+4]
|
|
pand mm0, mmx_mask1
|
|
|
|
movq mm5, mm2
|
|
|
|
movq mm7, [esi+8]
|
|
pand mm1, mmx_mask3
|
|
|
|
movq mm3, [esi+edx*4]
|
|
pand mm2, mmx_mask3
|
|
|
|
movq mm4, [esi+ebx*4]
|
|
pand mm3, mmx_mask3
|
|
|
|
psrl mm0, 1
|
|
pand mm4, mmx_mask3
|
|
|
|
psrl mm1, 3
|
|
pand mm7, mmx_mask1
|
|
|
|
movq mm6, [esi+12]
|
|
psrl mm2, 3
|
|
|
|
add esi, 16
|
|
psrl mm3, 3
|
|
|
|
paddb mm0, mm1
|
|
psrl mm4, 3
|
|
|
|
movq mm1, mm6
|
|
|
|
paddb mm2, mm3
|
|
paddb mm0, mm4
|
|
|
|
movq mm3, [esi+edx*4-8]
|
|
paddb mm0, mm2
|
|
|
|
movq mm4, [esi+ebx*4-8]
|
|
paddb mm0, [adj2]
|
|
|
|
pand mm6, mmx_mask3
|
|
|
|
movq [edi],mm0
|
|
pand mm5, mmx_mask3
|
|
|
|
psrl mm7, 1
|
|
pand mm3, mmx_mask3
|
|
|
|
psrl mm6, 3
|
|
pand mm4, mmx_mask3
|
|
|
|
psrl mm5, 3
|
|
paddb mm7, mm6
|
|
|
|
add edi, 16
|
|
psrl mm3, 3
|
|
|
|
psrl mm4, 3
|
|
paddb mm5, mm3
|
|
|
|
paddb mm7, mm4
|
|
dec ecx
|
|
|
|
paddb mm7, mm5
|
|
paddb mm7, [adj2]
|
|
|
|
movq [edi-8],mm7
|
|
|
|
jnz mmx_normal_blur_loop
|
|
mov of, edi
|
|
mov f, esi
|
|
};
|
|
f2=f+w; // update these bitches
|
|
f3=f-w;
|
|
}
|
|
#endif
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_2(f[0]) + DIV_8(f[1]) + DIV_8(f[-1]) + DIV_8(f2[0]) + DIV_8(f3[0]) + adj_tl2;
|
|
f++;
|
|
f2++;
|
|
f3++;
|
|
}
|
|
|
|
// right block
|
|
*of++=DIV_4(f[0])+DIV_4(f[-1])+DIV_4(f2[0])+DIV_4(f3[0]) + adj_tl1; f++;
|
|
}
|
|
}
|
|
|
|
// bottom block
|
|
if (at_bottom)
|
|
{
|
|
unsigned int *f2=f-w;
|
|
int adj_tl=0, adj_tl2=0;
|
|
if (roundmode) { adj_tl = 0x02020202; adj_tl2 = 0x03030303; }
|
|
int x;
|
|
// bottom left
|
|
*of++=DIV_2(f[0])+DIV_4(f[1])+DIV_4(f2[0]) + adj_tl; f++; f2++;
|
|
// bottom center
|
|
x=(w-2)/4;
|
|
while (x--)
|
|
{
|
|
of[0]=DIV_4(f[0]) + DIV_4(f[1]) + DIV_4(f[-1]) + DIV_4(f2[0]) + adj_tl2;
|
|
of[1]=DIV_4(f[1]) + DIV_4(f[2]) + DIV_4(f[0]) + DIV_4(f2[1]) + adj_tl2;
|
|
of[2]=DIV_4(f[2]) + DIV_4(f[3]) + DIV_4(f[1]) + DIV_4(f2[2]) + adj_tl2;
|
|
of[3]=DIV_4(f[3]) + DIV_4(f[4]) + DIV_4(f[2]) + DIV_4(f2[3]) + adj_tl2;
|
|
f+=4;
|
|
f2+=4;
|
|
of+=4;
|
|
}
|
|
x=(w-2)&3;
|
|
while (x--)
|
|
{
|
|
*of++=DIV_4(f[0]) + DIV_4(f[1]) + DIV_4(f[-1]) + DIV_4(f2[0]) + adj_tl2;
|
|
f++;
|
|
f2++;
|
|
}
|
|
// bottom right
|
|
*of++=DIV_2(f[0])+DIV_4(f[-1])+DIV_4(f2[0]) + adj_tl; f++; f2++;
|
|
}
|
|
}
|
|
|
|
#ifndef NO_MMX
|
|
__asm emms;
|
|
#endif
|
|
|
|
timingLeave(0);
|
|
}
|
|
|
|
int C_THISCLASS::smp_begin(int max_threads, char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h)
|
|
{
|
|
if (!enabled) return 0;
|
|
return max_threads;
|
|
}
|
|
|
|
|
|
int C_THISCLASS::smp_finish(char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h) // return value is that of render() for fbstuff etc
|
|
{
|
|
return !!enabled;
|
|
}
|
|
|
|
int C_THISCLASS::render(char visdata[2][2][576], int isBeat, int *framebuffer, int *fbout, int w, int h)
|
|
{
|
|
smp_begin(1,visdata,isBeat,framebuffer,fbout,w,h);
|
|
if (isBeat & 0x80000000) return 0;
|
|
|
|
smp_render(0,1,visdata,isBeat,framebuffer,fbout,w,h);
|
|
return smp_finish(visdata,isBeat,framebuffer,fbout,w,h);
|
|
}
|
|
|
|
|
|
C_RBASE *R_Blur(char *desc)
|
|
{
|
|
if (desc) { strcpy(desc,MOD_NAME); return NULL; }
|
|
return (C_RBASE *) new C_THISCLASS();
|
|
}
|
|
|
|
|
|
static C_THISCLASS *g_this;
|
|
|
|
static BOOL CALLBACK g_DlgProc(HWND hwndDlg, UINT uMsg, WPARAM wParam,LPARAM lParam)
|
|
{
|
|
switch (uMsg)
|
|
{
|
|
case WM_INITDIALOG:
|
|
if (g_this->enabled==2) CheckDlgButton(hwndDlg,IDC_RADIO3,BST_CHECKED);
|
|
else if (g_this->enabled==3) CheckDlgButton(hwndDlg,IDC_RADIO4,BST_CHECKED);
|
|
else if (g_this->enabled) CheckDlgButton(hwndDlg,IDC_RADIO2,BST_CHECKED);
|
|
else CheckDlgButton(hwndDlg,IDC_RADIO1,BST_CHECKED);
|
|
if (g_this->roundmode==0) CheckDlgButton(hwndDlg,IDC_ROUNDDOWN,BST_CHECKED);
|
|
else CheckDlgButton(hwndDlg,IDC_ROUNDUP,BST_CHECKED);
|
|
return 1;
|
|
case WM_COMMAND:
|
|
if (LOWORD(wParam) == IDC_RADIO1)
|
|
if (IsDlgButtonChecked(hwndDlg,IDC_RADIO1))
|
|
g_this->enabled=0;
|
|
if (LOWORD(wParam) == IDC_RADIO2)
|
|
if (IsDlgButtonChecked(hwndDlg,IDC_RADIO2))
|
|
g_this->enabled=1;
|
|
if (LOWORD(wParam) == IDC_RADIO3)
|
|
if (IsDlgButtonChecked(hwndDlg,IDC_RADIO3))
|
|
g_this->enabled=2;
|
|
if (LOWORD(wParam) == IDC_RADIO4)
|
|
if (IsDlgButtonChecked(hwndDlg,IDC_RADIO4))
|
|
g_this->enabled=3;
|
|
if (LOWORD(wParam) == IDC_ROUNDUP)
|
|
if (IsDlgButtonChecked(hwndDlg,IDC_ROUNDUP))
|
|
g_this->roundmode=1;
|
|
if (LOWORD(wParam) == IDC_ROUNDDOWN)
|
|
if (IsDlgButtonChecked(hwndDlg,IDC_ROUNDDOWN))
|
|
g_this->roundmode=0;
|
|
return 0;
|
|
return 0;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
|
|
HWND C_THISCLASS::conf(HINSTANCE hInstance, HWND hwndParent)
|
|
{
|
|
g_this = this;
|
|
return WASABI_API_CREATEDIALOG(IDD_CFG_BLUR,hwndParent,g_DlgProc);
|
|
}
|
|
|
|
#else
|
|
C_RBASE *R_Blur(char *desc) { return NULL; }
|
|
#endif |