OrdinaryMagician/vkdoom_m
OrdinaryMagician/vkdoom_m
1
Fork 0
Code Activity

Added SSE versions of bilinear filtering

Browse source
This commit is contained in:
Magnus Norddahl 2016-06-21 06:22:43 +02:00
commit c1b5ba5b90
3 changed files with 609 additions and 289 deletions
Download patch file Download diff file Expand all files Collapse all files

163
src/r_draw_rgba.cpp
View file

@ -58,7 +58,7 @@ extern float rw_lightstep;
extern int wallshade;
CVAR(Bool, r_multithreaded, true, 0)
CVAR(Bool, r_bilinear, false, 0)
CVAR(Bool, r_bilinear, true, 0)
#ifndef NO_SSE
@ -1680,43 +1680,70 @@ public:
xstep = _xstep;
ystep = _ystep;
if (_xbits == 6 && _ybits == 6)
{
// 64x64 is the most common case by far, so special case it.
do
{
uint32_t texdata;
fixed_t xmagnitude = abs((fixed_t)xstep) >> (32 - _xbits - FRACBITS);
fixed_t ymagnitude = abs((fixed_t)ystep) >> (32 - _ybits - FRACBITS);
fixed_t magnitude = xmagnitude + ymagnitude;
spot = ((xfrac >> (32 - 6 - 6))&(63 * 64)) + (yfrac >> (32 - 6));
texdata = source[spot];
if (texdata != 0)
bool magnifying = !r_bilinear || magnitude >> (FRACBITS - 1) == 0;
if (magnifying)
{
if (_xbits == 6 && _ybits == 6)
{
// 64x64 is the most common case by far, so special case it.
do
{
*dest = shade_bgra(texdata, light, shade_constants);
}
dest++;
xfrac += xstep;
yfrac += ystep;
} while (--count);
uint32_t texdata;
spot = ((xfrac >> (32 - 6 - 6))&(63 * 64)) + (yfrac >> (32 - 6));
texdata = source[spot];
*dest = alpha_blend(shade_bgra(texdata, light, shade_constants), *dest);
dest++;
xfrac += xstep;
yfrac += ystep;
} while (--count);
}
else
{
BYTE yshift = 32 - _ybits;
BYTE xshift = yshift - _xbits;
int xmask = ((1 << _xbits) - 1) << _ybits;
do
{
uint32_t texdata;
spot = ((xfrac >> xshift) & xmask) + (yfrac >> yshift);
texdata = source[spot];
*dest = alpha_blend(shade_bgra(texdata, light, shade_constants), *dest);
dest++;
xfrac += xstep;
yfrac += ystep;
} while (--count);
}
}
else
{
BYTE yshift = 32 - _ybits;
BYTE xshift = yshift - _xbits;
int xmask = ((1 << _xbits) - 1) << _ybits;
do
if (_xbits == 6 && _ybits == 6)
{
uint32_t texdata;
spot = ((xfrac >> xshift) & xmask) + (yfrac >> yshift);
texdata = source[spot];
if (texdata != 0)
// 64x64 is the most common case by far, so special case it.
do
{
*dest = shade_bgra(texdata, light, shade_constants);
}
dest++;
xfrac += xstep;
yfrac += ystep;
} while (--count);
*dest++ = alpha_blend(shade_bgra(sample_bilinear(source, xfrac, yfrac, 26, 26), light, shade_constants), *dest);
xfrac += xstep;
yfrac += ystep;
} while (--count);
}
else
{
BYTE yshift = 32 - _ybits;
BYTE xshift = yshift - _xbits;
int xmask = ((1 << _xbits) - 1) << _ybits;
do
{
*dest++ = alpha_blend(shade_bgra(sample_bilinear(source, xfrac, yfrac, 32 - _xbits, 32 - _ybits), light, shade_constants), *dest);
xfrac += xstep;
yfrac += ystep;
} while (--count);
}
}
}
};
@ -2439,6 +2466,8 @@ class Mvlinec1RGBACommand : public DrawerCommand
DWORD _texturefrac;
int _count;
const BYTE * RESTRICT _source;
const BYTE * RESTRICT _source2;
uint32_t _texturefracx;
BYTE * RESTRICT _dest;
int mvlinebits;
int _pitch;
@ -2452,6 +2481,8 @@ public:
_texturefrac = dc_texturefrac;
_count = dc_count;
_source = dc_source;
_source2 = dc_source2;
_texturefracx = dc_texturefracx;
_dest = dc_dest;
mvlinebits = ::mvlinebits;
_pitch = dc_pitch;
@ -2468,6 +2499,8 @@ public:
DWORD fracstep = _iscale * thread->num_cores;
DWORD frac = _texturefrac + _iscale * thread->skipped_by_thread(_dest_y);
const uint32 *source = (const uint32 *)_source;
const uint32 *source2 = (const uint32 *)_source2;
uint32_t texturefracx = _texturefracx;
uint32_t *dest = thread->dest_for_thread(_dest_y, _pitch, (uint32_t*)_dest);
int bits = mvlinebits;
int pitch = _pitch * thread->num_cores;
@ -2475,13 +2508,25 @@ public:
uint32_t light = calc_light_multiplier(_light);
ShadeConstants shade_constants = _shade_constants;
do
if (_source2 == nullptr)
{
uint32_t pix = source[frac >> bits];
*dest = alpha_blend(shade_bgra(pix, light, shade_constants), *dest);
frac += fracstep;
dest += pitch;
} while (--count);
do
{
uint32_t pix = source[frac >> bits];
*dest = alpha_blend(shade_bgra(pix, light, shade_constants), *dest);
frac += fracstep;
dest += pitch;
} while (--count);
}
else
{
do
{
*dest = alpha_blend(shade_bgra(sample_bilinear(source, source2, texturefracx, frac, bits), light, shade_constants), *dest);
frac += fracstep;
dest += pitch;
} while (--count);
}
}
};
@ -2496,6 +2541,8 @@ class Mvlinec4RGBACommand : public DrawerCommand
DWORD vplce[4];
DWORD vince[4];
const uint32 * RESTRICT bufplce[4];
const uint32 * RESTRICT bufplce2[4];
uint32_t buftexturefracx[4];
public:
Mvlinec4RGBACommand()
@ -2511,6 +2558,8 @@ public:
vplce[i] = ::vplce[i];
vince[i] = ::vince[i];
bufplce[i] = (const uint32 *)::bufplce[i];
bufplce2[i] = (const uint32_t *)::bufplce2[i];
buftexturefracx[i] = ::buftexturefracx[i];
}
}
@ -2541,15 +2590,29 @@ public:
local_vince[i] *= thread->num_cores;
}
do
if (bufplce2[0] == nullptr)
{
uint32_t pix;
pix = bufplce[0][(place = local_vplce[0]) >> bits]; dest[0] = alpha_blend(shade_bgra(pix, light0, shade_constants), dest[0]); local_vplce[0] = place + local_vince[0];
pix = bufplce[1][(place = local_vplce[1]) >> bits]; dest[1] = alpha_blend(shade_bgra(pix, light1, shade_constants), dest[1]); local_vplce[1] = place + local_vince[1];
pix = bufplce[2][(place = local_vplce[2]) >> bits]; dest[2] = alpha_blend(shade_bgra(pix, light2, shade_constants), dest[2]); local_vplce[2] = place + local_vince[2];
pix = bufplce[3][(place = local_vplce[3]) >> bits]; dest[3] = alpha_blend(shade_bgra(pix, light3, shade_constants), dest[3]); local_vplce[3] = place + local_vince[3];
dest += pitch;
} while (--count);
do
{
uint32_t pix;
pix = bufplce[0][(place = local_vplce[0]) >> bits]; dest[0] = alpha_blend(shade_bgra(pix, light0, shade_constants), dest[0]); local_vplce[0] = place + local_vince[0];
pix = bufplce[1][(place = local_vplce[1]) >> bits]; dest[1] = alpha_blend(shade_bgra(pix, light1, shade_constants), dest[1]); local_vplce[1] = place + local_vince[1];
pix = bufplce[2][(place = local_vplce[2]) >> bits]; dest[2] = alpha_blend(shade_bgra(pix, light2, shade_constants), dest[2]); local_vplce[2] = place + local_vince[2];
pix = bufplce[3][(place = local_vplce[3]) >> bits]; dest[3] = alpha_blend(shade_bgra(pix, light3, shade_constants), dest[3]); local_vplce[3] = place + local_vince[3];
dest += pitch;
} while (--count);
}
else
{
do
{
dest[0] = alpha_blend(shade_bgra(sample_bilinear(bufplce[0], bufplce2[0], buftexturefracx[0], place = local_vplce[0], bits), light0, shade_constants), dest[0]); local_vplce[0] = place + local_vince[0];
dest[1] = alpha_blend(shade_bgra(sample_bilinear(bufplce[1], bufplce2[1], buftexturefracx[1], place = local_vplce[1], bits), light1, shade_constants), dest[1]); local_vplce[1] = place + local_vince[1];
dest[2] = alpha_blend(shade_bgra(sample_bilinear(bufplce[2], bufplce2[2], buftexturefracx[2], place = local_vplce[2], bits), light2, shade_constants), dest[2]); local_vplce[2] = place + local_vince[2];
dest[3] = alpha_blend(shade_bgra(sample_bilinear(bufplce[3], bufplce2[3], buftexturefracx[3], place = local_vplce[3], bits), light3, shade_constants), dest[3]); local_vplce[3] = place + local_vince[3];
dest += pitch;
} while (--count);
}
}
};
@ -3719,10 +3782,7 @@ void R_DrawSpan_rgba()
#ifdef NO_SSE
DrawerCommandQueue::QueueCommand<DrawSpanRGBACommand>();
#else
if (!r_bilinear)
DrawerCommandQueue::QueueCommand<DrawSpanRGBA_SSE_Command>();
else
DrawerCommandQueue::QueueCommand<DrawSpanRGBACommand>();
DrawerCommandQueue::QueueCommand<DrawSpanRGBA_SSE_Command>();
#endif
}
@ -3776,10 +3836,7 @@ void vlinec4_rgba()
#ifdef NO_SSE
DrawerCommandQueue::QueueCommand<Vlinec4RGBACommand>();
#else
if (!r_bilinear)
DrawerCommandQueue::QueueCommand<Vlinec4RGBA_SSE_Command>();
else
DrawerCommandQueue::QueueCommand<Vlinec4RGBACommand>();
DrawerCommandQueue::QueueCommand<Vlinec4RGBA_SSE_Command>();
#endif
for (int i = 0; i < 4; i++)
vplce[i] += vince[i] * dc_count;