Added php script generating the 32 bit wall drawers

This commit is contained in:
Magnus Norddahl 2017-02-17 05:34:33 +01:00
parent ef22d10756
commit e42f914efa
1 changed files with 250 additions and 0 deletions

View File

@ -0,0 +1,250 @@
#!/usr/bin/php
/*
** Drawer commands for walls
** Copyright (c) 2016 Magnus Norddahl
**
** This software is provided 'as-is', without any express or implied
** warranty. In no event will the authors be held liable for any damages
** arising from the use of this software.
**
** Permission is granted to anyone to use this software for any purpose,
** including commercial applications, and to alter it and redistribute it
** freely, subject to the following restrictions:
**
** 1. The origin of this software must not be misrepresented; you must not
** claim that you wrote the original software. If you use this software
** in a product, an acknowledgment in the product documentation would be
** appreciated but is not required.
** 2. Altered source versions must be plainly marked as such, and must not be
** misrepresented as being the original software.
** 3. This notice may not be removed or altered from any source distribution.
**
*/
/*
Warning: this C++ source file has been auto-generated. Please modify the original php script that generated it.
*/
#pragma once
#include "swrenderer/drawers/r_draw_rgba.h"
#include "swrenderer/viewport/r_walldrawer.h"
namespace swrenderer
{
<?
GenerateDrawerCommand("DrawWall32Command", "opaque");
GenerateDrawerCommand("DrawWallMasked32Command", "masked");
GenerateDrawerCommand("DrawWallAddClamp32Command", "addclamp");
GenerateDrawerCommand("DrawWallSubClamp32Command", "subclamp");
GenerateDrawerCommand("DrawWallRevSubClamp32Command", "revsubclamp");
function GenerateDrawerCommand($className, $blendVariant)
{
?>
class <?=$className?> : public DrawerCommand
{
protected:
WallDrawerArgs args;
bool is_nearest_filter = false;
public:
<?=$className?>(const WallDrawerArgs &drawerargs) : args(drawerargs) { }
void Execute(DrawerThread *thread) override
{
auto shade_constants = args.ColormapConstants();
if (shade_constants.simple_shade)
{
<? LoopShade($blendVariant, true);?>
}
else
{
<? LoopShade($blendVariant, false);?>
}
}
};
<?
}
function LoopShade($blendVariant, $isSimpleShade)
{ ?>
if (is_nearest_filter)
{
<? Loop($blendVariant, $isSimpleShade, true);?>
}
else
{
<? Loop($blendVariant, $isSimpleShade, false);?>
}
<? }
function Loop($blendVariant, $isSimpleShade, $isNearestFilter)
{ ?>
int textureheight = args.TextureHeight();
// Shade constants
int light = args.Light();
__m128i mlight = _mm_set_epi16(light, light, light, 256, light, light, light, 256);
__m128i inv_light = _mm_set1_epi16(256 - light, 256 - light, 256 - light, 0, 256 - light, 256 - light, 256 - light, 0);
<? if ($isSimpleShade == false)
{ ?>
__m128i inv_desaturate = _mm_set1_epi16(256 - shade_constants.desaturate, 256 - shade_constants.desaturate, 256 - shade_constants.desaturate, 0, 256 - shade_constants.desaturate, 256 - shade_constants.desaturate, 256 - shade_constants.desaturate, 0);
__m128i shade_fade = _mm_set_epi16(shade_constants.fade_red, shade_constants.fade_green, shade_constants.fade_blue, shade_constants.fade_alpha, shade_constants.fade_red, shade_constants.fade_green, shade_constants.fade_blue, shade_constants.fade_alpha);
__m128i shade_light = _mm_set_epi16(shade_constants.light_red, shade_constants.light_green, shade_constants.light_blue, shade_constants.light_alpha, shade_constants.light_red, shade_constants.light_green, shade_constants.light_blue, shade_constants.light_alpha);
<? } ?>
int count = args.Count();
for (int index = 0; index < count; index++)
{
int offset = index * pitch * 4;
__m128i bgcolor = _mm_unpacklo_epi8(_mm_cvtsi32(dest[offset]), _mm_setzero_si128());
// Sample
<? Sample($isNearestFilter);?>
// Shade
<? Shade($isSimpleShade);?>
// Blend
<? Blend($blendVariant);?>
dest[offset] = _mm_cvtsi32(outcolor);
frac += fracstep;
}
<? }
function Sample($isNearestFilter)
{
if ($isNearestFilter == true)
{ ?>
int sample_index = ((frac >> FRACBITS) * textureheight) >> FRACBITS;
unsigned int ifgcolor = source[sample_index * 4];
__m128i fgcolor = _mm_unpacklo_epi16(_mm_unpacklo_epi8(_mm_cvtsi32(ifgcolor), _mm_setzero_si128()), _mm_setzero_si128());
<? }
else
{ ?>
unsigned int frac_y0 = (texturefracy >> FRACBITS) * textureheight;
unsigned int frac_y1 = ((texturefracy + one) >> FRACBITS) * textureheight;
unsigned int y0 = frac_y0 >> FRACBITS;
unsigned int y1 = frac_y1 >> FRACBITS;
unsigned int p00 = source[y0 * 4];
unsigned int p01 = source[y1 * 4];
unsigned int p10 = source2[y0 * 4];
unsigned int p11 = source2[y1 * 4];
unsigned int inv_b = texturefracx;
unsigned int a = (frac_y1 >> (FRACBITS - 4)) & 15;
unsigned int inv_a = 16 - a;
unsigned int b = 16 - inv_b;
unsigned int sred = (RPART(p00) * (a * b) + RPART(p01) * (inv_a * b) + RPART(p10) * (a * inv_b) + RPART(p11) * (inv_a * inv_b) + 127) >> 8;
unsigned int sgreen = (GPART(p00) * (a * b) + GPART(p01) * (inv_a * b) + GPART(p10) * (a * inv_b) + GPART(p11) * (inv_a * inv_b) + 127) >> 8;
unsigned int sblue = (BPART(p00) * (a * b) + BPART(p01) * (inv_a * b) + BPART(p10) * (a * inv_b) + BPART(p11) * (inv_a * inv_b) + 127) >> 8;
unsigned int salpha = (APART(p00) * (a * b) + APART(p01) * (inv_a * b) + APART(p10) * (a * inv_b) + APART(p11) * (inv_a * inv_b) + 127) >> 8;
unsigned int ifgcolor = (salpha << 24) | (sred << 16) | (sgreen << 8) | sblue;
__m128i fgcolor = _mm_unpacklo_epi16(_mm_unpacklo_epi8(_mm_cvtsi32(ifgcolor), _mm_setzero_si128()), _mm_setzero_si128());
<? }
}
function Shade($isSimpleShade)
{
if ($isSimpleShade == true)
{ ?>
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
<? }
else
{ ?>
int blue = BPART(ifgcolor);
int green = GPART(ifgcolor);
int red = RPART(ifgcolor);
__m128i intensity = _mm_set1_epi16(((red * 77 + green * 143 + blue * 37) >> 8) * desaturate);
fgcolor = _mm_srli_epi16(_mm_add_epi16(_mm_mullo_epi16(fgcolor, inv_desaturate), intensity), 8);
fgcolor = _mm_mullo_epi16(fgcolor, _mm_set1_epi16(light));
fgcolor = _mm_srli_epi16(_mm_add_epi16(_mm_mullo_epi16(shade_fade, inv_light), fgcolor), 8);
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
<? }
}
function Blend($blendVariant)
{
if ($blendVariant == "opaque")
{ ?>
__m128 outcolor = fgcolor;
outcolor = _mm_packus_epi16(outcolor, _mm_setzero128());
<? }
else if ($blendVariant == "masked")
{
CalcAlpha(); ?>
fgcolor = _mm_mullo_epi16(fgcolor, alpha);
bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha);
__m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8);
outcolor = _mm_packus_epi16(outcolor, _mm_setzero128());
outcolor = _mm_or_si128(outcolor, _mm_set_epi32(0xff000000));
<? }
else if ($blendVariant == "add" || $blendVariant == "addclamp")
{
CalcBlendColor(); ?>
__m128i out_lo = _mm_srai_epi16(_mm_add_epi32(fg_lo, bg_lo), 8);
__m128i out_hi = _mm_srai_epi16(_mm_add_epi32(fg_hi, bg_hi), 8);
__m128i outcolor = _mm_packs_epi32(fg_lo, fg_hi);
outcolor = _mm_packus_epi16(outcolor, _mm_setzero128());
outcolor = _mm_or_si128(outcolor, _mm_set_epi32(0xff000000));
<? }
else if ($blendVariant == "subclamp")
{
CalcBlendColor(); ?>
__m128i out_lo = _mm_srai_epi16(_mm_sub_epi32(fg_lo, bg_lo), 8);
__m128i out_hi = _mm_srai_epi16(_mm_sub_epi32(fg_hi, bg_hi), 8);
__m128i outcolor = _mm_packs_epi32(fg_lo, fg_hi);
outcolor = _mm_packus_epi16(outcolor, _mm_setzero128());
outcolor = _mm_or_si128(outcolor, _mm_set_epi32(0xff000000));
<? }
else if ($blendVariant == "revsubclamp")
{
CalcBlendColor(); ?>
__m128i out_lo = _mm_srai_epi16(_mm_sub_epi32(bg_lo, fg_lo), 8);
__m128i out_hi = _mm_srai_epi16(_mm_sub_epi32(bg_hi, fg_hi), 8);
__m128i outcolor = _mm_packs_epi32(fg_lo, fg_hi);
outcolor = _mm_packus_epi16(outcolor, _mm_setzero128());
outcolor = _mm_or_si128(outcolor, _mm_set_epi32(0xff000000));
<? }
}
function CalcAlpha()
{ ?>
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
<? }
function CalcBlendColor()
{
CalcAlpha();?>
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
bgcolor = _mm_mullo_epi16(bgcolor, bgalpha);
__m128i fg_lo = _mm_unpacklo_epi16(fgcolor, _mm_setzero_si128());
__m128i bg_lo = _mm_unpacklo_epi16(bgcolor, _mm_setzero_si128());
__m128i fg_hi = _mm_unpackhi_epi16(fgcolor, _mm_setzero_si128());
__m128i bg_hi = _mm_unpackhi_epi16(bgcolor, _mm_setzero_si128());
<? }
?>
}