gzdoom-gles/src/r_drawt.cpp
Randy Heit dda5ddd3c2 - Ported vlinetallasm4 to AMD64 assembly. Even with the increased number of
registers AMD64 provides, this routine still needs to be written as self-
  modifying code for maximum performance. The additional registers do allow
  for further optimization over the x86 version by allowing all four pixels
  to be in flight at the same time. The end result is that AMD64 ASM is about
  2.18 times faster than AMD64 C and about 1.06 times faster than x86 ASM.
  (For further comparison, AMD64 C and x86 C are practically the same for
  this function.) Should I port any more assembly to AMD64, mvlineasm4 is the
  most likely candidate, but it's not used enough at this point to bother.
  Also, this may or may not work with Linux at the moment, since it doesn't
  have the eh_handler metadata. Win64 is easier, since I just need to
  structure the function prologue and epilogue properly and use some
  assembler directives/macros to automatically generate the metadata. And
  that brings up another point: You need YASM to assemble the AMD64 code,
  because NASM doesn't support the Win64 metadata directives.
- Added an SSE version of DoBlending. This is strictly C intrinsics.
  VC++ still throws around unneccessary register moves. GCC seems to be
  pretty close to optimal, requiring only about 2 cycles/color. They're
  both faster than my hand-written MMX routine, so I don't need to feel
  bad about not hand-optimizing this for x64 builds.
- Removed an extra instruction from DoBlending_MMX, transposed two
  instructions, and unrolled it once, shaving off about 80 cycles from the
  time required to blend 256 palette entries. Why? Because I tried writing
  a C version of the routine using compiler intrinsics and was appalled by
  all the extra movq's VC++ added to the code. GCC was better, but still
  generated extra instructions. I only wanted a C version because I can't
  use inline assembly with VC++'s x64 compiler, and x64 assembly is a bit
  of a pain. (It's a pain because Linux and Windows have different calling
  conventions, and you need to maintain extra metadata for functions.) So,
  the assembly version stays and the C version stays out.
- Removed all the pixel doubling r_detail modes, since the one platform they
  were intended to assist (486) actually sees very little benefit from them.
- Rewrote CheckMMX in C and renamed it to CheckCPU.
- Fixed: CPUID function 0x80000005 is specified to return detailed L1 cache
  only for AMD processors, so we must not use it on other architectures, or
  we end up overwriting the L1 cache line size with 0 or some other number
  we don't actually understand.


SVN r1134 (trunk)
2008-08-09 03:13:43 +00:00

1214 lines
28 KiB
C++

/*
** r_drawt.cpp
** Faster column drawers for modern processors
**
**---------------------------------------------------------------------------
** Copyright 1998-2006 Randy Heit
** All rights reserved.
**
** Redistribution and use in source and binary forms, with or without
** modification, are permitted provided that the following conditions
** are met:
**
** 1. Redistributions of source code must retain the above copyright
** notice, this list of conditions and the following disclaimer.
** 2. 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.
** 3. The name of the author may not be used to endorse or promote products
** derived from this software without specific prior written permission.
**
** THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
**---------------------------------------------------------------------------
**
** These functions stretch columns into a temporary buffer and then
** map them to the screen. On modern machines, this is faster than drawing
** them directly to the screen.
**
** Will I be able to even understand any of this if I come back to it later?
** Let's hope so. :-)
*/
#include "templates.h"
#include "doomtype.h"
#include "doomdef.h"
#include "r_defs.h"
#include "r_draw.h"
#include "r_main.h"
#include "r_things.h"
#include "v_video.h"
// I should have commented this stuff better.
//
// dc_temp is the buffer R_DrawColumnHoriz writes into.
// dc_tspans points into it.
// dc_ctspan points into dc_tspans.
// But what is horizspan, and what is its relation with dc_ctspan?
BYTE dc_temp[MAXHEIGHT*4];
unsigned int dc_tspans[4][MAXHEIGHT];
unsigned int *dc_ctspan[4];
unsigned int *horizspan[4];
#ifdef X86_ASM
extern "C" void R_SetupShadedCol();
extern "C" void R_SetupAddCol();
extern "C" void R_SetupAddClampCol();
#endif
#ifndef X86_ASM
// Copies one span at hx to the screen at sx.
void rt_copy1col_c (int hx, int sx, int yl, int yh)
{
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4 + hx];
pitch = dc_pitch;
if (count & 1) {
*dest = *source;
source += 4;
dest += pitch;
}
if (count & 2) {
dest[0] = source[0];
dest[pitch] = source[4];
source += 8;
dest += pitch*2;
}
if (!(count >>= 2))
return;
do {
dest[0] = source[0];
dest[pitch] = source[4];
dest[pitch*2] = source[8];
dest[pitch*3] = source[12];
source += 16;
dest += pitch*4;
} while (--count);
}
// Copies all four spans to the screen starting at sx.
void STACK_ARGS rt_copy4cols_c (int sx, int yl, int yh)
{
int *source;
int *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
dest = (int *)(ylookup[yl] + sx + dc_destorg);
source = (int *)(&dc_temp[yl*4]);
pitch = dc_pitch/sizeof(int);
if (count & 1) {
*dest = *source;
source += 4/sizeof(int);
dest += pitch;
}
if (!(count >>= 1))
return;
do {
dest[0] = source[0];
dest[pitch] = source[4/sizeof(int)];
source += 8/sizeof(int);
dest += pitch*2;
} while (--count);
}
// Maps one span at hx to the screen at sx.
void rt_map1col_c (int hx, int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
colormap = dc_colormap;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4 + hx];
pitch = dc_pitch;
if (count & 1) {
*dest = colormap[*source];
source += 4;
dest += pitch;
}
if (!(count >>= 1))
return;
do {
dest[0] = colormap[source[0]];
dest[pitch] = colormap[source[4]];
source += 8;
dest += pitch*2;
} while (--count);
}
// Maps all four spans to the screen starting at sx.
void STACK_ARGS rt_map4cols_c (int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
colormap = dc_colormap;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4];
pitch = dc_pitch;
if (count & 1) {
dest[0] = colormap[source[0]];
dest[1] = colormap[source[1]];
dest[2] = colormap[source[2]];
dest[3] = colormap[source[3]];
source += 4;
dest += pitch;
}
if (!(count >>= 1))
return;
do {
dest[0] = colormap[source[0]];
dest[1] = colormap[source[1]];
dest[2] = colormap[source[2]];
dest[3] = colormap[source[3]];
dest[pitch] = colormap[source[4]];
dest[pitch+1] = colormap[source[5]];
dest[pitch+2] = colormap[source[6]];
dest[pitch+3] = colormap[source[7]];
source += 8;
dest += pitch*2;
} while (--count);
}
#endif
void rt_Translate1col(const BYTE *translation, int hx, int yl, int yh)
{
int count = yh - yl + 1;
BYTE *source = &dc_temp[yl*4 + hx];
// Things we do to hit the compiler's optimizer with a clue bat:
// 1. Parallelism is explicitly spelled out by using a separate
// C instruction for each assembly instruction. GCC lets me
// have four temporaries, but VC++ spills to the stack with
// more than two. Two is probably optimal, anyway.
// 2. The results of the translation lookups are explicitly
// stored in byte-sized variables. This causes the VC++ code
// to use byte mov instructions in most cases; for apparently
// random reasons, it will use movzx for some places. GCC
// ignores this and uses movzx always.
// Do 8 rows at a time.
for (int count8 = count >> 3; count8; --count8)
{
int c0, c1;
BYTE b0, b1;
c0 = source[0]; c1 = source[4];
b0 = translation[c0]; b1 = translation[c1];
source[0] = b0; source[4] = b1;
c0 = source[8]; c1 = source[12];
b0 = translation[c0]; b1 = translation[c1];
source[8] = b0; source[12] = b1;
c0 = source[16]; c1 = source[20];
b0 = translation[c0]; b1 = translation[c1];
source[16] = b0; source[20] = b1;
c0 = source[24]; c1 = source[28];
b0 = translation[c0]; b1 = translation[c1];
source[24] = b0; source[28] = b1;
source += 32;
}
// Finish by doing 1 row at a time.
for (count &= 7; count; --count, source += 4)
{
source[0] = translation[source[0]];
}
}
void rt_Translate4cols(const BYTE *translation, int yl, int yh)
{
int count = yh - yl + 1;
BYTE *source = &dc_temp[yl*4];
int c0, c1;
BYTE b0, b1;
// Do 2 rows at a time.
for (int count8 = count >> 1; count8; --count8)
{
c0 = source[0]; c1 = source[1];
b0 = translation[c0]; b1 = translation[c1];
source[0] = b0; source[1] = b1;
c0 = source[2]; c1 = source[3];
b0 = translation[c0]; b1 = translation[c1];
source[2] = b0; source[3] = b1;
c0 = source[4]; c1 = source[5];
b0 = translation[c0]; b1 = translation[c1];
source[4] = b0; source[5] = b1;
c0 = source[6]; c1 = source[7];
b0 = translation[c0]; b1 = translation[c1];
source[6] = b0; source[7] = b1;
source += 8;
}
// Do the final row if count was odd.
if (count & 1)
{
c0 = source[0]; c1 = source[1];
b0 = translation[c0]; b1 = translation[c1];
source[0] = b0; source[1] = b1;
c0 = source[2]; c1 = source[3];
b0 = translation[c0]; b1 = translation[c1];
source[2] = b0; source[3] = b1;
}
}
// Translates one span at hx to the screen at sx.
void rt_tlate1col (int hx, int sx, int yl, int yh)
{
rt_Translate1col(dc_translation, hx, yl, yh);
rt_map1col(hx, sx, yl, yh);
}
// Translates all four spans to the screen starting at sx.
void STACK_ARGS rt_tlate4cols (int sx, int yl, int yh)
{
rt_Translate4cols(dc_translation, yl, yh);
rt_map4cols(sx, yl, yh);
}
// Adds one span at hx to the screen at sx without clamping.
void rt_add1col (int hx, int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4 + hx];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD fg = colormap[*source];
DWORD bg = *dest;
fg = fg2rgb[fg];
bg = bg2rgb[bg];
fg = (fg+bg) | 0x1f07c1f;
*dest = RGB32k[0][0][fg & (fg>>15)];
source += 4;
dest += pitch;
} while (--count);
}
// Adds all four spans to the screen starting at sx without clamping.
void STACK_ARGS rt_add4cols_c (int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD fg = colormap[source[0]];
DWORD bg = dest[0];
fg = fg2rgb[fg];
bg = bg2rgb[bg];
fg = (fg+bg) | 0x1f07c1f;
dest[0] = RGB32k[0][0][fg & (fg>>15)];
fg = colormap[source[1]];
bg = dest[1];
fg = fg2rgb[fg];
bg = bg2rgb[bg];
fg = (fg+bg) | 0x1f07c1f;
dest[1] = RGB32k[0][0][fg & (fg>>15)];
fg = colormap[source[2]];
bg = dest[2];
fg = fg2rgb[fg];
bg = bg2rgb[bg];
fg = (fg+bg) | 0x1f07c1f;
dest[2] = RGB32k[0][0][fg & (fg>>15)];
fg = colormap[source[3]];
bg = dest[3];
fg = fg2rgb[fg];
bg = bg2rgb[bg];
fg = (fg+bg) | 0x1f07c1f;
dest[3] = RGB32k[0][0][fg & (fg>>15)];
source += 4;
dest += pitch;
} while (--count);
}
// Translates and adds one span at hx to the screen at sx without clamping.
void rt_tlateadd1col (int hx, int sx, int yl, int yh)
{
rt_Translate1col(dc_translation, hx, yl, yh);
rt_add1col(hx, sx, yl, yh);
}
// Translates and adds all four spans to the screen starting at sx without clamping.
void STACK_ARGS rt_tlateadd4cols (int sx, int yl, int yh)
{
rt_Translate4cols(dc_translation, yl, yh);
rt_add4cols(sx, yl, yh);
}
// Shades one span at hx to the screen at sx.
void rt_shaded1col (int hx, int sx, int yl, int yh)
{
DWORD *fgstart;
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
fgstart = &Col2RGB8[0][dc_color];
colormap = dc_colormap;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4 + hx];
pitch = dc_pitch;
do {
DWORD val = colormap[*source];
DWORD fg = fgstart[val<<8];
val = (Col2RGB8[64-val][*dest] + fg) | 0x1f07c1f;
*dest = RGB32k[0][0][val & (val>>15)];
source += 4;
dest += pitch;
} while (--count);
}
// Shades all four spans to the screen starting at sx.
void STACK_ARGS rt_shaded4cols_c (int sx, int yl, int yh)
{
DWORD *fgstart;
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
fgstart = &Col2RGB8[0][dc_color];
colormap = dc_colormap;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4];
pitch = dc_pitch;
do {
DWORD val;
val = colormap[source[0]];
val = (Col2RGB8[64-val][dest[0]] + fgstart[val<<8]) | 0x1f07c1f;
dest[0] = RGB32k[0][0][val & (val>>15)];
val = colormap[source[1]];
val = (Col2RGB8[64-val][dest[1]] + fgstart[val<<8]) | 0x1f07c1f;
dest[1] = RGB32k[0][0][val & (val>>15)];
val = colormap[source[2]];
val = (Col2RGB8[64-val][dest[2]] + fgstart[val<<8]) | 0x1f07c1f;
dest[2] = RGB32k[0][0][val & (val>>15)];
val = colormap[source[3]];
val = (Col2RGB8[64-val][dest[3]] + fgstart[val<<8]) | 0x1f07c1f;
dest[3] = RGB32k[0][0][val & (val>>15)];
source += 4;
dest += pitch;
} while (--count);
}
// Adds one span at hx to the screen at sx with clamping.
void rt_addclamp1col (int hx, int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4 + hx];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD a = fg2rgb[colormap[*source]] + bg2rgb[*dest];
DWORD b = a;
a |= 0x01f07c1f;
b &= 0x40100400;
a &= 0x3fffffff;
b = b - (b >> 5);
a |= b;
*dest = RGB32k[0][0][(a>>15) & a];
source += 4;
dest += pitch;
} while (--count);
}
// Adds all four spans to the screen starting at sx with clamping.
void STACK_ARGS rt_addclamp4cols_c (int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD a = fg2rgb[colormap[source[0]]] + bg2rgb[dest[0]];
DWORD b = a;
a |= 0x01f07c1f;
b &= 0x40100400;
a &= 0x3fffffff;
b = b - (b >> 5);
a |= b;
dest[0] = RGB32k[0][0][(a>>15) & a];
a = fg2rgb[colormap[source[1]]] + bg2rgb[dest[1]];
b = a;
a |= 0x01f07c1f;
b &= 0x40100400;
a &= 0x3fffffff;
b = b - (b >> 5);
a |= b;
dest[1] = RGB32k[0][0][(a>>15) & a];
a = fg2rgb[colormap[source[2]]] + bg2rgb[dest[2]];
b = a;
a |= 0x01f07c1f;
b &= 0x40100400;
a &= 0x3fffffff;
b = b - (b >> 5);
a |= b;
dest[2] = RGB32k[0][0][(a>>15) & a];
a = fg2rgb[colormap[source[3]]] + bg2rgb[dest[3]];
b = a;
a |= 0x01f07c1f;
b &= 0x40100400;
a &= 0x3fffffff;
b = b - (b >> 5);
a |= b;
dest[3] = RGB32k[0][0][(a>>15) & a];
source += 4;
dest += pitch;
} while (--count);
}
// Translates and adds one span at hx to the screen at sx with clamping.
void rt_tlateaddclamp1col (int hx, int sx, int yl, int yh)
{
rt_Translate1col(dc_translation, hx, yl, yh);
rt_addclamp1col(hx, sx, yl, yh);
}
// Translates and adds all four spans to the screen starting at sx with clamping.
void STACK_ARGS rt_tlateaddclamp4cols (int sx, int yl, int yh)
{
rt_Translate4cols(dc_translation, yl, yh);
rt_addclamp4cols(sx, yl, yh);
}
// Subtracts one span at hx to the screen at sx with clamping.
void rt_subclamp1col (int hx, int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4 + hx];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD a = (fg2rgb[colormap[*source]] | 0x40100400) - bg2rgb[*dest];
DWORD b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
*dest = RGB32k[0][0][(a>>15) & a];
source += 4;
dest += pitch;
} while (--count);
}
// Subtracts all four spans to the screen starting at sx with clamping.
void STACK_ARGS rt_subclamp4cols (int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD a = (fg2rgb[colormap[source[0]]] | 0x40100400) - bg2rgb[dest[0]];
DWORD b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[0] = RGB32k[0][0][(a>>15) & a];
a = (fg2rgb[colormap[source[1]]] | 0x40100400) - bg2rgb[dest[1]];
b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[1] = RGB32k[0][0][(a>>15) & a];
a = (fg2rgb[colormap[source[2]]] | 0x40100400) - bg2rgb[dest[2]];
b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[2] = RGB32k[0][0][(a>>15) & a];
a = (fg2rgb[colormap[source[3]]] | 0x40100400) - bg2rgb[dest[3]];
b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[3] = RGB32k[0][0][(a>>15) & a];
source += 4;
dest += pitch;
} while (--count);
}
// Translates and subtracts one span at hx to the screen at sx with clamping.
void rt_tlatesubclamp1col (int hx, int sx, int yl, int yh)
{
rt_Translate1col(dc_translation, hx, yl, yh);
rt_subclamp1col(hx, sx, yl, yh);
}
// Translates and subtracts all four spans to the screen starting at sx with clamping.
void STACK_ARGS rt_tlatesubclamp4cols (int sx, int yl, int yh)
{
rt_Translate4cols(dc_translation, yl, yh);
rt_subclamp4cols(sx, yl, yh);
}
// Subtracts one span at hx from the screen at sx with clamping.
void rt_revsubclamp1col (int hx, int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4 + hx];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD a = (bg2rgb[*dest] | 0x40100400) - fg2rgb[colormap[*source]];
DWORD b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
*dest = RGB32k[0][0][(a>>15) & a];
source += 4;
dest += pitch;
} while (--count);
}
// Subtracts all four spans from the screen starting at sx with clamping.
void STACK_ARGS rt_revsubclamp4cols (int sx, int yl, int yh)
{
BYTE *colormap;
BYTE *source;
BYTE *dest;
int count;
int pitch;
count = yh-yl;
if (count < 0)
return;
count++;
DWORD *fg2rgb = dc_srcblend;
DWORD *bg2rgb = dc_destblend;
dest = ylookup[yl] + sx + dc_destorg;
source = &dc_temp[yl*4];
pitch = dc_pitch;
colormap = dc_colormap;
do {
DWORD a = (bg2rgb[dest[0]] | 0x40100400) - fg2rgb[colormap[source[0]]];
DWORD b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[0] = RGB32k[0][0][(a>>15) & a];
a = (bg2rgb[dest[1]] | 0x40100400) - fg2rgb[colormap[source[1]]];
b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[1] = RGB32k[0][0][(a>>15) & a];
a = (bg2rgb[dest[2]] | 0x40100400) - fg2rgb[colormap[source[2]]];
b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[2] = RGB32k[0][0][(a>>15) & a];
a = (bg2rgb[dest[3]] | 0x40100400) - fg2rgb[colormap[source[3]]];
b = a;
b &= 0x40100400;
b = b - (b >> 5);
a &= b;
a |= 0x01f07c1f;
dest[3] = RGB32k[0][0][(a>>15) & a];
source += 4;
dest += pitch;
} while (--count);
}
// Translates and subtracts one span at hx from the screen at sx with clamping.
void rt_tlaterevsubclamp1col (int hx, int sx, int yl, int yh)
{
rt_Translate1col(dc_translation, hx, yl, yh);
rt_revsubclamp1col(hx, sx, yl, yh);
}
// Translates and subtracts all four spans from the screen starting at sx with clamping.
void STACK_ARGS rt_tlaterevsubclamp4cols (int sx, int yl, int yh)
{
rt_Translate4cols(dc_translation, yl, yh);
rt_revsubclamp4cols(sx, yl, yh);
}
// Copies all spans in all four columns to the screen starting at sx.
// sx should be dword-aligned.
void rt_draw4cols (int sx)
{
int x, bad;
unsigned int maxtop, minbot, minnexttop;
// Place a dummy "span" in each column. These don't get
// drawn. They're just here to avoid special cases in the
// max/min calculations below.
for (x = 0; x < 4; ++x)
{
dc_ctspan[x][0] = screen->GetHeight()+1;
dc_ctspan[x][1] = screen->GetHeight();
}
#ifdef X86_ASM
// Setup assembly routines for changed colormaps or other parameters.
if (hcolfunc_post4 == rt_shaded4cols)
{
R_SetupShadedCol();
}
else if (hcolfunc_post4 == rt_addclamp4cols || hcolfunc_post4 == rt_tlateaddclamp4cols)
{
R_SetupAddClampCol();
}
else if (hcolfunc_post4 == rt_add4cols || hcolfunc_post4 == rt_tlateadd4cols)
{
R_SetupAddCol();
}
#endif
for (;;)
{
// If a column is out of spans, mark it as such
bad = 0;
minnexttop = 0xffffffff;
for (x = 0; x < 4; ++x)
{
if (horizspan[x] >= dc_ctspan[x])
{
bad |= 1 << x;
}
else if ((horizspan[x]+2)[0] < minnexttop)
{
minnexttop = (horizspan[x]+2)[0];
}
}
// Once all columns are out of spans, we're done
if (bad == 15)
{
return;
}
// Find the largest shared area for the spans in each column
maxtop = MAX (MAX (horizspan[0][0], horizspan[1][0]),
MAX (horizspan[2][0], horizspan[3][0]));
minbot = MIN (MIN (horizspan[0][1], horizspan[1][1]),
MIN (horizspan[2][1], horizspan[3][1]));
// If there is no shared area with these spans, draw each span
// individually and advance to the next spans until we reach a shared area.
// However, only draw spans down to the highest span in the next set of
// spans. If we allow the entire height of a span to be drawn, it could
// prevent any more shared areas from being drawn in these four columns.
//
// Example: Suppose we have the following arrangement:
// A CD
// A CD
// B D
// B D
// aB D
// aBcD
// aBcD
// aBc
//
// If we draw the entire height of the spans, we end up drawing this first:
// A CD
// A CD
// B D
// B D
// B D
// B D
// B D
// B D
// B
//
// This leaves only the "a" and "c" columns to be drawn, and they are not
// part of a shared area, but if we can include B and D with them, we can
// get a shared area. So we cut off everything in the first set just
// above the "a" column and end up drawing this first:
// A CD
// A CD
// B D
// B D
//
// Then the next time through, we have the following arrangement with an
// easily shared area to draw:
// aB D
// aBcD
// aBcD
// aBc
if (bad != 0 || maxtop > minbot)
{
int drawcount = 0;
for (x = 0; x < 4; ++x)
{
if (!(bad & 1))
{
if (horizspan[x][1] < minnexttop)
{
hcolfunc_post1 (x, sx+x, horizspan[x][0], horizspan[x][1]);
horizspan[x] += 2;
drawcount++;
}
else if (minnexttop > horizspan[x][0])
{
hcolfunc_post1 (x, sx+x, horizspan[x][0], minnexttop-1);
horizspan[x][0] = minnexttop;
drawcount++;
}
}
bad >>= 1;
}
// Drawcount *should* always be non-zero. The reality is that some situations
// can make this not true. Unfortunately, I'm not sure what those situations are.
if (drawcount == 0)
{
return;
}
continue;
}
// Draw any span fragments above the shared area.
for (x = 0; x < 4; ++x)
{
if (maxtop > horizspan[x][0])
{
hcolfunc_post1 (x, sx+x, horizspan[x][0], maxtop-1);
}
}
// Draw the shared area.
hcolfunc_post4 (sx, maxtop, minbot);
// For each column, if part of the span is past the shared area,
// set its top to just below the shared area. Otherwise, advance
// to the next span in that column.
for (x = 0; x < 4; ++x)
{
if (minbot < horizspan[x][1])
{
horizspan[x][0] = minbot+1;
}
else
{
horizspan[x] += 2;
}
}
}
}
// Before each pass through a rendering loop that uses these routines,
// call this function to set up the span pointers.
void rt_initcols (void)
{
int y;
for (y = 3; y >= 0; y--)
horizspan[y] = dc_ctspan[y] = &dc_tspans[y][0];
}
// Stretches a column into a temporary buffer which is later
// drawn to the screen along with up to three other columns.
void R_DrawColumnHorizP_C (void)
{
int count = dc_count;
BYTE *dest;
fixed_t fracstep;
fixed_t frac;
if (count <= 0)
return;
{
int x = dc_x & 3;
unsigned int **span;
span = &dc_ctspan[x];
(*span)[0] = dc_yl;
(*span)[1] = dc_yh;
*span += 2;
dest = &dc_temp[x + 4*dc_yl];
}
fracstep = dc_iscale;
frac = dc_texturefrac;
{
const BYTE *source = dc_source;
if (count & 1) {
*dest = source[frac>>FRACBITS];
dest += 4;
frac += fracstep;
}
if (count & 2) {
dest[0] = source[frac>>FRACBITS];
frac += fracstep;
dest[4] = source[frac>>FRACBITS];
frac += fracstep;
dest += 8;
}
if (count & 4) {
dest[0] = source[frac>>FRACBITS];
frac += fracstep;
dest[4] = source[frac>>FRACBITS];
frac += fracstep;
dest[8] = source[frac>>FRACBITS];
frac += fracstep;
dest[12] = source[frac>>FRACBITS];
frac += fracstep;
dest += 16;
}
count >>= 3;
if (!count) return;
do
{
dest[0] = source[frac>>FRACBITS];
frac += fracstep;
dest[4] = source[frac>>FRACBITS];
frac += fracstep;
dest[8] = source[frac>>FRACBITS];
frac += fracstep;
dest[12] = source[frac>>FRACBITS];
frac += fracstep;
dest[16] = source[frac>>FRACBITS];
frac += fracstep;
dest[20] = source[frac>>FRACBITS];
frac += fracstep;
dest[24] = source[frac>>FRACBITS];
frac += fracstep;
dest[28] = source[frac>>FRACBITS];
frac += fracstep;
dest += 32;
} while (--count);
}
}
// [RH] Just fills a column with a given color
void R_FillColumnHorizP (void)
{
int count = dc_count;
BYTE color = dc_color;
BYTE *dest;
if (count <= 0)
return;
{
int x = dc_x & 3;
unsigned int **span = &dc_ctspan[x];
(*span)[0] = dc_yl;
(*span)[1] = dc_yh;
*span += 2;
dest = &dc_temp[x + 4*dc_yl];
}
if (count & 1) {
*dest = color;
dest += 4;
}
if (!(count >>= 1))
return;
do {
dest[0] = color;
dest[4] = color;
dest += 8;
} while (--count);
}
// Same as R_DrawMaskedColumn() except that it always uses R_DrawColumnHoriz().
void R_DrawMaskedColumnHoriz (const BYTE *column, const FTexture::Span *span)
{
while (span->Length != 0)
{
const int length = span->Length;
const int top = span->TopOffset;
// calculate unclipped screen coordinates for post
dc_yl = (sprtopscreen + spryscale * top) >> FRACBITS;
dc_yh = (sprtopscreen + spryscale * (top + length) - FRACUNIT) >> FRACBITS;
if (sprflipvert)
{
swap (dc_yl, dc_yh);
}
if (dc_yh >= mfloorclip[dc_x])
{
dc_yh = mfloorclip[dc_x] - 1;
}
if (dc_yl < mceilingclip[dc_x])
{
dc_yl = mceilingclip[dc_x];
}
if (dc_yl <= dc_yh)
{
if (sprflipvert)
{
dc_texturefrac = (dc_yl*dc_iscale) - (top << FRACBITS)
- FixedMul (centeryfrac, dc_iscale) - dc_texturemid;
const fixed_t maxfrac = length << FRACBITS;
while (dc_texturefrac >= maxfrac)
{
if (++dc_yl > dc_yh)
goto nextpost;
dc_texturefrac += dc_iscale;
}
fixed_t endfrac = dc_texturefrac + (dc_yh-dc_yl)*dc_iscale;
while (endfrac < 0)
{
if (--dc_yh < dc_yl)
goto nextpost;
endfrac -= dc_iscale;
}
}
else
{
dc_texturefrac = dc_texturemid - (top << FRACBITS)
+ (dc_yl*dc_iscale) - FixedMul (centeryfrac-FRACUNIT, dc_iscale);
while (dc_texturefrac < 0)
{
if (++dc_yl > dc_yh)
goto nextpost;
dc_texturefrac += dc_iscale;
}
fixed_t endfrac = dc_texturefrac + (dc_yh-dc_yl)*dc_iscale;
const fixed_t maxfrac = length << FRACBITS;
if (dc_yh < mfloorclip[dc_x]-1 && endfrac < maxfrac - dc_iscale)
{
dc_yh++;
}
else while (endfrac >= maxfrac)
{
if (--dc_yh < dc_yl)
goto nextpost;
endfrac -= dc_iscale;
}
}
dc_source = column + top;
dc_dest = ylookup[dc_yl] + dc_x + dc_destorg;
dc_count = dc_yh - dc_yl + 1;
hcolfunc_pre ();
}
nextpost:
span++;
}
if (sprflipvert)
{
unsigned int *front = horizspan[dc_x&3];
unsigned int *back = dc_ctspan[dc_x&3] - 2;
// Reorder the posts so that they get drawn top-to-bottom
// instead of bottom-to-top.
while (front < back)
{
swap (front[0], back[0]);
swap (front[1], back[1]);
front += 2;
back -= 2;
}
}
}