Page 1 of 1

Optimized MDL clipping (in software)

Posted: Thu Sep 12, 2013 6:00 am
by mankrip
This one is easy. Just replace the whole r_aclip.c (mdl_clip.c, in Makaqu) file with this:

Code: Select all

/*
Copyright (C) 1996-1997 Id Software, Inc.

This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.

*/
// r_aclip.c: clip routines for drawing Alias models directly to the screen

#include "quakedef.h"
#include "r_local.h"
#include "d_local.h"

static finalvert_t
	fv[2][8]
,	* clipfv // mankrip
	;
static auxvert_t
	av[8]
	;
// mankrip - begin
#define HORIZONTAL 0 // for clipplane
#define VERTICAL 1 // for clipplane
static int
	* clipfvi0
,	* clipfvi1
,	clipplane // for clipscale
,	cliplimit // for clipscale
	;
static float
	clipscale
	;
static vec3_t
	clipv0
,	clipv1
	;
// mankrip - end



/*
================
R_Alias_clip_z

pfv0 is the unclipped vertex, pfv1 is the z-clipped vertex
================
*/
void R_Alias_clip_z (finalvert_t *pfv0, finalvert_t *pfv1, finalvert_t *out)
{
	// mankrip - begin
	static auxvert_t
		avout
		;
	if (pfv0->v[1] >= pfv1->v[1])
	{
		clipfvi0 = pfv0->v;
		clipfvi1 = pfv1->v;
		// copying these integer vectors also implicitely typecasts them
		VectorCopy (av[pfv0 - &fv[0][0]].fv, clipv0);
		VectorCopy (av[pfv1 - &fv[0][0]].fv, clipv1);
	}
	else
	{
		clipfvi0 = pfv1->v;
		clipfvi1 = pfv0->v;
		// copying these integer vectors also implicitely typecasts them
		VectorCopy (av[pfv1 - &fv[0][0]].fv, clipv0);
		VectorCopy (av[pfv0 - &fv[0][0]].fv, clipv1);
	}
	clipscale = (ALIAS_Z_CLIP_PLANE - clipv0[2]) / (clipv1[2] - clipv0[2]);

	avout.fv[0] = (int) (clipv0[0] + (clipv1[0] - clipv0[0]) * clipscale);
	avout.fv[1] = (int) (clipv0[1] + (clipv1[1] - clipv0[1]) * clipscale);
	avout.fv[2] = ALIAS_Z_CLIP_PLANE;

	out->v[2] =	clipfvi0[2] + (int) ( (float) (clipfvi1[2] - clipfvi0[2]) * clipscale);
	out->v[3] =	clipfvi0[3] + (int) ( (float) (clipfvi1[3] - clipfvi0[3]) * clipscale);
	out->v[4] =	clipfvi0[4] + (int) ( (float) (clipfvi1[4] - clipfvi0[4]) * clipscale);
	// mankrip - end

	R_AliasProjectFinalVert (out, &avout);
}



// mankrip - begin
void R_Alias_clip_xy (finalvert_t *pfv0, finalvert_t *pfv1, finalvert_t *out)
{
	// for left/right/top/bottom variations, only clipscale changes
	// r_refdef.vrect.x for left cliplimit, r_refdef.vrectright for right cliplimit
	// r_refdef.vrect.y for top cliplimit, r_refdef.vrectbottom for bottom cliplimit
	// clipplane should be either zero (horizontal) or 1 (vertical)

	if (pfv0->v[1] >= pfv1->v[1])
	{
		clipfvi0 = pfv0->v;
		clipfvi1 = pfv1->v;
	}
	else
	{
		clipfvi0 = pfv1->v;
		clipfvi1 = pfv0->v;
	}
	clipscale = (float) (cliplimit - clipfvi0[clipplane]) / (float) (clipfvi1[clipplane] - clipfvi0[clipplane]);
	out->v[0] = clipfvi0[0] + (int) ( (float) (clipfvi1[0] - clipfvi0[0]) * clipscale + 0.5f);
	out->v[1] = clipfvi0[1] + (int) ( (float) (clipfvi1[1] - clipfvi0[1]) * clipscale + 0.5f);
	out->v[2] = clipfvi0[2] + (int) ( (float) (clipfvi1[2] - clipfvi0[2]) * clipscale + 0.5f);
	out->v[3] = clipfvi0[3] + (int) ( (float) (clipfvi1[3] - clipfvi0[3]) * clipscale + 0.5f);
	out->v[4] = clipfvi0[4] + (int) ( (float) (clipfvi1[4] - clipfvi0[4]) * clipscale + 0.5f);
	out->v[5] = clipfvi0[5] + (int) ( (float) (clipfvi1[5] - clipfvi0[5]) * clipscale + 0.5f);
}
// mankrip - end



int R_AliasClip (finalvert_t *in, finalvert_t *out, int flag, int count,
	void (*clip) (finalvert_t *pfv0, finalvert_t *pfv1, finalvert_t *out) )
{
	static int
		i
	,	j
	,	k
	,	flags
	,	oldflags
		;

	for (k = 0 , i = 0 , j = count - 1 ; i < count ; j = i, i++)
	{
		oldflags = in[j].flags & flag;
		flags    = in[i].flags & flag;

		if (flags && oldflags)
			continue;
		if (oldflags ^ flags)
		{
			clip (&in[j], &in[i], &out[k]);
			out[k].flags = 0;

			// check if it needs to be clipped again
			// mankrip - initial clipping is done right-to-left and bottom-to-top, so let's check the other way around now
			if (out[k].v[0] < r_refdef.vrect.x)		out[k].flags |= ALIAS_LEFT_CLIP;
			else
			if (out[k].v[0] > r_refdef.vrectright)	out[k].flags |= ALIAS_RIGHT_CLIP;
			if (out[k].v[1] < r_refdef.vrect.y)		out[k].flags |= ALIAS_TOP_CLIP;
			else
			if (out[k].v[1] > r_refdef.vrectbottom) out[k].flags |= ALIAS_BOTTOM_CLIP;

			k++;
		}
		if (!flags)
		{
			out[k] = in[i];
			k++;
		}
	}
	return k;
}



void R_AliasClipTriangle (mtriangle_t *ptri)
{
	static int
		i
	,	k
	,	pingpong
		;
	static unsigned
		clipflags
		;
	static mtriangle_t
		mtri
		;

	// copy vertexes and fix seam texture coordinates
	fv[0][0] = pfinalverts[ptri->vertindex[0]];
	fv[0][1] = pfinalverts[ptri->vertindex[1]];
	fv[0][2] = pfinalverts[ptri->vertindex[2]];
	// mankrip - begin
	if (!ptri->facesfront)
	{
		if (fv[0][0].flags & ALIAS_ONSEAM) fv[0][0].v[2] += r_affinetridesc.seamfixupX16;
		if (fv[0][1].flags & ALIAS_ONSEAM) fv[0][1].v[2] += r_affinetridesc.seamfixupX16;
		if (fv[0][2].flags & ALIAS_ONSEAM) fv[0][2].v[2] += r_affinetridesc.seamfixupX16;
	}
	// mankrip - end

	// clip
	clipflags = fv[0][0].flags | fv[0][1].flags | fv[0][2].flags;

	if (clipflags & ALIAS_Z_CLIP)
	{
		av[0] = pauxverts[ptri->vertindex[0]];
		av[1] = pauxverts[ptri->vertindex[1]];
		av[2] = pauxverts[ptri->vertindex[2]];

		k = R_AliasClip (fv[0], fv[1], ALIAS_Z_CLIP, 3, R_Alias_clip_z);
		if (k == 0)
			return;
		pingpong = 1;
		clipflags = fv[1][0].flags | fv[1][1].flags | fv[1][2].flags;
	}
	else
	{
		pingpong = 0;
		k = 3;
	}

	// mankrip - the viewmodel is always on the bottom of the screen, so let's clip the bottom first
	if (clipflags & ALIAS_BOTTOM_CLIP)
	{
		cliplimit = r_refdef.vrectbottom; // mankrip
		clipplane = VERTICAL; // mankrip
		k = R_AliasClip (fv[pingpong], fv[pingpong ^ 1], ALIAS_BOTTOM_CLIP, k, R_Alias_clip_xy);
		if (k == 0)
			return;
		pingpong ^= 1;
	}

	// mankrip - weapons such as the axe are right-handed, so let's clip the right before the left
	if (clipflags & ALIAS_RIGHT_CLIP)
	{
		cliplimit = r_refdef.vrectright; // mankrip
		clipplane = HORIZONTAL; // mankrip
		k = R_AliasClip (fv[pingpong], fv[pingpong ^ 1], ALIAS_RIGHT_CLIP, k, R_Alias_clip_xy);
		if (k == 0)
			return;
		pingpong ^= 1;
	}

	if (clipflags & ALIAS_TOP_CLIP)
	{
		cliplimit = r_refdef.vrect.y; // mankrip
		clipplane = VERTICAL; // mankrip
		k = R_AliasClip (fv[pingpong], fv[pingpong ^ 1], ALIAS_TOP_CLIP, k, R_Alias_clip_xy);
		if (k == 0)
			return;
		pingpong ^= 1;
	}

	if (clipflags & ALIAS_LEFT_CLIP)
	{
		cliplimit = r_refdef.vrect.x; // mankrip
		clipplane = HORIZONTAL; // mankrip
		k = R_AliasClip (fv[pingpong], fv[pingpong ^ 1], ALIAS_LEFT_CLIP, k, R_Alias_clip_xy);
		if (k == 0)
			return;
		pingpong ^= 1;
	}

	// mankrip - fix any rounding that went outside of the screen
	for (i = 0 ; i < k ; i++)
	{
		clipfvi0 = fv[pingpong][i].v;
		if (clipfvi0[0] > r_refdef.vrectright)
			clipfvi0[0] = r_refdef.vrectright;
		else
		if (clipfvi0[0] < r_refdef.vrect.x)
			clipfvi0[0] = r_refdef.vrect.x;

		if (clipfvi0[1] > r_refdef.vrectbottom)
			clipfvi0[1] = r_refdef.vrectbottom;
		else
		if (clipfvi0[1] < r_refdef.vrect.y)
			clipfvi0[1] = r_refdef.vrect.y;
		// mankrip - begin
		if (currententity->scrleft  > clipfvi0[0])
			currententity->scrleft  = clipfvi0[0];
		if (currententity->scrright < clipfvi0[0])
			currententity->scrright = clipfvi0[0];
		if (currententity->scrtop    > clipfvi0[1])
			currententity->scrtop    = clipfvi0[1];
		if (currententity->scrbottom < clipfvi0[1])
			currententity->scrbottom = clipfvi0[1];
		// mankrip - end

		fv[pingpong][i].flags = 0;
	}

	// draw triangles
	mtri.facesfront = ptri->facesfront;
	r_affinetridesc.ptriangles = &mtri;
	r_affinetridesc.pfinalverts = fv[pingpong];

	// FIXME: do all at once as trifan?
	mtri.vertindex[0] = 0;
	for (i = 1 ; i < k - 1 ; i++)
	{
		mtri.vertindex[1] = i;
		mtri.vertindex[2] = i+1;
		D_PolysetDraw_C (); // mankrip - transparencies - edited
	}
}
I've unrolled a number of loops and employed some other strategies to gain speed, but on the other hand, to make the code easier to maintain (and to read!) I've collapsed the four top/bottom/left/right clipping functions into one (R_Alias_clip_xy), which requires cliplimit and clipplane to be set before each call.

I haven't benchmarked the speed difference in comparison to the original C code, as my main goal with this was to understand how the MDL clipping works.

Re: Optimized MDL clipping (in software)

Posted: Fri Sep 13, 2013 2:38 am
by qbism
Well, in a quick timedemo comparison it works and is at least not slower. I did not do a vigorous benchmark with a lot of mdl flesh in the scene.