Clipbounds refactoring + VRAM and RAM savings

mars.h

@@ -63,7 +63,7 @@ void Mars_Sec_R_WallPrep(void) ATTR_DATA_CACHE_ALIGN;
 void Mars_Sec_R_SegCommands(void) ATTR_DATA_CACHE_ALIGN;
 void Mars_Sec_R_DrawPlanes(void) ATTR_DATA_CACHE_ALIGN;
 void Mars_Sec_R_PreDrawPlanes(void) ATTR_DATA_CACHE_ALIGN;
-void Mars_Sec_R_DrawSprites(int sprscreenhalf, int *sortedsprites) ATTR_DATA_CACHE_ALIGN;
+void Mars_Sec_R_DrawSprites(int sprscreenhalf) ATTR_DATA_CACHE_ALIGN;
 void Mars_Sec_R_DrawPSprites(int sprscreenhalf) ATTR_DATA_CACHE_ALIGN;
 void Mars_Sec_P_CheckSights(void) ATTR_DATA_CACHE_ALIGN;
 void Mars_Sec_wipe_doMelt(void);
@@ -125,11 +125,10 @@ static inline void Mars_R_EndDrawPlanes(void)
 }
 
 // r_phase8
-static inline void Mars_R_BeginDrawSprites(int sprscreenhalf, int *sortedsprites)
+static inline void Mars_R_BeginDrawSprites(int sprscreenhalf)
 {
 	Mars_R_SecWait();
 	MARS_SYS_COMM6 = sprscreenhalf;
-	*(volatile uintptr_t *)&MARS_SYS_COMM8 = (uintptr_t)sortedsprites;
 	MARS_SYS_COMM4 = MARS_SECCMD_R_DRAW_SPRITES;
 }
 

marshw.c

@@ -115,23 +115,28 @@ char Mars_UploadPalette(const uint8_t* palette)
 {
 	int	i;
 	unsigned short* cram = (unsigned short *)&MARS_CRAM;
-	int16_t br = mars_brightness;
+	int br = mars_brightness;
 
 	if ((MARS_SYS_INTMSK & MARS_SH2_ACCESS_VDP) == 0)
 		return 0;
 
 	for (i = 0; i < 256; i++) {
-		int16_t r = br + *palette++;
-		int16_t g = br + *palette++;
-		int16_t b = br + *palette++;
+		int r = br + *palette++;
+		int g = br + *palette++;
+		int b = br + *palette++;
 
 		if (r < 0) r = 0; else if (r > 255) r = 255;
 		if (g < 0) g = 0; else if (g > 255) g = 255;
 		if (b < 0) b = 0; else if (b > 255) b = 255;
 
-		unsigned short b1 = ((b >> 3) & 0x1f) << 10;
-		unsigned short g1 = ((g >> 3) & 0x1f) << 5;
-		unsigned short r1 = ((r >> 3) & 0x1f) << 0;
+		unsigned b1 = b;
+		unsigned g1 = g;
+		unsigned r1 = r;
+
+		b1 = ((b1 >> 3) & 0x1f) << 10;
+		g1 = ((g1 >> 3) & 0x1f) << 5;
+		r1 = ((r1 >> 3) & 0x1f) << 0;
+
 		cram[i] = r1 | g1 | b1;
 	}
 

marsnew.c

@@ -331,7 +331,7 @@ void Mars_Secondary(void)
 			Mars_Sec_R_DrawPlanes();
 			break;
 		case MARS_SECCMD_R_DRAW_SPRITES:
-			Mars_Sec_R_DrawSprites(MARS_SYS_COMM6, (int*)(*(volatile uintptr_t *)&MARS_SYS_COMM8));
+			Mars_Sec_R_DrawSprites(MARS_SYS_COMM6);
 			break;
 		case MARS_SECCMD_M_ANIMATE_FIRE:
 			Mars_Sec_M_AnimateFire();
@@ -516,7 +516,7 @@ void* I_RemapLumpPtr(void *ptr)
 ==================== 
 */ 
 
-static char zone[0x32000] __attribute__((aligned(16)));
+static char zone[0x33000] __attribute__((aligned(16)));
 byte *I_ZoneBase (int *size)
 {
 	*size = sizeof(zone);

marssound.c

@@ -90,7 +90,7 @@ int S_PaintChannel4IMA(void* mixer, int16_t* buffer, int32_t cnt, int32_t scale)
 int S_PaintChannel4IMA2x(void* mixer, int16_t* buffer, int32_t cnt, int32_t scale) ATTR_DATA_CACHE_ALIGN;
 void S_PaintChannel8(void* mixer, int16_t* buffer, int32_t cnt, int32_t scale) ATTR_DATA_CACHE_ALIGN;
 static int S_PaintChannel(sfxchannel_t *ch, int16_t* buffer, int painted) ATTR_DATA_CACHE_ALIGN;
-static void S_StartSoundEx(mobj_t *mobj, int sound_id, getsoundpos_t getpos) ATTR_DATA_CACHE_ALIGN;
+static void S_StartSoundEx(mobj_t *mobj, int sound_id, getsoundpos_t getpos);
 static void S_SpatializeAt(fixed_t*origin, mobj_t* listener, int* pvol, int* psep) ATTR_DATA_CACHE_ALIGN;
 static void S_Spatialize(mobj_t* mobj, int* pvol, int* psep, getsoundpos_t getpos) ATTR_DATA_CACHE_ALIGN;
 static void S_SpatializeAll(void) ATTR_DATA_CACHE_ALIGN;

p_ceilng.c

@@ -9,7 +9,7 @@
 /*================================================================== */
 /*================================================================== */
 
-ceiling_t	*activeceilings[MAXCEILINGS];
+ceiling_t	**activeceilings/*[MAXCEILINGS]*/ = NULL;
 
 /*================================================================== */
 /* */

p_map.c

@@ -21,7 +21,7 @@ typedef struct
 
 static fixed_t P_InterceptVector(divline_t* v2, divline_t* v1) ATTR_DATA_CACHE_ALIGN;
 boolean	PIT_UseLines(line_t* li, plineuse_t *lu) ATTR_DATA_CACHE_ALIGN;
-void P_UseLines(player_t* player) ATTR_DATA_CACHE_ALIGN __attribute__((noinline));
+void P_UseLines(player_t* player) __attribute__((noinline));
 boolean PIT_RadiusAttack(mobj_t* thing, pradiusattack_t *ra) ATTR_DATA_CACHE_ALIGN;
 void P_RadiusAttack(mobj_t* spot, mobj_t* source, int damage) ATTR_DATA_CACHE_ALIGN;
 fixed_t P_AimLineAttack(lineattack_t *la, mobj_t* t1, angle_t angle, fixed_t distance) ATTR_DATA_CACHE_ALIGN;

p_plats.c

@@ -8,7 +8,7 @@
 #include "doomdef.h"
 #include "p_local.h"
 
-plat_t	*activeplats[MAXPLATS];
+plat_t	**activeplats/*[MAXPLATS]*/ = NULL;
 
 /*================================================================== */
 /* */

p_setup.c

@@ -752,8 +752,15 @@ extern byte *debugscreen;
 
 void P_Init (void)
 {
-	P_InitSwitchList ();
+	anims = Z_Malloc(sizeof(*anims) * MAXANIMS, PU_STATIC, 0);
+	switchlist = Z_Malloc(sizeof(*switchlist)* MAXSWITCHES * 2, PU_STATIC, 0);
+	buttonlist = Z_Malloc(sizeof(*buttonlist) * MAXBUTTONS, PU_STATIC, 0);
+	linespeciallist = Z_Malloc(sizeof(*linespeciallist) * MAXLINEANIMS, PU_STATIC, 0);
+
+	activeplats = Z_Malloc(sizeof(*activeplats) * MAXPLATS, PU_STATIC, 0);
+	activeceilings = Z_Malloc(sizeof(*activeceilings) * MAXCEILINGS, PU_STATIC, 0);
 
+	P_InitSwitchList ();
 	P_InitPicAnims ();
 #ifndef MARS
 	pausepic = W_CacheLumpName ("PAUSED",PU_STATIC);

p_spec.c

@@ -47,7 +47,7 @@ static const animdef_t	animdefs[] =
 	{-1}
 };
 
-anim_t	anims[MAXANIMS], *lastanim;
+anim_t	*anims/*[MAXANIMS]*/, * lastanim;
 
 
 void P_InitPicAnims (void)
@@ -955,7 +955,7 @@ int EV_DoDonut(line_t *line)
 */
 
 int		numlinespecials = 0;
-line_t	*linespeciallist[MAXLINEANIMS];
+line_t	**linespeciallist = NULL;
 
 void P_SpawnSpecials (void)
 {

p_spec.h

@@ -32,15 +32,15 @@ typedef struct
 
 #define	MAXANIMS		32
 
-extern	anim_t	anims[MAXANIMS], *lastanim;
+extern	anim_t	*anims/*[MAXANIMS]*/, * lastanim;
 
 
 /* */
 /*	Animating line specials */
 /* */
 #define	MAXLINEANIMS		64
 extern	int		numlinespecials;
-extern	line_t	*linespeciallist[MAXLINEANIMS];
+extern	line_t	**linespeciallist/*[MAXLINEANIMS]*/;
 
 
 /*	Define values for map objects */
@@ -176,8 +176,8 @@ typedef struct
 #define	MAXBUTTONS	16		/* 4 players, 4 buttons each at once, max. */
 #define BUTTONTIME	15		/* 1 second */
 
-extern	VINT		switchlist[MAXSWITCHES * 2];
-extern	button_t buttonlist[MAXBUTTONS];
+extern	VINT		*switchlist/*[MAXSWITCHES * 2]*/;
+extern	button_t* buttonlist/*[MAXBUTTONS]*/;
 
 void	P_ChangeSwitchTexture(line_t *line,int useAgain);
 void 	P_InitSwitchList(void);
@@ -226,7 +226,7 @@ typedef struct
 #define	PLATSPEED	(FRACUNIT*THINKERS_TICS)
 #define	MAXPLATS	30
 
-extern	plat_t	*activeplats[MAXPLATS];
+extern	plat_t	**activeplats/*[MAXPLATS]*/;
 
 void	T_PlatRaise(plat_t	*plat);
 int		EV_DoPlat(line_t *line,plattype_e type,int amount);
@@ -310,7 +310,7 @@ typedef struct
 #define	CEILSPEED		FRACUNIT*THINKERS_TICS
 #define MAXCEILINGS		30
 
-extern	ceiling_t	*activeceilings[MAXCEILINGS];
+extern	ceiling_t	**activeceilings/*[MAXCEILINGS]*/;
 
 int		EV_DoCeiling (line_t *line, ceiling_e  type);
 void	T_MoveCeiling (ceiling_t *ceiling);

p_switch.c

@@ -16,9 +16,9 @@ static const switchlist_t alphSwitchList[] =
 	{"SW1WOOD",		"SW2WOOD"},
 };
 
-VINT		switchlist[MAXSWITCHES * 2];
+VINT		*switchlist/*[MAXSWITCHES * 2]*/ = NULL;
 int			numswitches;
-button_t	buttonlist[MAXBUTTONS];
+button_t	*buttonlist/*[MAXBUTTONS]*/ = NULL;
 
 /*
 ===============

r_data.c

@@ -392,6 +392,7 @@ void R_InitMathTables(void)
 	viewangletox = (VINT *)I_AllocWorkBuffer(sizeof(*viewangletox) * (FINEANGLES / 2));
 	distscale = (fixed_t *)I_AllocWorkBuffer(sizeof(*distscale) * SCREENWIDTH);
 	yslope = (fixed_t *)I_AllocWorkBuffer(sizeof(*yslope) * SCREENHEIGHT);
+	xtoviewangle = (uint16_t *)I_AllocWorkBuffer(sizeof(*xtoviewangle) * (SCREENWIDTH+1));
 	tempviewangletox = (VINT *)I_WorkBuffer();
 
 	// Use tangent table to generate viewangletox:
@@ -470,6 +471,12 @@ void R_InitMathTables(void)
 	{
 		fuzzoffset[i] = fuzzoffset[i] < 0 ? -fuzzunit : fuzzunit;
 	}
+
+	// enable caching for LUTs
+	viewangletox = (void *)(((intptr_t)viewangletox) & ~0x20000000);
+	distscale = (void *)(((intptr_t)distscale) & ~0x20000000);
+	yslope = (void *)(((intptr_t)yslope) & ~0x20000000);
+	xtoviewangle = (void *)(((intptr_t)xtoviewangle) & ~0x20000000);
 }
 
 

r_local.h

@@ -335,7 +335,7 @@ extern	VINT	*viewangletox/*[FINEANGLES/2]*/;
 
 /* The xtoviewangleangle[] table maps a screen pixel to the lowest viewangle */
 /* that maps back to x ranges from clipangle to -clipangle */
-extern	uint16_t		xtoviewangle[SCREENWIDTH+1];
+extern	uint16_t		*xtoviewangle/*[SCREENWIDTH+1]*/;
 
 #ifdef MARS
 extern	const fixed_t		finetangent_[FINEANGLES/4];
@@ -458,7 +458,7 @@ typedef struct
 
 	int			b_bottomheight;
 	int			b_texturemid;
-	/* !!! THE FOLLOWING SECTION MUST BE LARGE ENOUGH */
+	/* !!! THE SECTION ABOVE MUST BE LARGE ENOUGH */
 	/* !!! TO ACCOMODATE VISSPRITE_T STRUCTURE, GETS */
 	/* !!! OVERWRITTEN AFTER PHASE 7 - END */
 
@@ -497,7 +497,7 @@ typedef struct
 		mapvertex_t		v2;
 	};
 
-	byte 		*sil;
+	uint16_t 		*clipbounds;
 } viswall_t;
 
 typedef struct
@@ -526,7 +526,7 @@ typedef struct vissprite_s
 
 #define	MAXVISSPRITES	MAXWALLCMDS
 
-#define	MAXOPENINGS		SCREENWIDTH*14
+#define	MAXOPENINGS		SCREENWIDTH*16
 
 #define	MAXVISSSEC		128
 
@@ -600,9 +600,9 @@ __attribute__((aligned(16)))
 	/* */
 	/* openings / misc refresh memory */
 	/* */
-	unsigned short	* volatile openings/*[MAXOPENINGS]*/;
-	unsigned short	* volatile lastopening;
 	unsigned short	* volatile segclip, * volatile lastsegclip;
+
+	int * volatile gsortedsprites;
 } viewdef_t;
 
 extern	viewdef_t	vd;

r_main.c

@@ -77,7 +77,7 @@ fixed_t *distscale/*[SCREENWIDTH]*/;
 
 VINT *viewangletox/*[FINEANGLES/2]*/;
 
-uint16_t xtoviewangle[SCREENWIDTH+1];
+uint16_t *xtoviewangle/*[SCREENWIDTH+1]*/;
 
 /* */
 /* performance counters */
@@ -660,8 +660,8 @@ static void R_Setup (int displayplayer, visplane_t *visplanes_,
 
 	vd.viswallextras = viswallex_ + 1;
 
-	vd.openings = tempbuf;
-	tempbuf += MAXOPENINGS;
+	// re-use the openings array in VRAM
+	vd.gsortedsprites = (void *)(((intptr_t)vd.visplanes[1].open + 3) & ~3);
 
 	vd.vissprites = (void *)vd.viswalls;
 
@@ -681,8 +681,6 @@ static void R_Setup (int displayplayer, visplane_t *visplanes_,
 	vd.vissprite_p = vd.vissprites;
 	vd.lastsprite_p = vd.vissprite_p;
 
-	vd.lastopening = vd.openings;
-
 	vd.vissectors = vissectors_;
 	vd.lastvissector = vd.vissectors;	/* no subsectors visible yet */
 
@@ -918,12 +916,9 @@ void R_RenderPlayerView(int displayplayer)
 	t_segs = I_GetFRTCounter() - t_segs;
 
 	Mars_ClearCacheLine(&vd.lastsegclip);
-	Mars_ClearCacheLine(&vd.lastopening);
 
 	if (vd.lastsegclip - vd.segclip > MAXOPENINGS)
 		I_Error("lastsegclip > MAXOPENINGS: %d", vd.lastsegclip - vd.segclip);
-	if (vd.lastopening - vd.openings > MAXOPENINGS)
-		I_Error("lastopening > MAXOPENINGS: %d", vd.lastopening - vd.openings);
 	if (vd.lastvissector - vd.vissectors > MAXVISSSEC)
 		I_Error("lastvissector > MAXVISSSEC: %d", vd.lastvissector - vd.vissectors);