massive acc speedup, removed fill error, hardcursor OK

git-svn-id: file:///srv/svn/repos/haiku/trunk/current@7075 a95241bf-73f2-0310-859d-f6bbb57e9c96

src/add-ons/accelerants/neomagic/InitAccelerant.c

@@ -4,7 +4,7 @@
 
 	Other authors:
 	Mark Watson,
-	Rudolf Cornelissen 10/2002-1/2004.
+	Rudolf Cornelissen 10/2002-3/2004.
 */
 
 #define MODULE_BIT 0x00800000
@@ -110,7 +110,6 @@ report success or failure.
 */
 status_t INIT_ACCELERANT(int the_fd) {
 	status_t result;
-	int pointer_reservation; //mem reserved for pointer
 	int cnt; 				 //used for iteration through the overlay buffers
 
 	if (1) {
@@ -161,15 +160,13 @@ status_t INIT_ACCELERANT(int the_fd) {
 	si->cursor.y = 0;
 
 	/*
-	Put the frame buffer immediately following the cursor data. We store this
+	Put the frame buffer at the beginning of the cardRAM because the acc engine
+	does not support offsets, or we lack info to get that ability. We store this
 	info in a frame_buffer_config structure to make it convienient to return
 	to the app_server later.
 	*/
-	pointer_reservation = 0;
+	si->fbc.frame_buffer = si->framebuffer;
-	if (si->settings.hardcursor) pointer_reservation = si->ps.curmem_size;
+	si->fbc.frame_buffer_dma = si->framebuffer_pci;
-
-	si->fbc.frame_buffer = (void *)((char *)si->framebuffer+pointer_reservation);
-	si->fbc.frame_buffer_dma = (void *)((char *)si->framebuffer_pci+pointer_reservation);
 
 	/* count of issued parameters or commands */
 	si->engine.last_idle = si->engine.count = 0;

src/add-ons/accelerants/neomagic/Overlay.c

@@ -1,4 +1,4 @@
-/* Written by Rudolf Cornelissen 05-2002/06-2003 */
+/* Written by Rudolf Cornelissen 05-2002/03-2004 */
 
 /* Note on 'missing features' in BeOS 5.0.3 and DANO:
  * BeOS needs to define more colorspaces! It would be nice if BeOS would support the FourCC 'definitions'
@@ -176,9 +176,10 @@ const overlay_buffer *ALLOCATE_OVERLAY_BUFFER(color_space cs, uint16 width, uint
 
 		/* calculate first free RAM adress in card:
 		 * Driver setup is as follows: 
-		 * card base: 		- hardware cursor bitmap (if used),
+		 * card base: 		- screen memory,
-		 * directly above	- screen memory for both heads */
+		 * free space:      - used for overlay,
-		adress2 = (((uint32)((uint8*)si->fbc.frame_buffer)) +	/* cursor already included here */
+		 * card top:        - hardware cursor bitmap (if used) */
+		adress2 = (((uint32)((uint8*)si->fbc.frame_buffer)) +	/* cursor not yet included here */
 			(si->fbc.bytes_per_row * si->dm.virtual_height));	/* size in bytes of screen(s) */
 		LOG(4,("Overlay: first free cardRAM virtual adress $%08x\n", adress2));
 
@@ -189,6 +190,9 @@ const overlay_buffer *ALLOCATE_OVERLAY_BUFFER(color_space cs, uint16 width, uint
 
 		/* calculate virtual memory adress that would be needed for a new bitmap */
 		adress = (((uint32)((uint8*)si->framebuffer)) + (si->ps.memory_size * 1024));
+		/* take cursor into account (if it exists) */
+		if(si->settings.hardcursor) adress -= si->ps.curmem_size;
+		LOG(4,("Overlay: last free cardRAM virtual adress $%08x\n", (adress - 1)));
 		for (cnt = 0; cnt <= offset; cnt++)
 		{
 			adress -= si->overlay.myBufInfo[cnt].size;
@@ -273,6 +277,8 @@ const overlay_buffer *ALLOCATE_OVERLAY_BUFFER(color_space cs, uint16 width, uint
 
 		/* calculate physical memory adress (for dma use) */
 		adress = (((uint32)((uint8*)si->framebuffer_pci)) + (si->ps.memory_size * 1024));
+		/* take cursor into account (if it exists) */
+		if(si->settings.hardcursor) adress -= si->ps.curmem_size;
 		for (cnt = 0; cnt <= offset; cnt++)
 		{
 			adress -= si->overlay.myBufInfo[cnt].size;

src/add-ons/accelerants/neomagic/engine/nm_acc.c

@@ -23,11 +23,11 @@ blit
 status_t nm_acc_wait_idle()
 {
 	/* wait until engine completely idle */
-	while (ACCR(STATUS) & 0x00000001)
+	/* Note:
-	{
+	 * because we have no FIFO functionality we have to make sure we return ASAP(!).
-		/* snooze a bit so I do not hammer the bus */
+	 * snoozing() for just 1 microSecond already more than halfs the engine
-		snooze (100); 
+	 * performance... */
-	}
+	while (ACCR(STATUS) & 0x00000001);
 
 	return B_OK;
 }
@@ -101,11 +101,13 @@ status_t nm_acc_init()
 
 	/* setup buffer startadress */
 	/* fixme when/if possible:
-	 * not possible on all cards or not enough specs known :-/ (tried NM2160)
+	 * not possible on all cards or not enough specs known. (tried NM2160)
-	 * workaround: place cursor bitmap _after_ screenbuffer instead of vice versa. */
+	 * workaround: place cursor bitmap at _end_ of cardRAM instead of in _beginning_. */
 
 	/* setup clipping */
-	/* apparantly not needed. */
+	si->engine.control |= (1 << 26);
+	ACCW(CLIPLT, 0);
+	ACCW(CLIPRB, ((si->dm.virtual_height << 16) | (si->dm.virtual_width & 0x0000ffff)));
 
 	return B_OK;
 }
@@ -123,7 +125,7 @@ status_t nm_acc_blit(uint16 xs,uint16 ys,uint16 xd,uint16 yd,uint16 w,uint16 h)
     {
 		/* start with upper left corner */
 		/* use ROP GXcopy (b16-19), and use XY coord. system (b24-25) */
-		ACCW(BLTCNTL, si->engine.control | 0x830c0000);
+		ACCW(CONTROL, si->engine.control | 0x830c0000);
 		/* send command and exexute */
 		ACCW(SRCSTARTOFF, ((ys << 16) | (xs & 0x0000ffff)));
 		ACCW(DSTSTARTOFF, ((yd << 16) | (xd & 0x0000ffff)));
@@ -133,7 +135,7 @@ status_t nm_acc_blit(uint16 xs,uint16 ys,uint16 xd,uint16 yd,uint16 w,uint16 h)
     {
 		/* start with lower right corner */
 		/* use ROP GXcopy (b16-19), and use XY coord. system (b24-25) */
-		ACCW(BLTCNTL, (si->engine.control | 0x830c0013));
+		ACCW(CONTROL, (si->engine.control | 0x830c0013));
 		/* send command and exexute */
 		ACCW(SRCSTARTOFF, (((ys + h) << 16) | ((xs + w) & 0x0000ffff)));
 		ACCW(DSTSTARTOFF, (((yd + h) << 16) | ((xd + w) & 0x0000ffff)));
@@ -153,7 +155,7 @@ status_t nm_acc_setup_rectangle(uint32 color)
 	nm_acc_wait_idle();
 
 	/* use ROP GXcopy (b16-19), use XY coord. system (b24-25), do foreground color (b3) */
-	ACCW(BLTCNTL, (si->engine.control | 0x830c0008));
+	ACCW(CONTROL, (si->engine.control | 0x830c0008));
 	/* setup color */
 	ACCW(FGCOLOR, color);
 
@@ -162,6 +164,12 @@ status_t nm_acc_setup_rectangle(uint32 color)
 
 status_t nm_acc_rectangle(uint32 xs,uint32 xe,uint32 ys,uint32 yl)
 {
+	/* The engine does not take kindly if we try to let it fill a rect with
+	 * zero width. Dano's app_server occasionally tries to let us do that though!
+	 * Testable with BeRoMeter 1.2.6, all Ellipses tests. Effect of zero width fill:
+	 * horizontal lines across the entire screen at top and bottom of ellipses. */
+	if (xe == xs) return B_OK;
+
 	/* make sure the previous command (if any) is completed */
 //	does not work yet:
 //	nm_acc_wait_fifo(2);
@@ -185,7 +193,7 @@ status_t nm_acc_setup_rect_invert()
 	nm_acc_wait_idle();
 
 	/* use ROP GXinvert (b16-19), use XY coord. system (b24-25), do foreground color (b3) */
-	ACCW(BLTCNTL, (si->engine.control | 0x83050008));
+	ACCW(CONTROL, (si->engine.control | 0x83050008));
 	/* reset color */
 	ACCW(FGCOLOR, 0);
 
@@ -194,6 +202,10 @@ status_t nm_acc_setup_rect_invert()
 
 status_t nm_acc_rectangle_invert(uint32 xs,uint32 xe,uint32 ys,uint32 yl)
 {
+	/* The engine probably also does not take kindly if we try to let it invert a
+	 * rect with zero width... (see nm_acc_rectangle() routine for explanation.) */
+	if (xe == xs) return B_OK;
+
 	/* make sure the previous command (if any) is completed */
 //	does not work yet:
 //	nm_acc_wait_fifo(4);

src/add-ons/accelerants/neomagic/engine/nm_crtc.c

@@ -569,14 +569,17 @@ status_t nm_crtc_cursor_init()
 {
 	int i;
 	uint32 * fb;
-	/* cursor bitmap will be stored at the start of the framebuffer */
+	uint32 curadd, curreg;
-	uint32 curadd = 0, curreg;
+
+	/* the cursor is at the end of cardRAM */
+	curadd = ((si->ps.memory_size * 1024) - si->ps.curmem_size);
 
 	/* set cursor bitmap adress on a 1kb boundary, and move the bits around
 	 * so they get placed at the correct registerbits */
-	//fixme: NM2380 must have an extra bit for > 4Mb???
 	curreg = (((curadd >> 10) & 0x000f) << 8);
 	curreg |= (((curadd >> 10) & 0x0ff0) >> 4);
+	/* NM2380 must have an extra bit for > 4Mb: assuming it to be on b12... */
+	curreg |= ((curadd >> 10) & 0x1000);
 
 	if (si->ps.card_type < NM2200)
 		CR1W(CURADDRESS, curreg);
@@ -650,8 +653,9 @@ status_t nm_crtc_cursor_define(uint8* andMask,uint8* xorMask)
 	uint8 y;
 	uint8 * cursor;
 
-	/*get a pointer to the cursor*/
+	/* get a pointer to the cursor: it's at the end of cardRAM */
 	cursor = (uint8*) si->framebuffer;
+	cursor += ((si->ps.memory_size * 1024) - si->ps.curmem_size);
 
 	/*draw the cursor*/
 	for(y=0;y<16;y++)

src/add-ons/accelerants/neomagic/engine/nm_general.c

@@ -47,7 +47,7 @@ status_t nm_general_powerup()
 {
 	status_t status;
 
-	LOG(1,("POWERUP: Neomagic (open)BeOS Accelerant 0.06-3 running.\n"));
+	LOG(1,("POWERUP: Neomagic (open)BeOS Accelerant 0.06-4 running.\n"));
 
 	/* detect card type and power it up */
 	switch(CFGR(DEVID))