Help
Add this page to your collection
Show collection (0 pages)
Suggest pagesAros/Developer/Docs/Libraries/CGFX
This page may need to be
Contents |
[edit] Introduction
cybergraphics.library is a 15, 16, 24b or 32it graphics board extension for the native graphics.library. Up to OS3.1 there are no possibilities to add custom boards into the display database in a system friendly fashion and so it instead patched some of the original graphics functions in order to integrate custom graphics boards into the system.
Nearly every standard 8bit application will work on this 15/16/24 bit screen. But it only can use the additional features, if it knows of CyberGraphics, of course.
[edit] Writing Pixels
For writing pixels, reading Pixels and blitting, there are the functions FillPixelArray, MovePixelArray, ReadPixelArray, ScalePixelArray, SwapPixelArray, WritePixelArray and WriteRGBPixel.
For double buffering you use standard Double Buffering functions like the functions around AllocScreenBuffer.
Apps using the original cgfx functions (and attributes for functions like GetCyberMapAttr) should never see this values.
Yes, however only GetCyberMapAttr() processes this. Other places pass them on to applications.
a CYBRMATTR_PIXFMT_ALPHA was introduced, too, which means "do not map, I know about the 0RGB32, etc. pixfmts". In case you want to keep that.
While driver_GetCyberMapAttr() for example seems to map the 0RGB32, BGR032, ... correctly to ARGB32, BGRA32, ... so that (legacy) apps get only values they expect. In other places (which callc hidd2cyber_pixfmt()) this is missing/was forgotten to be fixed.
You could (*) add a BOOL param to hidd2cyber_pixfmt() function which indicates whether or not to map the 0RGB32, etc. to ARGB32, etc. And then pass FALSE in most functions which call hidd2cyber_pixfmt().
[edit] Bitmaps
Starting with OS 3.x two new graphics calls were added to support bitmap creation and removal: AllocBitmap()/gfx and FreeBitmap()/gfx. With this calls the first step was made to allow custom bitmap formats. Before it was no problem at all to directly render into the planes or to make assumptions about the structure of the bitmap data. To get more information about the bitmap data an information retrieval call was added:
GetBitmapAttr()/gfx. With this one you could get information on the width, height, depth and certain flags of the bitmap. This was the first step into the RTG direction. As soon as GetBitmapAttr(bm,BMA_FLAGS) didn't return BMF_STANDARD, the bitmap was not direct accessible anymore and could only be accessed by attachment to a rastport and subsequent rendering calls or by using blit calls with a standard amiga bitmap as destination.
Unfortunately many programs didn't follow these guide lines and didn't check for the BMF_STANDARD flag with OS3.0 and above. Cybergrphx version of GetBitMapAttr (bitmap, P96BMA_MEMORY) has to use LockBitMapAddress() with LBMI_BASEADDRESS and don't forget use LBMI_BYTESPERROW either.
LONG modulo; APTR buffer; APTR lock = LockBitMapTags(bitmap, LBMI_BASEADDRESS, &buffer, LBMI_BYTESPERROW, &modulo, TAG_DONE); if (lock) { /* Manipulate bitmap here */ UnlockBitMapTags(lock); }
With CyberGraphX you also have to add BMF_MINPLANES in the flags field of AllocBitmap()/gfx. It is also possible now to allocate device dependant pixel format bitmaps and standard 15/16/24/32 bit depth image maps. Therefore new bits in the 32bit flags field parameter of AllocBitmap() have been added. If you specify the BMB_SPECIALFMT bit (see includes), the upper 8 bits of the flags longword contain information on the pixel format of the requested bitmap.
[edit] Lock BitMaps
Once allocated, you can attach the returned bitmap to a rastport and do rendering calls. It is NOT possible to attach this bitmap to a screen by supplying it as custom bitmap with OpenScreenTagList() !!
You are not allowed to directly poke into the bitmap image data as long as the bitmap isn't locked !. The location and contents of the image data is subject to change and is only valid when it's locked by using the available locking calls. LockBitmapTags()/UnLockBitmap() has been added for this purpose. You HAVE TO supply a taglist with LockBitmapTags() which contains pointers to longwords which are filled with valid data if the call returns with a non-zero value. Only if a non-zero value is returned you are able to access the bitmap directly ! Check the address you get back with the LBMI_BASEADDRESS tag. This is the base address you can do your rendering to.
which flags do I need for AllocBitMap to create a lockable bitmap? anything else resulted in a "no lock" message
#include <proto/graphics.h>
#include <cybergraphx/cybergraphics.h>
#include <proto/cybergraphics.h>
#include <stdio.h>
struct BitMap *bitmap;
APTR handle;
IPTR buffer;
int main(void)
{
bitmap = AllocBitMap(400, 300, 32,
/*BMF_MINPLANES | BMF_DISPLAYABLE | */ BMF_SPECIALFMT | SHIFT_PIXFMT(PIXFMT_ABGR32), NULL);
if (bitmap)
{
//handle = LockBitMapTags(bitmap, LBMI_BASEADDRESS, &buffer, TAG_DONE);
handle = LockBitMapTags(bitmap, TAG_DONE);
if (handle)
{
printf("buffer %x\n", (unsigned int)buffer);
UnLockBitMap(handle);
}
else
{
puts("no lock");
}
FreeBitMap(bitmap);
}
else
{
puts("no bitmap");
}
return 0;
}
[edit] OffScreen BitMap
If using a windowed application running on a 15-bit (or higher) Workbench. The application uses an offscreen BitMap for rendering, the contents are then copied to the window with a ClipBlit(). Assuming my offscreen BitMap is a friend of the Workbench screen Bitmap and is at least 15 bits deep, how do I arbitrarily set the colour I want to draw with? Basically you will need to GetColorMap(),followed by a number of SetRGB32CM().
win->WScreen->ViewPort.Colormap
So I can take a pointer to that and feed it to ObtainPen() to get my custom drawing colour, right? Better use ObtainBestPen().
All drawing functions work on a bitmap. A rastport is basically just some structure which contains the graphics context (foreground color, line pattern, drawmode). Most drawing functions require a rastport. (a rastport does not automatically come with it's own bitmap!).
You create some rastport. You create some bitmap. You connect the two (put a pointer to the bitmap into the rastport). Then you render into rastport and it will go into your bitmap:
struct RastPort rp; InitRastPort(&rp); [...] /* Use rastport */ [...] DeinitRastPort(&rp);
struct RastPort *rp;
if ((rp = CreateRastPort()))
{
[...]
/* use rastport */
FreeRastPort(rp);
}
allocating bitmap in same format as screen (wb) bitmap for best performance:
bm = AllocBitMap(width, height, depth, BMF_MINPLANES, screen->RastPort.BitMap);
Connecting rastport with bitmap:
rp->BitMap = bm;
Note tough that this way you don't get any clipping at all, so do not render outside of the bitmap dimensions or it will crash (not necessarily happening under AROS/hosted/x11 as there x11 clips everything)).
If you really need clipping as well then you would have to install a layerinfo plus some layer connected with your bitmap and then use layer->rp as rastport.
You need to differentiate between indexed and truecolor by force. To draw single pixel in indexed mode there is graphics.library/WritePixel(). To draw single pixel in truecolor mode using an RGB value there's cybergraphics.library/WriteRGBPixel.
To draw multiple indexed pixels there's graphics.library/WritePixelLine8() or WritePixelArray8() or WriteChunkyPixels(). On truecolor screens drawing multiple *still indexed* pixels there's cybergraphics.library/WriteLUTPixelArray().
To draw RGB pixels on truecolors screens you can use cybergraphics.library/WritePixelArray(rectfmt = RECTFMT_RGB or RECTFMT_ARGB or RECTFMT_RGBA).
RGB = R8 G8 B8 R8 G8 B8 ... ARGB = A8 R8 G8 B8 A8 R8 G8 B8 ... ...
in memory, independant of endianess of computer.
The pixel buffer will be converted to the destination bitmap pixel format automatically but this conversion may be slow as we have only one general slow conversion routine.
Better would be to use RECTFMT_RAW which means same pixel format as destination bitmap. Then you would have to do the conversion yourself before calling WritePixelArray(RECTFMT_RAW).
Unfortunately there's no special function to do something WritePixelArray()-like-but-with-mask.
line drawing: well, that's done throug Move(rp, x1, y1) and Draw(rp, x2, y2). At the moment this is tough only with tricks usuable using a specific RGB color of your choice on a truecolor screen. By default color is pen/indexed based, both on indexed screens and truecolor screens where color is set with SetAPen().
There's no such thing for polygon functions, unfortunately. At the moment those are rendered old-style in software by graphics.library into a 1-plane bitmap and then put into destination rp/bitmap using BltPattern().
By the way, the BitMap you get with the AllocScreenBuffer command is the same as the screen's, so it's suitable for accelerated operations. You would just link the bitmap in the screenbuffer to a RastPort to use with the graphics library commands that need one. Again, that's on the Amiga. It would be nice to get this working in AROS as well.
[edit] Buffering (Double or Multi)
[edit] DBufInfo dbi
struct DBufInfo
STRUCTURE DBufInfo,0
APTR dbi_Link1
ULONG dbi_Count1
STRUCT dbi_SafeMessage,MN_SIZE
APTR dbi_UserData1
APTR dbi_Link2
ULONG dbi_Count2
STRUCT dbi_DispMessage,MN_SIZE
APTR dbi_UserData2
ULONG dbi_MatchLong
APTR dbi_CopPtr1
APTR dbi_CopPtr2
APTR dbi_CopPtr3
UWORD dbi_BeamPos1
UWORD dbi_BeamPos2
LABEL dbi_SIZEOF
Only these are to be used dbi_SafeMessage, dbi_DispMessage, dbi_UserData1 and dbi_UserData2
The following fragment shows proper double buffering synchronization: int SafeToChange=TRUE, SafeToWrite=TRUE, CurBuffer=1; struct MsgPort *ports[2]; /* reply ports for DispMessage and SafeMessage */ struct BitMap *BmPtrs[2]; struct DBufInfo *myDBI; ... allocate bitmap pointers, DBufInfo, set up viewports, etc. myDBI->dbi_SafeMessage.mn_ReplyPort=ports[0]; myDBI->dbi_DispMessage.mn_ReplyPort=ports[1]; while (! done) { if (! SafeToWrite) while(! GetMsg(ports[0])) Wait(1l<<(ports[0]->mp_SigBit)); SafeToWrite=TRUE; ... render to bitmap # CurBuffer. if (! SafeToChange) while(! GetMsg(ports[1])) Wait(1l<<(ports[1]->mp_SigBit)); SafeToChange=TRUE; WaitBlit(); /* be sure rendering has finished */ ChangeVPBitMap(vp,BmPtrs[CurBuffer],myDBI); SafeToChange=FALSE; SafeToWrite=FALSE; CurBuffer ^=1; /* toggle current buffer */ } if (! SafeToChange) /* cleanup pending messages */ while(! GetMsg(ports[1])) Wait(1l<<(ports[1]->mp_SigBit)); if (! SafeToWrite) /* cleanup */ while(! GetMsg(ports[0])) Wait(1l<<(ports[0]->mp_SigBit));
[edit] ScrollVport
here's my double buffer algorithm for cybergraphics screens: I allocate a screen twice the height (non draggable), do all my off-screen rendering in the non-viewable section (the bottom half usually) and when I'm done I do one huge blit (ClipBlit) to copy everything in the bottom half up to the top, viewable area of the screen. This works pretty quickly even on big screens with high depth.
/* multibuffering use ScrollVPort() on double height screen and sync with WaitBOV() but AVOID WaitTOF() */ /* struct Screen *screen = OpenScreenTags() with SA_HEIGHT equal to 960 Then screen->Viewport->RasInfo->RyOffset = 480; ScrollVport(screen->ViewPort); WaitBOVP(screen->ViewPort); check overscan set RTG modes do not use sprites (except mouse) use BltBitMap functions instead */ fb->bm = bm_create_disp(width, height*2, depth); newScreen.LeftEdge = 0; newScreen.TopEdge = 0; newScreen.Width = width; newScreen.Height = height; newScreen.Depth = depth; newScreen.DetailPen = 0; newScreen.BlockPen = 0; newScreen.ViewModes = NULL; newScreen.Type = CUSTOMSCREEN | SCREENQUIET | CUSTOMBITMAP; newScreen.Font = NULL; newScreen.DefaultTitle = NULL; newScreen.Gadgets = NULL; newScreen.CustomBitMap = fb->bm->bitmap; fb->screen = OpenScreen(&newScreen); /* To get bitmap pointer */ bloqueo=(APTR)LockBitMapTags(*rastportptr->BitMap,LBMI_BASEADDRESS, &direccionbase,TAG_DONE); /* made the double buffer changing the RasInfo->RyOffset value, and executing ScrollVPort and WaitBOVP */ fb->screen->ViewPort.RasInfo->RyOffset = fb->frame*fb->height; fb->frame ^= 1; fb->frameOffset = fb->frame*fb->height; ScrollVPort(&fb->screen->ViewPort); WaitBOVP(&fb->screen->ViewPort);
[edit] ChangeScreenBuffer (Aros Broken?)
Sources here
// Flip. while (ChangeScreenBuffer(video_screen, screenBuffer[drawBuffer]) == 0) { /* do nothing! */ ; }
The idea being that ChangeScreenBuffer() returns zero if it can't do the flip and probably does the signal checking stuff internally.
Struct ScreenBuffer
sb[0] = AllocScreenBuffer(screen, NULL, SB_SCREEN_BITMAP); sb[1] = AllocScreenBuffer(screen, NULL, SB_COPY_BITMAP); ports[0] = CreateMsgPort(); ports[1] = CreateMsgPort(); UBYTE toggle=0; // sb[toggle]->sb_DBufInfo->dbi_SafeMessage.mn_ReplyPort = ports[0]; sb[toggle]->sb_DBufInfo->dbi_DispMessage.mn_ReplyPort = ports[1]; Wait(1l << (ports[1]->mp_SigBit)); ChangeScreenBuffer(screen, sb[toggle]); toggle ^= 1; // Wait(1l << (ports[0]->mp_SigBit));
There is also a Commodore example doublebuffer.c to look at as well as CyberAnim13.lha.
Instead you should use WritePixelArray() or WriteLUTPixelArray() from CGX API or WriteChunkyPixels() from graphisc.library. Please note WriteChunkyPixels() is available only in Kickstart 3.1 so you have to add fallback routine for 3.0 users.
Please note you can not write directly to bitmap which dont have BMF_STANDARD flag set!
I recommend taking a look at CGX devkit. It explains what restrictions you have with RTG systems, and what you should take into account (especially that BMF_STANDARD flag in bitmaps).
[edit] ScreenMode
Cybergrafix supports some very fine functions to select a screenmode, for example BestCModeIDTagList, which gets the best screenmode for a specified tag list of screen attributes.
[edit] Pixel Formats
The following pixel formats are available:
PIXFMT_LUT8,PIXFMT_RGB15,PIXFMT_BGR15,PIXFMT_RGB15PC,PIXFMT_BGR15PC, PIXFMT_RGB16,PIXFMT_BGR16,PIXFMT_RGB16PC,PIXFMT_BGR16PC,PIXFMT_RGB24, PIXFMT_BGR24,PIXFMT_ARGB32,PIXFMT_BGRA32,PIXFMT_RGBA32
Many of this pixel formats are device specific and should not be used, recommended formats are PIXFMT_LUT8, PIXFMT_RGB16, PIXFMT_RGB24 and PIXFMT_ARGB32.
Get the value of the LBMI_PIXFMT ULONG field to get information on the colormodel you have to use for image rendering. ALL models have to be supported! The other fields give you information on the layout of the bitmap data. It should be no problem, to directly render into the bitmap anymore. Keep in mind that all this values are only valid until a subsequent call to UnLockBitmap(). Afterwards you have to lock the bitmap again. Don't hold this lock for longer than one frame ! You may use the standard graphics.library bitmap blitting calls to copy the bitmap contents to another bitmap. Copying truecolour bitmaps into indexed colour chunky bitmaps is not supported by now.
locking the bitmap to write directly into gfx card memory will require that you use cgx functions. Otherwise you'll have to use WritePixelArray like functions. Opening cgx library is good since it works with classic miggies with rtg (both cgx & p96 will work)
how to fix direct bitmap access. First i want to remove extended pixelformat definitions:
#define PIXFMT_ABGR32 100UL #define PIXFMT_0RGB32 101UL #define PIXFMT_BGR032 102UL #define PIXFMT_RGB032 103UL #define PIXFMT_0BGR32 104UL
They can confuse programs because they don't know about them. Additionally, 0RGB32 (for example) is effectively the same as ARGB32, which is the standard definition. The only question is ABGR32 format. It is expected to be other-endian version of RGBA32. And i'd like to check if some GFX cards are actually using ABGR32.
On such cards monitorclass test will report that none of known pixelformats are supported. Monitorclass does not work with extended pixelformats, as a test. If not, i'll leave ABGR32 format in the specification but give it number 14, next to RGBA32. This will be much more comfortable for monitorclass (it uses a pixelformat-indexed array in one place).
I'm just afraid to introduce a collision with some potential official CGX expansion from MorphOS authors. And unfortunately looks like i won't be able to estabilish an agreement with them about official reserving of this pixelformat.
Currently i know one driver that uses RGBA32 format. It's Windows-hosted driver. Windows GDI uses this pixelformat internally. If we imagine Windows on bigendian CPU, it will become ABGR32. At the other hand, there will never be any more bigendian CPUs in the world. Additionally, this is fixable by the software.
Getting rid of bad pixelformat codes in CGX will almost do the job. The other thing will be endianess conversion removal in some software. Pixelformat describes how pixels are laid out in RAM, so there's no need to swap bytes.
1. Rendering directly to bitmap works, but only if gfx driver supports it. Nvidia driver does, most others don't.
2. Pixel and rectformats in AROS cybergraphics always refer to the layout in memory. So RECTFMT_ARGB means 0xAA 0xRR 0xGG 0xBB. And this is so no matter if running on big endian or little endian machine. This is important to know for when running on little endian (x86) machine, because then if you have an ULONG pixel array/buffer where you write ARGB pixels into it using pixel (ULONG) accesses then in memory it will look like 0xBB 0xGG 0xRR 0xAA and you would need to use RECTFMT_BGRA.
If you have a screen using PIXFMT_RGB16PC, but WORD pixelbuffers then on AROS little endian you would need to use RECTFMT_RGB16 and only on AROS big endinan RECTFMT_RGB16PC.
Since this actual meaning of rectfmts/pixfmts is sometimes not very nice intuitive (#if AROS_BIG_ENDIAN do_this #else do_that) there are plans to modify libraries/cybergraphics.h so that it switches meaning of rectfmts pixelformats from being "based" on memory layout to being based on pixel access.
Btw, there's also RECTFMT_RAW which means same format as bitmap format.
Could AROS's CGFX implementation not just use an internal lookup for RECTFMTS that is "correct" for whatever arch?
Programmer writes RACTFMT_BGR32 -> CGFX interprets it before performing the normal action? or would it be too much of a slow down?
Is using WritePixelArray any faster to an offscreen bitmap (one not allocated to be visible) ?
Is there some function that returns whether a locking of the bitmap works ?
No. The cgx autodocs say that LockbitMap will return 0 if the bitmap could not be locked. It does not say that a bitmap which was successfully locked once, can be successfully locked every time. So a function telling whether bitmaps are lockable or not would not really help.
I would make the code check return value of LockBitMap(). if it is 0 (failure) do the workaround (WritePixelArray(), otherwise do direct bitmap access. Or if this for some reason is not optimal/working, make it configurable whether program shall (try to) use LockBitMap or not.
Otherwise for vice this means that I will have to do a 'test' lock and see if it works, if it works the code can use the lock code and otherwise it needs to use the WritePixelArray code.
Even though that may work with current drivers, I think it would be better to check the retval each time at every LockBitMap() call.
It would be so much easier if aros had some backup code for this. For example if a bitmap lock is not supported but it is called anyway, aros allocates a buffer, copies the bitmap into it, gives the caller the address of the buffer, caller does whatever, when the lock gets released aros copies back the bitmap (or if it's faster, only the changes), and releases the buffer.
As said, this would be slow. But if you want you can try that out yourself by writing your own MyLockBitMap() functions which works like you described.
It's possible, but it would be very slow as it is unknown what the LockBitMap() caller will do. What parts of the bitmap he will access. Whether he will read pixels, write pixels or do both. So the fallback code would need to alloc a pixel buffer and read the whole bitmap into it.
hardware accel: gfx drivers could override/reimplement PutAlphaImage() method if they want. But I do not know how much sense that makes or if it's possible at all as with WritePixelArrayAlpha() you have an ARGB pixel buffer which will likely be in RAM. Gfx cards if they can accelerate this might need the pixels to be in VRAM. I don't know.
This is an endianess thing. On AROS big endian it's exactly the same as in AOS/MOS. Because pixel access/component access/memory layout on big endian machine does not make a difference. If you write a 0xAARRGGBB ULONG to memory it will be 0xAA 0xRR 0xGG 0xBB in memory, too. Same thing. No problem.
On little endian machine it was something of a design decision whether the RECTFMTs/PIXFMTs should reflect pixel access based meaning (unit == pixel) or memory layout based meaning (unit == byte or component). Because RECTFMT_ARGB does not mean that one by force always uses ULONG pixel buffers and only ULONG pixel accesses. One may also use byte/component access. Or read/write such pixel buffers to/from disk.
But if you do not like that it can easily be fixed by adding something like this to your code:
- if !AROS_BIG_ENDIAN
- define PIXFMT_LUT8 0UL
- define PIXFMT_RGB15 3UL /* RGB15PC */
- define PIXFMT_BGR15 4UL /* BGR15PC */
- define PIXFMT_RGB15PC 1UL /* RGB15 */
- define PIXFMT_BGR15PC 2UL /* BGR15 */
- define PIXFMT_RGB16 7UL /* RGB16PC */
- define PIXFMT_BGR16 8UL /* BGR16PC */
- define PIXFMT_RGB16PC 5UL /* RGB16 */
- define PIXFMT_BGR16PC 6UL /* BGR16 */
- define PIXFMT_RGB24 10UL /* BGR24 */
- define PIXFMT_BGR24 9UL /* RGB24 */
- define PIXFMT_ARGB32 12UL /* BGRA32 */
- define PIXFMT_BGRA32 11UL /* ARGB32 */
- define PIXFMT_RGBA32 113UL /* ABGR32 */
- endif
At least for the moment. As said it's anyway planned to add such an option to libraries/cybergraphics.h. Then you would do a #define CGX_PIXEL_BASED_FMT_IDS (or some better name) before including libraries cybergraphics.h and be done.
Is there any way to avoid the excessive blinking and have the display behave more like normal ?
AROS at the moment has only one general non optimized RGB format conversion routine which is of course slow. There are no special one for common cases like ARGB32 to RGB15. You could do the conversion yourself and then use RECTFMT_RAW.
gray bars/blinking. Do you WritePixelArray() directly into window RastPort? It's unlikely original code (if using LockBitMap()) did that, so you shouldn't either, if that's not a whole "frame" which is written at once. Either collect pixels in a whole frame pixel buffer and do WritePixelArray() to window, once a whole frame is "ready". Or WritePixelArray() into a offscreen bitmap and BltBitMapRastPort() that to the window, once a whole frame is "ready".
Or it could be the double buffering functions (AllocScreenBuffer(), ChangeScreenBuffer(), ...) which are not yet working in AROS.
Could AROS's CGFX implementation not just use an internal lookup for RECTFMTS that is "correct" for whatever arch?
Programmer writes RACTFMT_BGR32 -> CGFX interprets it before performing the normal action? or would it be too much of a slow down? There's no need to change anything in the CGFX implementation. Only some optional redefining of the pixfmt/rectfmt ids in libraries/cybergraphics.h
What exactly does WritePixelArrayAlpha do on AROS? From BM__Hidd_BitMap__PutAlphaImage() in workbench/hidds/graphics BM_Class.c. That's the fallback code when the gfx drivers do not override reeimplement it.
[edit] CGFX Fix For Cairo
Cybergrafix supports a function to get some information about the (modified) Bitmap structure, but (sadly) you do not get the address of the video memory. So you can't write directly to the video memory.
For historical reasons related to the Cybergraphics API, graphics.hidd will set the alpha value to 0 when reading 24->32 bits or writing 32->24 bits. pixman will (correctly) interpret the pixels as being transparent. SO before we feed the pixel data in to cairo, we have to fix it by forcing the alpha value to 0xff.
Its complicated further by the fact that we have to handle the non-native endian RGB format as well. The two available formats are (effectively) xRGB and xBGR. For xBGR, we do an endianness swap then shift down before adding the alpha; that is xBGR->RBGx->0RGB->ARGB.
One day graphics.library will give us a way to request this from the graphics.hidd format conversion routines, and this won't be necessary.
Our CyberGraphX API has BltTemplateAlpha() and WritePixelArrayAlpha(), and they work. Other functions do not process alpha channel, because they are not intended to do it. For CGX ARGB pixelformat actually means 0RGB.
I think this is specifically about hardware alpha handling:
Fix up Hardware & Software Alpha channel support in graphics.library and gfx.hidd
- Must enable at least one hardware alpha capable driver to demonstrate functionality.
- Must also provide a test application demonstrating it in use.
I think currently all drivers implement PutAlphaImage via a software method. I borrowed the implementation in nouveau from nvidia if I remember correctly and the implementation is:
For each pixel of alpha enabled image, check if alpha is different than 0 or 0xff, if yes read the pixel from VRAM, apply new color with alpha and write it back. Reading from VRAM is terribly slow, thus our Decoration, which uses alpha images, seems very slow when you had several windows open and moving them around.
[edit] Examples
#include <cybergraphx/cybergraphics.h> #include <proto/intuition.h> #include <proto/cybergraphics.h> #include <stdio.h> #include <string.h> #define NUM_PIXFMT 14 static char *pixfmt_str[14]= { "LUT8", "RGB15", "BGR15", "RGB15PC", "BGR15PC", "RGB16", "BGR16", "RGB16PC", "BGR16PC", "RGB24", "BGR24", "ARGB32", "BGRA32", "RGBA32" }; int main(void) { struct Screen *scr = IntuitionBase->ActiveScreen; if (scr) { struct BitMap *bm = scr->RastPort.BitMap; LONG pixfmt; pixfmt = GetCyberMapAttr(bm, CYBRMATTR_PIXFMT); printf("Pixel Format: #%ld (%s)\n", pixfmt, ((pixfmt >= 0) && (pixfmt < NUM_PIXFMT)) ? pixfmt_str[pixfmt] : "<unknown>"); } return 0; }
#include <dos/dos.h> #include <intuition/intuition.h> #include <graphics/gfx.h> #include <cybergraphx/cybergraphics.h> #include <proto/exec.h> #include <proto/dos.h> #include <proto/graphics.h> #include <proto/cybergraphics.h> #include <proto/intuition.h> #include <math.h> #include <stdio.h> #include <stdlib.h> #define SCREENWIDTH 300 #define SCREENHEIGHT 200 #define SCREENCY (SCREENHEIGHT / 2) /***********************************************************************************/ struct IntuitionBase *IntuitionBase; struct GfxBase *GfxBase; struct Library *CyberGfxBase; struct Screen *scr; struct Window *win; struct RastPort *rp; ULONG cgfx_coltab[256]; UBYTE Keys[128]; /***********************************************************************************/ static void cleanup(char *msg) { if (msg) { printf("WritePixelArray: %s\n",msg); } if (win) CloseWindow(win); if (scr) UnlockPubScreen(0, scr); if (CyberGfxBase) CloseLibrary(CyberGfxBase); if (GfxBase) CloseLibrary((struct Library *)GfxBase); if (IntuitionBase) CloseLibrary((struct Library *)IntuitionBase); exit(0); } /***********************************************************************************/ static void openlibs(void) { if (!(IntuitionBase = (struct IntuitionBase *)OpenLibrary("intuition.library", 39))) { cleanup("Can't open intuition.library V39!"); } if (!(GfxBase = (struct GfxBase *)OpenLibrary("graphics.library", 39))) { cleanup("Can't open graphics.library V39!"); } if (!(CyberGfxBase = OpenLibrary("cybergraphics.library",0))) { cleanup("Can't open cybergraphics.library!"); } } /***********************************************************************************/ static void getvisual(void) { if (!(scr = LockPubScreen(NULL))) { cleanup("Can't lock pub screen!"); } if (GetBitMapAttr(scr->RastPort.BitMap, BMA_DEPTH) <= 8) { cleanup("Need hi or true color screen!"); } } /***********************************************************************************/ static void makewin(void) { win = OpenWindowTags(NULL, WA_CustomScreen , (IPTR)scr, WA_InnerWidth , SCREENWIDTH, WA_InnerHeight , SCREENHEIGHT, WA_Title , (IPTR)"WritePixelArray: Move mouse!", WA_DragBar , TRUE, WA_DepthGadget , TRUE, WA_CloseGadget , TRUE, WA_Activate , TRUE, WA_IDCMP , IDCMP_CLOSEWINDOW | IDCMP_RAWKEY, TAG_DONE); if (!win) cleanup("Can't open window"); rp = win->RPort; } /***********************************************************************************/ #define KC_LEFT 0x4F #define KC_RIGHT 0x4E #define KC_UP 0x4C #define KC_DOWN 0x4D #define KC_ESC 0x45 /***********************************************************************************/ static void getevents(void) { struct IntuiMessage *msg; while ((msg = (struct IntuiMessage *)GetMsg(win->UserPort))) { switch(msg->Class) { case IDCMP_CLOSEWINDOW: Keys[KC_ESC] = 1; break; case IDCMP_RAWKEY: { WORD code = msg->Code & ~IECODE_UP_PREFIX; Keys[code] = (code == msg->Code) ? 1 : 0; } break; } ReplyMsg((struct Message *)msg); } } /***********************************************************************************/ static void action(void) { static LONG tab[SCREENWIDTH * SCREENHEIGHT]; LONG x, y; LONG ar1, ar2, ar3, ar4; LONG ag1, ag2, ag3, ag4; LONG ab1, ab2, ab3, ab4; LONG r1, r2, r3, r4; LONG g1, g2, g3, g4; LONG b1, b2, b3, b4; LONG tr1, tg1, tb1; LONG tr2, tg2, tb2; LONG tr3, tg3, tb3; LONG tr4, tg4, tb4; LONG ttr1, ttg1, ttb1; LONG ttr2, ttg2, ttb2; LONG tttr, tttg, tttb; LONG col; ar1 = 0xFF; ag1 = 0xFF; ab1 = 0xFF; ar2 = 0xFF; ag2 = 0x00; ab2 = 0x00; ar3 = 0x00; ag3 = 0xFF; ab3 = 0x00; ar4 = 0x00; ag4 = 0x00; ab4 = 0xFF; r1 = 0xFF; g1 = 0xFF; b1 = 0xFF; r2 = 0xFF; g2 = 0x00; b2 = 0x00; r3 = 0x00; g3 = 0xFF; b3 = 0x00; r4 = 0x00; g4 = 0x00; b4 = 0xFF; while(!Keys[KC_ESC]) { x = scr->MouseX; if (x < 0) x = 0; else if (x >= scr->Width) x = scr->Width - 1; r1 = ar1 + (ar2 - ar1) * x / (scr->Width - 1); g1 = ag1 + (ag2 - ag1) * x / (scr->Width - 1); b1 = ab1 + (ab2 - ab1) * x / (scr->Width - 1); r2 = ar2 + (ar3 - ar2) * x / (scr->Width - 1); g2 = ag2 + (ag3 - ag2) * x / (scr->Width - 1); b2 = ab2 + (ab3 - ab2) * x / (scr->Width - 1); r3 = ar3 + (ar4 - ar3) * x / (scr->Width - 1); g3 = ag3 + (ag4 - ag3) * x / (scr->Width - 1); b3 = ab3 + (ab4 - ab3) * x / (scr->Width - 1); r4 = ar4 + (ar1 - ar4) * x / (scr->Width - 1); g4 = ag4 + (ag1 - ag4) * x / (scr->Width - 1); b4 = ab4 + (ab1 - ab4) * x / (scr->Width - 1); for(y = 0; y < SCREENHEIGHT; y ++) { for(x = 0; x < SCREENWIDTH; x++) { tr1 = r1 + (r2 - r1) * x / (SCREENWIDTH - 1); tg1 = g1 + (g2 - g1) * x / (SCREENWIDTH - 1); tb1 = b1 + (b2 - b1) * x / (SCREENWIDTH - 1); tr2 = r3 + (r4 - r3) * x / (SCREENWIDTH - 1); tg2 = g3 + (g4 - g3) * x / (SCREENWIDTH - 1); tb2 = b3 + (b4 - b3) * x / (SCREENWIDTH - 1); tr3 = r1 + (r3 - r1) * y / (SCREENHEIGHT - 1); tg3 = g1 + (g3 - g1) * y / (SCREENHEIGHT - 1); tb3 = b1 + (b3 - b1) * y / (SCREENHEIGHT - 1); tr4 = r2 + (r4 - r2) * y / (SCREENHEIGHT - 1); tg4 = g2 + (g4 - g2) * y / (SCREENHEIGHT - 1); tb4 = b2 + (b4 - b2) * y / (SCREENHEIGHT - 1); ttr1 = tr1 + (tr2 - tr1) * y / (SCREENHEIGHT - 1); ttg1 = tg1 + (tg2 - tg1) * y / (SCREENHEIGHT - 1); ttb1 = tg1 + (tg2 - tg1) * y / (SCREENHEIGHT - 1); ttr2 = tr3 + (tr4 - tr3) * x / (SCREENWIDTH - 1); ttg2 = tg3 + (tg4 - tg3) * x / (SCREENWIDTH - 1); ttb2 = tb3 + (tb4 - tb3) * x / (SCREENWIDTH - 1); tttr = (ttr1 + ttr2) / 2; tttg = (ttg1 + ttg2) / 2; tttb = (ttb1 + ttb2) / 2; #if AROS_BIG_ENDIAN col = (tttr << 16) + (tttg << 8) + tttb; #else col = (tttb << 24) + (tttg << 16) + (tttr << 8); #endif //kprintf("col[%d,%d] = %08x\n", x,y,col); tab[y * SCREENWIDTH + x] = col; } /* for(y = 0; y < SCREENHEIGHT; y ++) */ } /* for(y = 0; y < SCREENHEIGHT; y ++) */ WritePixelArray(tab, 0, 0, SCREENWIDTH * sizeof(LONG), win->RPort, win->BorderLeft, win->BorderTop, SCREENWIDTH, SCREENHEIGHT, RECTFMT_ARGB); getevents(); } /* while(!Keys[KC_ESC]) */ } /***********************************************************************************/ int main(void) { openlibs(); getvisual(); makewin(); action(); cleanup(0); return 0; /* keep compiler happy */ } /***********************************************************************************/
On the other hand to draw in the screen you could also use a generic function like: WritePixelArray8(rastportptr, 0, (bufferactual-1)*(pantalla->Height/numerodebuffers ),319,199,(UBYTE *)chunkybuffer, NULL);
WritePixelArray ( ) ? Look up on the cybergfx functions, they are very easy and do a lot of pixel conversions for you.
http://thomas-rapp.homepage.t-online.de/examples/dtwin.c
struct BitMap *copy_bitmap (struct BitMap *oldbm); BOOL load_pic (char *filename,struct BitMap **bitmap_ptr,ULONG **palette_ptr) { Object *o; long ncols = 0; struct BitMap *bitmap; ULONG *palette; o = NewDTObject (filename, DTA_GroupID,GID_PICTURE, PDTA_Remap,FALSE, TAG_END); if (!o) return (FALSE); DoDTMethod (o,NULL,NULL,DTM_PROCLAYOUT,NULL,TRUE); GetDTAttrs (o, PDTA_NumColors, (long unsigned int)&ncols, PDTA_CRegs,(long unsigned int) &palette, PDTA_DestBitMap,(long unsigned int) &bitmap, TAG_END); if (ncols) { long size = ncols * 3 * sizeof(ULONG); *palette_ptr = (ULONG*)AllocVec (size,0); *palette_ptr = AllocVec (size,0); if (*palette_ptr) memcpy (*palette_ptr,palette,size); } *bitmap_ptr = copy_bitmap (bitmap); DisposeDTObject (o); return (TRUE); }
The tags ... PDTA_Remap,TRUE,PDTA_Screen,scr,PDTA_DestMode,PMODE_V43 ... should enable palette remapping.
/* gcc -noixemul -Wall -O2 truecolortest.c -o truecolortest quit */ /* * Quick'n'dirty example on how to use MorphOS truecolor rendering * * Written by Harry "Piru" Sintonen <sintonen@iki.fi>. * Public Domain. * */ #include <graphics/rastport.h> #include <graphics/rpattr.h> #include <graphics/modeid.h> #include <intuition/screens.h> #include <proto/intuition.h> #include <proto/graphics.h> #include <proto/dos.h> #define MKRGB(r,g,b) (((r) << 16) | ((g) << 8) | (b)) #define WIDTH 640 #define HEIGHT 480 #define DEPTH 24 int main(void) { struct Screen *scr; ULONG modeid; int ret = RETURN_ERROR; modeid = BestModeID(BIDTAG_NominalWidth, WIDTH, BIDTAG_NominalHeight, HEIGHT, BIDTAG_Depth, DEPTH, TAG_DONE); if (modeid == INVALID_ID) { Printf("could not find modeid for %lux%lux%lu mode\n", WIDTH, HEIGHT, DEPTH); return RETURN_WARN; } scr = OpenScreenTags(NULL, SA_DisplayID, modeid, SA_Width, WIDTH, SA_Height, HEIGHT, SA_Depth, DEPTH, SA_Quiet, TRUE, TAG_DONE); if (scr) { struct BitMap *cookiebm; cookiebm = AllocBitMap(WIDTH, HEIGHT, 1, 0, NULL); if (cookiebm) { struct RastPort tmprp, *cookierp = &tmprp; struct RastPort *rp = &scr->RastPort; STRPTR txt; WORD len; LONG xmid = WIDTH / 2; LONG ymid = HEIGHT / 2; /* Init cookie rastport */ InitRastPort(cookierp); cookierp->BitMap = cookiebm; SetABPenDrMd(cookierp, 1, 0, JAM1); /* Set screen background */ SetRPAttrs(rp, RPTAG_PenMode, FALSE, RPTAG_FgColor, MKRGB(0xaa, 0xbb, 0xcc), TAG_DONE); RectFill(rp, 0, 0, WIDTH - 1, HEIGHT - 1); /* Draw filled ellipse to cookiebm */ SetRast(cookierp, 0); DrawEllipse(cookierp, xmid, ymid, 200, 200); Flood(cookierp, 0, xmid, ymid); /* Blast the cookie cut image to display */ SetRPAttrs(rp, RPTAG_DrMd, JAM1, RPTAG_PenMode, FALSE, RPTAG_FgColor, MKRGB(222, 100, 70), TAG_DONE); BltTemplate(cookiebm->Planes[0], 0, GetBitMapAttr(cookiebm, BMA_WIDTH) / 8, rp, 0, 0, WIDTH, HEIGHT); /* Draw filled box */ SetRPAttrs(rp, RPTAG_PenMode, FALSE, RPTAG_FgColor, MKRGB(40, 70, 90), TAG_DONE); RectFill(rp, xmid - 100, ymid - 100, xmid + 100, ymid + 100); /* Put some text on screen, too */ SetRPAttrs(rp, RPTAG_DrMd, JAM1, RPTAG_PenMode, FALSE, RPTAG_FgColor, MKRGB(255, 244, 244), TAG_DONE); txt = "MorphOS rules!"; len = strlen(txt); Move(rp, xmid - TextLength(rp, txt, len) / 2, ymid); Text(rp, txt, len); /* Wait a bit */ Delay(5 * 50); ret = RETURN_OK; FreeBitMap(cookiebm); } else Printf("no memory for %lux%lux1 cookie bitmap\n", WIDTH, HEIGHT); CloseScreen(scr); } else Printf("could not open %lux%lux%lu screen\n",WIDTH, HEIGHT, DEPTH); return ret; }
[edit] Reference
AllocCModeListTagList -- get an exec list with requested modes.
AllocCModeListTags -- Varargs stub for AllocCModeListTagList
result = AllocCModeListTagList( TagItems )
D0 A1
APTR AllocCModeListTagList( struct TagItem * );
result = AllocCModeListTags( Tag1, ... )
APTR AllocCModeListTags( Tag Tag1, ... );
Allocates a list structure which contains all requested modes (All nodes are of type CyberModeNode). See defines for more information about the structure of this nodes.
Tags available are:
CYBRMREQ_MinWidth (ULONG) - The minimum display width to let the user choose. Default is 320.
CYBRMREQ_MaxWidth (ULONG) - The maximum display width to let the user choose. Default is 1600.
CYBRMREQ_MinHeight (ULONG) - The minimum display height to let the user choose. Default is 240.
CYBRMREQ_MaxHeight (ULONG) - The maximum display height to let the user choose. Default is 1200.
CYBRMREQ_MinDepth (UWORD) - The minimum display depth to let the user choose. Default is 8.
CYBRMREQ_MaxDepth (UWORD) - The maximum display depth to let the user choose. Default is 32.
CYBRMREQ_CModelArray (UWORD *) - Array of color models which should be available for screenmode selection. Currently supported colormodels are:
PIXFMT_LUT8
PIXFMT_RGB15
PIXFMT_BGR15
PIXFMT_RGB15PC
PIXFMT_BGR15PC
PIXFMT_RGB16
PIXFMT_BGR16
PIXFMT_RGB16PC
PIXFMT_BGR16PC
PIXFMT_RGB24
PIXFMT_BGR24
PIXFMT_ARGB32
PIXFMT_BGRA32
PIXFMT_RGBA32
default is all colormodels available, nothing filtered
The result - 0 if no modes are available, a pointer to a exec list if there
are modes which fit your needs.
where A is AlphaChannel, R is Red, B is Blue, G is Green and the number is bits per...
BestCModeListTagList ()
BestCModeIDTagList( TagItems )
void FreeCModeList ( struct List * );
GetBitmapAttr()
CModeRequestTagList ()
ULONG GetCyberMapAttr(struct BitMap *bitMap, ULONG attribute)
attribute - a number telling cybergraphics which attribute of the bitmap should be returned:
CYBRMATTR_XMOD returns bytes per row of the supplied bitmap
CYBRMATTR_BPPIX returns number of bytes per pixel
CYBRMATTR_PIXFMT return the pixel format of the bitmap
CYBRMATTR_WIDTH return width of the bitmap in pixels
CYBRMATTR_HEIGHT return the height in lines
CYBRMATTR_DEPTH returns bits per pixel
CYBRMATTR_ISCYBERGFX returns TRUE if supplied bitmap is a CyberGraphics one
CYBRMATTR_ISLINEARMEM returns TRUE if the related display buffer supports linear memory access
ULONG GetCyberIDAttr(ULONG attribute, ULONG DisplayModeID)
attribute - A number telling cybergraphics which attribute of the displaymode id should be returned:
CYBRIDATTR_PIXFMT return the pixel format of the supplied screenmode id
CYBRIDATTR_WIDTH returns visible width in pixels
CYBRIDATTR_HEIGHT returns visible height in lines
CYBRIDATTR_DEPTH returns bits per pixel
CYBRIDATTR_BPPIX returns bytes per pixel
ULONG ReadRGBPixel(struct RastPort *rp, UWORD x, UWORD y)
LONG WriteRGBPixel(struct RastPort *rp, UWORD x, UWORD y, ULONG pixel)
ULONG ReadPixelArray(IPTR dst, UWORD destx, UWORD desty, UWORD dstmod, struct RastPort *rp, UWORD srcx,
UWORD srcy, UWORD width, UWORD height, UBYTE dstformat)
destRect - pointer to an array of pixels where to write the pixel
data to. The pixel format is specified in DestFormat
(DestX,DestY) - starting point in the destination rectangle
DestMod - The number of bytes per row in the destination rectangle.
RastPort - pointer to a RastPort structure
(SrcX,SrcY) - starting point in the RastPort
(SizeX,SizeY) - size of the rectangle that should be transfered
DestFormat - pixel format in the destination rectangle
Currently supported formats are:
RECTFMT_RGB 3 bytes per pixel, one byte red, one blue
and one byte green component
RECTFMT_RGBA 4 bytes per pixel, one byte red, one blue,
one byte green component and the last
byte is alpha channel information which
is 0 if the board does not support alpha
channel
RECTFMT_ARGB 4 bytes per pixel, one byte red, one blue,
one byte green component and the first
byte is alpha channel information. If the
board does not support alpha channel a
0 is returned for alpha channel information
ULONG WritePixelArray(IPTR src, UWORD srcx, UWORD srcy, UWORD srcmod, struct RastPort *rp, UWORD destx,
UWORD desty, UWORD width, UWORD height, UBYTE srcformat)
ULONG MovePixelArray(UWORD SrcX, UWORD SrcY, struct RastPort *RastPort, UWORD DstX, UWORD DstY,
UWORD SizeX, UWORD SizeY)
ULONG InvertPixelArray(struct RastPort *rp, UWORD destx, UWORD desty, UWORD width, UWORD height)
ULONG FillPixelArray(struct RastPort *rp, UWORD destx, UWORD desty, UWORD width, UWORD height, ULONG pixel)
void DoCDrawMethodTagList(struct Hook *hook, struct RastPort *rp, struct TagItem *tags)
void CVideoCtrlTagList(struct ViewPort *vp, struct TagItem *tags)
IPTR LockBitMapTagList(IPTR bitmap, struct TagItem *tags)
Tags LBMI_WIDTH (ULONG *) - points to a longword which contains the bitmap
width after a succesful call
LBMI_HEIGHT (ULONG *) - points to a longword which contains the bitmap
height after a succesful call
LBMI_DEPTH (ULONG *) - points to a longword which contains the bitmap
depth after a successful call
LBMI_PIXFMT (ULONG *) - points to a longword which contains the used
pixel format.
Possibly returned colormodels are:
PIXFMT_LUT8
PIXFMT_RGB15
PIXFMT_BGR15
PIXFMT_RGB15PC
PIXFMT_BGR15PC
PIXFMT_RGB16
PIXFMT_BGR16
PIXFMT_RGB16PC
PIXFMT_BGR16PC
PIXFMT_RGB24
PIXFMT_BGR24
PIXFMT_ARGB32
PIXFMT_BGRA32
PIXFMT_RGBA32
LBMI_BYTESPERPIX (ULONG *) - points to a longword which contains the
amount of bytes per pixel data.
LBMI_BYTESPERROW (ULONG *) - points to a longword which contains the
number of bytes per row for one bitmap
line
LBMI_BASEADDRESS (ULONG *) - points to a longword which contains the
bitmap base address. THIS ADDRESS IS ONLY
VALID INSIDE OF THE LOCK/UNLOCKBITMAP
CALL!!!!!!!!!
void UnLockBitMap(IPTR Handle)
void UnLockBitMapTagList(IPTR Handle, struct TagItem *Tags)
ULONG ExtractColor(struct RastPort *RastPort, struct BitMap *SingleMap, ULONG Colour, ULONG sX, ULONG sY,
ULONG Width, ULONG Height)
LONG ScalePixelArray(IPTR srcRect, UWORD SrcW, UWORD SrcH, UWORD SrcMod, struct RastPort *RastPort,
UWORD DestX, UWORD DestY, UWORD DestW, UWORD DestH, UBYTE SrcFormat)
LONG WriteLUTPixelArray(IPTR srcRect, UWORD SrcX, UWORD SrcY, UWORD SrcMod, struct RastPort *rp,
IPTR CTable, UWORD DestX, UWORD DestY, UWORD SizeX, UWORD SizeY, UBYTE CTabFormat)
ULONG WritePixelArrayAlpha(IPTR src, UWORD srcx, UWORD srcy, UWORD srcmod, struct RastPort *rp,
UWORD destx, UWORD desty, UWORD width, UWORD height, ULONG globalalpha)
void BltTemplateAlpha(IPTR src, LONG srcx, LONG srcmod, struct RastPort *rp, LONG destx, LONG desty,
LONG width, LONG height)
