[ongoing] DosBox AGA (v0.74)

[gulliver]

Originally Posted by NovaCoder

Yep I'll be adding RTG support when it runs properly, it's pretty easy to add rtg

[Reido]

Originally Posted by NovaCoder

I've recently had some time to kill so I thought I'd revisit my old AGA port.

Great news, looking forward to testing!

[AMike]

Originally Posted by NovaCoder

I'll upload something to the zone for testing on real hardware when I get it working properly. Unfortunately I don't have a real Amiga anymore so I can only test with WinUAE these days.

Great that you are back. Looking forward for the RTG version.

[turrican3]

Originally Posted by NovaCoder

I've recently had some time to kill so I thought I'd revisit my old AGA port. After spending time analysing the 0.74 source code I can see that the render pipeline is expensive. I'm currently working on a new build using a native Amiga render pipeline to see if I can get it usable for some (slow-paced) DOS games.

This will be based on v6.02 with various fixes from newer versions mixed in where performance isn't impacted too much.

I'll upload something to the zone for testing on real hardware when I get it working properly. Unfortunately I don't have a real Amiga anymore so I can only test with WinUAE these days.

hi Nova,
could you add the support of a second mouse,please ???
I'm happy to see you back !!

[NovaCoder]

Thanks guys

Thought the coders might like to see the before and after shots

Original SDL based renderer:

PHP Code:


enum SCREEN_TYPES    { 
    SCREEN_SURFACE,
    SCREEN_SURFACE_DDRAW,
    SCREEN_OVERLAY,
    SCREEN_OPENGL
};

enum PRIORITY_LEVELS {
    PRIORITY_LEVEL_LOWER,
    PRIORITY_LEVEL_NORMAL,
    PRIORITY_LEVEL_HIGHER,
    PRIORITY_LEVEL_HIGHEST
};


struct SDL_Block {
    bool active;                            //If this isn't set don't draw
    bool updating;
    struct {
        Bit32u width;
        Bit32u height;
        Bitu flags;
        GFX_Modes mode;
        double scalex,scaley;
        GFX_ResetCallBack reset;
    } draw;
    bool wait_on_error;
    struct {
        Bit32u width,height,bpp;
        bool fixed;
        bool fullscreen;
        bool doublebuf;
        SCREEN_TYPES type;
        SCREEN_TYPES want_type;
        double hwscale;
    } desktop;
    struct {
        PRIORITY_LEVELS focus;
        PRIORITY_LEVELS nofocus;
    } priority;
    SDL_Rect clip;
    SDL_Surface * surface;
    SDL_Overlay * overlay;
    SDL_cond *cond;
    struct {
        bool autolock;
        bool autoenable;
        bool requestlock;
        bool locked;
        Bitu sensitivity;
    } mouse;
};

static SDL_Block sdl;


<SNIP>

bool GFX_StartUpdate(Bit8u * & pixels,Bitu & pitch) {
    if (!sdl.active || sdl.updating) return false;
    sdl.updating=true;
    switch (sdl.desktop.type) {
    case SCREEN_SURFACE:
        if (SDL_MUSTLOCK(sdl.surface)) {
            if (SDL_LockSurface(sdl.surface)) {
//                LOG_MSG("SDL Lock failed");
                sdl.updating=false;
                return false;
            }
        }
        pixels=(Bit8u *)sdl.surface->pixels;
        pixels+=sdl.clip.y*sdl.surface->pitch;
        pixels+=sdl.clip.x*sdl.surface->format->BytesPerPixel;
        pitch=sdl.surface->pitch;
        return true;
#if defined(HAVE_DDRAW_H) && defined(WIN32)
    case SCREEN_SURFACE_DDRAW:
        if (SDL_LockSurface(sdl.blit.surface)) {
//            LOG_MSG("SDL Lock failed");
            sdl.updating=false;
            return false;
        }
        pixels=(Bit8u *)sdl.blit.surface->pixels;
        pitch=sdl.blit.surface->pitch;
        return true;
#endif
    case SCREEN_OVERLAY:
        SDL_LockYUVOverlay(sdl.overlay);
        pixels=(Bit8u *)*(sdl.overlay->pixels);
        pitch=*(sdl.overlay->pitches);
        return true;
#if C_OPENGL
    case SCREEN_OPENGL:
        pixels=(Bit8u *)sdl.opengl.framebuf;
        pitch=sdl.opengl.pitch;
        return true;
#endif
    }
    return false;
}

void GFX_EndUpdate(void) {
    int ret;
    if (!sdl.updating) return;
    sdl.updating=false;
    switch (sdl.desktop.type) {
    case SCREEN_SURFACE:
        if (SDL_MUSTLOCK(sdl.surface)) {
            SDL_UnlockSurface(sdl.surface);
        }
        SDL_Flip(sdl.surface);
        break;
#if defined(HAVE_DDRAW_H) && defined(WIN32)
    case SCREEN_SURFACE_DDRAW:
        if (SDL_MUSTLOCK(sdl.blit.surface)) {
            SDL_UnlockSurface(sdl.blit.surface);
        }
        ret=IDirectDrawSurface3_Blt(
            sdl.surface->hwdata->dd_writebuf,&sdl.blit.rect,
            sdl.blit.surface->hwdata->dd_surface,0,
            DDBLT_WAIT, NULL);
        switch (ret) {
        case DD_OK:
            break;
        case DDERR_SURFACELOST:
            IDirectDrawSurface3_Restore(sdl.blit.surface->hwdata->dd_surface);
            break;
        default:
            LOG_MSG("DDRAW:Failed to blit, error %X",ret);
        }
        SDL_Flip(sdl.surface);
        break;
#endif
    case SCREEN_OVERLAY:
        SDL_UnlockYUVOverlay(sdl.overlay);
        SDL_DisplayYUVOverlay(sdl.overlay,&sdl.clip);
        break;
#if C_OPENGL
    case SCREEN_OPENGL:
        glBindTexture(GL_TEXTURE_2D, sdl.opengl.texture);
        glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 
                sdl.draw.width, sdl.draw.height, GL_BGRA_EXT,
                GL_UNSIGNED_INT_8_8_8_8_REV, sdl.opengl.framebuf);
        glCallList(sdl.opengl.displaylist);
        SDL_GL_SwapBuffers();
        break;
#endif

    }
}


void GFX_SetPalette(Bitu start,Bitu count,GFX_PalEntry * entries) {
    /* I should probably not change the GFX_PalEntry :) */
    if (sdl.surface->flags & SDL_HWPALETTE) {
        if (!SDL_SetPalette(sdl.surface,SDL_PHYSPAL,(SDL_Color *)entries,start,count)) {
            E_Exit("SDL:Can't set palette");
        }
    } else {
        if (!SDL_SetPalette(sdl.surface,SDL_LOGPAL,(SDL_Color *)entries,start,count)) {
            E_Exit("SDL:Can't set palette");
        }
    }
}


static Bit8u laltstate = SDL_KEYUP;
static Bit8u raltstate = SDL_KEYUP;


void GFX_Events() {
    SDL_Event event;
    while (SDL_PollEvent(&event)) {
        switch (event.type) {
        case SDL_ACTIVEEVENT:
            if (event.active.state & SDL_APPINPUTFOCUS) {
                if (event.active.gain) {
                    if (sdl.desktop.fullscreen && !sdl.mouse.locked)
                        CaptureMouse();    
                    SetPriority(sdl.priority.focus);
                } else {
                    if (sdl.mouse.locked) 
                        CaptureMouse();    
                    SetPriority(sdl.priority.nofocus);
                }
            }
            break;
        case SDL_MOUSEMOTION:
            HandleMouseMotion(&event.motion);
            break;
        case SDL_MOUSEBUTTONDOWN:
        case SDL_MOUSEBUTTONUP:
            HandleMouseButton(&event.button);
            break;
        case SDL_VIDEORESIZE:
//            HandleVideoResize(&event.resize);
            break;
        case SDL_QUIT:
            throw(0);
            break;
#ifdef WIN32
        case SDL_KEYDOWN:
        case SDL_KEYUP:
            // ignore event alt+tab
            if (event.key.keysym.sym==SDLK_LALT) laltstate = event.key.type;
            if (event.key.keysym.sym==SDLK_RALT) raltstate = event.key.type;
            if (((event.key.keysym.sym==SDLK_TAB)) &&
                ((laltstate==SDL_KEYDOWN) || (raltstate==SDL_KEYDOWN))) break;
#endif

        default:
            void MAPPER_CheckEvent(SDL_Event * event);
            MAPPER_CheckEvent(&event);
        }
    }
} 


New native AGA version:

PHP Code:


/** Global Hardware window */
static struct Window *_hardwareWindow = NULL;


/** Hardware screen */
static struct Screen *_hardwareScreen = NULL;


// Hardware double buffering.
static struct ScreenBuffer *_hardwareScreenBuffer[2];
static BYTE _currentScreenBuffer = 0;



// AGA C2P.
static void *c2p[2] = {NULL, NULL};

static UWORD emptypointer[] = {
  0x0000, 0x0000,    /* reserved, must be NULL */
  0x0000, 0x0000,     /* 1 row of image data */
  0x0000, 0x0000    /* reserved, must be NULL */
};

 

// These settings are for PAL HIGH RES!
#define AGA_VIDEO_DEPTH         8
#define AGA_MAX_VIDEO_WIDTH     640
#define AGA_MAX_VIDEO_HEIGHT    512



struct AGA_Block {
    bool active;                            //If this isn't set don't draw
    struct {
        Bit32u width;
        Bit32u height;
    } draw;
    
    Bit8u * surface;
};

static AGA_Block aga;




void GFX_SetTitle(Bits cycles,Bits frameskip,bool paused){
    // TODO - will be used by RTG version.
}



static struct Screen* CreateHardwareScreen(ULONG modeId, int width, int height) {
    return OpenScreenTags(NULL,
                     SA_Depth, AGA_VIDEO_DEPTH,
                     SA_DisplayID, modeId,
                     SA_Top, 0,
                     SA_Left, 0,
                     SA_Width, width,
                     SA_Height, height,
                     SA_Type, CUSTOMSCREEN,
                     SA_Quiet, TRUE,
                     SA_ShowTitle, FALSE,
                     SA_Draggable, FALSE,
                     SA_Exclusive, TRUE,
                     SA_AutoScroll, FALSE,
                     TAG_END);
}

static struct Window* CreateHardwareWindow(int width, int height) {
    return OpenWindowTags(NULL,
                          WA_Left, 0,
                        WA_Top, 0,
                        WA_Width, width,
                        WA_Height, height,
                        SA_AutoScroll, FALSE,
                        WA_CustomScreen, (ULONG)_hardwareScreen,
                        WA_Backdrop, TRUE,
                        WA_Borderless, TRUE,
                        WA_DragBar, FALSE,
                        WA_Activate, TRUE,
                        WA_SimpleRefresh, TRUE,
                        WA_NoCareRefresh, TRUE,
                        WA_ReportMouse, TRUE,
                        WA_RMBTrap, TRUE,
                          WA_IDCMP, IDCMP_RAWKEY|IDCMP_MOUSEMOVE|IDCMP_MOUSEBUTTONS,
                          TAG_END);  
}

static void GFX_ResetMode() {
    aga.draw.width=0;
    aga.draw.height=0;

    if (aga.surface) {
        free(aga.surface);
        aga.surface = NULL;
    }

    if (_hardwareWindow) {
        ClearPointer(_hardwareWindow);
        CloseWindow(_hardwareWindow);
        _hardwareWindow = NULL;
    }

    if (_hardwareScreenBuffer[0]) {
        ChangeScreenBuffer (_hardwareScreen, _hardwareScreenBuffer[0]);
        WaitTOF();
        WaitTOF();
        FreeScreenBuffer (_hardwareScreen, _hardwareScreenBuffer[0]);
        _hardwareScreenBuffer[0] = NULL;
    }

    if (_hardwareScreenBuffer[1]) {
        FreeScreenBuffer (_hardwareScreen, _hardwareScreenBuffer[1]);
        _hardwareScreenBuffer[1] = NULL;
    }


    if (_hardwareScreen) {
        CloseScreen(_hardwareScreen);
        _hardwareScreen = NULL;
    }

    if (c2p[0]) {
        c2p8_deinit_stub(c2p[0]);
        c2p[0] = NULL;
    }

    if (c2p[1]) {
        c2p8_deinit_stub(c2p[1]);
        c2p[1] = NULL;
    }
    
    aga.active = false;
}
 

void GFX_SetSize(Bitu width, Bitu height) {
    ULONG modeId = INVALID_ID;
    DisplayInfoHandle handle;
    struct DisplayInfo dispinfo;
    struct DimensionInfo dimsinfo;
    
#ifdef C_DEBUG        
    LOG_DEBUG("GFX_SetSize(width=%d, height=%d)\n", width, height);
#endif

    // Check args.
    if (width == 0) {
        LOG_ERROR("Width cannot be zero");
        E_Exit("Failed to create a rendering output");
    }
    
    if (height == 0) {
        LOG_ERROR("Height cannot be zero");
        E_Exit("Failed to create a rendering output");
    }
    
       if (width > AGA_MAX_VIDEO_WIDTH) {
        LOG_ERROR("Width cannot be greater than %d", AGA_MAX_VIDEO_WIDTH);
        E_Exit("Failed to create a rendering output");
    }

    if (height > AGA_MAX_VIDEO_HEIGHT) {
        LOG_ERROR("Height cannot be greater than %d", AGA_MAX_VIDEO_HEIGHT);
        E_Exit("Failed to create a rendering output");
    }
    
    if ((aga.draw.width == width) && (aga.draw.height == height)) {
        // Nothing to do...
#ifdef C_DEBUG
        LOG_DEBUG("GFX_SetSize: Skipping pointless resolution switch");
#endif
        return;
    }

    // Free any existing mode.
    GFX_ResetMode();
    
    
    // Reset.
    aga.draw.width=width;
    aga.draw.height=height;

    // Automatically choose the best mode.
    modeId = BestModeID(BIDTAG_NominalWidth, width,
                    BIDTAG_NominalHeight, height,
                    BIDTAG_DesiredWidth, width,
                    BIDTAG_DesiredHeight, height,
                    BIDTAG_Depth, AGA_VIDEO_DEPTH,
                    BIDTAG_MonitorID, PAL_MONITOR_ID,
                    TAG_END);


    // Verify the mode choosen.
    if (modeId != INVALID_ID) {
        if ((handle = FindDisplayInfo(modeId)) == NULL) {
            LOG_ERROR("Couldn't find Display Info for requested mode");
             E_Exit("Failed to create a rendering output");
        }

        if (GetDisplayInfoData(handle, (UBYTE *)&dispinfo, sizeof(dispinfo), DTAG_DISP,0) == 0) {
            LOG_ERROR("Couldn't get Display Info Data for requested mode");
             E_Exit("Failed to create a rendering output");
        }

        if (GetDisplayInfoData(handle, (UBYTE *)&dimsinfo, sizeof(dimsinfo), DTAG_DIMS, 0) == 0) {
            LOG_ERROR("Couldn't get Display Info Data for requested mode");
             E_Exit("Failed to create a rendering output");
        }

        if (dimsinfo.MaxDepth != AGA_VIDEO_DEPTH) {
           modeId = INVALID_ID;
        }

        if ((dimsinfo.Nominal.MaxX + 1) != width) {
           modeId = INVALID_ID;
        }

        if ((dimsinfo.Nominal.MaxY + 1) < height) {
           modeId = INVALID_ID;
        }
    }

    if (modeId == INVALID_ID) {
        LOG_ERROR("Couldn't find a Screen Mode for the requested resolution (%d * %d)\n", width, height);
        E_Exit("Failed to create a rendering output");
    }
    
    
    _hardwareScreen = CreateHardwareScreen(modeId, width, height);
    if (!_hardwareScreen ) {
        LOG_ERROR("Couldn't create a Hardware Screen for the requested mode");
        E_Exit("Failed to create a rendering output");
    }     
    
    // Setup double buffering.
    _hardwareScreenBuffer[0] = AllocScreenBuffer (_hardwareScreen, NULL, SB_SCREEN_BITMAP);
    if (!_hardwareScreenBuffer[0]) {
        LOG_ERROR("Couldn't set up double buffering for the requested mode");
        E_Exit("Failed to create a rendering output");
    }
    
    _hardwareScreenBuffer[1] = AllocScreenBuffer (_hardwareScreen, NULL, 0);    
    if (!_hardwareScreenBuffer[1]) {
        LOG_ERROR("Couldn't set up double buffering for the requested mode");
        E_Exit("Failed to create a rendering output");
    }


    _currentScreenBuffer = 1;

    // Setup C2P.    
    c2p[0] = c2p8_reloc_stub(_hardwareScreenBuffer[0]->sb_BitMap);
    c2p[1] = c2p8_reloc_stub(_hardwareScreenBuffer[1]->sb_BitMap);
    
    
    // Create the hardware window.
    _hardwareWindow = CreateHardwareWindow(width, height);
    if (!_hardwareWindow) {
        LOG_ERROR("Couldn't create a Hardware Window for the requested mode");
        E_Exit("Failed to create a rendering output");
    }
    
    SetPointer(_hardwareWindow, emptypointer, 1, 1, 0, 0);
    
    // Create a new chunky back buffer.
    aga.surface = (Bit8u *)malloc(width * height);
    if (!aga.surface) {
        LOG_ERROR("Couldn't allocate buffer for requested mode");
        E_Exit("Failed to create a rendering output");
    }

    memset(aga.surface, 0, width * height);
    
    
    // Good to go!
    aga.active = true;
}

void GFX_UpdatePalette(Bitu numberOfEntries, Bitu paletteDirtyStart,  GFX_PalEntry * currentPalette) {

    int i;
    ULONG  agaPalette[numberOfEntries*3 + 2];
    
    if (aga.active) {
        for ( i=0; i < numberOfEntries; i++ ) {
            agaPalette[i*3+1] = currentPalette[paletteDirtyStart + i].r<<24;
            agaPalette[i*3+2] = currentPalette[paletteDirtyStart + i].g<<24;
            agaPalette[i*3+3] = currentPalette[paletteDirtyStart + i].b<<24;
        }
    
        agaPalette[0] = (numberOfEntries << 16) + paletteDirtyStart;
    
        // Terminator: NEEDED
        agaPalette[((numberOfEntries * 3) + 1)] = 0x00000000;
    
        LoadRGB32(&_hardwareScreen->ViewPort, agaPalette);
    }
}

bool GFX_StartUpdate(Bit8u * & pixels) {
    if (!aga.active) {
        // Just in case GFX_SetSize() hasn't been called yet.
       return false;
    }

     pixels = aga.surface;

     return true;
}

void GFX_EndUpdate(void) {
    c2p8_stub(c2p[_currentScreenBuffer], _hardwareScreenBuffer[_currentScreenBuffer]->sb_BitMap, (UBYTE*)aga.surface, (aga.draw.width * aga.draw.height));

    if (ChangeScreenBuffer(_hardwareScreen, _hardwareScreenBuffer[_currentScreenBuffer])) {
        // Flip.
       _currentScreenBuffer = _currentScreenBuffer ^ 1;
    }
}


void GFX_Events() {
    if (aga.active) {       
        struct IntuiMessage *imsg;
        
        while (imsg = (struct IntuiMessage *)GetMsg(_hardwareWindow->UserPort)) {
    
            ReplyMsg((struct Message *)imsg);
     
            switch (imsg->Class) {  
                case IDCMP_MOUSEMOVE:
                    // TODO
                    break;
     
    
                case IDCMP_MOUSEBUTTONS:
                    // TODO
                    break;
 
 
                case IDCMP_RAWKEY:
                    int code = imsg->Code;
    
    
                    bool pressed = (code & IECODE_UP_PREFIX) ? false : true;
    
                    code &= ~IECODE_UP_PREFIX;
                    
                    
    
                    KEYBOARD_AddKey(keycode_lookup[code], pressed);
            }
        }
    }    
}

static void GUI_ShutDown(Section * sec) {
    LOG_DEBUG("GUI shutdown called");

    // Free any existing mode.
    GFX_ResetMode();
} 


[turrican3]

Sorry for not to be a coder (really),
could you explain what you did ???

[NovaCoder]

Originally Posted by turrican3

Sorry for not to be a coder (really),
could you explain what you did ???

I hacked it

Basically I removed all unnecessary code and cleaned it up to make it look pretty (which should improve performance hopefully).

In the zone is a rough Alpha build of version 0.69, can someone with real hardware compare it to the version 0.74 in the same ZIP file. I've included the DOTT DOS VGA demo in the same ZIP file.

Both config files 0.69 an 0.74 are the same and both are built to target 040's.

Note that sound and mouse input is disabled on 0.69 at the moment (very early build)

[Reido]

Thanks Nova! I'll test this this evening A1260/80mhz 64mb

Video of 0.69 build:
[ Show youtube player ]

Video of 0.74 build:
[ Show youtube player ]

Video of original DosBox__AGA
[ Show youtube player ]

[trixster]

I'll do some more testing this weekend too

[NovaCoder]

Cool

I've managed to get the executable size down from 24 MB to about 8 MB and I should be able to get it to about 6 MB.

It's still very rough and buggy BTW

[1time]

NovaCoder: really nice work. This will be really appreciated!

[turrican3]

hi nova i tested it with winuae,I just replaced your old dosboxaga with the new one.
error 80000008.
A privilege error ??? I tried without fpu but no luck.

[Reido]

Tested, see post #108

[NovaCoder]

Originally Posted by Reido

Tested, see post #108

Thanks for the videos, that really helps me to see how they run. Screen update routine looks very different and seems to be skipping less frames on v0.69 but the core looks slightly slower than 0.74 for some reason.

I'll see if I can do anything to make it faster, I'm sure there's some more speed in the code that can be unlocked.


Update: OK, tried again with v0.74.010. In the Zone (compiled for 060 this time).

On a real Amiga you will need to play around with the number of cycles, if you set it too high it will cause stalling (note - max and auto aren't working properly).

Original exe was 26MB, now down to 2MB

[NovaCoder]

v0.74.011 now in the zone with optimized cpu loop with gives me better results in WinUAE.

I've managed to break the EGA code somehow, only VGA is working.

I think to be able to run something like DOTT properly you'll need to set it at least 3000 cycles but if your CPU stalls it will actually run better at less than 3000.

[Reido]

Top work Nova! I'm away for a day or two so should be able to test soon.

[NovaCoder]

Originally Posted by Reido

Top work Nova! I'm away for a day or two so should be able to test soon.

No worries, mate

I'll keep uploading newer versions to the Zone as I go..

I've got an idea about optimizing the core using GCC 6, not much else I can do after that to improve performance.

Check out the improvement to video performance.

[AMike]

I've made a quick'n dirty test with the latest version - looks good But we need more horsepower.

[NovaCoder]

Cool, is it any faster than than the build on AmiNet?

You are correct about needing more horsepower! It dawned on my last night (I'm a bit slow) that even if I've improved the rendering performance by 100% it will only mean a real world improvement of 33% because of the frame skipping (3 frame skips seems perfect to me, the original AmiNet release was set to 10 which is too high). And that 33% render performance will be less overall because rendering the image might only be 10% of the entire frame time as most of the time will be taken to actually emulate the X86 CPU.

I've heard rumors that the Vampire A1200 exists now? Maybe that will finally provide enough CPU power to make this port usable.

I've done pretty much everything I can to make this thing more efficient for real hardware but I'm out of ideas now. I'm just impressed with myself that I reduced the executable size down from over 20 MB of the the original SDL build to 2 MB using native code which is much more 'Amiga' like

I've tried using GGC 6 to improve performance but can't get it to build anything that runs

[Pyromania]

Go NovaCoder, go!