- Add a new UploadToScratch kaa hook for putting the data for a single

    pixmap into temporary offscreen storage. Subsequent UploadToScratch may
    clobber the data of previous ones. This allows hardware acceleration of
    composite operations on glyphs.
- Add a new UploadToScreen kaa hook for doing the actual moving of data to
    framebuffer. This would allow us to do things like hostdata blits or
    memcpy to agp and then blit.
- Add an UploadToScreen on ATI which is just memcpy, but which will be
    replaced with a hostdata blit soon.
- Add UploadToScratch on ATI and reserve 64k of scratch space. This
    provided a 3x speedup of rgb24text on my Radeon.

6 files changed, 99 insertions, 8 deletions

diff --git a/hw/kdrive/ati/ati.c b/hw/kdrive/ati/ati.c
index bdd3c4941..1e6fa18b8 100644
--- a/hw/kdrive/ati/ati.c
+++ b/hw/kdrive/ati/ati.c
@@ -269,9 +269,7 @@ ATIScreenInit(KdScreenInfo *screen)
 	if (atic->use_fbdev) {
 		success = fbdevScreenInitialize(screen,
 						&atis->backend_priv.fbdev);
-		screen->memory_size = min(atic->backend_priv.fbdev.fix.smem_len,
-		    8192 * screen->fb[0].byteStride);
-		/*screen->memory_size = atic->backend_priv.fbdev.fix.smem_len;*/
+		screen->memory_size = atic->backend_priv.fbdev.fix.smem_len;
 		screen->off_screen_base =
 		    atic->backend_priv.fbdev.var.yres_virtual *
 		    screen->fb[0].byteStride;
@@ -285,12 +283,25 @@ ATIScreenInit(KdScreenInfo *screen)
 		    &atis->backend_priv.vesa);
 	}
 #endif
+
 	if (!success) {
 		screen->driver = NULL;
 		xfree(atis);
 		return FALSE;
 	}
 
+	/* Reserve a scratch area.  It'll be used for storing glyph data during
+	 * Composite operations, because glyphs aren't in real pixmaps and thus
+	 * can't be migrated.
+	 */
+	atis->scratch_size = 65536;	/* big enough for 128x128@32bpp */
+	if (screen->off_screen_base + atis->scratch_size > screen->memory_size)
+		atis->scratch_size = 0;
+	else {
+		atis->scratch_offset = screen->off_screen_base;
+		screen->off_screen_base += atis->scratch_size;
+	}
+
 	return TRUE;
 }
 
diff --git a/hw/kdrive/ati/ati.h b/hw/kdrive/ati/ati.h
index cacc264dd..5d10d9253 100644
--- a/hw/kdrive/ati/ati.h
+++ b/hw/kdrive/ati/ati.h
@@ -147,6 +147,9 @@ typedef struct _ATIScreenInfo {
 	Bool using_dri;
 	Bool using_dma;
 
+	int scratch_offset;
+	int scratch_size;
+
 #ifdef USE_DRI
 	drmSize         registerSize;
 	drmHandle       registerHandle;
diff --git a/hw/kdrive/ati/ati_draw.c b/hw/kdrive/ati/ati_draw.c
index 1e2879443..250d1b337 100644
--- a/hw/kdrive/ati/ati_draw.c
+++ b/hw/kdrive/ati/ati_draw.c
@@ -378,6 +378,53 @@ RadeonSwitchTo3D(void)
 	ADVANCE_RING();
 }
 
+static Bool
+ATIUploadToScreen(PixmapPtr pDst, char *src, int src_pitch)
+{
+	int i;
+	char *dst;
+	int dst_pitch;
+	int bytes;
+
+	dst = pDst->devPrivate.ptr;
+	dst_pitch = pDst->devKind;
+	bytes = src_pitch < dst_pitch ? src_pitch : dst_pitch;
+
+	KdCheckSync(pDst->drawable.pScreen);
+
+	for (i = 0; i < pDst->drawable.height; i++) {
+		memcpy(dst, src, bytes);
+		dst += dst_pitch;
+		src += src_pitch;
+	}
+
+	return TRUE;
+}
+
+
+static Bool
+ATIUploadToScratch(PixmapPtr pSrc, PixmapPtr pDst)
+{
+	KdScreenPriv(pSrc->drawable.pScreen);
+	ATIScreenInfo(pScreenPriv);
+	int dst_pitch;
+
+	dst_pitch = (pSrc->drawable.width * pSrc->drawable.bitsPerPixel / 8 +
+	    atis->kaa.offscreenPitch - 1) & ~(atis->kaa.offscreenPitch - 1);
+	
+	if (dst_pitch * pSrc->drawable.height > atis->scratch_size)
+		ATI_FALLBACK(("Pixmap too large for scratch (%d,%d)\n",
+		    pSrc->drawable.width, pSrc->drawable.height));
+
+	memcpy(pDst, pSrc, sizeof(*pDst));
+	pDst->devKind = dst_pitch;
+	pDst->devPrivate.ptr = atis->scratch_offset +
+	    pScreenPriv->screen->memory_base;
+
+	return ATIUploadToScreen(pDst, pSrc->devPrivate.ptr, pSrc->devKind);
+}
+
+
 #endif /* USE_DRI */
 
 static Bool
@@ -392,7 +439,7 @@ R128GetDatatypePict(CARD32 format, CARD32 *type)
 		return TRUE;
 	}
 
-	ErrorF ("Unsupported format: %x\n", format);
+	ATI_FALLBACK(("Unsupported format: %x\n", format));
 
 	return FALSE;
 }
@@ -420,7 +467,7 @@ ATIGetDatatypeBpp(int bpp, CARD32 *type)
 		*type = R128_DATATYPE_ARGB_8888;
 		return TRUE;
 	default:
-		ErrorF("Unsupported bpp: %x\n", bpp);
+		ATI_FALLBACK(("Unsupported bpp: %x\n", bpp));
 		return FALSE;
 	}
 }
@@ -496,6 +543,9 @@ ATIDrawInit(ScreenPtr pScreen)
 			atis->kaa.DoneBlend = R128DoneBlendMMIO;
 		}
 	}
+	atis->kaa.UploadToScreen = ATIUploadToScreen;
+	if (atis->scratch_size != 0)
+		atis->kaa.UploadToScratch = ATIUploadToScratch;
 	atis->kaa.DoneSolid = ATIDoneSolid;
 	atis->kaa.DoneCopy = ATIDoneCopy;
 	atis->kaa.flags = KAA_OFFSCREEN_PIXMAPS;
diff --git a/hw/kdrive/src/kaa.c b/hw/kdrive/src/kaa.c
index a13e258aa..5e67368a8 100644
--- a/hw/kdrive/src/kaa.c
+++ b/hw/kdrive/src/kaa.c
@@ -131,12 +131,12 @@ kaaPixmapAllocArea (PixmapPtr pPixmap)
 static void
 kaaMoveInPixmap (PixmapPtr pPixmap)
 {
+    ScreenPtr	pScreen = pPixmap->drawable.pScreen;
+    KaaScreenPriv (pScreen);
     int dst_pitch, src_pitch, bytes;
     unsigned char *dst, *src;
     int i;
 
-    KdCheckSync (pPixmap->drawable.pScreen);
-    
     DBG_MIGRATE (("-> 0x%08x (0x%x) (%dx%d)\n",
 		  pPixmap->drawable.id,
 		  KaaGetPixmapPriv(pPixmap)->area ? 
@@ -150,11 +150,19 @@ kaaMoveInPixmap (PixmapPtr pPixmap)
     if (!kaaPixmapAllocArea (pPixmap))
 	return;
     
+    if (pKaaScr->info->UploadToScreen)
+    {
+	if (pKaaScr->info->UploadToScreen(pPixmap, src, src_pitch))
+	    return;
+    }
+
     dst = pPixmap->devPrivate.ptr;
     dst_pitch = pPixmap->devKind;
     
     bytes = src_pitch < dst_pitch ? src_pitch : dst_pitch;
 
+    KdCheckSync (pPixmap->drawable.pScreen);
+
     i = pPixmap->drawable.height;
     while (i--) {
 	memcpy (dst, src, bytes);
diff --git a/hw/kdrive/src/kaapict.c b/hw/kdrive/src/kaapict.c
index 88bebbf91..d8badbd08 100644
--- a/hw/kdrive/src/kaapict.c
+++ b/hw/kdrive/src/kaapict.c
@@ -319,6 +319,7 @@ kaaTryDriverBlend(CARD8		op,
     int nbox;
     int src_off_x, src_off_y, dst_off_x, dst_off_y;
     PixmapPtr pSrcPix, pDstPix;
+    struct _Pixmap srcScratch;
 
     xDst += pDst->pDrawable->x;
     yDst += pDst->pDrawable->y;
@@ -339,7 +340,19 @@ kaaTryDriverBlend(CARD8		op,
     
     pSrcPix = kaaGetOffscreenPixmap (pSrc->pDrawable, &src_off_x, &src_off_y);
     pDstPix = kaaGetOffscreenPixmap (pDst->pDrawable, &dst_off_x, &dst_off_y);
-    if (!pSrcPix || !pDstPix) {
+
+    if (!pDstPix) {
+	REGION_UNINIT(pDst->pDrawable->pScreen, &region);
+	return 0;
+    }
+
+    if (!pSrcPix && pKaaScr->info->UploadToScratch) {
+	if ((*pKaaScr->info->UploadToScratch) ((PixmapPtr) pSrc->pDrawable,
+					       &srcScratch))
+	    pSrcPix = &srcScratch;
+    }
+
+    if (!pSrcPix) {
 	REGION_UNINIT(pDst->pDrawable->pScreen, &region);
 	return 0;
     }
diff --git a/hw/kdrive/src/kdrive.h b/hw/kdrive/src/kdrive.h
index bee6393b7..bb9bc238d 100644
--- a/hw/kdrive/src/kdrive.h
+++ b/hw/kdrive/src/kdrive.h
@@ -363,6 +363,12 @@ typedef struct _KaaScreenInfo {
 			     int	width,
 			     int	height);
     void	(*DoneComposite) (void);
+
+    Bool        (*UploadToScreen) (PixmapPtr		pDst,
+				   char			*src,
+				   int			src_pitch);
+    Bool        (*UploadToScratch) (PixmapPtr		pSrc,
+				   PixmapPtr		pDst);
 } KaaScreenInfoRec, *KaaScreenInfoPtr;
 
 #define KAA_OFFSCREEN_PIXMAPS (1 << 0)