Merge branch 'master' into exa-damagetrack

Conflicts:

	exa/exa_accel.c
	exa/exa_migration.c

7 files changed, 169 insertions, 65 deletions

diff --git a/exa/Makefile.am b/exa/Makefile.am
index 98db0271f..e2f7ed302 100644
--- a/exa/Makefile.am
+++ b/exa/Makefile.am
@@ -12,7 +12,7 @@ INCLUDES = \
 	$(XORG_INCS) \
 	-I$(srcdir)/../miext/cw
 
-AM_CFLAGS = $(XORG_CFLAGS) @SERVER_DEFINES@ @LOADER_DEFINES@ $(DIX_CFLAGS)
+AM_CFLAGS = $(XORG_CFLAGS) $(DIX_CFLAGS)
 
 libexa_la_SOURCES = \
 	exa.c \
diff --git a/exa/exa.c b/exa/exa.c
index 3e6ac76dc..195457cd5 100644
--- a/exa/exa.c
+++ b/exa/exa.c
@@ -32,6 +32,10 @@
 #include <dix-config.h>
 #endif
 
+#ifdef MITSHM
+#include "shmint.h"
+#endif
+
 #include <stdlib.h>
 
 #include "exa_priv.h"
@@ -169,7 +173,7 @@ exaLog2(int val)
 {
     int bits;
 
-    if (!val)
+    if (val <= 0)
 	return 0;
     for (bits = 0; val != 0; bits++)
 	val >>= 1;
@@ -549,6 +553,9 @@ exaDriverInit (ScreenPtr		pScreen,
                ExaDriverPtr	pScreenInfo)
 {
     ExaScreenPrivPtr pExaScr;
+#ifdef RENDER
+    PictureScreenPtr ps;
+#endif
 
     if (pScreenInfo->exa_major != EXA_VERSION_MAJOR ||
 	pScreenInfo->exa_minor > EXA_VERSION_MINOR)
@@ -562,7 +569,7 @@ exaDriverInit (ScreenPtr		pScreen,
     }
 
 #ifdef RENDER
-    PictureScreenPtr	ps = GetPictureScreenIfSet(pScreen);
+    ps = GetPictureScreenIfSet(pScreen);
 #endif
     if (exaGeneration != serverGeneration)
     {
@@ -629,8 +636,17 @@ exaDriverInit (ScreenPtr		pScreen,
     }
 #endif
 
+#ifdef COMPOSITE
     miDisableCompositeWrapper(pScreen);
+#endif
 
+#ifdef MITSHM
+    /* Re-register with the MI funcs, which don't allow shared pixmaps.
+     * Shared pixmaps are almost always a performance loss for us, but this
+     * still allows for SHM PutImage.
+     */
+    ShmRegisterFuncs(pScreen, NULL);
+#endif
     /*
      * Hookup offscreen pixmaps
      */
@@ -649,6 +665,10 @@ exaDriverInit (ScreenPtr		pScreen,
 
         pExaScr->SavedDestroyPixmap = pScreen->DestroyPixmap;
 	pScreen->DestroyPixmap = exaDestroyPixmap;
+
+	LogMessage(X_INFO, "EXA(%d): Offscreen pixmap area of %d bytes\n",
+		   pScreen->myNum,
+		   pExaScr->info->memorySize - pExaScr->info->offScreenBase);
     }
     else
     {
@@ -667,6 +687,22 @@ exaDriverInit (ScreenPtr		pScreen,
         }
     }
 
+    LogMessage(X_INFO, "EXA(%d): Driver registered support for the following"
+	       " operations:\n", pScreen->myNum);
+    assert(pScreenInfo->PrepareSolid != NULL);
+    LogMessage(X_INFO, "        Solid\n");
+    assert(pScreenInfo->PrepareCopy != NULL);
+    LogMessage(X_INFO, "        Copy\n");
+    if (pScreenInfo->PrepareComposite != NULL) {
+	LogMessage(X_INFO, "        Composite (RENDER acceleration)\n");
+    }
+    if (pScreenInfo->UploadToScreen != NULL) {
+	LogMessage(X_INFO, "        UploadToScreen\n");
+    }
+    if (pScreenInfo->DownloadFromScreen != NULL) {
+	LogMessage(X_INFO, "        DownloadFromScreen\n");
+    }
+
     return TRUE;
 }
 
diff --git a/exa/exa.h b/exa/exa.h
index 2cb200854..96465a775 100644
--- a/exa/exa.h
+++ b/exa/exa.h
@@ -36,6 +36,7 @@
 #include "windowstr.h"
 #include "gcstruct.h"
 #include "picturestr.h"
+#include "fb.h"
 
 #define EXA_VERSION_MAJOR   2
 #define EXA_VERSION_MINOR   0
@@ -715,4 +716,12 @@ exaGetPixmapSize(PixmapPtr pPix);
 void
 exaEnableDisableFBAccess (int index, Bool enable);
 
+/**
+ * Returns TRUE if the given planemask covers all the significant bits in the
+ * pixel values for pDrawable.
+ */
+#define EXA_PM_IS_SOLID(_pDrawable, _pm) \
+	(((_pm) & FbFullMask((_pDrawable)->depth)) == \
+	 FbFullMask((_pDrawable)->depth))
+
 #endif /* EXA_H */
diff --git a/exa/exa_accel.c b/exa/exa_accel.c
index 102973ac6..d2fe2e034 100644
--- a/exa/exa_accel.c
+++ b/exa/exa_accel.c
@@ -54,12 +54,12 @@ exaFillSpans(DrawablePtr pDrawable, GCPtr pGC, int n,
 
     pixmaps[0].as_dst = TRUE;
     pixmaps[0].as_src = FALSE;
-    pixmaps[0].pPix = exaGetDrawablePixmap (pDrawable);
+    pixmaps[0].pPix = pPixmap = exaGetDrawablePixmap (pDrawable);
 
     if (pExaScr->swappedOut ||
 	pGC->fillStyle != FillSolid ||
-	pDrawable->width > pExaScr->info->maxX ||
-	pDrawable->height > pExaScr->info->maxY)
+	pPixmap->drawable.width > pExaScr->info->maxX ||
+	pPixmap->drawable.height > pExaScr->info->maxY)
     {
 	exaDoMigration (pixmaps, 1, FALSE);
 	ExaCheckFillSpans (pDrawable, pGC, n, ppt, pwidth, fSorted);
@@ -393,19 +393,19 @@ exaCopyNtoN (DrawablePtr    pSrcDrawable,
 
     pixmaps[0].as_dst = TRUE;
     pixmaps[0].as_src = FALSE;
-    pixmaps[0].pPix = exaGetDrawablePixmap (pDstDrawable);
+    pixmaps[0].pPix = pDstPixmap = exaGetDrawablePixmap (pDstDrawable);
     pixmaps[1].as_dst = FALSE;
     pixmaps[1].as_src = TRUE;
-    pixmaps[1].pPix = exaGetDrawablePixmap (pSrcDrawable);
+    pixmaps[1].pPix = pSrcPixmap = exaGetDrawablePixmap (pSrcDrawable);
 
     /* Respect maxX/maxY in a trivial way: don't set up drawing when we might
      * violate the limits.  The proper solution would be a temporary pixmap
      * adjusted so that the drawing happened within limits.
      */
-    if (pSrcDrawable->width > pExaScr->info->maxX ||
-	pSrcDrawable->height > pExaScr->info->maxY ||
-	pDstDrawable->width > pExaScr->info->maxX ||
-	pDstDrawable->height > pExaScr->info->maxY)
+    if (pSrcPixmap->drawable.width > pExaScr->info->maxX ||
+	pSrcPixmap->drawable.height > pExaScr->info->maxY ||
+	pDstPixmap->drawable.width > pExaScr->info->maxX ||
+	pDstPixmap->drawable.height > pExaScr->info->maxY)
     {
 	exaDoMigration (pixmaps, 2, FALSE);
 	goto fallback;
@@ -414,7 +414,8 @@ exaCopyNtoN (DrawablePtr    pSrcDrawable,
     }
 
     /* Mixed directions must be handled specially if the card is lame */
-    if (pExaScr->info->flags & EXA_TWO_BITBLT_DIRECTIONS && (dx*dy) < 0) {
+    if (pExaScr->info->flags & EXA_TWO_BITBLT_DIRECTIONS &&
+	reverse != upsidedown) {
 	if (!exaCopyNtoNTwoDir(pSrcDrawable, pDstDrawable, pGC, pbox, nbox,
 			       dx, dy))
 	    goto fallback;
@@ -424,7 +425,7 @@ exaCopyNtoN (DrawablePtr    pSrcDrawable,
     if ((pSrcPixmap = exaGetOffscreenPixmap (pSrcDrawable, &src_off_x, &src_off_y)) &&
 	(pDstPixmap = exaGetOffscreenPixmap (pDstDrawable, &dst_off_x, &dst_off_y)) &&
 	(*pExaScr->info->PrepareCopy) (pSrcPixmap, pDstPixmap,
-				       dx, dy,
+				       reverse ? -1 : 1, upsidedown ? -1 : 1,
 				       pGC ? pGC->alu : GXcopy,
 				       pGC ? pGC->planemask : FB_ALLONES))
     {
@@ -643,12 +644,12 @@ exaPolyFillRect(DrawablePtr pDrawable,
 
     pixmaps[0].as_dst = TRUE;
     pixmaps[0].as_src = FALSE;
-    pixmaps[0].pPix = exaGetDrawablePixmap (pDrawable);
+    pixmaps[0].pPix = pPixmap = exaGetDrawablePixmap (pDrawable);
  
     if (pExaScr->swappedOut ||
 	pGC->fillStyle != FillSolid ||
-	pDrawable->width > pExaScr->info->maxX ||
-	pDrawable->height > pExaScr->info->maxY)
+	pPixmap->drawable.width > pExaScr->info->maxX ||
+	pPixmap->drawable.height > pExaScr->info->maxY)
     {
 	exaDoMigration (pixmaps, 1, FALSE);
 	ExaCheckPolyFillRect (pDrawable, pGC, nrect, prect);
@@ -767,14 +768,14 @@ exaSolidBoxClipped (DrawablePtr	pDrawable,
 
     pixmaps[0].as_dst = TRUE;
     pixmaps[0].as_src = FALSE;
-    pixmaps[0].pPix = exaGetDrawablePixmap (pDrawable);
+    pixmaps[0].pPix = pPixmap = exaGetDrawablePixmap (pDrawable);
 
     /* We need to initialize x/yoff for tracking damage in the fallback case */
     pPixmap = exaGetOffscreenPixmap (pDrawable, &xoff, &yoff);
 
     if (pExaScr->swappedOut ||
-	pDrawable->width > pExaScr->info->maxX ||
-	pDrawable->height > pExaScr->info->maxY)
+	pPixmap->drawable.width > pExaScr->info->maxX ||
+	pPixmap->drawable.height > pExaScr->info->maxY)
     {
 	exaDoMigration (pixmaps, 1, FALSE);
 	goto fallback;
@@ -1042,10 +1043,10 @@ exaFillRegionSolid (DrawablePtr	pDrawable,
 
     pixmaps[0].as_dst = TRUE;
     pixmaps[0].as_src = FALSE;
-    pixmaps[0].pPix = exaGetDrawablePixmap (pDrawable);
+    pixmaps[0].pPix = pPixmap = exaGetDrawablePixmap (pDrawable);
  
-    if (pDrawable->width > pExaScr->info->maxX ||
-	pDrawable->height > pExaScr->info->maxY)
+    if (pPixmap->drawable.width > pExaScr->info->maxX ||
+	pPixmap->drawable.height > pExaScr->info->maxY)
     {
 	exaDoMigration (pixmaps, 1, FALSE);
 	goto fallback;
@@ -1117,7 +1118,7 @@ exaFillRegionTiled (DrawablePtr	pDrawable,
 
     pixmaps[0].as_dst = TRUE;
     pixmaps[0].as_src = FALSE;
-    pixmaps[0].pPix = exaGetDrawablePixmap (pDrawable);
+    pixmaps[0].pPix = pPixmap = exaGetDrawablePixmap (pDrawable);
     pixmaps[1].as_dst = FALSE;
     pixmaps[1].as_src = TRUE;
     pixmaps[1].pPix = pTile;
@@ -1125,8 +1126,8 @@ exaFillRegionTiled (DrawablePtr	pDrawable,
     /* We need to initialize x/yoff for tracking damage in the fallback case */
     pPixmap = exaGetOffscreenPixmap (pDrawable, &xoff, &yoff);
 
-    if (pDrawable->width > pExaScr->info->maxX ||
-	pDrawable->height > pExaScr->info->maxY ||
+    if (pPixmap->drawable.width > pExaScr->info->maxX ||
+	pPixmap->drawable.height > pExaScr->info->maxY ||
 	tileWidth > pExaScr->info->maxX ||
 	tileHeight > pExaScr->info->maxY)
     {
diff --git a/exa/exa_offscreen.c b/exa/exa_offscreen.c
index d5864dc57..fd3599e8c 100644
--- a/exa/exa_offscreen.c
+++ b/exa/exa_offscreen.c
@@ -414,7 +414,6 @@ exaOffscreenInit (ScreenPtr pScreen)
     if (!area)
 	return FALSE;
 
-
     area->state = ExaOffscreenAvail;
     area->base_offset = pExaScr->info->offScreenBase;
     area->offset = area->base_offset;
@@ -423,10 +422,6 @@ exaOffscreenInit (ScreenPtr pScreen)
     area->next = NULL;
     area->score = 0;
 
-#if DEBUG_OFFSCREEN
-    ErrorF("============ initial memory block of %d\n", area->size);
-#endif
-
     /* Add it to the free areas */
     pExaScr->info->offScreenAreas = area;
 
diff --git a/exa/exa_priv.h b/exa/exa_priv.h
index c725b4da0..80cf60903 100644
--- a/exa/exa_priv.h
+++ b/exa/exa_priv.h
@@ -138,14 +138,6 @@ extern int exaPixmapPrivateIndex;
 /** Align an offset to a power-of-two alignment */
 #define EXA_ALIGN2(offset, align) (((offset) + (align) - 1) & ~((align) - 1))
 
-/**
- * Returns TRUE if the given planemask covers all the significant bits in the
- * pixel values for pDrawable.
- */
-#define EXA_PM_IS_SOLID(_pDrawable, _pm) \
-	(((_pm) & FbFullMask((_pDrawable)->depth)) == \
-	 FbFullMask((_pDrawable)->depth))
-
 #define EXA_PIXMAP_SCORE_MOVE_IN    10
 #define EXA_PIXMAP_SCORE_MAX	    20
 #define EXA_PIXMAP_SCORE_MOVE_OUT   -10
diff --git a/exa/exa_render.c b/exa/exa_render.c
index 790a359a3..a73582000 100644
--- a/exa/exa_render.c
+++ b/exa/exa_render.c
@@ -282,8 +282,12 @@ exaTryDriverSolidFill(PicturePtr	pSrc,
 	return -1;
     }
 
-    exaGetPixelFromRGBA(&pixel, red, green, blue, alpha,
-			pDst->format);
+    if (!exaGetPixelFromRGBA(&pixel, red, green, blue, alpha,
+			pDst->format))
+    {
+	REGION_UNINIT(pDst->pDrawable->pScreen, &region);
+	return -1;
+    }
 
     if (!(*pExaScr->info->PrepareSolid) (pDstPix, GXcopy, 0xffffffff, pixel))
     {
@@ -333,16 +337,21 @@ exaTryDriverComposite(CARD8		op,
     struct _Pixmap scratch;
     ExaMigrationRec pixmaps[3];
 
+    pSrcPix = exaGetDrawablePixmap(pSrc->pDrawable);
+    pDstPix = exaGetDrawablePixmap(pDst->pDrawable);
+    if (pMask)
+	pMaskPix = exaGetDrawablePixmap(pMask->pDrawable);
+
     /* Bail if we might exceed coord limits by rendering from/to these.  We
      * should really be making some scratch pixmaps with offsets and coords
      * adjusted to deal with this, but it hasn't been done yet.
      */
-    if (pSrc->pDrawable->width > pExaScr->info->maxX ||
-	pSrc->pDrawable->height > pExaScr->info->maxY ||
-	pDst->pDrawable->width > pExaScr->info->maxX ||
-	pDst->pDrawable->height > pExaScr->info->maxY || 
-	(pMask && (pMask->pDrawable->width > pExaScr->info->maxX ||
-		   pMask->pDrawable->height > pExaScr->info->maxY)))
+    if (pSrcPix->drawable.width > pExaScr->info->maxX ||
+	pSrcPix->drawable.height > pExaScr->info->maxY ||
+	pDstPix->drawable.width > pExaScr->info->maxX ||
+	pDstPix->drawable.height > pExaScr->info->maxY || 
+	(pMask && (pMaskPix->drawable.width > pExaScr->info->maxX ||
+		   pMaskPix->drawable.height > pExaScr->info->maxY)))
     {
 	return -1;
     }
@@ -450,24 +459,72 @@ exaTryDriverComposite(CARD8		op,
     return 1;
 }
 
+/**
+ * exaTryMagicTwoPassCompositeHelper implements PictOpOver using two passes of
+ * simpler operations PictOpOutReverse and PictOpAdd. Mainly used for component
+ * alpha and limited 1-tmu cards.
+ *
+ * From http://anholt.livejournal.com/32058.html:
+ *
+ * The trouble is that component-alpha rendering requires two different sources
+ * for blending: one for the source value to the blender, which is the
+ * per-channel multiplication of source and mask, and one for the source alpha
+ * for multiplying with the destination channels, which is the multiplication
+ * of the source channels by the mask alpha. So the equation for Over is:
+ *
+ * dst.A = src.A * mask.A + (1 - (src.A * mask.A)) * dst.A
+ * dst.R = src.R * mask.R + (1 - (src.A * mask.R)) * dst.R
+ * dst.G = src.G * mask.G + (1 - (src.A * mask.G)) * dst.G
+ * dst.B = src.B * mask.B + (1 - (src.A * mask.B)) * dst.B
+ *
+ * But we can do some simpler operations, right? How about PictOpOutReverse,
+ * which has a source factor of 0 and dest factor of (1 - source alpha). We
+ * can get the source alpha value (srca.X = src.A * mask.X) out of the texture
+ * blenders pretty easily. So we can do a component-alpha OutReverse, which
+ * gets us:
+ *
+ * dst.A = 0 + (1 - (src.A * mask.A)) * dst.A
+ * dst.R = 0 + (1 - (src.A * mask.R)) * dst.R
+ * dst.G = 0 + (1 - (src.A * mask.G)) * dst.G
+ * dst.B = 0 + (1 - (src.A * mask.B)) * dst.B
+ *
+ * OK. And if an op doesn't use the source alpha value for the destination
+ * factor, then we can do the channel multiplication in the texture blenders
+ * to get the source value, and ignore the source alpha that we wouldn't use.
+ * We've supported this in the Radeon driver for a long time. An example would
+ * be PictOpAdd, which does:
+ *
+ * dst.A = src.A * mask.A + dst.A
+ * dst.R = src.R * mask.R + dst.R
+ * dst.G = src.G * mask.G + dst.G
+ * dst.B = src.B * mask.B + dst.B
+ *
+ * Hey, this looks good! If we do a PictOpOutReverse and then a PictOpAdd right
+ * after it, we get:
+ *
+ * dst.A = src.A * mask.A + ((1 - (src.A * mask.A)) * dst.A)
+ * dst.R = src.R * mask.R + ((1 - (src.A * mask.R)) * dst.R)
+ * dst.G = src.G * mask.G + ((1 - (src.A * mask.G)) * dst.G)
+ * dst.B = src.B * mask.B + ((1 - (src.A * mask.B)) * dst.B)
+ */
+
 static int
-exaTryComponentAlphaHelper(CARD8 op,
-			   PicturePtr pSrc,
-			   PicturePtr pMask,
-			   PicturePtr pDst,
-			   INT16 xSrc,
-			   INT16 ySrc,
-			   INT16 xMask,
-			   INT16 yMask,
-			   INT16 xDst,
-			   INT16 yDst,
-			   CARD16 width,
-			   CARD16 height)
+exaTryMagicTwoPassCompositeHelper(CARD8 op,
+				  PicturePtr pSrc,
+				  PicturePtr pMask,
+				  PicturePtr pDst,
+				  INT16 xSrc,
+				  INT16 ySrc,
+				  INT16 xMask,
+				  INT16 yMask,
+				  INT16 xDst,
+				  INT16 yDst,
+				  CARD16 width,
+				  CARD16 height)
 {
     ExaScreenPriv (pDst->pDrawable->pScreen);
 
     assert(op == PictOpOver);
-    assert(pMask->componentAlpha == TRUE);
 
     if (pExaScr->info->CheckComposite &&
 	(!(*pExaScr->info->CheckComposite)(PictOpOutReverse, pSrc, pMask,
@@ -533,7 +590,8 @@ exaComposite(CARD8	op,
 	if (op == PictOpSrc)
 	{
 	    if (pSrc->pDrawable->width == 1 &&
-		pSrc->pDrawable->height == 1 && pSrc->repeat)
+		pSrc->pDrawable->height == 1 && pSrc->repeat &&
+		pSrc->repeatType == RepeatNormal)
 	    {
 		ret = exaTryDriverSolidFill(pSrc, pDst, xSrc, ySrc, xDst, yDst,
 					    width, height);
@@ -573,21 +631,34 @@ exaComposite(CARD8	op,
 	    pMask->repeat = 0;
 
     if (pExaScr->info->PrepareComposite &&
+	(!pSrc->repeat || pSrc->repeat == RepeatNormal) &&
+	(!pMask || !pMask->repeat || pMask->repeat == RepeatNormal) &&
 	!pSrc->alphaMap && (!pMask || !pMask->alphaMap) && !pDst->alphaMap)
     {
+	Bool isSrcSolid;
+
 	ret = exaTryDriverComposite(op, pSrc, pMask, pDst, xSrc, ySrc, xMask,
 				    yMask, xDst, yDst, width, height);
 	if (ret == 1)
 	    goto done;
 
+	/* For generic masks and solid src pictures, mach64 can do Over in two
+	 * passes, similar to the component-alpha case.
+	 */
+	isSrcSolid = pSrc->pDrawable->width == 1 &&
+		     pSrc->pDrawable->height == 1 &&
+		     pSrc->repeat;
+
 	/* If we couldn't do the Composite in a single pass, and it was a
 	 * component-alpha Over, see if we can do it in two passes with
 	 * an OutReverse and then an Add.
 	 */
-	if (ret == -1 && pMask && pMask->componentAlpha && op == PictOpOver) {
-	    ret = exaTryComponentAlphaHelper(op, pSrc, pMask, pDst, xSrc, ySrc,
-					     xMask, yMask, xDst, yDst,
-					     width, height);
+	if (ret == -1 && op == PictOpOver && pMask &&
+	    (pMask->componentAlpha || isSrcSolid)) {
+	    ret = exaTryMagicTwoPassCompositeHelper(op, pSrc, pMask, pDst,
+						    xSrc, ySrc,
+						    xMask, yMask, xDst, yDst,
+						    width, height);
 	    if (ret == 1)
 		goto done;
 	}