Add an accelerated path for rotation Render operations in XAA.

The now-generic (kind of) EXA code will be cleaned up and moved to generic
files in a later commit.

7 files changed, 231 insertions, 19 deletions

diff --git a/src/i830.h b/src/i830.h
index 66cfba1e..58e4ec34 100644
--- a/src/i830.h
+++ b/src/i830.h
@@ -365,6 +365,18 @@ typedef struct _I830Rec {
    Bool cursorOn;
 #ifdef I830_USE_XAA
    XAAInfoRecPtr AccelInfoRec;
+
+   /* additional XAA accelerated Composite support */
+   CompositeProcPtr saved_composite;
+   Bool (*xaa_check_composite)(int op, PicturePtr pSrc, PicturePtr pMask,
+			       PicturePtr pDst);
+   Bool (*xaa_prepare_composite)(int op, PicturePtr pSrc, PicturePtr pMask,
+				 PicturePtr pDst, PixmapPtr pSrcPixmap,
+				 PixmapPtr pMaskPixmap, PixmapPtr pDstPixmap);
+   void (*xaa_composite)(PixmapPtr pDst, int xSrc, int ySrc,
+			 int xMask, int yMask, int xDst, int yDst,
+			 int w, int h);
+   void (*xaa_done_composite)(PixmapPtr pDst);
 #endif
    xf86CursorInfoPtr CursorInfoRec;
    CloseScreenProcPtr CloseScreen;
@@ -560,6 +572,8 @@ extern Bool I830DRILock(ScrnInfoPtr pScrn);
 extern Bool I830DRISetVBlankInterrupt (ScrnInfoPtr pScrn, Bool on);
 #endif
 
+unsigned long intel_get_pixmap_offset(PixmapPtr pPix);
+unsigned long intel_get_pixmap_pitch(PixmapPtr pPix);
 extern Bool I830AccelInit(ScreenPtr pScreen);
 extern void I830SetupForScreenToScreenCopy(ScrnInfoPtr pScrn, int xdir,
 					   int ydir, int rop,
@@ -634,7 +648,6 @@ DisplayModePtr i830_ddc_get_modes(xf86OutputPtr output);
 /* i830_tv.c */
 void i830_tv_init(ScrnInfoPtr pScrn);
 
-#ifdef I830_USE_EXA
 extern Bool I830EXACheckComposite(int, PicturePtr, PicturePtr, PicturePtr);
 extern Bool I830EXAPrepareComposite(int, PicturePtr, PicturePtr, PicturePtr, 
 				PixmapPtr, PixmapPtr, PixmapPtr);
@@ -647,6 +660,9 @@ extern Bool I965EXAPrepareComposite(int, PicturePtr, PicturePtr, PicturePtr,
 				PixmapPtr, PixmapPtr, PixmapPtr);
 extern void I965EXAComposite(PixmapPtr pDst, int srcX, int srcY, int maskX, 
 			int maskY, int dstX, int dstY, int width, int height);
+void IntelEXAComposite(PixmapPtr pDst, int srcX, int srcY,
+		       int maskX, int maskY, int dstX, int dstY, int w, int h);
+void IntelEXADoneComposite(PixmapPtr pDst);
 
 extern Bool
 I830EXACheckComposite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
@@ -663,7 +679,7 @@ i830_get_transformed_coordinates(int x, int y, PictTransformPtr transform,
 
 extern const int I830PatternROP[16];
 extern const int I830CopyROP[16];
-#endif
+
 /* Flags for memory allocation function */
 #define FROM_ANYWHERE			0x00000000
 #define FROM_POOL_ONLY			0x00000001
diff --git a/src/i830_accel.c b/src/i830_accel.c
index 5fdd1012..db3168a8 100644
--- a/src/i830_accel.c
+++ b/src/i830_accel.c
@@ -60,6 +60,36 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "i810_reg.h"
 #include "i830_debug.h"
 
+unsigned long
+intel_get_pixmap_offset(PixmapPtr pPix)
+{
+    ScreenPtr pScreen = pPix->drawable.pScreen;
+    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    I830Ptr pI830 = I830PTR(pScrn);
+
+#ifdef I830_USE_EXA
+    if (pI830->useEXA)
+	return exaGetPixmapOffset(pPix);
+#endif
+    return (unsigned long)pPix->devPrivate.ptr - (unsigned long)pI830->FbBase;
+}
+
+unsigned long
+intel_get_pixmap_pitch(PixmapPtr pPix)
+{
+    ScreenPtr pScreen = pPix->drawable.pScreen;
+    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    I830Ptr pI830 = I830PTR(pScrn);
+
+#ifdef I830_USE_EXA
+    if (pI830->useEXA)
+	return exaGetPixmapPitch(pPix);
+#endif
+#ifdef I830_USE_XAA
+    return (unsigned long)pPix->devKind;
+#endif
+}
+
 int
 I830WaitLpRing(ScrnInfoPtr pScrn, int n, int timeout_millis)
 {
diff --git a/src/i830_exa.c b/src/i830_exa.c
index b0d886b0..02d2dcfe 100644
--- a/src/i830_exa.c
+++ b/src/i830_exa.c
@@ -314,7 +314,7 @@ i830_get_transformed_coordinates(int x, int y, PictTransformPtr transform,
  *
  * This function is shared between i830 and i915 generation code.
  */
-static void
+void
 IntelEXAComposite(PixmapPtr pDst, int srcX, int srcY, int maskX, int maskY,
 		 int dstX, int dstY, int w, int h)
 {
@@ -396,7 +396,7 @@ IntelEXAComposite(PixmapPtr pDst, int srcX, int srcY, int maskX, int maskY,
     }
 }
 
-static void
+void
 IntelEXADoneComposite(PixmapPtr pDst)
 {
 #if ALWAYS_SYNC
diff --git a/src/i830_exa_render.c b/src/i830_exa_render.c
index 98a444f1..bf521b45 100644
--- a/src/i830_exa_render.c
+++ b/src/i830_exa_render.c
@@ -256,8 +256,8 @@ I830TextureSetup(PicturePtr pPict, PixmapPtr pPix, int unit)
     int w, h, i;
     CARD32 wrap_mode = TEXCOORDMODE_CLAMP; 
 
-    offset = exaGetPixmapOffset(pPix);
-    pitch = exaGetPixmapPitch(pPix);
+    offset = intel_get_pixmap_offset(pPix);
+    pitch = intel_get_pixmap_pitch(pPix);
     w = pPict->pDrawable->width;
     h = pPict->pDrawable->height;
     pI830->scale_units[unit][0] = pPix->drawable.width;
@@ -381,8 +381,8 @@ I830EXAPrepareComposite(int op, PicturePtr pSrcPicture,
     CARD32 dst_format, dst_offset, dst_pitch;
 
     I830GetDestFormat(pDstPicture, &dst_format);
-    dst_offset = exaGetPixmapOffset(pDst);
-    dst_pitch = exaGetPixmapPitch(pDst);
+    dst_offset = intel_get_pixmap_offset(pDst);
+    dst_pitch = intel_get_pixmap_pitch(pDst);
 
     pI830->last_3d = LAST_3D_RENDER;
 
diff --git a/src/i830_xaa.c b/src/i830_xaa.c
index 13ba7432..be1fb831 100644
--- a/src/i830_xaa.c
+++ b/src/i830_xaa.c
@@ -52,6 +52,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "xaarop.h"
 #include "i830.h"
 #include "i810_reg.h"
+#include "mipict.h"
 
 #ifndef DO_SCANLINE_IMAGE_WRITE
 #define DO_SCANLINE_IMAGE_WRITE 0
@@ -91,11 +92,26 @@ static void I830SubsequentImageWriteScanline(ScrnInfoPtr pScrn, int bufno);
 #endif
 static void I830RestoreAccelState(ScrnInfoPtr pScrn);
 
+void
+i830_xaa_composite(CARD8	op,
+		   PicturePtr	pSrc,
+		   PicturePtr	pMask,
+		   PicturePtr	pDst,
+		   INT16	xSrc,
+		   INT16	ySrc,
+		   INT16	xMask,
+		   INT16	yMask,
+		   INT16	xDst,
+		   INT16	yDst,
+		   CARD16	width,
+		   CARD16	height);
+
 Bool
 I830XAAInit(ScreenPtr pScreen)
 {
     XAAInfoRecPtr infoPtr;
     ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    PictureScreenPtr ps = GetPictureScreenIfSet(pScreen);
     I830Ptr pI830 = I830PTR(pScrn);
     int i;
     int width = 0;
@@ -220,7 +236,37 @@ I830XAAInit(ScreenPtr pScreen)
 
     I830SelectBuffer(pScrn, I830_SELECT_FRONT);
 
-    return XAAInit(pScreen, infoPtr);
+    if (!XAAInit(pScreen, infoPtr))
+	return FALSE;
+
+    if (ps != NULL) {
+	if (IS_I915G(pI830) || IS_I915GM(pI830) ||
+	    IS_I945G(pI830) || IS_I945GM(pI830))
+	{
+	    pI830->xaa_check_composite = I915EXACheckComposite;
+	    pI830->xaa_prepare_composite = I915EXAPrepareComposite;
+	    pI830->xaa_composite = IntelEXAComposite;
+	    pI830->xaa_done_composite = IntelEXADoneComposite;
+	} else if (IS_I865G(pI830) || IS_I855(pI830) ||
+		   IS_845G(pI830) || IS_I830(pI830)) {
+	    pI830->xaa_check_composite = I830EXACheckComposite;
+	    pI830->xaa_prepare_composite = I830EXAPrepareComposite;
+	    pI830->xaa_composite = IntelEXAComposite;
+	    pI830->xaa_done_composite = IntelEXADoneComposite;
+	} else if (IS_I965G(pI830)) {
+	    pI830->xaa_check_composite = I965EXACheckComposite;
+	    pI830->xaa_prepare_composite = I965EXAPrepareComposite;
+	    pI830->xaa_composite = I965EXAComposite;
+	    pI830->xaa_done_composite = IntelEXADoneComposite;
+	} else {
+	    return TRUE;
+	}
+
+	pI830->saved_composite = ps->Composite;
+	ps->Composite = i830_xaa_composite;
+    }
+
+    return TRUE;
 }
 
 #ifdef XF86DRI
@@ -694,6 +740,126 @@ I830SubsequentImageWriteScanline(ScrnInfoPtr pScrn, int bufno)
 #endif /* DO_SCANLINE_IMAGE_WRITE */
 /* Support for multiscreen */
 
+/**
+ * Special case acceleration for Render acceleration of rotation operations
+ * by xf86Rotate.c
+ */
+void
+i830_xaa_composite(CARD8	op,
+		   PicturePtr	pSrc,
+		   PicturePtr	pMask,
+		   PicturePtr	pDst,
+		   INT16	xSrc,
+		   INT16	ySrc,
+		   INT16	xMask,
+		   INT16	yMask,
+		   INT16	xDst,
+		   INT16	yDst,
+		   CARD16	width,
+		   CARD16	height)
+{
+    ScreenPtr pScreen = pDst->pDrawable->pScreen;
+    ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+    I830Ptr pI830 = I830PTR(pScrn);
+    xf86CrtcConfigPtr xf86_config = XF86_CRTC_CONFIG_PTR(pScrn);
+    PictureScreenPtr ps;
+    PixmapPtr pSrcPixmap, pDstPixmap;
+    RegionRec region;
+    BoxPtr pbox;
+    int nbox;
+    int	i;
+
+    /* Throw out cases that aren't going to be our rotation first */
+    if (pMask != NULL || op != PictOpSrc || pSrc->pDrawable == NULL)
+	goto fallback;
+
+    if (pSrc->pDrawable->type != DRAWABLE_PIXMAP ||
+	pDst->pDrawable->type != DRAWABLE_PIXMAP)
+    {
+	goto fallback;
+    }
+    pSrcPixmap = (PixmapPtr)pSrc->pDrawable;
+    pDstPixmap = (PixmapPtr)pDst->pDrawable;
+
+    /* Check if the dest is one of our shadow pixmaps */
+    for (i = 0; i < xf86_config->num_crtc; i++) {
+	xf86CrtcPtr crtc = xf86_config->crtc[i];
+
+	if (crtc->rotatedPixmap == pDstPixmap)
+	    break;
+    }
+    if (i == xf86_config->num_crtc)
+	goto fallback;
+
+    if (pSrcPixmap != pScreen->GetScreenPixmap(pScreen))
+	goto fallback;
+
+    /* OK, so we've got a Render operation on one of our shadow pixmaps, with
+     * the source being the real framebuffer.  We know that both of these are
+     * in framebuffer, with no x/y offsets, i.e. normal pixmaps like our EXA-
+     * based Render acceleration code expects.
+     */
+    assert(pSrcPixmap->drawable.x == 0);
+    assert(pSrcPixmap->drawable.y == 0);
+    assert(pDstPixmap->drawable.x == 0);
+    assert(pDstPixmap->drawable.y == 0);
+
+    if (!miComputeCompositeRegion (&region, pSrc, NULL, pDst,
+				   xSrc, ySrc, 0, 0, xDst, yDst,
+				   width, height))
+	return;
+
+    if (!pI830->xaa_check_composite(op, pSrc, NULL, pDst)) {
+	REGION_UNINIT(pScreen, &region);
+	goto fallback;
+    }
+
+    if (!pI830->xaa_prepare_composite(op, pSrc, NULL, pDst,
+				      pSrcPixmap, NULL, pDstPixmap))
+    {
+	REGION_UNINIT(pScreen, &region);
+	goto fallback;
+    }
+
+    nbox = REGION_NUM_RECTS(&region);
+    pbox = REGION_RECTS(&region);
+
+    xSrc -= xDst;
+    ySrc -= yDst;
+
+    while (nbox--)
+    {
+	pI830->xaa_composite(pDstPixmap,
+			     pbox->x1 + xSrc,
+			     pbox->y1 + ySrc,
+			     0, 0,
+			     pbox->x1,
+			     pbox->y1,
+			     pbox->x2 - pbox->x1,
+			     pbox->y2 - pbox->y1);
+	pbox++;
+    }
+
+    REGION_UNINIT(pDst->pDrawable->pScreen, &region);
+
+    pI830->xaa_done_composite(pDstPixmap);
+    i830MarkSync(pScrn);
+
+    return;
+
+fallback:
+    /* Fallback path: Call down to the next level (XAA) */
+    ps = GetPictureScreenIfSet(pScreen);
+
+    ps->Composite = pI830->saved_composite;
+
+    ps->Composite(op, pSrc, pMask, pDst, xSrc, ySrc, xMask, yMask, xDst, yDst,
+		  width, height);
+
+    pI830->saved_composite = ps->Composite;
+    ps->Composite = i830_xaa_composite;
+}
+
 static void
 I830RestoreAccelState(ScrnInfoPtr pScrn)
 {
diff --git a/src/i915_exa_render.c b/src/i915_exa_render.c
index 8705c6de..2fb41ad1 100644
--- a/src/i915_exa_render.c
+++ b/src/i915_exa_render.c
@@ -248,8 +248,8 @@ I915TextureSetup(PicturePtr pPict, PixmapPtr pPix, int unit)
     int w, h, i;
     CARD32 wrap_mode = TEXCOORDMODE_CLAMP_BORDER;
 
-    offset = exaGetPixmapOffset(pPix);
-    pitch = exaGetPixmapPitch(pPix);
+    offset = intel_get_pixmap_offset(pPix);
+    pitch = intel_get_pixmap_pitch(pPix);
     w = pPict->pDrawable->width;
     h = pPict->pDrawable->height;
     pI830->scale_units[unit][0] = pPix->drawable.width;
@@ -318,8 +318,8 @@ I915EXAPrepareComposite(int op, PicturePtr pSrcPicture,
     pI830->last_3d = LAST_3D_RENDER;
 
     I915GetDestFormat(pDstPicture, &dst_format);
-    dst_offset = exaGetPixmapOffset(pDst);
-    dst_pitch = exaGetPixmapPitch(pDst);
+    dst_offset = intel_get_pixmap_offset(pDst);
+    dst_pitch = intel_get_pixmap_pitch(pDst);
     FS_LOCALS(20);
 
     if (!I915TextureSetup(pSrcPicture, pSrc, 0))
diff --git a/src/i965_exa_render.c b/src/i965_exa_render.c
index 1dbccc68..99bd6284 100644
--- a/src/i965_exa_render.c
+++ b/src/i965_exa_render.c
@@ -379,13 +379,13 @@ I965EXAPrepareComposite(int op, PicturePtr pSrcPicture,
 
     pI830->last_3d = LAST_3D_RENDER;
 
-    src_offset = exaGetPixmapOffset(pSrc);
-    src_pitch = exaGetPixmapPitch(pSrc);
-    dst_offset = exaGetPixmapOffset(pDst);
-    dst_pitch = exaGetPixmapPitch(pDst);
+    src_offset = intel_get_pixmap_offset(pSrc);
+    src_pitch = intel_get_pixmap_pitch(pSrc);
+    dst_offset = intel_get_pixmap_offset(pDst);
+    dst_pitch = intel_get_pixmap_pitch(pDst);
     if (pMask) {
-	mask_offset = exaGetPixmapOffset(pMask);
-	mask_pitch = exaGetPixmapPitch(pMask);
+	mask_offset = intel_get_pixmap_offset(pMask);
+	mask_pitch = intel_get_pixmap_pitch(pMask);
     }
     pI830->scale_units[0][0] = pSrc->drawable.width;
     pI830->scale_units[0][1] = pSrc->drawable.height;