i915: WIP swap rework

3 files changed, 144 insertions, 169 deletions

diff --git a/src/gallium/winsys/dri/intel/intel_context.h b/src/gallium/winsys/dri/intel/intel_context.h
index ced18da143..18b96217ac 100644
--- a/src/gallium/winsys/dri/intel/intel_context.h
+++ b/src/gallium/winsys/dri/intel/intel_context.h
@@ -32,6 +32,7 @@
 #include "drm.h"
 
 #include "pipe/p_debug.h"
+#include "util/u_blit.h"
 
 #include "intel_screen.h"
 #include "i915_drm.h"
@@ -82,6 +83,11 @@ struct intel_context
     * Configuration cache
     */
    driOptionCache optionCache;
+
+   /**
+    * Blit state for swapbuffers
+    */
+   struct blit_state *blit;
 };
 
 
diff --git a/src/gallium/winsys/dri/intel/intel_swapbuffers.c b/src/gallium/winsys/dri/intel/intel_swapbuffers.c
index f751f97524..6e10235e89 100644
--- a/src/gallium/winsys/dri/intel/intel_swapbuffers.c
+++ b/src/gallium/winsys/dri/intel/intel_swapbuffers.c
@@ -28,16 +28,13 @@
 #include "intel_screen.h"
 #include "intel_context.h"
 #include "intel_swapbuffers.h"
-
+#include "intel_batchbuffer.h"
 #include "intel_reg.h"
 
-#include "pipe/p_context.h"
 #include "state_tracker/st_public.h"
 #include "state_tracker/st_context.h"
-#include "state_tracker/st_cb_fbo.h"
-
 #include "intel_drm/ws_dri_bufmgr.h"
-#include "intel_batchbuffer.h"
+
 
 /**
  * Display a colorbuffer surface in an X window.
@@ -49,158 +46,130 @@
  */
 void
 intelDisplaySurface(__DRIdrawablePrivate *dPriv,
-                    struct pipe_surface *surf,
-                    const drm_clip_rect_t *rect)
+		struct pipe_texture *tex,
+		const drm_clip_rect_t *rect)
 {
-   struct intel_screen *intelScreen = intel_screen(dPriv->driScreenPriv);
-   struct intel_context *intel = intelScreen->dummyContext;
-
-   DBG(SWAP, "%s\n", __FUNCTION__);
-
-   if (!intel) {
-      /* XXX this is where some kind of extra/meta context could be useful */
-      return;
-   }
-
-   if (intel->last_swap_fence) {
-      driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, TRUE);
-      driFenceUnReference(&intel->last_swap_fence);
-      intel->last_swap_fence = NULL;
-   }
-   intel->last_swap_fence = intel->first_swap_fence;
-   intel->first_swap_fence = NULL;
-
-   /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
-    * should work regardless.
-    */
-   LOCK_HARDWARE(intel);
-   /* if this drawable isn't currently bound the LOCK_HARDWARE done on the
-    * current context (which is what intelScreenContext should return) might
-    * not get a contended lock and thus cliprects not updated (tests/manywin)
-    */
-   if (intel_context(dPriv->driContextPriv) != intel)
-      DRI_VALIDATE_DRAWABLE_INFO(intel->driScreen, dPriv);
-
-
-   if (dPriv && dPriv->numClipRects) {
-      const int srcWidth = surf->width;
-      const int srcHeight = surf->height;
-      const int nbox = dPriv->numClipRects;
-      const drm_clip_rect_t *pbox = dPriv->pClipRects;
-      const int pitch = intelScreen->front.pitch / intelScreen->front.cpp;
-      const int cpp = intelScreen->front.cpp;
-      const int srcpitch = surf->stride / cpp;
-      int BR13, CMD;
-      int i;
-
-      ASSERT(surf->buffer);
-      ASSERT(surf->cpp == cpp);
-
-      DBG(SWAP, "screen pitch %d  src surface pitch %d\n",
-	  pitch, surf->stride);
-
-      if (cpp == 2) {
-	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24);
-	 CMD = XY_SRC_COPY_BLT_CMD;
-      }
-      else {
-	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25);
-	 CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
-		XY_SRC_COPY_BLT_WRITE_RGB);
-      }
-
-      for (i = 0; i < nbox; i++, pbox++) {
-	 drm_clip_rect_t box;
-	 drm_clip_rect_t sbox;
-
-	 if (pbox->x1 > pbox->x2 ||
-	     pbox->y1 > pbox->y2 ||
-	     pbox->x2 > intelScreen->front.width ||
-	     pbox->y2 > intelScreen->front.height) {
-            /* invalid cliprect, skip it */
-	    continue;
-         }
-
-	 box = *pbox;
-
-	 if (rect) {
-            /* intersect cliprect with user-provided src rect */
-	    drm_clip_rect_t rrect;
-
-	    rrect.x1 = dPriv->x + rect->x1;
-	    rrect.y1 = (dPriv->h - rect->y1 - rect->y2) + dPriv->y;
-	    rrect.x2 = rect->x2 + rrect.x1;
-	    rrect.y2 = rect->y2 + rrect.y1;
-	    if (rrect.x1 > box.x1)
-	       box.x1 = rrect.x1;
-	    if (rrect.y1 > box.y1)
-	       box.y1 = rrect.y1;
-	    if (rrect.x2 < box.x2)
-	       box.x2 = rrect.x2;
-	    if (rrect.y2 < box.y2)
-	       box.y2 = rrect.y2;
-
-	    if (box.x1 > box.x2 || box.y1 > box.y2)
-	       continue;
-	 }
-
-	 /* restrict blit to size of actually rendered area */
-	 if (box.x2 - box.x1 > srcWidth)
-	    box.x2 = srcWidth + box.x1;
-	 if (box.y2 - box.y1 > srcHeight)
-	    box.y2 = srcHeight + box.y1;
-
-	 DBG(SWAP, "box x1 x2 y1 y2 %d %d %d %d\n",
-	     box.x1, box.x2, box.y1, box.y2);
-
-	 sbox.x1 = box.x1 - dPriv->x;
-	 sbox.y1 = box.y1 - dPriv->y;
-
-         assert(box.x1 < box.x2);
-         assert(box.y1 < box.y2);
-
-         /* XXX this could be done with pipe->surface_copy() */
-	 /* XXX should have its own batch buffer */
-	 if (!BEGIN_BATCH(8, 2)) {
-	    /*
-	     * Since we share this batch buffer with a context
-	     * we can't flush it since that risks a GPU lockup
-	     */
-	    assert(0);
-	    continue;
-	 }
-
-	 OUT_BATCH(CMD);
-	 OUT_BATCH(BR13);
-	 OUT_BATCH((box.y1 << 16) | box.x1);
-	 OUT_BATCH((box.y2 << 16) | box.x2);
-
-	 OUT_RELOC(intelScreen->front.buffer,
-		   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
-		   DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
-	 OUT_BATCH((sbox.y1 << 16) | sbox.x1);
-	 OUT_BATCH((srcpitch * cpp) & 0xffff);
-	 OUT_RELOC(dri_bo(surf->buffer),
-                   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
-		   DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
-
-      }
-
-      if (intel->first_swap_fence)
-	 driFenceUnReference(&intel->first_swap_fence);
-      intel->first_swap_fence = intel_be_batchbuffer_flush(intel->base.batch);
-   }
-
-   UNLOCK_HARDWARE(intel);
-
-   if (intel->lastStamp != dPriv->lastStamp) {
-      intelUpdateWindowSize(dPriv);
-      intel->lastStamp = dPriv->lastStamp;
-   }
+	struct intel_screen *intelScreen = intel_screen(dPriv->driScreenPriv);
+	struct intel_context *intel = intelScreen->dummyContext;
+
+	DBG(SWAP, "%s\n", __FUNCTION__);
+
+	if (!intel) {
+		/* XXX this is where some kind of extra/meta context could be useful */
+		return;
+	} else if (!intel->blit) {
+		intel->blit = util_create_blit(intel->st->pipe, intel->st->cso_context);
+	}
+
+	if (intel->last_swap_fence) {
+		driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, TRUE);
+		driFenceUnReference(&intel->last_swap_fence);
+		intel->last_swap_fence = NULL;
+	}
+	intel->last_swap_fence = intel->first_swap_fence;
+	intel->first_swap_fence = NULL;
+
+	/* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
+	 * should work regardless.
+	 */
+	LOCK_HARDWARE(intel);
+	/* if this drawable isn't currently bound the LOCK_HARDWARE done on the
+	 * current context (which is what intelScreenContext should return) might
+	 * not get a contended lock and thus cliprects not updated (tests/manywin)
+	 */
+	if (intel_context(dPriv->driContextPriv) != intel)
+		DRI_VALIDATE_DRAWABLE_INFO(intel->driScreen, dPriv);
+
+	if (dPriv && dPriv->numClipRects) {
+		const int srcWidth = tex->width[0];
+		const int srcHeight = tex->height[0];
+		const int nbox = dPriv->numClipRects;
+		const drm_clip_rect_t *pbox = dPriv->pClipRects;
+		int i;
+
+		assert(surf->buffer);
+		assert(surf->cpp == cpp);
+
+
+		for (i = 0; i < nbox; i++, pbox++) {
+			drm_clip_rect_t box;
+			drm_clip_rect_t sbox;
+
+			if (pbox->x1 > pbox->x2 ||
+					pbox->y1 > pbox->y2 ||
+					pbox->x2 > intelScreen->front.width ||
+					pbox->y2 > intelScreen->front.height) {
+				/* invalid cliprect, skip it */
+				continue;
+			}
+
+			box = *pbox;
+
+			if (rect) {
+				/* intersect cliprect with user-provided src rect */
+				drm_clip_rect_t rrect;
+
+				rrect.x1 = dPriv->x + rect->x1;
+				rrect.y1 = (dPriv->h - rect->y1 - rect->y2) + dPriv->y;
+				rrect.x2 = rect->x2 + rrect.x1;
+				rrect.y2 = rect->y2 + rrect.y1;
+				if (rrect.x1 > box.x1)
+					box.x1 = rrect.x1;
+				if (rrect.y1 > box.y1)
+					box.y1 = rrect.y1;
+				if (rrect.x2 < box.x2)
+					box.x2 = rrect.x2;
+				if (rrect.y2 < box.y2)
+					box.y2 = rrect.y2;
+
+				if (box.x1 > box.x2 || box.y1 > box.y2)
+					continue;
+			}
+
+			/* restrict blit to size of actually rendered area */
+			if (box.x2 - box.x1 > srcWidth)
+				box.x2 = srcWidth + box.x1;
+			if (box.y2 - box.y1 > srcHeight)
+				box.y2 = srcHeight + box.y1;
+
+			DBG(SWAP, "\tdbox x1 x2 y1 y2 %d %d %d %d\n",
+					box.x1, box.x2, box.y1, box.y2);
+
+			sbox.x1 = box.x1 - dPriv->x;
+			sbox.y1 = box.y1 - dPriv->y;
+			sbox.x2 = box.x2 - dPriv->x;
+			sbox.y2 = box.y2 - dPriv->y;
+			DBG(SWAP, "\tsbox x1 x2 y1 y2 %d %d %d %d\n",
+					sbox.x1, sbox.x2, sbox.y1, sbox.y2);
+
+			assert(box.x1 < box.x2);
+			assert(box.y1 < box.y2);
+
+			util_blit_pixels_tex(intel->blit,
+					tex,
+					sbox.x1, sbox.y1,
+					sbox.x2, sbox.y2,
+					intelScreen->front.surface,
+					box.x1, box.y1,
+					box.x2, box.y2,
+					0.0,
+					PIPE_TEX_MIPFILTER_NEAREST);
+		}
+
+		if (intel->first_swap_fence)
+			driFenceUnReference(&intel->first_swap_fence);
+		intel->first_swap_fence = intel_be_batchbuffer_flush(intel->base.batch);
+	}
+
+
+	UNLOCK_HARDWARE(intel);
+
+	if (intel->lastStamp != dPriv->lastStamp) {
+		intelUpdateWindowSize(dPriv);
+		intel->lastStamp = dPriv->lastStamp;
+	}
 }
 
-
-
 /**
  * This will be called whenever the currently bound window is moved/resized.
  */
@@ -212,27 +181,27 @@ intelUpdateWindowSize(__DRIdrawablePrivate *dPriv)
    st_resize_framebuffer(intelfb->stfb, dPriv->w, dPriv->h);
 }
 
-
-
+/**
+ * Called from glx code
+ */
 void
 intelSwapBuffers(__DRIdrawablePrivate * dPriv)
 {
    struct intel_framebuffer *intel_fb = intel_framebuffer(dPriv);
-   struct pipe_surface *back_surf;
+   struct pipe_texture *back_tex;
 
    assert(intel_fb);
    assert(intel_fb->stfb);
 
-   back_surf = st_get_framebuffer_surface(intel_fb->stfb,
-                                          ST_SURFACE_BACK_LEFT);
-   if (back_surf) {
+   back_tex = st_get_framebuffer_texture(intel_fb->stfb,
+                                         ST_SURFACE_BACK_LEFT);
+   if (back_tex) {
       st_notify_swapbuffers(intel_fb->stfb);
-      intelDisplaySurface(dPriv, back_surf, NULL);
+      intelDisplaySurface(dPriv, back_tex, NULL);
       st_notify_swapbuffers_complete(intel_fb->stfb);
    }
 }
 
-
 /**
  * Called via glXCopySubBufferMESA() to copy a subrect of the back
  * buffer to the front buffer/screen.
@@ -241,14 +210,14 @@ void
 intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h)
 {
    struct intel_framebuffer *intel_fb = intel_framebuffer(dPriv);
-   struct pipe_surface *back_surf;
+   struct pipe_texture *back_tex;
 
    assert(intel_fb);
    assert(intel_fb->stfb);
 
-   back_surf = st_get_framebuffer_surface(intel_fb->stfb,
-                                          ST_SURFACE_BACK_LEFT);
-   if (back_surf) {
+   back_tex = st_get_framebuffer_texture(intel_fb->stfb,
+                                         ST_SURFACE_BACK_LEFT);
+   if (back_tex) {
       drm_clip_rect_t rect;
       rect.x1 = x;
       rect.y1 = y;
@@ -256,6 +225,6 @@ intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h)
       rect.y2 = h;
 
       st_notify_swapbuffers(intel_fb->stfb);
-      intelDisplaySurface(dPriv, back_surf, &rect);
+      intelDisplaySurface(dPriv, back_tex, &rect);
    }
 }
diff --git a/src/gallium/winsys/dri/intel/intel_swapbuffers.h b/src/gallium/winsys/dri/intel/intel_swapbuffers.h
index 46c9bab3af..88c479f81d 100644
--- a/src/gallium/winsys/dri/intel/intel_swapbuffers.h
+++ b/src/gallium/winsys/dri/intel/intel_swapbuffers.h
@@ -29,11 +29,11 @@
 #define INTEL_SWAPBUFFERS_H
 
 
-struct pipe_surface;
+struct pipe_texture;
 
 
 extern void intelDisplaySurface(__DRIdrawablePrivate * dPriv,
-                                struct pipe_surface *surf,
+                                struct pipe_texture *surf,
                                 const drm_clip_rect_t * rect);
 
 extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);