i915: amalgamate composite into a single primitive list

Combine all the calls to composite between prepare_composite and
done_composite into a single primitive list, rather than a primitive
call per composite().

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

5 files changed, 43 insertions, 6 deletions

diff --git a/src/i830.h b/src/i830.h
index 2c875f33..bf0fc49d 100644
--- a/src/i830.h
+++ b/src/i830.h
@@ -295,6 +295,7 @@ typedef struct intel_screen_private {
 
 	CloseScreenProcPtr CloseScreen;
 
+	void (*vertex_flush) (struct intel_screen_private *intel);
 	void (*batch_flush_notify) (ScrnInfoPtr scrn);
 
 	uxa_driver_t *uxa_driver;
@@ -357,6 +358,9 @@ typedef struct intel_screen_private {
 		uint32_t dst_format;
 	} i915_render_state;
 
+	uint32_t prim_offset;
+	uint32_t prim_count;
+
 	/* 965 render acceleration state */
 	struct gen4_render_state *gen4_render_state;
 
@@ -474,6 +478,7 @@ Bool i915_prepare_composite(int op, PicturePtr sourcec, PicturePtr mask,
 			    PixmapPtr maskPixmap, PixmapPtr destPixmap);
 void i915_composite(PixmapPtr dest, int srcX, int srcY,
 		    int maskX, int maskY, int dstX, int dstY, int w, int h);
+void i915_vertex_flush(intel_screen_private *intel);
 void i915_batch_flush_notify(ScrnInfoPtr scrn);
 void i830_batch_flush_notify(ScrnInfoPtr scrn);
 /* i965_render.c */
diff --git a/src/i830_batchbuffer.c b/src/i830_batchbuffer.c
index c23b0b85..0fe81d03 100644
--- a/src/i830_batchbuffer.c
+++ b/src/i830_batchbuffer.c
@@ -165,6 +165,9 @@ void intel_batch_submit(ScrnInfoPtr scrn)
 	if (intel->batch_used == 0)
 		return;
 
+	if (intel->vertex_flush)
+		intel->vertex_flush(intel);
+
 	/* Emit a padding dword if we aren't going to be quad-word aligned. */
 	if ((intel->batch_used & 4) == 0) {
 		*(uint32_t *) (intel->batch_ptr + intel->batch_used) = MI_NOOP;
diff --git a/src/i830_driver.c b/src/i830_driver.c
index db8af063..6ec6f519 100644
--- a/src/i830_driver.c
+++ b/src/i830_driver.c
@@ -1227,11 +1227,12 @@ I830ScreenInit(int scrnIndex, ScreenPtr screen, int argc, char **argv)
 		return FALSE;
 	}
 
-	if (IS_I965G(intel))
+	if (IS_I965G(intel)) {
 		intel->batch_flush_notify = i965_batch_flush_notify;
-	else if (IS_I9XX(intel))
+	} else if (IS_I9XX(intel)) {
+		intel->vertex_flush = i915_vertex_flush;
 		intel->batch_flush_notify = i915_batch_flush_notify;
-	else
+	} else
 		intel->batch_flush_notify = i830_batch_flush_notify;
 
 	miInitializeBackingStore(screen);
diff --git a/src/i830_uxa.c b/src/i830_uxa.c
index d4028959..f68ec701 100644
--- a/src/i830_uxa.c
+++ b/src/i830_uxa.c
@@ -467,6 +467,10 @@ static void i830_uxa_done_copy(PixmapPtr dest)
 void i830_done_composite(PixmapPtr dest)
 {
 	ScrnInfoPtr scrn = xf86Screens[dest->drawable.pScreen->myNum];
+	intel_screen_private *intel = intel_get_screen_private(scrn);
+
+	if (intel->vertex_flush)
+		intel->vertex_flush(intel);
 
 	i830_debug_flush(scrn);
 }
@@ -1051,6 +1055,9 @@ Bool i830_uxa_init(ScreenPtr screen)
 	intel->uxa_driver->uxa_major = 1;
 	intel->uxa_driver->uxa_minor = 0;
 
+	intel->prim_offset = 0;
+	intel->prim_count = 0;
+
 	/* Solid fill */
 	intel->uxa_driver->check_solid = i830_uxa_check_solid;
 	intel->uxa_driver->prepare_solid = i830_uxa_prepare_solid;
diff --git a/src/i915_render.c b/src/i915_render.c
index 59d92481..7c204b62 100644
--- a/src/i915_render.c
+++ b/src/i915_render.c
@@ -708,6 +708,9 @@ static void i915_emit_composite_setup(ScrnInfoPtr scrn)
 
 	    FS_END();
 	}
+
+	intel->prim_offset = 0;
+	intel->prim_count = 0;
 }
 
 /* Emit the vertices for a single composite rectangle.
@@ -850,9 +853,10 @@ i915_emit_composite_primitive(PixmapPtr dest,
 
 	num_floats = 3 * per_vertex;
 
-	ATOMIC_BATCH(1 + num_floats);
+	ATOMIC_BATCH(num_floats);
+
+	intel->prim_count += num_floats;
 
-	OUT_BATCH(PRIM3D_INLINE | PRIM3D_RECTLIST | (num_floats - 1));
 	OUT_BATCH_F(intel->dst_coord_adjust + dstX + w);
 	OUT_BATCH_F(intel->dst_coord_adjust + dstY + h);
 	if (! intel->render_source_is_solid) {
@@ -926,13 +930,30 @@ i915_composite(PixmapPtr dest, int srcX, int srcY, int maskX, int maskY,
 	if (intel->needs_render_state_emit)
 		i915_emit_composite_setup(scrn);
 
+	if (intel->prim_offset == 0) {
+		intel->prim_offset = intel->batch_used;
+		ATOMIC_BATCH(1);
+		OUT_BATCH(PRIM3D_INLINE | PRIM3D_RECTLIST);
+		ADVANCE_BATCH();
+	}
+
 	i915_emit_composite_primitive(dest, srcX, srcY, maskX, maskY, dstX,
 				      dstY, w, h);
 
 	intel_batch_end_atomic(scrn);
 }
 
-void i915_batch_flush_notify(ScrnInfoPtr scrn)
+void
+i915_vertex_flush(intel_screen_private *intel)
+{
+	if (intel->prim_offset) {
+		*(uint32_t *) (intel->batch_ptr + intel->prim_offset) |= intel->prim_count - 1;
+		intel->prim_offset = 0;
+	}
+}
+
+void
+i915_batch_flush_notify(ScrnInfoPtr scrn)
 {
 	intel_screen_private *intel = intel_get_screen_private(scrn);