radeon_state.c

内核linux2.4.20,可跟rtlinux3.2打补丁 组成实时linux系统,编译内核

	unsigned int flags = clear->flags;
	int i;
	RING_LOCALS;
	DRM_DEBUG( "%s\n", __FUNCTION__ );

	if ( dev_priv->page_flipping && dev_priv->current_page == 1 ) {
		unsigned int tmp = flags;

		flags &= ~(RADEON_FRONT | RADEON_BACK);
		if ( tmp & RADEON_FRONT ) flags |= RADEON_BACK;
		if ( tmp & RADEON_BACK )  flags |= RADEON_FRONT;
	}

	for ( i = 0 ; i < nbox ; i++ ) {
		int x = pbox[i].x1;
		int y = pbox[i].y1;
		int w = pbox[i].x2 - x;
		int h = pbox[i].y2 - y;

		DRM_DEBUG( "dispatch clear %d,%d-%d,%d flags 0x%x\n",
			   x, y, w, h, flags );

		if ( flags & (RADEON_FRONT | RADEON_BACK) ) {
			BEGIN_RING( 4 );

			/* Ensure the 3D stream is idle before doing a
			 * 2D fill to clear the front or back buffer.
			 */
			RADEON_WAIT_UNTIL_3D_IDLE();

			OUT_RING( CP_PACKET0( RADEON_DP_WRITE_MASK, 0 ) );
			OUT_RING( clear->color_mask );

			ADVANCE_RING();

			/* Make sure we restore the 3D state next time.
			 */
			dev_priv->sarea_priv->dirty |= (RADEON_UPLOAD_CONTEXT |
							RADEON_UPLOAD_MASKS);
		}

		if ( flags & RADEON_FRONT ) {
			BEGIN_RING( 6 );

			OUT_RING( CP_PACKET3( RADEON_CNTL_PAINT_MULTI, 4 ) );
			OUT_RING( RADEON_GMC_DST_PITCH_OFFSET_CNTL |
				  RADEON_GMC_BRUSH_SOLID_COLOR |
				  (dev_priv->color_fmt << 8) |
				  RADEON_GMC_SRC_DATATYPE_COLOR |
				  RADEON_ROP3_P |
				  RADEON_GMC_CLR_CMP_CNTL_DIS );

			OUT_RING( dev_priv->front_pitch_offset );
			OUT_RING( clear->clear_color );

			OUT_RING( (x << 16) | y );
			OUT_RING( (w << 16) | h );

			ADVANCE_RING();
		}

		if ( flags & RADEON_BACK ) {
			BEGIN_RING( 6 );

			OUT_RING( CP_PACKET3( RADEON_CNTL_PAINT_MULTI, 4 ) );
			OUT_RING( RADEON_GMC_DST_PITCH_OFFSET_CNTL |
				  RADEON_GMC_BRUSH_SOLID_COLOR |
				  (dev_priv->color_fmt << 8) |
				  RADEON_GMC_SRC_DATATYPE_COLOR |
				  RADEON_ROP3_P |
				  RADEON_GMC_CLR_CMP_CNTL_DIS );

			OUT_RING( dev_priv->back_pitch_offset );
			OUT_RING( clear->clear_color );

			OUT_RING( (x << 16) | y );
			OUT_RING( (w << 16) | h );

			ADVANCE_RING();

		}

		if ( flags & RADEON_DEPTH ) {
			drm_radeon_depth_clear_t *depth_clear =
			   &dev_priv->depth_clear;

			if ( sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS ) {
				radeon_emit_state( dev_priv );
			}

			/* FIXME: Render a rectangle to clear the depth
			 * buffer.  So much for those "fast Z clears"...
			 */
			BEGIN_RING( 23 );

			RADEON_WAIT_UNTIL_2D_IDLE();

			OUT_RING( CP_PACKET0( RADEON_PP_CNTL, 1 ) );
			OUT_RING( 0x00000000 );
			OUT_RING( depth_clear->rb3d_cntl );
			OUT_RING( CP_PACKET0( RADEON_RB3D_ZSTENCILCNTL, 0 ) );
			OUT_RING( depth_clear->rb3d_zstencilcntl );
			OUT_RING( CP_PACKET0( RADEON_RB3D_PLANEMASK, 0 ) );
			OUT_RING( 0x00000000 );
			OUT_RING( CP_PACKET0( RADEON_SE_CNTL, 0 ) );
			OUT_RING( depth_clear->se_cntl );

			OUT_RING( CP_PACKET3( RADEON_3D_DRAW_IMMD, 10 ) );
			OUT_RING( RADEON_VTX_Z_PRESENT );
			OUT_RING( (RADEON_PRIM_TYPE_RECT_LIST |
				   RADEON_PRIM_WALK_RING |
				   RADEON_MAOS_ENABLE |
				   RADEON_VTX_FMT_RADEON_MODE |
				   (3 << RADEON_NUM_VERTICES_SHIFT)) );

			OUT_RING( depth_boxes[i].ui[CLEAR_X1] );
			OUT_RING( depth_boxes[i].ui[CLEAR_Y1] );
			OUT_RING( depth_boxes[i].ui[CLEAR_DEPTH] );

			OUT_RING( depth_boxes[i].ui[CLEAR_X1] );
			OUT_RING( depth_boxes[i].ui[CLEAR_Y2] );
			OUT_RING( depth_boxes[i].ui[CLEAR_DEPTH] );

			OUT_RING( depth_boxes[i].ui[CLEAR_X2] );
			OUT_RING( depth_boxes[i].ui[CLEAR_Y2] );
			OUT_RING( depth_boxes[i].ui[CLEAR_DEPTH] );

			ADVANCE_RING();

			/* Make sure we restore the 3D state next time.
			 */
			dev_priv->sarea_priv->dirty |= (RADEON_UPLOAD_CONTEXT |
							RADEON_UPLOAD_SETUP |
							RADEON_UPLOAD_MASKS);
		}
	}

	/* Increment the clear counter.  The client-side 3D driver must
	 * wait on this value before performing the clear ioctl.  We
	 * need this because the card's so damned fast...
	 */
	dev_priv->sarea_priv->last_clear++;

	BEGIN_RING( 4 );

	RADEON_CLEAR_AGE( dev_priv->sarea_priv->last_clear );
	RADEON_WAIT_UNTIL_IDLE();

	ADVANCE_RING();
}

static void radeon_cp_dispatch_swap( drm_device_t *dev )
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
	int nbox = sarea_priv->nbox;
	drm_clip_rect_t *pbox = sarea_priv->boxes;
	int i;
	RING_LOCALS;
	DRM_DEBUG( "%s\n", __FUNCTION__ );

#if RADEON_PERFORMANCE_BOXES
	/* Do some trivial performance monitoring...
	 */
	radeon_cp_performance_boxes( dev_priv );
#endif

	/* Wait for the 3D stream to idle before dispatching the bitblt.
	 * This will prevent data corruption between the two streams.
	 */
	BEGIN_RING( 2 );

	RADEON_WAIT_UNTIL_3D_IDLE();

	ADVANCE_RING();

	for ( i = 0 ; i < nbox ; i++ ) {
		int x = pbox[i].x1;
		int y = pbox[i].y1;
		int w = pbox[i].x2 - x;
		int h = pbox[i].y2 - y;

		DRM_DEBUG( "dispatch swap %d,%d-%d,%d\n",
			   x, y, w, h );

		BEGIN_RING( 7 );

		OUT_RING( CP_PACKET3( RADEON_CNTL_BITBLT_MULTI, 5 ) );
		OUT_RING( RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
			  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
			  RADEON_GMC_BRUSH_NONE |
			  (dev_priv->color_fmt << 8) |
			  RADEON_GMC_SRC_DATATYPE_COLOR |
			  RADEON_ROP3_S |
			  RADEON_DP_SRC_SOURCE_MEMORY |
			  RADEON_GMC_CLR_CMP_CNTL_DIS |
			  RADEON_GMC_WR_MSK_DIS );

		OUT_RING( dev_priv->back_pitch_offset );
		OUT_RING( dev_priv->front_pitch_offset );

		OUT_RING( (x << 16) | y );
		OUT_RING( (x << 16) | y );
		OUT_RING( (w << 16) | h );

		ADVANCE_RING();
	}

	/* Increment the frame counter.  The client-side 3D driver must
	 * throttle the framerate by waiting for this value before
	 * performing the swapbuffer ioctl.
	 */
	dev_priv->sarea_priv->last_frame++;

	BEGIN_RING( 4 );

	RADEON_FRAME_AGE( dev_priv->sarea_priv->last_frame );
	RADEON_WAIT_UNTIL_2D_IDLE();

	ADVANCE_RING();
}

static void radeon_cp_dispatch_flip( drm_device_t *dev )
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	RING_LOCALS;
	DRM_DEBUG( "%s: page=%d\n", __FUNCTION__, dev_priv->current_page );

#if RADEON_PERFORMANCE_BOXES
	/* Do some trivial performance monitoring...
	 */
	radeon_cp_performance_boxes( dev_priv );
#endif

	BEGIN_RING( 6 );

	RADEON_WAIT_UNTIL_3D_IDLE();
	RADEON_WAIT_UNTIL_PAGE_FLIPPED();

	OUT_RING( CP_PACKET0( RADEON_CRTC_OFFSET, 0 ) );

	if ( dev_priv->current_page == 0 ) {
		OUT_RING( dev_priv->back_offset );
		dev_priv->current_page = 1;
	} else {
		OUT_RING( dev_priv->front_offset );
		dev_priv->current_page = 0;
	}

	ADVANCE_RING();

	/* Increment the frame counter.  The client-side 3D driver must
	 * throttle the framerate by waiting for this value before
	 * performing the swapbuffer ioctl.
	 */
	dev_priv->sarea_priv->last_frame++;

	BEGIN_RING( 2 );

	RADEON_FRAME_AGE( dev_priv->sarea_priv->last_frame );

	ADVANCE_RING();
}

static void radeon_cp_dispatch_vertex( drm_device_t *dev,
				       drm_buf_t *buf )
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
	int format = sarea_priv->vc_format;
	int offset = dev_priv->agp_buffers_offset + buf->offset;
	int size = buf->used;
	int prim = buf_priv->prim;
	int i = 0;
	RING_LOCALS;
	DRM_DEBUG( "%s: nbox=%d\n", __FUNCTION__, sarea_priv->nbox );

	if ( 0 )
		radeon_print_dirty( "dispatch_vertex", sarea_priv->dirty );

	if ( buf->used ) {
		buf_priv->dispatched = 1;

		if ( sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS ) {
			radeon_emit_state( dev_priv );
		}

		do {
			/* Emit the next set of up to three cliprects */
			if ( i < sarea_priv->nbox ) {
				radeon_emit_clip_rect( dev_priv,
						       &sarea_priv->boxes[i] );
			}

			/* Emit the vertex buffer rendering commands */
			BEGIN_RING( 5 );

			OUT_RING( CP_PACKET3( RADEON_3D_RNDR_GEN_INDX_PRIM, 3 ) );
			OUT_RING( offset );
			OUT_RING( size );
			OUT_RING( format );
			OUT_RING( prim | RADEON_PRIM_WALK_LIST |
				  RADEON_COLOR_ORDER_RGBA |
				  RADEON_VTX_FMT_RADEON_MODE |
				  (size << RADEON_NUM_VERTICES_SHIFT) );

			ADVANCE_RING();

			i++;
		} while ( i < sarea_priv->nbox );
	}

	if ( buf_priv->discard ) {
		buf_priv->age = dev_priv->sarea_priv->last_dispatch;

		/* Emit the vertex buffer age */
		BEGIN_RING( 2 );
		RADEON_DISPATCH_AGE( buf_priv->age );
		ADVANCE_RING();

		buf->pending = 1;
		buf->used = 0;
		/* FIXME: Check dispatched field */
		buf_priv->dispatched = 0;
	}

	dev_priv->sarea_priv->last_dispatch++;

	sarea_priv->dirty &= ~RADEON_UPLOAD_CLIPRECTS;
	sarea_priv->nbox = 0;
}


static void radeon_cp_dispatch_indirect( drm_device_t *dev,
					 drm_buf_t *buf,
					 int start, int end )
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
	RING_LOCALS;
	DRM_DEBUG( "indirect: buf=%d s=0x%x e=0x%x\n",
		   buf->idx, start, end );

	if ( start != end ) {
		int offset = (dev_priv->agp_buffers_offset
			      + buf->offset + start);
		int dwords = (end - start + 3) / sizeof(u32);

		/* Indirect buffer data must be an even number of
		 * dwords, so if we've been given an odd number we must
		 * pad the data with a Type-2 CP packet.
		 */
		if ( dwords & 1 ) {
			u32 *data = (u32 *)
				((char *)dev_priv->buffers->handle
				 + buf->offset + start);
			data[dwords++] = RADEON_CP_PACKET2;
		}

		buf_priv->dispatched = 1;

		/* Fire off the indirect buffer */
		BEGIN_RING( 3 );

		OUT_RING( CP_PACKET0( RADEON_CP_IB_BASE, 1 ) );
		OUT_RING( offset );
		OUT_RING( dwords );

		ADVANCE_RING();
	}

	if ( buf_priv->discard ) {
		buf_priv->age = dev_priv->sarea_priv->last_dispatch;

		/* Emit the indirect buffer age */
		BEGIN_RING( 2 );
		RADEON_DISPATCH_AGE( buf_priv->age );
		ADVANCE_RING();

		buf->pending = 1;
		buf->used = 0;
		/* FIXME: Check dispatched field */
		buf_priv->dispatched = 0;
	}

	dev_priv->sarea_priv->last_dispatch++;
}

static void radeon_cp_dispatch_indices( drm_device_t *dev,
					drm_buf_t *buf,
					int start, int end,
					int count )
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
	int format = sarea_priv->vc_format;
	int offset = dev_priv->agp_buffers_offset;
	int prim = buf_priv->prim;
	u32 *data;
	int dwords;
	int i = 0;
	RING_LOCALS;
	DRM_DEBUG( "indices: s=%d e=%d c=%d\n", start, end, count );

	if ( 0 )
		radeon_print_dirty( "dispatch_indices", sarea_priv->dirty );

	if ( start != end ) {
		buf_priv->dispatched = 1;

		if ( sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS ) {
			radeon_emit_state( dev_priv );
		}

		dwords = (end - start + 3) / sizeof(u32);

		data = (u32 *)((char *)dev_priv->buffers->handle
			       + buf->offset + start);

		data[0] = CP_PACKET3( RADEON_3D_RNDR_GEN_INDX_PRIM, dwords-2 );

		data[1] = offset;
		data[2] = RADEON_MAX_VB_VERTS;
		data[3] = format;
		data[4] = (prim | RADEON_PRIM_WALK_IND |
			   RADEON_COLOR_ORDER_RGBA |
			   RADEON_VTX_FMT_RADEON_MODE |
			   (count << RADEON_NUM_VERTICES_SHIFT) );

		if ( count & 0x1 ) {
			data[dwords-1] &= 0x0000ffff;
		}

		do {
			/* Emit the next set of up to three cliprects */
			if ( i < sarea_priv->nbox ) {
				radeon_emit_clip_rect( dev_priv,
						       &sarea_priv->boxes[i] );
			}

			radeon_cp_dispatch_indirect( dev, buf, start, end );

			i++;
		} while ( i < sarea_priv->nbox );
	}

	if ( buf_priv->discard ) {
		buf_priv->age = dev_priv->sarea_priv->last_dispatch;

		/* Emit the vertex buffer age */
		BEGIN_RING( 2 );
		RADEON_DISPATCH_AGE( buf_priv->age );
		ADVANCE_RING();

		buf->pending = 1;
		/* FIXME: Check dispatched field */
		buf_priv->dispatched = 0;
	}

	dev_priv->sarea_priv->last_dispatch++;

	sarea_priv->dirty &= ~RADEON_UPLOAD_CLIPRECTS;
	sarea_priv->nbox = 0;
}

#define RADEON_MAX_TEXTURE_SIZE (RADEON_BUFFER_SIZE - 8 * sizeof(u32))

static int radeon_cp_dispatch_texture( drm_device_t *dev,
				       drm_radeon_texture_t *tex,
				       drm_radeon_tex_image_t *image )
{
	drm_radeon_private_t *dev_priv = dev->dev_private;
	drm_buf_t *buf;
	drm_radeon_buf_priv_t *buf_priv;
	u32 format;
	u32 *buffer;
	u8 *data;
	int size, dwords, tex_width, blit_width;
	u32 y, height;
	int ret = 0, i;
	RING_LOCALS;

	/* FIXME: Be smarter about this...
	 */
	buf = radeon_freelist_get( dev );
	if ( !buf ) return -EAGAIN;

	DRM_DEBUG( "tex: ofs=0x%x p=%d f=%d x=%hd y=%hd w=%hd h=%hd\n",
		   tex->offset >> 10, tex->pitch, tex->format,
		   image->x, image->y, image->width, image->height );

	buf_priv = buf->dev_private;