[Mesa-dev] [PATCH] i965: implement (un)mapImage

[Chris Wilson]

Quoting Julien Isorce (2017-11-09 17:03:13)
> v2: add early return if (flag & MAP_INTERNAL_MASK)
> v3: take input rect into account and test with kmscube and piglit.
> v4: handle wraparound and bo reference.
> 
> Already implemented for Gallium drivers.
> 
> Useful for gbm_bo_(un)map.
> 
> Tests:
>   By porting wayland/weston/clients/simple-dmabuf-drm.c to GBM.
>   kmscube --mode=rgba
>   kmscube --mode=nv12-1img
>   kmscube --mode=nv12-2img
>   piglit ext_image_dma_buf_import-refcount -auto
>   piglit ext_image_dma_buf_import-transcode-nv12-as-r8-gr88 -auto
>   piglit ext_image_dma_buf_import-sample_rgb -fmt=AR24 -alpha-one -auto
>   piglit ext_image_dma_buf_import-sample_rgb -fmt=XR24 -auto
>   piglit ext_image_dma_buf_import-sample_yuv -fmt=NV12 -auto
>   piglit ext_image_dma_buf_import-sample_yuv -fmt=YU12 -auto
>   piglit ext_image_dma_buf_import-sample_yuv -fmt=YV12 -auto
> 
> Signed-off-by: Julien Isorce <jisorce at oblong.com>
> ---
>  src/mesa/drivers/dri/i965/intel_screen.c | 75 +++++++++++++++++++++++++++++++-
>  1 file changed, 73 insertions(+), 2 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/intel_screen.c b/src/mesa/drivers/dri/i965/intel_screen.c
> index cdc36ad..6074ee8 100644
> --- a/src/mesa/drivers/dri/i965/intel_screen.c
> +++ b/src/mesa/drivers/dri/i965/intel_screen.c
> @@ -755,6 +755,77 @@ intel_create_image(__DRIscreen *dri_screen,
>                                 loaderPrivate);
>  }
>  
> +static void *
> +intel_map_image(__DRIcontext *context, __DRIimage *image,
> +                int x0, int y0, int width, int height,
> +                unsigned int flags, int *stride, void **map_info)
> +{
> +   struct brw_context *brw = NULL;
> +   struct brw_bo *bo = NULL;
> +   void *raw_data = NULL;
> +
> +   if (!context || !image || !stride || !map_info || *map_info)
> +      return NULL;
> +
> +   if (x0 < 0 || x0 > image->width || width > image->width - x0)
> +      return NULL;

Second careful thought here is that you want x0 >= image->width so that
we correctly exclude a 0 width request on the boundary. Or reject 0
width/height.

> +
> +   if (y0 < 0 || y0 > image->height || height > image->height - y0)
> +        return NULL;
> +
> +   if (flags & MAP_INTERNAL_MASK)
> +      return NULL;
> +
> +   brw = context->driverPrivate;
> +   bo = image->bo;
> +
> +   assert(brw);
> +   assert(bo);
> +
> +   brw_bo_reference(bo);
> +
> +   /* DRI flags and GL_MAP.*_BIT flags are the same, so just pass them on. */
> +   raw_data = brw_bo_map(brw, bo, flags);
> +
> +   if (raw_data) {
> +      GLuint pix_w = 1;
> +      GLuint pix_h = 1;
> +      GLint pix_bytes = 1;
> +
> +      *map_info = raw_data;
> +      *stride = image->pitch;
> +
> +      _mesa_get_format_block_size(image->format, &pix_w, &pix_h);
> +      pix_bytes = _mesa_get_format_bytes(image->format);
> +
> +      assert(pix_w);
> +      assert(pix_h);
> +      assert(pix_bytes > 0);
> +
> +      raw_data += ((x0 / pix_w) * pix_bytes) + (y0 / pix_h) * image->pitch;

Ok, this matches up with similar code for intel_miptree_map_gtt().

> +   } else {
> +      brw_bo_unreference(bo);
> +   }

Going back to Ben's patch, the rationale for ref/unref was that we could
map the bo independently of the image; i.e. later on image->bo may be
pointing to something else.

So *map_info = brw_bo_reference(bo); in intel_map_image() and
brw_bo_unreference(map_info) here.

Taking the suggestions from elsewhere;
	raw_data = brw_bo_map(brw, image->bo, flags);
	if (!raw_data)
		return NULL;

	...
	_mesa_get_format_block_size();
	_mesa_get_format_bytes;
	raw_data += x0 / pix_w * pix_bytes + y0 / pix_h * image->pitch;

	*stride = image->pitch;
	*map_info = brw_bo_reference(image->bo);
	return raw_data;
-Chris