[Mesa-dev] [PATCH v2] i965: Implement CopyTexSubImage2D via BLORP (and use it by default).

[Kenneth Graunke]

On 01/28/2013 02:34 PM, Kenneth Graunke wrote:
> The BLT engine has many limitations.  Currently, it can only blit
> X-tiled buffers (since we don't have a kernel API to whack the BLT
> tiling mode register), which means all depth/stencil operations get
> punted to meta code, which can be very CPU-intensive.
>
> Even if we used the BLT engine, it can't blit between buffers with
> different tiling modes, such as an X-tiled non-MSAA ARGB8888 texture
> and a Y-tiled CMS ARGB8888 renderbuffer.  This is a fundamental
> limitation, and the only way around that is to use BLORP.
>
> Previously, BLORP only handled BlitFramebuffer.  This patch adds an
> additional frontend for doing CopyTexSubImage.  It also makes it the
> default.  This is partly to increase testing and avoid hiding bugs,
> and partly because the BLORP path can already handle more cases.  With
> trivial extensions, it should be able to handle everything the BLT can.
>
> This helps PlaneShift massively, which tries to CopyTexSubImage2D
> between depth buffers whenever a player casts a spell.  Since these
> are Y-tiled, we hit meta and software ReadPixels paths, eating 99% CPU
> while delivering ~1 FPS.  This is particularly bad in an MMO setting
> because people cast spells all the time.
>
> It also helps Xonotic in 4X MSAA mode.  At default power management
> settings, I measured a 6.35138% +/- 0.672548% performance boost (n=5).
> (This data is from v1 of the patch.)
>
> No Piglit regressions on Ivybridge.  I have not tested Sandybridge.
>
> v2: Create a fake intel_renderbuffer to wrap the destination texture
>      image and then reuse do_blorp_blit rather than reimplementing most
>      of it.  Remove unnecessary clipping code and conditional rendering
>      check.
>
> Cc: Paul Berry <stereotype441 at gmail.com>
> Cc: Chad Versace <chad.versace at linux.intel.com>
> ---
>   src/mesa/drivers/dri/i965/brw_blorp_blit.cpp | 64 ++++++++++++++++++++++++++++
>   src/mesa/drivers/dri/i965/brw_context.h      |  8 ++++
>   src/mesa/drivers/dri/intel/intel_fbo.c       | 30 +++++++++++++
>   src/mesa/drivers/dri/intel/intel_fbo.h       |  4 ++
>   src/mesa/drivers/dri/intel/intel_tex_copy.c  | 32 ++++++++++----
>   5 files changed, 130 insertions(+), 8 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
> index bc7916a..e5f12fe 100644
> --- a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
> @@ -295,6 +295,70 @@ try_blorp_blit(struct intel_context *intel,
>      return true;
>   }
>
> +bool
> +brw_blorp_copytexsubimage(struct intel_context *intel,
> +                          struct gl_renderbuffer *src_rb,
> +                          struct gl_texture_image *dst_image,
> +                          int srcX0, int srcY0,
> +                          int dstX0, int dstY0,
> +                          int width, int height)
> +{
> +   struct gl_context *ctx = &intel->ctx;
> +
> +   /* BLORP is not supported before Gen6. */
> +   if (intel->gen < 6)
> +      return false;
> +
> +   /* BLORP requires the formats to match.  In the future, we should relax
> +    * this for simple format mismatches like XRGB <-> ARGB.
> +    */
> +   if (_mesa_get_srgb_format_linear(src_rb->Format) !=
> +       _mesa_get_srgb_format_linear(dst_image->TexFormat))
> +      return false;
> +
> +   /* Sync up the state of window system buffers.  We need to do this before
> +    * we go looking for the buffers.
> +    */
> +   intel_prepare_render(intel);
> +
> +   int srcY1 = srcY0 + height;
> +   int dstX1 = dstX0 + width;
> +   int dstY1 = dstY0 + height;
> +
> +   /* Source clipping shouldn't be necessary, since copytexsubimage (in
> +    * src/mesa/main/teximage.c) calls _mesa_clip_copytexsubimage() which
> +    * takes care of it.
> +    *
> +    * Destination clipping shouldn't be necessary since the restrictions on
> +    * glCopyTexSubImage prevent the user from specifying a destination rectangle
> +    * that falls outside the bounds of the destination texture.
> +    * See error_check_subtexture_dimensions().
> +    */
> +
> +   /* Account for the fact that in the system framebuffer, the origin is at
> +    * the lower left.
> +    */
> +   bool mirror_y = false;
> +   if (_mesa_is_winsys_fbo(ctx->ReadBuffer)) {
> +      GLint tmp = src_rb->Height - srcY0;
> +      srcY0 = src_rb->Height - srcY1;
> +      srcY1 = tmp;
> +      mirror_y = true;
> +   }
> +
> +   /* Create a fake/wrapper renderbuffer to allow us to use do_blorp_blit(). */
> +   struct intel_renderbuffer *dst_irb =
> +      intel_create_fake_renderbuffer_wrapper(intel, dst_image);
> +   if (!dst_irb)
> +      return false;

Uh.  I forgot to send out the prerequisite patch.

This one also doesn't delete the fake intel_renderbuffer.

Working on a v3.  Consider this NAK'd.