-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 04/07/2011 04:49 PM, Kenneth Graunke wrote:
> Since we currently do not support Geometry Shaders, there's no need to
> allocate any URB space to the GS unit. Give all of it to the VS, and
> document some of the units, restrictions, and calculations.
> ---
> src/mesa/drivers/dri/i965/brw_context.h | 2 +-
> src/mesa/drivers/dri/i965/gen6_urb.c | 18 +++++++++++-------
> 2 files changed, 12 insertions(+), 8 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_context.h
> b/src/mesa/drivers/dri/i965/brw_context.h
> index d63e8a4..6be7188 100644
> --- a/src/mesa/drivers/dri/i965/brw_context.h
> +++ b/src/mesa/drivers/dri/i965/brw_context.h
> @@ -560,7 +560,7 @@ struct brw_context
> * a number of 1024-bit (128-byte) units. Should be >= 1.
> */
> GLuint vs_size;
> -/* GLuint gs_size; */
> + GLuint gs_size;
>
> GLuint vs_start;
> GLuint gs_start;
> diff --git a/src/mesa/drivers/dri/i965/gen6_urb.c
> b/src/mesa/drivers/dri/i965/gen6_urb.c
> index c3819f9..83a6b97 100644
> --- a/src/mesa/drivers/dri/i965/gen6_urb.c
> +++ b/src/mesa/drivers/dri/i965/gen6_urb.c
> @@ -45,15 +45,19 @@ prepare_urb( struct brw_context *brw )
> max_urb_entry = 256;
> }
>
> - brw->urb.nr_vs_entries = max_urb_entry;
> - brw->urb.nr_gs_entries = max_urb_entry;
> -
> /* CACHE_NEW_VS_PROG */
> + /* According to volume 2a, nr_vs_entries must be a multiple of 4 in the
> + * range [24, 256] on GT2 and [24, 128] on GT1.
> + */
> + brw->urb.nr_vs_entries = max_urb_entry;
> brw->urb.vs_size = MAX2(brw->vs.prog_data->urb_entry_size, 1);
It occurs to me that this might not work. Is it possible for
urb_entry_size * max_urb_entry to be larger than will fit?
32k / 256 = 128 bytes
On SNB, it looks like the max entry size is (16 attributes + 4 header
regs) * (16 bytes per entry) = 320 bytes. Right?
>
> - if (2 * brw->urb.vs_size > urb_size)
> - brw->urb.nr_vs_entries = brw->urb.nr_gs_entries =
> - (urb_size ) / (2 * brw->urb.vs_size);
> + /* CACHE_NEW_GS_PROG */
> + /* nr_gs_entries must be a multiple of 4 in the range [0, 256] on GT2
> + * and [0, 254] on GT1. We don't believe it either.
> + */
> + brw->urb.nr_gs_entries = 0;
> + brw->urb.gs_size = 1; /* wrong, but with 0 GS entries it doesn't matter */
> }
>
> static void
> @@ -70,7 +74,7 @@ upload_urb(struct brw_context *brw)
> OUT_BATCH(_3DSTATE_URB << 16 | (3 - 2));
> OUT_BATCH(((brw->urb.vs_size - 1) << GEN6_URB_VS_SIZE_SHIFT) |
> ((brw->urb.nr_vs_entries) << GEN6_URB_VS_ENTRIES_SHIFT));
> - OUT_BATCH(((brw->urb.vs_size - 1) << GEN6_URB_GS_SIZE_SHIFT) |
> + OUT_BATCH(((brw->urb.gs_size - 1) << GEN6_URB_GS_SIZE_SHIFT) |
> ((brw->urb.nr_gs_entries) << GEN6_URB_GS_ENTRIES_SHIFT));
> ADVANCE_BATCH();
> }
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/
iEYEARECAAYFAk2fKpcACgkQX1gOwKyEAw8/bgCeMJ6bpSJMtlXKy4TOW5Y/9FOO
A0IAn25zPtHx+t0Zx5vT5cbdHii/67Qh
=aXhL
-----END PGP SIGNATURE-----
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
