[Mesa-dev] [PATCH] i965/fs/generator: Don't use the address immediate for MOV_INDIRECT

[Matt Turner]

On Fri, Oct 28, 2016 at 2:53 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
> The address immediate field is only 9 bits and, since the value is in
> bytes, the highest GRF we can point to with it is g15.  This makes it
> pretty close to useless for MOV_INDIRECT.  There were already piles of
> restrictions preventing us from using it prior to Broadwell, so let's get
> rid of the gen8+ code path entirely.
>
> Signed-off-by: Jason Ekstrand <jason at jlekstrand.net>
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97779
> Cc: "12.0 13.0" <mesa-stable at lists.freedesktop.org>
> ---
>  src/mesa/drivers/dri/i965/brw_fs_generator.cpp | 55 +++++++++++++-------------
>  1 file changed, 28 insertions(+), 27 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> index d25d26a..7130bf5 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> @@ -386,33 +386,34 @@ fs_generator::generate_mov_indirect(fs_inst *inst,
>           retype(spread(indirect_byte_offset, 2), BRW_REGISTER_TYPE_UW);
>
>        struct brw_reg ind_src;
> -      if (devinfo->gen < 8) {
> -         /* From the Haswell PRM section "Register Region Restrictions":
> -          *
> -          *    "The lower bits of the AddressImmediate must not overflow to
> -          *    change the register address.  The lower 5 bits of Address
> -          *    Immediate when added to lower 5 bits of address register gives
> -          *    the sub-register offset. The upper bits of Address Immediate
> -          *    when added to upper bits of address register gives the register
> -          *    address. Any overflow from sub-register offset is dropped."
> -          *
> -          * This restriction is only listed in the Haswell PRM but emperical
> -          * testing indicates that it applies on all older generations and is
> -          * lifted on Broadwell.
> -          *
> -          * Since the indirect may cause us to cross a register boundary, this
> -          * makes the base offset almost useless.  We could try and do
> -          * something clever where we use a actual base offset if
> -          * base_offset % 32 == 0 but that would mean we were generating
> -          * different code depending on the base offset.  Instead, for the
> -          * sake of consistency, we'll just do the add ourselves.
> -          */
> -         brw_ADD(p, addr, indirect_byte_offset, brw_imm_uw(imm_byte_offset));
> -         ind_src = brw_VxH_indirect(0, 0);
> -      } else {
> -         brw_MOV(p, addr, indirect_byte_offset);
> -         ind_src = brw_VxH_indirect(0, imm_byte_offset);
> -      }
> +
> +      /* There are a number of reasons why we don't use the base offset here.
> +       * One reason is that the field is only 9 bits which means we can only
> +       * use it on the first 16 GRFs.  Also, from the Haswell PRM section

s/on/to access/

> +       * "Register Region Restrictions":
> +       *
> +       *    "The lower bits of the AddressImmediate must not overflow to
> +       *    change the register address.  The lower 5 bits of Address
> +       *    Immediate when added to lower 5 bits of address register gives
> +       *    the sub-register offset. The upper bits of Address Immediate
> +       *    when added to upper bits of address register gives the register
> +       *    address. Any overflow from sub-register offset is dropped."
> +       *
> +       * Since the indirect may cause us to cross a register boundary, this
> +       * makes the base offset almost useless.  We could try and do something
> +       * clever where we use a actual base offset if base_offset % 32 == 0 but
> +       * that would mean we were generating different code depending on the
> +       * base offset.  Instead, for the sake of consistency, we'll just do the
> +       * add ourselves.  This restriction is only listed in the Haswell PRM
> +       * but emperical testing indicates that it applies on all older

empirical

> +       * generations and is lifted on Broadwell.
> +       *
> +       * In the end, while base_offset is nice to look at in the generated
> +       * code, using it saves us 0 instructions and would require quite a bit
> +       * of case-by-case work.  It's just not worth it.
> +       */
> +      brw_ADD(p, addr, indirect_byte_offset, brw_imm_uw(imm_byte_offset));
> +      ind_src = brw_VxH_indirect(0, 0);

I think you can move the declaration of ind_src here as well.

Reviewed-by: Matt Turner <mattst88 at gmail.com>