[Mesa-dev] [PATCHv3 09/20] i965/fs: Get 64-bit indirect moves working on IVB.

Samuel Iglesias Gonsálvez siglesias at igalia.com
Fri Feb 10 08:06:29 UTC 2017


On Thu, 2017-02-09 at 10:16 -0800, Francisco Jerez wrote:
> ---
> This replaces "[PATCH v2 09/20] i965/fs: indirect addressing with
> doubles is not supported in IVB/BYT".
> 
> src/mesa/drivers/dri/i965/brw_fs_generator.cpp | 27
> ++++++++++++++++++++++++--
>  1 file changed, 25 insertions(+), 2 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> index ea4a3fe1399..0e2dbe23571 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> @@ -440,7 +440,7 @@ fs_generator::generate_mov_indirect(fs_inst
> *inst,
>        brw_MOV(p, dst, reg);
>     } else {
>        /* Prior to Broadwell, there are only 8 address registers. */
> -      assert(inst->exec_size == 8 || devinfo->gen >= 8);
> +      assert(inst->exec_size <= 8 || devinfo->gen >= 8);
>  
>        /* We use VxH indirect addressing, clobbering a0.0 through
> a0.7. */
>        struct brw_reg addr = vec8(brw_address_reg(0));
> @@ -478,7 +478,30 @@ fs_generator::generate_mov_indirect(fs_inst
> *inst,
>         * 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));
> +      if (devinfo->gen >= 8 || devinfo->is_haswell ||
> type_sz(reg.type) < 8) {
> +         brw_ADD(p, addr, indirect_byte_offset,
> brw_imm_uw(imm_byte_offset));
> +      } else {
> +         /* IVB reads two address register components per channel
> for
> +          * indirectly addressed 64-bit sources, so we need to
> initialize
> +          * adjacent address components to consecutive dwords of the
> source
> +          * region by emitting two separate ADD instructions.  Found
> +          * empirically.
> +          */
> +         assert(inst->exec_size <= 4);
> +         brw_push_insn_state(p);
> +         brw_set_default_exec_size(p, cvt(inst->exec_size) - 1);
> +
> +         brw_ADD(p, spread(addr, 2), indirect_byte_offset,
> +                 brw_imm_uw(imm_byte_offset));
> +         brw_inst_set_no_dd_clear(devinfo, brw_last_inst, true);
> +
> +         brw_ADD(p, spread(suboffset(addr, 1), 2),
> indirect_byte_offset,
> +                 brw_imm_uw(imm_byte_offset + 4));
> +         brw_inst_set_no_dd_check(devinfo, brw_last_inst, true);
> +
> +         brw_pop_insn_state(p);
> +      }
> +

Oh, you did this change in the generator. The solution I was developing
last weeks was to do this in a lowering pass. Your change is better!

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias at igalia.com>

I am going to add it to my rc3 branch.

Thanks!

Sam

>        struct brw_reg ind_src = brw_VxH_indirect(0, 0);
>  
>        brw_inst *mov = brw_MOV(p, dst, retype(ind_src, dst.type));
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170210/fb356980/attachment-0001.sig>


More information about the mesa-dev mailing list