[Mesa-stable] [PATCH v3 2/2] i965/vec4: load dvec3/4 uniforms first in the push constant buffer

Samuel Iglesias Gonsálvez siglesias at igalia.com
Thu May 4 08:42:47 UTC 2017


On Wed, 2017-05-03 at 16:47 -0700, Francisco Jerez wrote:
> Samuel Iglesias Gonsálvez <siglesias at igalia.com> writes:
> 
> > Reorder the uniforms to load first the dvec4-aligned variables
> > in the push constant buffer and then push the vec4-aligned ones.
> > 
> > This fixes a bug were the dvec3/4 might be loaded one part on a GRF
> > and
> > the rest in next GRF, so the region parameters to read that could
> > break
> > the HW rules.
> > 
> > v2:
> > - Fix broken logic.
> > - Add a comment to explain what should be needed to optimise the
> > usage
> > of the push constant buffer slots, as this patch does not pack the
> > uniforms.
> > 
> > Signed-off-by: Samuel Iglesias Gonsálvez <siglesias at igalia.com>
> > Cc: "17.1" <mesa-stable at lists.freedesktop.org>
> > ---
> >  src/intel/compiler/brw_vec4.cpp | 97
> > +++++++++++++++++++++++++++++++----------
> >  1 file changed, 74 insertions(+), 23 deletions(-)
> > 
> > diff --git a/src/intel/compiler/brw_vec4.cpp
> > b/src/intel/compiler/brw_vec4.cpp
> > index 0909ddb586..18bfd48fa1 100644
> > --- a/src/intel/compiler/brw_vec4.cpp
> > +++ b/src/intel/compiler/brw_vec4.cpp
> > @@ -583,16 +583,44 @@ vec4_visitor::split_uniform_registers()
> >     }
> >  }
> >  
> > +/* This function returns the register number where we placed the
> > uniform */
> > +static int
> > +set_push_constant_loc(const int nr_uniforms, int
> > *new_uniform_count,
> > +                      const int src, const int size,
> > +                      int *new_loc, int *new_chan,
> > +                      int *new_chans_used)
> > +{
> > +   int dst;
> > +   /* Find the lowest place we can slot this uniform in. */
> > +   for (dst = 0; dst < nr_uniforms; dst++) {
> > +      if (new_chans_used[dst] + size <= 4)
> > +         break;
> > +   }
> > +
> > +   assert(dst < nr_uniforms);
> > +
> > +   new_loc[src] = dst;
> > +   new_chan[src] = new_chans_used[dst];
> > +   new_chans_used[dst] += size;
> > +
> > +   *new_uniform_count = MAX2(*new_uniform_count, dst + 1);
> > +   return dst;
> > +}
> > +
> >  void
> >  vec4_visitor::pack_uniform_registers()
> >  {
> >     uint8_t chans_used[this->uniforms];
> >     int new_loc[this->uniforms];
> >     int new_chan[this->uniforms];
> > +   bool is_aligned_to_dvec4[this->uniforms];
> > +   int new_chans_used[this->uniforms];
> >  
> >     memset(chans_used, 0, sizeof(chans_used));
> >     memset(new_loc, 0, sizeof(new_loc));
> >     memset(new_chan, 0, sizeof(new_chan));
> > +   memset(new_chans_used, 0, sizeof(new_chans_used));
> > +   memset(is_aligned_to_dvec4, 0, sizeof(is_aligned_to_dvec4));
> >  
> >     /* Find which uniform vectors are actually used by the
> > program.  We
> >      * expect unused vector elements when we've moved array access
> > out
> > @@ -631,10 +659,19 @@ vec4_visitor::pack_uniform_registers()
> >  
> >              unsigned channel = BRW_GET_SWZ(inst->src[i].swizzle,
> > c) + 1;
> >              unsigned used = MAX2(chans_used[reg], channel *
> > channel_size);
> > -            if (used <= 4)
> > -               chans_used[reg] = used;
> > -            else
> > -               chans_used[reg + 1] = used - 4;
> > +            /* FIXME: Marked all channels as used, so each uniform
> > will
> > +             * fully use one or two vec4s. If we want to pack
> > them, we need
> > +             * to, among other changes, set chans_used[reg] =
> > used;
> > +             * chans_used[reg+1] = used - 4; and fix the swizzle
> > at the
> > +             * end in order to set the proper location.
> > +             */
> > +            if (used <= 4) {
> > +               chans_used[reg] = 4;
> 
> Uhm...  So this change prevents the uniform packing pass from
> actually
> packing anything?  Might affect more applications negatively than
> broken
> FP64 would.  Are you planning to send a v3 that fixes the issue
> without
> disabling the optimization?

Yes, I am planning to send a v3 of this patch with the optimization in-
place.

>   May be worth holding this off until then.
> Even if that means it will miss the v17.1 release it will probably
> make
> it for the next bug-fix release.
> 

OK, thanks!

Sam


> > +            } else {
> > +               is_aligned_to_dvec4[reg] = true;
> > +               is_aligned_to_dvec4[reg + 1] = true;
> > +               chans_used[reg + 1] = 4;
> > +            }
> >           }
> >        }
> >  
> > @@ -659,42 +696,56 @@ vec4_visitor::pack_uniform_registers()
> >  
> >     int new_uniform_count = 0;
> >  
> > +   /* As the uniforms are going to be reordered, take the data
> > from a temporary
> > +    * copy of the original param[].
> > +    */
> > +   gl_constant_value **param = ralloc_array(NULL,
> > gl_constant_value*,
> > +                                            stage_prog_data-
> > >nr_params);
> > +   memcpy(param, stage_prog_data->param,
> > +          sizeof(gl_constant_value*) * stage_prog_data-
> > >nr_params);
> > +
> >     /* Now, figure out a packing of the live uniform vectors into
> > our
> > -    * push constants.
> > +    * push constants. Start with dvec{3,4} because they are
> > aligned to
> > +    * dvec4 size (2 vec4).
> >      */
> >     for (int src = 0; src < uniforms; src++) {
> >        int size = chans_used[src];
> >  
> > -      if (size == 0)
> > +      if (size == 0 || !is_aligned_to_dvec4[src])
> >           continue;
> >  
> > -      int dst;
> > -      /* Find the lowest place we can slot this uniform in. */
> > -      for (dst = 0; dst < src; dst++) {
> > -         if (chans_used[dst] + size <= 4)
> > -            break;
> > +      int dst = set_push_constant_loc(uniforms,
> > &new_uniform_count,
> > +                                      src, size, new_loc,
> > new_chan,
> > +                                      new_chans_used);
> > +      if (dst != src) {
> > +         /* Move the references to the data */
> > +         for (int j = 0; j < size; j++) {
> > +            stage_prog_data->param[dst * 4 + new_chan[src] + j] =
> > +               param[src * 4 + j];
> > +         }
> >        }
> > +   }
> >  
> > -      if (src == dst) {
> > -         new_loc[src] = dst;
> > -         new_chan[src] = 0;
> > -      } else {
> > -         new_loc[src] = dst;
> > -         new_chan[src] = chans_used[dst];
> > +   /* Continue with the rest of data, which is aligned to vec4. */
> > +   for (int src = 0; src < uniforms; src++) {
> > +      int size = chans_used[src];
> > +
> > +      if (size == 0 || is_aligned_to_dvec4[src])
> > +         continue;
> >  
> > +      int dst = set_push_constant_loc(uniforms,
> > &new_uniform_count,
> > +                                      src, size, new_loc,
> > new_chan,
> > +                                      new_chans_used);
> > +      if (dst != src) {
> >           /* Move the references to the data */
> >           for (int j = 0; j < size; j++) {
> >              stage_prog_data->param[dst * 4 + new_chan[src] + j] =
> > -               stage_prog_data->param[src * 4 + j];
> > +               param[src * 4 + j];
> >           }
> > -
> > -         chans_used[dst] += size;
> > -         chans_used[src] = 0;
> >        }
> > -
> > -      new_uniform_count = MAX2(new_uniform_count, dst + 1);
> >     }
> >  
> > +   ralloc_free(param);
> >     this->uniforms = new_uniform_count;
> >  
> >     /* Now, update the instructions for our repacked uniforms. */
> > -- 
> > 2.11.0
-------------- 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-stable/attachments/20170504/f3f9647e/attachment-0001.sig>


More information about the mesa-stable mailing list