fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-01-01 16:10:09 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

intel/fs/gen7+: Implement discard/demote for SIMD32 programs.

Browse source 
At this point this simply involves fixing the initialization of the
sample mask flag register to take the right dispatch mask from the
thread payload, and fixing sample_mask_reg() to return f1.1 for the
second half of a SIMD32 thread.  This improves Manhattan 3.1
performance by 2.4%±0.31% (N>40) on my ICL with SIMD32 enabled
relative to falling back to SIMD16 for the shaders that use discard.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This commit is contained in:
Francisco Jerez 2020-01-04 16:16:24 -08:00
parent 04c7d3d4b1
commit 8d3b86e34a
2 changed files with 14 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
src/intel/compiler/brw_fs.cpp
View file

@ -4285,8 +4285,8 @@ sample_mask_reg(const fs_builder &bld)
if (v->stage != MESA_SHADER_FRAGMENT) {
return brw_imm_ud(0xffffffff);
} else if (brw_wm_prog_data(v->stage_prog_data)->uses_kill) {
assert(bld.group() < 16 && bld.dispatch_width() <= 16);
return brw_flag_subreg(sample_mask_flag_subreg(v));
assert(bld.dispatch_width() <= 16);
return brw_flag_subreg(sample_mask_flag_subreg(v) + bld.group() / 16);
} else {
assert(v->devinfo->gen >= 6 && bld.dispatch_width() <= 16);
return retype(brw_vec1_grf((bld.group() >= 16 ? 2 : 1), 7),
@ -8171,11 +8171,15 @@ fs_visitor::run_fs(bool allow_spilling, bool do_rep_send)
* Initialize it with the dispatched pixels.
*/
if (wm_prog_data->uses_kill) {
const fs_reg dispatch_mask =
devinfo->gen >= 6 ? brw_vec1_grf(1, 7) : brw_vec1_grf(0, 0);
bld.exec_all().group(1, 0)
.MOV(sample_mask_reg(bld),
retype(dispatch_mask, BRW_REGISTER_TYPE_UW));
const unsigned lower_width = MIN2(dispatch_width, 16);
for (unsigned i = 0; i < dispatch_width / lower_width; i++) {
const fs_reg dispatch_mask =
devinfo->gen >= 6 ? brw_vec1_grf((i ? 2 : 1), 7) :
brw_vec1_grf(0, 0);
bld.exec_all().group(1, 0)
.MOV(sample_mask_reg(bld.group(lower_width, i)),
retype(dispatch_mask, BRW_REGISTER_TYPE_UW));
}
}
emit_nir_code();

4
src/intel/compiler/brw_fs_nir.cpp
View file

@ -3562,7 +3562,9 @@ fs_visitor::nir_emit_fs_intrinsic(const fs_builder &bld,
emit_discard_jump();
}
limit_dispatch_width(16, "Fragment discard/demote not implemented in SIMD32 mode.\n");
if (devinfo->gen < 7)
limit_dispatch_width(
16, "Fragment discard/demote not implemented in SIMD32 mode.\n");
break;
}