fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-27 04:00:10 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

mesa/st: make double subgroup lowering more precise

Browse source 
Really don't touch anything else.

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37164>
This commit is contained in:
Georg Lehmann 2025-09-03 09:46:04 +02:00 • committed by Marge Bot
parent 9bc14a0047
commit c7d5108373
1 changed files with 30 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

36
src/mesa/state_tracker/st_glsl_to_nir.cpp
View file

@ -207,6 +207,26 @@ filter_64_bit_instr(const nir_instr *const_instr, UNUSED const void *data)
return lower;
}
static bool
filter_double_subgroup(const nir_intrinsic_instr *intr, UNUSED const void *data)
{
switch(intr->intrinsic) {
case nir_intrinsic_vote_feq:
return intr->src[0].ssa->bit_size == 64;
case nir_intrinsic_reduce:
case nir_intrinsic_exclusive_scan:
case nir_intrinsic_inclusive_scan: {
if (intr->src[0].ssa->bit_size != 64)
return false;
unsigned op = nir_intrinsic_reduction_op(intr);
nir_alu_type type = nir_op_infos[op].output_type;
return type == nir_type_float;
}
default:
return false;
}
}
/* Second third of converting glsl_to_nir. This creates uniforms, gathers
* info on varyings, etc after NIR link time opts have been applied.
*/
@ -277,16 +297,20 @@ st_glsl_to_nir_post_opts(struct st_context *st, struct gl_program *prog,
/* 64-bit subgroup ops like vote_feq and inclusive_scan are secretly
* fp64 operations, so lower them first to make the ALU operation
* appear for nir_lower_doubles to lower after.
*
* Create GL_KHR_shader_subgroup like uvec4 ballots, drivers have to further
* lower that on their own.
*/
nir_lower_subgroups_options subgroup_opts = {0};
subgroup_opts.subgroup_size = nir->options->subgroup_size;
subgroup_opts.ballot_bit_size = nir->options->ballot_bit_size;
subgroup_opts.ballot_components = nir->options->ballot_components;
subgroup_opts.lower_fp64 = true;
subgroup_opts.filter = filter_double_subgroup;
subgroup_opts.subgroup_size = 0;
subgroup_opts.ballot_bit_size = 32;
subgroup_opts.ballot_components = 4;
subgroup_opts.lower_vote_feq = true;
subgroup_opts.lower_reduce = true;
if (subgroup_opts.ballot_bit_size) {
if (nir->options->lower_doubles_options & nir_lower_fp64_full_software)
NIR_PASS(lowered_64bit_ops, nir, nir_lower_subgroups, &subgroup_opts);
}
/* nir_lower_doubles is not prepared for vector ops, so if the backend doesn't
* request lower_alu_to_scalar until now, lower all 64 bit ops, and try to