OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nir: Add a new subgroups lowering pass

Browse Source 
This commit pulls nir_lower_read_invocations_to_scalar along with most
of the guts of nir_opt_intrinsics (which mostly does subgroup lowering)
into a new nir_lower_subgroups pass.  There are various other bits of
subgroup lowering that we're going to want to do so it makes a bit more
sense to keep it all together in one pass.  We also move it in i965 to
happen after nir_lower_system_values to ensure that because we want to
handle the subgroup mask system value intrinsics here.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This commit is contained in:
Jason Ekstrand
2017-08-22 13:23:59 -07:00
parent 1ca3a94427
commit 28da82f978
8 changed files with 208 additions and 186 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
src/compiler/nir/nir_opt_intrinsics.c
View File

@@ -28,26 +28,6 @@
* \file nir_opt_intrinsics.c
*/
static nir_ssa_def *
high_subgroup_mask(nir_builder *b,
nir_ssa_def *count,
uint64_t base_mask)
{
/* group_mask could probably be calculated more efficiently but we want to
* be sure not to shift by 64 if the subgroup size is 64 because the GLSL
* shift operator is undefined in that case. In any case if we were worried
* about efficency this should probably be done further down because the
* subgroup size is likely to be known at compile time.
*/
nir_ssa_def *subgroup_size = nir_load_subgroup_size(b);
nir_ssa_def *all_bits = nir_imm_int64(b, ~0ull);
nir_ssa_def *shift = nir_isub(b, nir_imm_int(b, 64), subgroup_size);
nir_ssa_def *group_mask = nir_ushr(b, all_bits, shift);
nir_ssa_def *higher_bits = nir_ishl(b, nir_imm_int64(b, base_mask), count);
return nir_iand(b, higher_bits, group_mask);
}
static bool
opt_intrinsics_impl(nir_function_impl *impl)
{
@@ -66,22 +46,14 @@ opt_intrinsics_impl(nir_function_impl *impl)
switch (intrin->intrinsic) {
case nir_intrinsic_vote_any:
case nir_intrinsic_vote_all: {
nir_const_value *val = nir_src_as_const_value(intrin->src[0]);
if (!val && !b.shader->options->lower_vote_trivial)
continue;
replacement = nir_ssa_for_src(&b, intrin->src[0], 1);
case nir_intrinsic_vote_all:
if (nir_src_as_const_value(intrin->src[0]))
replacement = nir_ssa_for_src(&b, intrin->src[0], 1);
break;
}
case nir_intrinsic_vote_eq: {
nir_const_value *val = nir_src_as_const_value(intrin->src[0]);
if (!val && !b.shader->options->lower_vote_trivial)
continue;
replacement = nir_imm_int(&b, NIR_TRUE);
case nir_intrinsic_vote_eq:
if (nir_src_as_const_value(intrin->src[0]))
replacement = nir_imm_int(&b, NIR_TRUE);
break;
}
case nir_intrinsic_ballot: {
assert(b.shader->options->max_subgroup_size != 0);
if (b.shader->options->max_subgroup_size > 32 ||
@@ -100,37 +72,6 @@ opt_intrinsics_impl(nir_function_impl *impl)
nir_imm_int(&b, 0));
break;
}
case nir_intrinsic_load_subgroup_eq_mask:
case nir_intrinsic_load_subgroup_ge_mask:
case nir_intrinsic_load_subgroup_gt_mask:
case nir_intrinsic_load_subgroup_le_mask:
case nir_intrinsic_load_subgroup_lt_mask: {
if (!b.shader->options->lower_subgroup_masks)
break;
nir_ssa_def *count = nir_load_subgroup_invocation(&b);
switch (intrin->intrinsic) {
case nir_intrinsic_load_subgroup_eq_mask:
replacement = nir_ishl(&b, nir_imm_int64(&b, 1ull), count);
break;
case nir_intrinsic_load_subgroup_ge_mask:
replacement = high_subgroup_mask(&b, count, ~0ull);
break;
case nir_intrinsic_load_subgroup_gt_mask:
replacement = high_subgroup_mask(&b, count, ~1ull);
break;
case nir_intrinsic_load_subgroup_le_mask:
replacement = nir_inot(&b, nir_ishl(&b, nir_imm_int64(&b, ~1ull), count));
break;
case nir_intrinsic_load_subgroup_lt_mask:
replacement = nir_inot(&b, nir_ishl(&b, nir_imm_int64(&b, ~0ull), count));
break;
default:
unreachable("you seriously can't tell this is unreachable?");
}
break;
}
default:
break;
}