fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-01-01 05:20:09 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

nir/serialize: Make more space for intrinsic_op allowing 1024 ops

Browse source 
We are close to the limit of 512 intrinsics, make more space to
be able to support up to 1024 intrinsics.

Take one bit from packed_const_indices, they shouldn't suffer in
a common case.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13456>
This commit is contained in:
Danylo Piliaiev 2021-10-20 17:36:27 +03:00 • committed by Marge Bot
parent dc74285d32
commit b7c7abded7
1 changed files with 13 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

28
src/compiler/nir/nir_serialize.c
View file

@ -575,14 +575,12 @@ union packed_dest {
};
enum intrinsic_const_indices_encoding {
/* Use the 9 bits of packed_const_indices to store 1-9 indices.
* 1 9-bit index, or 2 4-bit indices, or 3 3-bit indices, or
* 4 2-bit indices, or 5-9 1-bit indices.
/* Use packed_const_indices to store tightly packed indices.
*
* The common case for load_ubo is 0, 0, 0, which is trivially represented.
* The common cases for load_interpolated_input also fit here, e.g.: 7, 3
*/
const_indices_9bit_all_combined,
const_indices_all_combined,
const_indices_8bit, /* 8 bits per element */
const_indices_16bit, /* 16 bits per element */
@ -641,9 +639,9 @@ union packed_instr {
} deref_var;
struct {
unsigned instr_type:4;
unsigned intrinsic:9;
unsigned intrinsic:10;
unsigned const_indices_encoding:2;
unsigned packed_const_indices:9;
unsigned packed_const_indices:8;
unsigned dest:8;
} intrinsic;
struct {
@ -1167,11 +1165,11 @@ read_deref(read_ctx *ctx, union packed_instr header)
static void
write_intrinsic(write_ctx *ctx, const nir_intrinsic_instr *intrin)
{
/* 9 bits for nir_intrinsic_op */
STATIC_ASSERT(nir_num_intrinsics <= 512);
/* 10 bits for nir_intrinsic_op */
STATIC_ASSERT(nir_num_intrinsics <= 1024);
unsigned num_srcs = nir_intrinsic_infos[intrin->intrinsic].num_srcs;
unsigned num_indices = nir_intrinsic_infos[intrin->intrinsic].num_indices;
assert(intrin->intrinsic < 512);
assert(intrin->intrinsic < 1024);
union packed_instr header;
header.u32 = 0;
@ -1187,11 +1185,11 @@ write_intrinsic(write_ctx *ctx, const nir_intrinsic_instr *intrin)
max_bits = MAX2(max_bits, max);
}
if (max_bits * num_indices <= 9) {
header.intrinsic.const_indices_encoding = const_indices_9bit_all_combined;
if (max_bits * num_indices <= 8) {
header.intrinsic.const_indices_encoding = const_indices_all_combined;
/* Pack all const indices into 6 bits. */
unsigned bit_size = 9 / num_indices;
/* Pack all const indices into 8 bits. */
unsigned bit_size = 8 / num_indices;
for (unsigned i = 0; i < num_indices; i++) {
header.intrinsic.packed_const_indices |=
intrin->const_index[i] << (i * bit_size);
@ -1262,8 +1260,8 @@ read_intrinsic(read_ctx *ctx, union packed_instr header)
if (num_indices) {
switch (header.intrinsic.const_indices_encoding) {
case const_indices_9bit_all_combined: {
unsigned bit_size = 9 / num_indices;
case const_indices_all_combined: {
unsigned bit_size = 8 / num_indices;
unsigned bit_mask = u_bit_consecutive(0, bit_size);
for (unsigned i = 0; i < num_indices; i++) {
intrin->const_index[i] =