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

Revert "nir: replace lower_ffma and fuse_ffma with has_ffma"

Browse Source 
This reverts commit 939ddf3f67.

Intel has a separate pass for fusing FFMAs selectively.  We split
these flags in commit 1b72c31e1f and
the reasoning still stands.  The patch being reverted was just a
cleanup, so there should be no issue with reverting it.

Acked-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6849>
This commit is contained in:
Kenneth Graunke
2020-09-24 08:46:31 -07:00
parent d8cdcd4adf
commit 140f53e646
19 changed files with 84 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/compiler/nir/nir.h
View File

@@ -3055,9 +3055,12 @@ typedef enum {
typedef struct nir_shader_compiler_options {
bool lower_fdiv;
bool has_ffma16;
bool has_ffma32;
bool has_ffma64;
bool lower_ffma16;
bool lower_ffma32;
bool lower_ffma64;
bool fuse_ffma16;
bool fuse_ffma32;
bool fuse_ffma64;
bool lower_flrp16;
bool lower_flrp32;
/** Lowers flrp when it does not support doubles */

6
src/compiler/nir/nir_lower_flrp.c
View File

@@ -370,11 +370,11 @@ convert_flrp_instruction(nir_builder *bld,
unsigned bit_size = nir_dest_bit_size(alu->dest.dest);
if (bit_size == 16)
have_ffma = bld->shader->options->has_ffma16;
have_ffma = !bld->shader->options->lower_ffma16;
else if (bit_size == 32)
have_ffma = bld->shader->options->has_ffma32;
have_ffma = !bld->shader->options->lower_ffma32;
else if (bit_size == 64)
have_ffma = bld->shader->options->has_ffma64;
have_ffma = !bld->shader->options->lower_ffma64;
else
unreachable("invalid bit_size");

18
src/compiler/nir/nir_opt_algebraic.py
View File

@@ -193,13 +193,13 @@ optimizations.extend([
(('fadd', a, ('fneg', ('ffract', a))), ('ffloor', a), '!options->lower_ffloor'),
(('ffract', a), ('fsub', a, ('ffloor', a)), 'options->lower_ffract'),
(('fceil', a), ('fneg', ('ffloor', ('fneg', a))), 'options->lower_fceil'),
(('ffma@16', a, b, c), ('fadd', ('fmul', a, b), c), '!options->has_ffma16'),
(('ffma@32', a, b, c), ('fadd', ('fmul', a, b), c), '!options->has_ffma32'),
(('ffma@64', a, b, c), ('fadd', ('fmul', a, b), c), '!options->has_ffma64'),
(('ffma@16', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma16'),
(('ffma@32', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma32'),
(('ffma@64', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma64'),
# Always lower inexact ffma, because it will be fused back by late optimizations (nir_opt_algebraic_late).
(('~ffma@16', a, b, c), ('fadd', ('fmul', a, b), c), 'options->has_ffma16'),
(('~ffma@32', a, b, c), ('fadd', ('fmul', a, b), c), 'options->has_ffma32'),
(('~ffma@64', a, b, c), ('fadd', ('fmul', a, b), c), 'options->has_ffma64'),
(('~ffma@16', a, b, c), ('fadd', ('fmul', a, b), c), 'options->fuse_ffma16'),
(('~ffma@32', a, b, c), ('fadd', ('fmul', a, b), c), 'options->fuse_ffma32'),
(('~ffma@64', a, b, c), ('fadd', ('fmul', a, b), c), 'options->fuse_ffma64'),
(('~fmul', ('fadd', ('iand', ('ineg', ('b2i', 'a@bool')), ('fmul', b, c)), '#d'), '#e'),
('bcsel', a, ('fmul', ('fadd', ('fmul', b, c), d), e), ('fmul', d, e))),
@@ -2032,9 +2032,9 @@ late_optimizations = [
(('fneg', a), ('fsub', 0.0, a), 'options->lower_negate'),
(('ineg', a), ('isub', 0, a), 'options->lower_negate'),
(('iabs', a), ('imax', a, ('ineg', a)), 'options->lower_iabs'),
(('~fadd@16', ('fmul', a, b), c), ('ffma', a, b, c), 'options->has_ffma16'),
(('~fadd@32', ('fmul', a, b), c), ('ffma', a, b, c), 'options->has_ffma32'),
(('~fadd@64', ('fmul', a, b), c), ('ffma', a, b, c), 'options->has_ffma64'),
(('~fadd@16', ('fmul', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma16'),
(('~fadd@32', ('fmul', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma32'),
(('~fadd@64', ('fmul', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma64'),
# These are duplicated from the main optimizations table. The late
# patterns that rearrange expressions like x - .5 < 0 to x < .5 can create