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

nir: split fuse_ffma into fuse_ffma16/32/64

Browse Source 
AMD wants different behavior for each bit size

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6756>
This commit is contained in:
Marek Olšák
2020-09-16 21:48:18 -04:00
committed by Marge Bot
parent c58b46edf0
commit 21174dedec
10 changed files with 39 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/compiler/nir/nir.h
View File

@@ -3056,7 +3056,9 @@ typedef enum {
typedef struct nir_shader_compiler_options {
bool lower_fdiv;
bool lower_ffma;
bool fuse_ffma;
bool fuse_ffma16;
bool fuse_ffma32;
bool fuse_ffma64;
bool lower_flrp16;
bool lower_flrp32;
/** Lowers flrp when it does not support doubles */

8
src/compiler/nir/nir_opt_algebraic.py
View File

@@ -195,7 +195,9 @@ optimizations.extend([
(('fceil', a), ('fneg', ('ffloor', ('fneg', a))), 'options->lower_fceil'),
(('ffma', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'),
# Always lower inexact ffma, because it will be fused back by late optimizations (nir_opt_algebraic_late).
(('~ffma', a, b, c), ('fadd', ('fmul', a, b), c), 'options->fuse_ffma'),
(('~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))),
@@ -2028,7 +2030,9 @@ 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', ('fmul', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma'),
(('~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