[Mesa-dev] [PATCH 3/5] nir/i965: add before ffma algebraic opts

[Timothy Arceri]

This shuffles constants down in the reverse of what the previous
patch does and applies some simpilifications that may be made
possible from doing so.

Shader-db results BDW:

total instructions in shared programs: 12978405 -> 12975412 (-0.02%)
instructions in affected programs: 279862 -> 276869 (-1.07%)
helped: 1230
HURT: 125

total cycles in shared programs: 246622124 -> 246625712 (0.00%)
cycles in affected programs: 11311912 -> 11315500 (0.03%)
helped: 1666
HURT: 1342
---
 src/compiler/nir/nir.h                |  1 +
 src/compiler/nir/nir_opt_algebraic.py | 23 +++++++++++++++++++++++
 src/mesa/drivers/dri/i965/brw_nir.c   |  6 ++++++
 3 files changed, 30 insertions(+)

diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 8bbc41d..8d6f7d1 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -2569,6 +2569,7 @@ bool nir_lower_phis_to_regs_block(nir_block *block);
 bool nir_lower_ssa_defs_to_regs_block(nir_block *block);
 
 bool nir_opt_algebraic(nir_shader *shader);
+bool nir_opt_algebraic_before_ffma(nir_shader *shader);
 bool nir_opt_algebraic_late(nir_shader *shader);
 bool nir_opt_constant_folding(nir_shader *shader);
 
diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 797fc4b..a1c18cc 100644
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -513,6 +513,27 @@ for op in ['flt', 'fge', 'feq', 'fne',
        ('bcsel', 'a', (op, 'd', 'b'), (op, 'd', 'c'))),
    ]
 
+# This section contains "late" optimizations that should be run before
+# creating ffmas and calling regular optimizations for the final time.
+# Optimizations should go here if they help code generation and conflict
+# with the regular optimizations.
+before_ffma_optimizations = [
+   # Propagate constants down multiplication chains
+   (('fmul(is_used_once)', ('fmul(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('fmul', ('fmul', a, c), b)),
+   (('imul(is_used_once)', ('imul(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('imul', ('imul', a, c), b)),
+   (('fadd(is_used_once)', ('fadd(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('fadd', ('fadd', a, c), b)),
+   (('iadd(is_used_once)', ('iadd(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('iadd', ('iadd', a, c), b)),
+
+   (('~fadd', ('fmul', a, b), ('fmul', a, c)), ('fmul', a, ('fadd', b, c))),
+   (('iadd', ('imul', a, b), ('imul', a, c)), ('imul', a, ('iadd', b, c))),
+   (('~fadd', ('fneg', a), a), 0.0),
+   (('iadd', ('ineg', a), a), 0),
+   (('iadd', ('ineg', a), ('iadd', a, b)), b),
+   (('iadd', a, ('iadd', ('ineg', a), b)), b),
+   (('~fadd', ('fneg', a), ('fadd', a, b)), b),
+   (('~fadd', a, ('fadd', ('fneg', a), b)), b),
+]
+
 # This section contains "late" optimizations that should be run after the
 # regular optimizations have finished.  Optimizations should go here if
 # they help code generation but do not necessarily produce code that is
@@ -539,5 +560,7 @@ late_optimizations = [
 ]
 
 print nir_algebraic.AlgebraicPass("nir_opt_algebraic", optimizations).render()
+print nir_algebraic.AlgebraicPass("nir_opt_algebraic_before_ffma",
+                                  before_ffma_optimizations).render()
 print nir_algebraic.AlgebraicPass("nir_opt_algebraic_late",
                                   late_optimizations).render()
diff --git a/src/mesa/drivers/dri/i965/brw_nir.c b/src/mesa/drivers/dri/i965/brw_nir.c
index a5912a0..999e1d2 100644
--- a/src/mesa/drivers/dri/i965/brw_nir.c
+++ b/src/mesa/drivers/dri/i965/brw_nir.c
@@ -604,6 +604,12 @@ brw_postprocess_nir(nir_shader *nir, const struct brw_compiler *compiler,
    bool progress; /* Written by OPT and OPT_V */
    (void)progress;
 
+
+   do {
+      progress = false;
+      OPT(nir_opt_algebraic_before_ffma);
+   } while (progress);
+
    nir = nir_optimize(nir, compiler, is_scalar);
 
    if (devinfo->gen >= 6) {
-- 
2.9.3