nir/algebraic: rename lower_bitshift to lower_bitops

Optimizations that insert bitshift or bitwise operations should not be
applied on GPUs that don't support integer operations.
The .lower_bitshift could be used to control the bitshift related ones,
but there was also one bitwise optimization uncovered.
Since only lima and freedreno use this option and the use case is that
no bit operations are wanted, let's rename it to .lower_bitops and use
it to control all bitops related optimizations.

Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>

diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 088b8d7d5e81b3a3ab49988b928ef7c430ca8a78..7671f8caffbd17474e078a41f6b5d03c9bacfcb8 100644 (file)
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -2544,8 +2544,8 @@ typedef struct nir_shader_compiler_options {
    /** enables rules to lower idiv by power-of-two: */
    bool lower_idiv;
 
-   /** enable rules to avoid bit shifts */
-   bool lower_bitshift;
+   /** enable rules to avoid bit ops */
+   bool lower_bitops;
 
    /** enables rules to lower isign to imin+imax */
    bool lower_isign;

diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 9fc3be8f8b0930d6afa5772b04255432fd9f6e21..09ccc69665778334ea055c0e8958ec540362942e 100644 (file)
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -83,9 +83,9 @@ def lowered_sincos(c):
 
 optimizations = [
 
-   (('imul', a, '#b@32(is_pos_power_of_two)'), ('ishl', a, ('find_lsb', b)), '!options->lower_bitshift'),
-   (('imul', a, '#b@32(is_neg_power_of_two)'), ('ineg', ('ishl', a, ('find_lsb', ('iabs', b)))), '!options->lower_bitshift'),
-   (('ishl', a, '#b@32'), ('imul', a, ('ishl', 1, b)), 'options->lower_bitshift'),
+   (('imul', a, '#b@32(is_pos_power_of_two)'), ('ishl', a, ('find_lsb', b)), '!options->lower_bitops'),
+   (('imul', a, '#b@32(is_neg_power_of_two)'), ('ineg', ('ishl', a, ('find_lsb', ('iabs', b)))), '!options->lower_bitops'),
+   (('ishl', a, '#b@32'), ('imul', a, ('ishl', 1, b)), 'options->lower_bitops'),
 
    (('unpack_64_2x32_split_x', ('imul_2x32_64(is_used_once)', a, b)), ('imul', a, b)),
    (('unpack_64_2x32_split_x', ('umul_2x32_64(is_used_once)', a, b)), ('imul', a, b)),
@@ -95,7 +95,7 @@ optimizations = [
    (('idiv', a, 1), a),
    (('umod', a, 1), 0),
    (('imod', a, 1), 0),
-   (('udiv', a, '#b@32(is_pos_power_of_two)'), ('ushr', a, ('find_lsb', b)), '!options->lower_bitshift'),
+   (('udiv', a, '#b@32(is_pos_power_of_two)'), ('ushr', a, ('find_lsb', b)), '!options->lower_bitops'),
    (('idiv', a, '#b@32(is_pos_power_of_two)'), ('imul', ('isign', a), ('ushr', ('iabs', a), ('find_lsb', b))), 'options->lower_idiv'),
    (('idiv', a, '#b@32(is_neg_power_of_two)'), ('ineg', ('imul', ('isign', a), ('ushr', ('iabs', a), ('find_lsb', ('iabs', b))))), 'options->lower_idiv'),
    (('umod', a, '#b(is_pos_power_of_two)'),    ('iand', a, ('isub', b, 1))),
@@ -509,7 +509,7 @@ optimizations = [
    (('ine', ('ineg', ('b2i32', 'a@1')), ('ineg', ('b2i32', 'b@1'))), ('ine', a, b)),
    (('b2i32', ('ine', 'a@1', 'b@1')), ('b2i32', ('ixor', a, b))),
 
-   (('iand', ('ieq', 'a@32', 0), ('ieq', 'b@32', 0)), ('ieq', ('ior', 'a@32', 'b@32'), 0)),
+   (('iand', ('ieq', 'a@32', 0), ('ieq', 'b@32', 0)), ('ieq', ('ior', 'a@32', 'b@32'), 0), '!options->lower_bitops'),
 
    # These patterns can result when (a < b || a < c) => (a < min(b, c))
    # transformations occur before constant propagation and loop-unrolling.

diff --git a/src/gallium/drivers/freedreno/a2xx/ir2_nir.c b/src/gallium/drivers/freedreno/a2xx/ir2_nir.c
index 8226d271a38b895f7bad63c7770cef05b6561162..eefd597dbf31fa1681fbf48b0c456418b34c2df5 100644 (file)
--- a/src/gallium/drivers/freedreno/a2xx/ir2_nir.c
+++ b/src/gallium/drivers/freedreno/a2xx/ir2_nir.c
@@ -39,7 +39,7 @@ static const nir_shader_compiler_options options = {
        /* .fdot_replicates = true, it is replicated, but it makes things worse */
        .lower_all_io_to_temps = true,
        .vertex_id_zero_based = true, /* its not implemented anyway */
-       .lower_bitshift = true,
+       .lower_bitops = true,
        .lower_rotate = true,
 };
 

diff --git a/src/gallium/drivers/lima/lima_program.c b/src/gallium/drivers/lima/lima_program.c
index 610782306e91ef64813650f4069a23dd2754cbee..efd870251fb06c3db8ea67ed6a9558cdbad74251 100644 (file)
--- a/src/gallium/drivers/lima/lima_program.c
+++ b/src/gallium/drivers/lima/lima_program.c
@@ -51,7 +51,7 @@ static const nir_shader_compiler_options vs_nir_options = {
    .lower_ftrunc = true,
    /* could be implemented by clamp */
    .lower_fsat = true,
-   .lower_bitshift = true,
+   .lower_bitops = true,
    .lower_rotate = true,
    .lower_sincos = true,
 };