* easier to not have them when we're doing optimizations.
*/
+static void
+alu_src_consume_abs(nir_alu_src *src)
+{
+ src->abs = true;
+}
+
+static void
+alu_src_consume_negate(nir_alu_src *src)
+{
+ /* If abs is set on the source, the negate goes away */
+ if (!src->abs)
+ src->negate = !src->negate;
+}
+
static bool
-nir_lower_to_source_mods_block(nir_block *block)
+nir_lower_to_source_mods_block(nir_block *block,
+ nir_lower_to_source_mods_flags options)
{
+ bool progress = false;
+
nir_foreach_instr(instr, block) {
if (instr->type != nir_instr_type_alu)
continue;
nir_alu_instr *alu = nir_instr_as_alu(instr);
+ bool lower_abs = (nir_op_infos[alu->op].num_inputs < 3) ||
+ (options & nir_lower_triop_abs);
+
for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) {
if (!alu->src[i].src.is_ssa)
continue;
switch (nir_alu_type_get_base_type(nir_op_infos[alu->op].input_types[i])) {
case nir_type_float:
- if (parent->op != nir_op_fmov)
+ if (!(options & nir_lower_float_source_mods))
+ continue;
+ if (parent->op != nir_op_fabs && parent->op != nir_op_fneg)
continue;
break;
case nir_type_int:
- if (parent->op != nir_op_imov)
+ if (!(options & nir_lower_int_source_mods))
+ continue;
+ if (parent->op != nir_op_iabs && parent->op != nir_op_ineg)
continue;
break;
default:
if (!parent->src[0].src.is_ssa)
continue;
+ if (!lower_abs && (parent->op == nir_op_fabs ||
+ parent->op == nir_op_iabs))
+ continue;
+
nir_instr_rewrite_src(instr, &alu->src[i].src, parent->src[0].src);
- if (alu->src[i].abs) {
- /* abs trumps both neg and abs, do nothing */
- } else {
- alu->src[i].negate = (alu->src[i].negate != parent->src[0].negate);
- alu->src[i].abs |= parent->src[0].abs;
- }
+
+ /* Apply any modifiers that come from the parent opcode */
+ if (parent->op == nir_op_fneg || parent->op == nir_op_ineg)
+ alu_src_consume_negate(&alu->src[i]);
+ if (parent->op == nir_op_fabs || parent->op == nir_op_iabs)
+ alu_src_consume_abs(&alu->src[i]);
+
+ /* Apply modifiers from the parent source */
+ if (parent->src[0].negate)
+ alu_src_consume_negate(&alu->src[i]);
+ if (parent->src[0].abs)
+ alu_src_consume_abs(&alu->src[i]);
for (int j = 0; j < 4; ++j) {
if (!nir_alu_instr_channel_used(alu, i, j))
alu->src[i].swizzle[j] = parent->src[0].swizzle[alu->src[i].swizzle[j]];
}
- if (list_empty(&parent->dest.dest.ssa.uses) &&
- list_empty(&parent->dest.dest.ssa.if_uses))
+ if (list_is_empty(&parent->dest.dest.ssa.uses) &&
+ list_is_empty(&parent->dest.dest.ssa.if_uses))
nir_instr_remove(&parent->instr);
- }
- switch (alu->op) {
- case nir_op_fsat:
- alu->op = nir_op_fmov;
- alu->dest.saturate = true;
- break;
- case nir_op_ineg:
- alu->op = nir_op_imov;
- alu->src[0].negate = !alu->src[0].negate;
- break;
- case nir_op_fneg:
- alu->op = nir_op_fmov;
- alu->src[0].negate = !alu->src[0].negate;
- break;
- case nir_op_iabs:
- alu->op = nir_op_imov;
- alu->src[0].abs = true;
- alu->src[0].negate = false;
- break;
- case nir_op_fabs:
- alu->op = nir_op_fmov;
- alu->src[0].abs = true;
- alu->src[0].negate = false;
- break;
- default:
- break;
+ progress = true;
}
/* We've covered sources. Now we're going to try and saturate the
nir_type_float)
continue;
- if (!list_empty(&alu->dest.dest.ssa.if_uses))
+ if (!(options & nir_lower_float_source_mods))
+ continue;
+
+ if (!list_is_empty(&alu->dest.dest.ssa.if_uses))
continue;
bool all_children_are_sat = true;
- nir_foreach_use(&alu->dest.dest.ssa, child_src) {
+ nir_foreach_use(child_src, &alu->dest.dest.ssa) {
assert(child_src->is_ssa);
nir_instr *child = child_src->parent_instr;
if (child->type != nir_instr_type_alu) {
continue;
}
- if (child_alu->op != nir_op_fsat &&
- !(child_alu->op == nir_op_fmov && child_alu->dest.saturate)) {
+ if (child_alu->op != nir_op_fsat) {
all_children_are_sat = false;
continue;
}
continue;
alu->dest.saturate = true;
+ progress = true;
- nir_foreach_use(&alu->dest.dest.ssa, child_src) {
+ nir_foreach_use(child_src, &alu->dest.dest.ssa) {
assert(child_src->is_ssa);
- nir_instr *child = child_src->parent_instr;
- assert(child->type == nir_instr_type_alu);
- nir_alu_instr *child_alu = nir_instr_as_alu(child);
+ nir_alu_instr *child_alu = nir_instr_as_alu(child_src->parent_instr);
- child_alu->op = nir_op_fmov;
+ child_alu->op = nir_op_mov;
child_alu->dest.saturate = false;
/* We could propagate the dest of our instruction to the
* destinations of the uses here. However, one quick round of
}
}
- return true;
+ return progress;
}
-void
-nir_lower_to_source_mods(nir_shader *shader)
+static bool
+nir_lower_to_source_mods_impl(nir_function_impl *impl,
+ nir_lower_to_source_mods_flags options)
{
- nir_foreach_function(shader, function) {
+ bool progress = false;
+
+ nir_foreach_block(block, impl) {
+ progress |= nir_lower_to_source_mods_block(block, options);
+ }
+
+ if (progress)
+ nir_metadata_preserve(impl, nir_metadata_block_index |
+ nir_metadata_dominance);
+
+ return progress;
+}
+
+bool
+nir_lower_to_source_mods(nir_shader *shader,
+ nir_lower_to_source_mods_flags options)
+{
+ bool progress = false;
+
+ nir_foreach_function(function, shader) {
if (function->impl) {
- nir_foreach_block(block, function->impl) {
- nir_lower_to_source_mods_block(block);
- }
+ progress |= nir_lower_to_source_mods_impl(function->impl, options);
}
}
+
+ return progress;
}
projects / mesa.git / blobdiff