Previously, lower_atomics was non-SSA only. We assert-failed if the
destination of an atomic operation intrinsic was an SSA def and we used
temporary registers for computing offsets. This commit changes both of
these behaviors. We now use SSA values for computing offsets (so we can
optimize them) and we handle SSA destinations. We also move the pass to
run before we go out of SSA on i965 as it now generates SSA values.
---
src/glsl/nir/nir_lower_atomics.c | 50 ++++++++++++++++++++------------
src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 6 ++--
2 files changed, 35 insertions(+), 21 deletions(-)
diff --git a/src/glsl/nir/nir_lower_atomics.c b/src/glsl/nir/nir_lower_atomics.c
index b43e4fe..6362640 100644
--- a/src/glsl/nir/nir_lower_atomics.c
+++ b/src/glsl/nir/nir_lower_atomics.c
@@ -59,23 +59,19 @@ lower_instr(nir_intrinsic_instr *instr, nir_function_impl
*impl)
void *mem_ctx = ralloc_parent(instr);
- /* TODO support SSA */
- assert(!instr->dest.is_ssa);
-
nir_intrinsic_instr *new_instr = nir_intrinsic_instr_create(mem_ctx, op);
- new_instr->dest = nir_dest_copy(instr->dest, mem_ctx);
new_instr->const_index[0] =
(int) instr->variables[0]->var->data.atomic.buffer_index;
nir_load_const_instr *offset_const = nir_load_const_instr_create(mem_ctx);
offset_const->num_components = 1;
offset_const->value.u[0] = instr->variables[0]->var->data.atomic.offset;
- offset_const->dest.reg.reg = nir_local_reg_create(impl);
- offset_const->dest.reg.reg->num_components = 1;
+ offset_const->dest.is_ssa = true;
+ nir_ssa_def_init(&offset_const->instr, &offset_const->dest.ssa, 1, NULL);
nir_instr_insert_before(&instr->instr, &offset_const->instr);
- nir_register *offset_reg = offset_const->dest.reg.reg;
+ nir_ssa_def *offset_def = &offset_const->dest.ssa;
if (instr->variables[0]->deref.child != NULL) {
assert(instr->variables[0]->deref.child->deref_type ==
@@ -91,31 +87,49 @@ lower_instr(nir_intrinsic_instr *instr, nir_function_impl
*impl)
nir_load_const_instr_create(mem_ctx);
atomic_counter_size->num_components = 1;
atomic_counter_size->value.u[0] = ATOMIC_COUNTER_SIZE;
- atomic_counter_size->dest.reg.reg = nir_local_reg_create(impl);
- atomic_counter_size->dest.reg.reg->num_components = 1;
+ atomic_counter_size->dest.is_ssa = true;
+ nir_ssa_def_init(&atomic_counter_size->instr,
+ &atomic_counter_size->dest.ssa, 1, NULL);
nir_instr_insert_before(&instr->instr, &atomic_counter_size->instr);
nir_alu_instr *mul = nir_alu_instr_create(mem_ctx, nir_op_imul);
- mul->dest.dest.reg.reg = nir_local_reg_create(impl);
- mul->dest.dest.reg.reg->num_components = 1;
+ mul->dest.dest.is_ssa = true;
+ nir_ssa_def_init(&mul->instr, &mul->dest.dest.ssa, 1, NULL);
mul->dest.write_mask = 0x1;
mul->src[0].src = nir_src_copy(deref_array->indirect, mem_ctx);
- mul->src[1].src.reg.reg = atomic_counter_size->dest.reg.reg;
+ mul->src[1].src.is_ssa = true;
+ mul->src[1].src.ssa = &atomic_counter_size->dest.ssa;
nir_instr_insert_before(&instr->instr, &mul->instr);
nir_alu_instr *add = nir_alu_instr_create(mem_ctx, nir_op_iadd);
- add->dest.dest.reg.reg = nir_local_reg_create(impl);
- add->dest.dest.reg.reg->num_components = 1;
+ add->dest.dest.is_ssa = true;
+ nir_ssa_def_init(&add->instr, &add->dest.dest.ssa, 1, NULL);
add->dest.write_mask = 0x1;
- add->src[0].src.reg.reg = mul->dest.dest.reg.reg;
- add->src[1].src.reg.reg = offset_const->dest.reg.reg;
+ add->src[0].src.is_ssa = true;
+ add->src[0].src.ssa = &mul->dest.dest.ssa;
+ add->src[1].src.is_ssa = true;
+ add->src[1].src.ssa = &offset_const->dest.ssa;
nir_instr_insert_before(&instr->instr, &add->instr);
- offset_reg = add->dest.dest.reg.reg;
+ offset_def = &add->dest.dest.ssa;
}
}
- new_instr->src[0].reg.reg = offset_reg;
+ new_instr->src[0].is_ssa = true;
+ new_instr->src[0].ssa = offset_def;;
+
+ if (instr->dest.is_ssa) {
+ new_instr->dest.is_ssa = true;
+ nir_ssa_def_init(&new_instr->instr, &new_instr->dest.ssa,
+ instr->dest.ssa.num_components, NULL);
+ nir_src new_dest_src = {
+ .is_ssa = true,
+ .ssa = &new_instr->dest.ssa,
+ };
+ nir_ssa_def_rewrite_uses(&instr->dest.ssa, new_dest_src, mem_ctx);
+ } else {
+ new_instr->dest = nir_dest_copy(instr->dest, mem_ctx);
+ }
nir_instr_insert_before(&instr->instr, &new_instr->instr);
nir_instr_remove(&instr->instr);
diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index 04e09be..060912e 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -69,6 +69,9 @@ fs_visitor::emit_nir_code()
nir_remove_dead_variables(nir);
nir_validate_shader(nir);
+ nir_lower_atomics(nir);
+ nir_validate_shader(nir);
+
nir_lower_to_source_mods(nir);
nir_validate_shader(nir);
nir_copy_prop(nir);
@@ -84,9 +87,6 @@ fs_visitor::emit_nir_code()
nir_lower_system_values(nir);
nir_validate_shader(nir);
- nir_lower_atomics(nir);
- nir_validate_shader(nir);
-
/* emit the arrays used for inputs and outputs - load/store intrinsics will
* be converted to reads/writes of these arrays
*/
--
2.2.0
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
