#include "nir.h"
#include "nir_builder.h"
#include "nir_control_flow.h"
+#include "nir_vla.h"
-struct inline_functions_state {
- struct set *inlined;
- nir_builder builder;
- bool progress;
-};
+void nir_inline_function_impl(struct nir_builder *b,
+ const nir_function_impl *impl,
+ nir_ssa_def **params)
+{
+ nir_function_impl *copy = nir_function_impl_clone(b->shader, impl);
+
+ /* Insert a nop at the cursor so we can keep track of where things are as
+ * we add/remove stuff from the CFG.
+ */
+ nir_intrinsic_instr *nop =
+ nir_intrinsic_instr_create(b->shader, nir_intrinsic_nop);
+ nir_builder_instr_insert(b, &nop->instr);
+
+ exec_list_append(&b->impl->locals, ©->locals);
+ exec_list_append(&b->impl->registers, ©->registers);
+
+ nir_foreach_block(block, copy) {
+ nir_foreach_instr_safe(instr, block) {
+ switch (instr->type) {
+ case nir_instr_type_intrinsic: {
+ nir_intrinsic_instr *load = nir_instr_as_intrinsic(instr);
+ if (load->intrinsic != nir_intrinsic_load_param)
+ break;
+
+ unsigned param_idx = nir_intrinsic_param_idx(load);
+ assert(param_idx < impl->function->num_params);
+ assert(load->dest.is_ssa);
+ nir_ssa_def_rewrite_uses(&load->dest.ssa,
+ nir_src_for_ssa(params[param_idx]));
+
+ /* Remove any left-over load_param intrinsics because they're soon
+ * to be in another function and therefore no longer valid.
+ */
+ nir_instr_remove(&load->instr);
+ break;
+ }
+
+ case nir_instr_type_jump:
+ /* Returns have to be lowered for this to work */
+ assert(nir_instr_as_jump(instr)->type != nir_jump_return);
+ break;
+
+ default:
+ break;
+ }
+ }
+ }
+
+ /* Pluck the body out of the function and place it here */
+ nir_cf_list body;
+ nir_cf_list_extract(&body, ©->body);
+ nir_cf_reinsert(&body, nir_before_instr(&nop->instr));
+
+ b->cursor = nir_instr_remove(&nop->instr);
+}
static bool inline_function_impl(nir_function_impl *impl, struct set *inlined);
static bool
-inline_functions_block(nir_block *block, void *void_state)
+inline_functions_block(nir_block *block, nir_builder *b,
+ struct set *inlined)
{
- struct inline_functions_state *state = void_state;
-
- nir_builder *b = &state->builder;
-
+ bool progress = false;
/* This is tricky. We're iterating over instructions in a block but, as
* we go, the block and its instruction list are being split into
* pieces. However, this *should* be safe since foreach_safe always
* properly get moved to the next block when it gets split, and we
* continue iterating there.
*/
- nir_foreach_instr_safe(block, instr) {
+ nir_foreach_instr_safe(instr, block) {
if (instr->type != nir_instr_type_call)
continue;
- state->progress = true;
+ progress = true;
nir_call_instr *call = nir_instr_as_call(instr);
assert(call->callee->impl);
- inline_function_impl(call->callee->impl, state->inlined);
-
- nir_function_impl *callee_copy =
- nir_function_impl_clone(call->callee->impl);
-
- exec_list_append(&b->impl->locals, &callee_copy->locals);
- exec_list_append(&b->impl->registers, &callee_copy->registers);
-
- b->cursor = nir_before_instr(&call->instr);
-
- /* Add copies of all in parameters */
- assert(call->num_params == callee_copy->num_params);
- for (unsigned i = 0; i < callee_copy->num_params; i++) {
- /* Only in or inout parameters */
- if (call->callee->params[i].param_type == nir_parameter_out)
- continue;
-
- nir_copy_deref_var(b, nir_deref_var_create(b->shader,
- callee_copy->params[i]),
- call->params[i]);
- }
-
- /* Pluck the body out of the function and place it here */
- nir_cf_list body;
- nir_cf_list_extract(&body, &callee_copy->body);
- nir_cf_reinsert(&body, b->cursor);
-
- b->cursor = nir_before_instr(&call->instr);
-
- /* Add copies of all out parameters and the return */
- assert(call->num_params == callee_copy->num_params);
- for (unsigned i = 0; i < callee_copy->num_params; i++) {
- /* Only out or inout parameters */
- if (call->callee->params[i].param_type == nir_parameter_in)
- continue;
-
- nir_copy_deref_var(b, call->params[i],
- nir_deref_var_create(b->shader,
- callee_copy->params[i]));
- }
- if (!glsl_type_is_void(call->callee->return_type)) {
- nir_copy_deref_var(b, call->return_deref,
- nir_deref_var_create(b->shader,
- callee_copy->return_var));
+ /* Make sure that the function we're calling is already inlined */
+ inline_function_impl(call->callee->impl, inlined);
+
+ b->cursor = nir_instr_remove(&call->instr);
+
+ /* Rewrite all of the uses of the callee's parameters to use the call
+ * instructions sources. In order to ensure that the "load" happens
+ * here and not later (for register sources), we make sure to convert it
+ * to an SSA value first.
+ */
+ const unsigned num_params = call->num_params;
+ NIR_VLA(nir_ssa_def *, params, num_params);
+ for (unsigned i = 0; i < num_params; i++) {
+ params[i] = nir_ssa_for_src(b, call->params[i],
+ call->callee->params[i].num_components);
}
- nir_instr_remove(&call->instr);
+ nir_inline_function_impl(b, call->callee->impl, params);
}
- return true;
+ return progress;
}
static bool
if (_mesa_set_search(inlined, impl))
return false; /* Already inlined */
- struct inline_functions_state state;
-
- state.inlined = inlined;
- state.progress = false;
- nir_builder_init(&state.builder, impl);
+ nir_builder b;
+ nir_builder_init(&b, impl);
- nir_foreach_block(impl, inline_functions_block, &state);
+ bool progress = false;
+ nir_foreach_block_safe(block, impl) {
+ progress |= inline_functions_block(block, &b, inlined);
+ }
- if (state.progress) {
+ if (progress) {
/* SSA and register indices are completely messed up now */
nir_index_ssa_defs(impl);
nir_index_local_regs(impl);
nir_metadata_preserve(impl, nir_metadata_none);
+ } else {
+ nir_metadata_preserve(impl, nir_metadata_all);
}
_mesa_set_add(inlined, impl);
- return state.progress;
+ return progress;
}
+/** A pass to inline all functions in a shader into their callers
+ *
+ * For most use-cases, function inlining is a multi-step process. The general
+ * pattern employed by SPIR-V consumers and others is as follows:
+ *
+ * 1. nir_lower_variable_initializers(shader, nir_var_function_temp)
+ *
+ * This is needed because local variables from the callee are simply added
+ * to the locals list for the caller and the information about where the
+ * constant initializer logically happens is lost. If the callee is
+ * called in a loop, this can cause the variable to go from being
+ * initialized once per loop iteration to being initialized once at the
+ * top of the caller and values to persist from one invocation of the
+ * callee to the next. The simple solution to this problem is to get rid
+ * of constant initializers before function inlining.
+ *
+ * 2. nir_lower_returns(shader)
+ *
+ * nir_inline_functions assumes that all functions end "naturally" by
+ * execution reaching the end of the function without any return
+ * instructions causing instant jumps to the end. Thanks to NIR being
+ * structured, we can't represent arbitrary jumps to various points in the
+ * program which is what an early return in the callee would have to turn
+ * into when we inline it into the caller. Instead, we require returns to
+ * be lowered which lets us just copy+paste the callee directly into the
+ * caller.
+ *
+ * 3. nir_inline_functions(shader)
+ *
+ * This does the actual function inlining and the resulting shader will
+ * contain no call instructions.
+ *
+ * 4. nir_opt_deref(shader)
+ *
+ * Most functions contain pointer parameters where the result of a deref
+ * instruction is passed in as a parameter, loaded via a load_param
+ * intrinsic, and then turned back into a deref via a cast. Function
+ * inlining will get rid of the load_param but we are still left with a
+ * cast. Running nir_opt_deref gets rid of the intermediate cast and
+ * results in a whole deref chain again. This is currently required by a
+ * number of optimizations and lowering passes at least for certain
+ * variable modes.
+ *
+ * 5. Loop over the functions and delete all but the main entrypoint.
+ *
+ * In the Intel Vulkan driver this looks like this:
+ *
+ * foreach_list_typed_safe(nir_function, func, node, &nir->functions) {
+ * if (func != entry_point)
+ * exec_node_remove(&func->node);
+ * }
+ * assert(exec_list_length(&nir->functions) == 1);
+ *
+ * While nir_inline_functions does get rid of all call instructions, it
+ * doesn't get rid of any functions because it doesn't know what the "root
+ * function" is. Instead, it's up to the individual driver to know how to
+ * decide on a root function and delete the rest. With SPIR-V,
+ * spirv_to_nir returns the root function and so we can just use == whereas
+ * with GL, you may have to look for a function named "main".
+ *
+ * 6. nir_lower_variable_initializers(shader, ~nir_var_function_temp)
+ *
+ * Lowering constant initializers on inputs, outputs, global variables,
+ * etc. requires that we know the main entrypoint so that we know where to
+ * initialize them. Otherwise, we would have to assume that anything
+ * could be a main entrypoint and initialize them at the start of every
+ * function but that would clearly be wrong if any of those functions were
+ * ever called within another function. Simply requiring a single-
+ * entrypoint function shader is the best way to make it well-defined.
+ */
bool
nir_inline_functions(nir_shader *shader)
{
- struct set *inlined = _mesa_set_create(NULL, _mesa_hash_pointer,
- _mesa_key_pointer_equal);
+ struct set *inlined = _mesa_pointer_set_create(NULL);
bool progress = false;
- nir_foreach_function(shader, function) {
+ nir_foreach_function(function, shader) {
if (function->impl)
progress = inline_function_impl(function->impl, inlined) || progress;
}
projects / mesa.git / blobdiff