From: Louis-Francis Ratté-Boulianne Date: Sat, 12 Oct 2019 04:19:50 +0000 (-0400) Subject: glsl/linker: handle array/struct members for DisableXfbPacking X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=4a329bea44fca8607a3e4538b18fd93864d99c18;p=mesa.git glsl/linker: handle array/struct members for DisableXfbPacking When varying packing is disabled for transform feedback and a xfb declaration points to an array element or structure member, the element/member should be aligned to the start of a slot as well. If that's not the case, a new varying is created and the element/member value is copied. There might a way to further optimize the number of slots allocated or the number of copies necessary if the performance cost is problematic. For example, in cases where simply padding the top level variable might correctly align all the captured values. Signed-off-by: Louis-Francis Ratté-Boulianne Reviewed-by: Alyssa Rosenzweig Acked-by: Daniel Stone Part-of: --- diff --git a/src/compiler/Makefile.sources b/src/compiler/Makefile.sources index 2482de2a799..48c5d6e7cf2 100644 --- a/src/compiler/Makefile.sources +++ b/src/compiler/Makefile.sources @@ -125,6 +125,7 @@ LIBGLSL_FILES = \ glsl/lower_output_reads.cpp \ glsl/lower_shared_reference.cpp \ glsl/lower_ubo_reference.cpp \ + glsl/lower_xfb_varying.cpp \ glsl/opt_algebraic.cpp \ glsl/opt_array_splitting.cpp \ glsl/opt_conditional_discard.cpp \ diff --git a/src/compiler/glsl/ir_optimization.h b/src/compiler/glsl/ir_optimization.h index 11695c5aba2..20766db3354 100644 --- a/src/compiler/glsl/ir_optimization.h +++ b/src/compiler/glsl/ir_optimization.h @@ -143,6 +143,9 @@ bool lower_quadop_vector(exec_list *instructions, bool dont_lower_swz); bool lower_const_arrays_to_uniforms(exec_list *instructions, unsigned stage, unsigned max_uniform_components); bool lower_clip_cull_distance(struct gl_shader_program *prog, gl_linked_shader *shader); +ir_variable * lower_xfb_varying(void *mem_ctx, + gl_linked_shader *shader, + const char *old_var_name); void lower_output_reads(unsigned stage, exec_list *instructions); bool lower_packing_builtins(exec_list *instructions, int op_mask); void lower_shared_reference(struct gl_context *ctx, diff --git a/src/compiler/glsl/link_varyings.cpp b/src/compiler/glsl/link_varyings.cpp index 4b63f7693f6..4757825c244 100644 --- a/src/compiler/glsl/link_varyings.cpp +++ b/src/compiler/glsl/link_varyings.cpp @@ -1373,6 +1373,21 @@ tfeedback_decl::find_candidate(gl_shader_program *prog, return this->matched_candidate; } +/** + * Force a candidate over the previously matched one. It happens when a new + * varying needs to be created to match the xfb declaration, for example, + * to fullfil an alignment criteria. + */ +void +tfeedback_decl::set_lowered_candidate(const tfeedback_candidate *candidate) +{ + this->matched_candidate = candidate; + + /* The subscript part is no longer relevant */ + this->is_subscripted = false; + this->array_subscript = 0; +} + /** * Parse all the transform feedback declarations that were passed to @@ -2779,6 +2794,41 @@ assign_varying_locations(struct gl_context *ctx, return false; } + /* A new output varying is needed, and the matched candidate should be + * replaced, if varying packing is disabled for xfb and the current + * declaration is not aligned within the top level varying + * (e.g. vec3_arr[1]). + */ + const unsigned dmul = + matched_candidate->type->without_array()->is_64bit() ? 2 : 1; + if (disable_xfb_packing && + !tfeedback_decls[i].is_aligned(dmul, matched_candidate->offset)) { + ir_variable *new_var; + tfeedback_candidate *new_candidate = NULL; + + new_var = lower_xfb_varying(mem_ctx, producer, tfeedback_decls[i].name()); + if (new_var == NULL) { + ralloc_free(hash_table_ctx); + return false; + } + + /* Create new candidate and replace matched_candidate */ + new_candidate = rzalloc(mem_ctx, tfeedback_candidate); + new_candidate->toplevel_var = new_var; + new_candidate->toplevel_var->data.is_unmatched_generic_inout = 1; + new_candidate->type = new_var->type; + new_candidate->offset = 0; + _mesa_hash_table_insert(tfeedback_candidates, + ralloc_strdup(mem_ctx, new_var->name), + new_candidate); + + tfeedback_decls[i].set_lowered_candidate(new_candidate); + matched_candidate = new_candidate; + } + + /* Mark as xfb varying */ + matched_candidate->toplevel_var->data.is_xfb = 1; + /* Mark xfb varyings as always active */ matched_candidate->toplevel_var->data.always_active_io = 1; diff --git a/src/compiler/glsl/link_varyings.h b/src/compiler/glsl/link_varyings.h index b802250819e..6f4bcdc79c5 100644 --- a/src/compiler/glsl/link_varyings.h +++ b/src/compiler/glsl/link_varyings.h @@ -104,6 +104,7 @@ public: const void *mem_ctx) const; const tfeedback_candidate *find_candidate(gl_shader_program *prog, hash_table *tfeedback_candidates); + void set_lowered_candidate(const tfeedback_candidate *candidate); bool is_next_buffer_separator() const { @@ -123,6 +124,11 @@ public: return !this->next_buffer_separator && !this->skip_components; } + bool is_aligned(unsigned dmul, unsigned offset) const + { + return (dmul * (this->array_subscript + offset)) % 4 == 0; + } + const char *name() const { return this->orig_name; diff --git a/src/compiler/glsl/lower_xfb_varying.cpp b/src/compiler/glsl/lower_xfb_varying.cpp new file mode 100644 index 00000000000..d460bbd5cae --- /dev/null +++ b/src/compiler/glsl/lower_xfb_varying.cpp @@ -0,0 +1,222 @@ +/* + * Copyright ©2019 Collabora Ltd. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +/** + * \file lower_xfb_varying.cpp + * + */ + +#include "ir.h" +#include "main/mtypes.h" +#include "glsl_symbol_table.h" +#include "util/strndup.h" + +namespace { + +/** + * Visitor that splices varying packing code before every return. + */ +class lower_xfb_var_splicer : public ir_hierarchical_visitor +{ +public: + explicit lower_xfb_var_splicer(void *mem_ctx, + const exec_list *instructions); + + virtual ir_visitor_status visit_leave(ir_return *ret); + virtual ir_visitor_status visit_leave(ir_function_signature *sig); + +private: + /** + * Memory context used to allocate new instructions for the shader. + */ + void * const mem_ctx; + + /** + * Instructions that should be spliced into place before each return. + */ + const exec_list *instructions; +}; + +} /* anonymous namespace */ + + +lower_xfb_var_splicer::lower_xfb_var_splicer(void *mem_ctx, const exec_list *instructions) + : mem_ctx(mem_ctx), instructions(instructions) +{ +} + +ir_visitor_status +lower_xfb_var_splicer::visit_leave(ir_return *ret) +{ + foreach_in_list(ir_instruction, ir, this->instructions) { + ret->insert_before(ir->clone(this->mem_ctx, NULL)); + } + return visit_continue; +} + +/** Insert a copy-back assignment at the end of the main() function */ +ir_visitor_status +lower_xfb_var_splicer::visit_leave(ir_function_signature *sig) +{ + if (strcmp(sig->function_name(), "main") != 0) + return visit_continue; + + if (((ir_instruction*)sig->body.get_tail())->ir_type == ir_type_return) + return visit_continue; + + foreach_in_list(ir_instruction, ir, this->instructions) { + sig->body.push_tail(ir->clone(this->mem_ctx, NULL)); + } + + return visit_continue; +} + +static char* +get_field_name(const char *name) +{ + const char *first_dot = strchr(name, '.'); + const char *first_square_bracket = strchr(name, '['); + int name_size = 0; + + if (!first_square_bracket && !first_dot) + name_size = strlen(name); + else if ((!first_square_bracket || + (first_dot && first_dot < first_square_bracket))) + name_size = first_dot - name; + else + name_size = first_square_bracket - name; + + return strndup(name, name_size); +} + +/* Generate a new name given the old xfb declaration string by replacing dots + * with '_', brackets with '@' and appending "-xfb" */ +static char * +generate_new_name(void *mem_ctx, const char *name) +{ + char *new_name; + unsigned i = 0; + + new_name = ralloc_strdup(mem_ctx, name); + while (new_name[i]) { + if (new_name[i] == '.') { + new_name[i] = '_'; + } else if (new_name[i] == '[' || new_name[i] == ']') { + new_name[i] = '@'; + } + i++; + } + + if (!ralloc_strcat(&new_name, "-xfb")) { + ralloc_free(new_name); + return NULL; + } + + return new_name; +} + +/* Get the dereference for the given variable name. The method is called + * recursively to parse array indices and struct members. */ +static bool +get_deref(void *ctx, + const char *name, + struct gl_linked_shader *shader, + ir_dereference **deref, + const glsl_type **type) +{ + if (name[0] == '\0') { + /* End */ + return (*deref != NULL); + } else if (name[0] == '[') { + /* Array index */ + char *endptr = NULL; + unsigned index; + + index = strtol(name + 1, &endptr, 10); + assert(*type != NULL && (*type)->is_array() && endptr[0] == ']'); + *deref = new(ctx) ir_dereference_array(*deref, new(ctx) ir_constant(index)); + *type = (*type)->without_array(); + return get_deref(ctx, endptr + 1, shader, deref, type); + } else if (name[0] == '.') { + /* Struct member */ + char *field = get_field_name(name + 1); + + assert(*type != NULL && (*type)->is_struct() && field != NULL); + *deref = new(ctx) ir_dereference_record(*deref, field); + *type = (*type)->field_type(field); + assert(*type != glsl_type::error_type); + name += 1 + strlen(field); + free(field); + return get_deref(ctx, name, shader, deref, type); + } else { + /* Top level variable */ + char *field = get_field_name(name); + ir_variable *toplevel_var; + + toplevel_var = shader->symbols->get_variable(field); + name += strlen(field); + free(field); + if (toplevel_var == NULL) { + return false; + } + + *deref = new (ctx) ir_dereference_variable(toplevel_var); + *type = toplevel_var->type; + return get_deref(ctx, name, shader, deref, type); + } +} + +ir_variable * +lower_xfb_varying(void *mem_ctx, + struct gl_linked_shader *shader, + const char *old_var_name) +{ + exec_list new_instructions; + char *new_var_name; + ir_dereference *deref = NULL; + const glsl_type *type = NULL; + + if (!get_deref(mem_ctx, old_var_name, shader, &deref, &type)) { + if (deref) { + delete deref; + } + return NULL; + } + + new_var_name = generate_new_name(mem_ctx, old_var_name); + ir_variable *new_variable + = new(mem_ctx) ir_variable(type, new_var_name, ir_var_shader_out); + new_variable->data.assigned = true; + new_variable->data.used = true; + shader->ir->push_head(new_variable); + ralloc_free(new_var_name); + + ir_dereference *lhs = new(mem_ctx) ir_dereference_variable(new_variable); + ir_assignment *new_assignment = new(mem_ctx) ir_assignment(lhs, deref); + new_instructions.push_tail(new_assignment); + + lower_xfb_var_splicer splicer(mem_ctx, &new_instructions); + visit_list_elements(&splicer, shader->ir); + + return new_variable; +} diff --git a/src/compiler/glsl/meson.build b/src/compiler/glsl/meson.build index 1e2b3da1b4b..ea3e7beaa0c 100644 --- a/src/compiler/glsl/meson.build +++ b/src/compiler/glsl/meson.build @@ -175,6 +175,7 @@ files_libglsl = files( 'lower_output_reads.cpp', 'lower_shared_reference.cpp', 'lower_ubo_reference.cpp', + 'lower_xfb_varying.cpp', 'opt_algebraic.cpp', 'opt_array_splitting.cpp', 'opt_conditional_discard.cpp',