From 32aaf89823de11e98cb59d5ec78c66cd3e74bcd4 Mon Sep 17 00:00:00 2001 From: Kenneth Graunke Date: Mon, 15 Nov 2010 14:35:46 -0800 Subject: glsl: Rename various ir_* files to lower_* and opt_*. This helps distinguish between lowering passes, optimization passes, and other compiler code. --- src/glsl/SConscript | 48 ++++++++++++++++++++++++------------------------ 1 file changed, 24 insertions(+), 24 deletions(-) (limited to 'src/glsl/SConscript') diff --git a/src/glsl/SConscript b/src/glsl/SConscript index c183e208d63..f9cc823a2d5 100644 --- a/src/glsl/SConscript +++ b/src/glsl/SConscript @@ -27,54 +27,54 @@ sources = [ 'glsl_types.cpp', 'glsl_symbol_table.cpp', 'hir_field_selection.cpp', - 'ir_algebraic.cpp', 'ir_basic_block.cpp', 'ir_clone.cpp', - 'ir_constant_expression.cpp', - 'ir_constant_folding.cpp', - 'ir_constant_propagation.cpp', - 'ir_constant_variable.cpp', - 'ir_copy_propagation.cpp', 'ir.cpp', - 'ir_dead_code.cpp', - 'ir_dead_code_local.cpp', - 'ir_dead_functions.cpp', - 'ir_div_to_mul_rcp.cpp', - 'ir_explog_to_explog2.cpp', 'ir_expression_flattening.cpp', 'ir_function_can_inline.cpp', 'ir_function.cpp', - 'ir_function_inlining.cpp', 'ir_hierarchical_visitor.cpp', 'ir_hv_accept.cpp', - 'ir_if_simplification.cpp', - 'ir_if_to_cond_assign.cpp', 'ir_import_prototypes.cpp', - 'ir_lower_jumps.cpp', - 'ir_mat_op_to_vec.cpp', - 'ir_mod_to_fract.cpp', - 'ir_noop_swizzle.cpp', 'ir_print_visitor.cpp', 'ir_reader.cpp', 'ir_rvalue_visitor.cpp', 'ir_set_program_inouts.cpp', - 'ir_structure_splitting.cpp', - 'ir_sub_to_add_neg.cpp', - 'ir_swizzle_swizzle.cpp', - 'ir_tree_grafting.cpp', 'ir_validate.cpp', 'ir_variable.cpp', 'ir_variable_refcount.cpp', - 'ir_vec_index_to_cond_assign.cpp', - 'ir_vec_index_to_swizzle.cpp', 'linker.cpp', 'link_functions.cpp', 'loop_analysis.cpp', 'loop_controls.cpp', 'loop_unroll.cpp', + 'lower_div_to_mul_rcp.cpp', + 'lower_explog_to_explog2.cpp', + 'lower_if_to_cond_assign.cpp', + 'lower_jumps.cpp', + 'lower_mat_op_to_vec.cpp', + 'lower_mod_to_fract.cpp', 'lower_noise.cpp', + 'lower_sub_to_add_neg.cpp', 'lower_variable_index_to_cond_assign.cpp', + 'lower_vec_index_to_cond_assign.cpp', + 'lower_vec_index_to_swizzle.cpp', + 'opt_algebraic.cpp', + 'opt_constant_folding.cpp', + 'opt_noop_swizzle.cpp', 'opt_redundant_jumps.cpp', + 'opt_constant_expression.cpp', + 'opt_constant_propagation.cpp', + 'opt_constant_variable.cpp', + 'opt_copy_propagation.cpp', + 'opt_dead_code.cpp', + 'opt_dead_code_local.cpp', + 'opt_dead_functions.cpp', + 'opt_function_inlining.cpp', + 'opt_if_simplification.cpp', + 'opt_structure_splitting.cpp', + 'opt_swizzle_swizzle.cpp', + 'opt_tree_grafting.cpp', 's_expression.cpp', ] -- cgit v1.2.3 From 9b4b70e7e28e5371ca5fddd6f8c7e09d8efd5ce0 Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Mon, 15 Nov 2010 18:59:39 -0700 Subject: glsl: remove opt_constant_expression.cpp from SConscript And alphabetize the opt_* files. --- src/glsl/SConscript | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) (limited to 'src/glsl/SConscript') diff --git a/src/glsl/SConscript b/src/glsl/SConscript index f9cc823a2d5..20cb91b6077 100644 --- a/src/glsl/SConscript +++ b/src/glsl/SConscript @@ -61,9 +61,6 @@ sources = [ 'lower_vec_index_to_swizzle.cpp', 'opt_algebraic.cpp', 'opt_constant_folding.cpp', - 'opt_noop_swizzle.cpp', - 'opt_redundant_jumps.cpp', - 'opt_constant_expression.cpp', 'opt_constant_propagation.cpp', 'opt_constant_variable.cpp', 'opt_copy_propagation.cpp', @@ -72,6 +69,8 @@ sources = [ 'opt_dead_functions.cpp', 'opt_function_inlining.cpp', 'opt_if_simplification.cpp', + 'opt_noop_swizzle.cpp', + 'opt_redundant_jumps.cpp', 'opt_structure_splitting.cpp', 'opt_swizzle_swizzle.cpp', 'opt_tree_grafting.cpp', -- cgit v1.2.3 From 1935bdca44ec1f0a980b6b0c024ff2e4c7032dfe Mon Sep 17 00:00:00 2001 From: Vinson Lee Date: Mon, 15 Nov 2010 20:54:12 -0800 Subject: glsl: Add ir_constant_expression.cpp to SConscript. This was accidentally removed in commit 32aaf89823de11e98cb59d5ec78c66cd3e74bcd4. Fixes SCons builds. --- src/glsl/SConscript | 1 + 1 file changed, 1 insertion(+) (limited to 'src/glsl/SConscript') diff --git a/src/glsl/SConscript b/src/glsl/SConscript index 20cb91b6077..4cb2d8bcb36 100644 --- a/src/glsl/SConscript +++ b/src/glsl/SConscript @@ -29,6 +29,7 @@ sources = [ 'hir_field_selection.cpp', 'ir_basic_block.cpp', 'ir_clone.cpp', + 'ir_constant_expression.cpp', 'ir.cpp', 'ir_expression_flattening.cpp', 'ir_function_can_inline.cpp', -- cgit v1.2.3 From 63684a9ae7a66f68df1f2c68cd9358e5622122a3 Mon Sep 17 00:00:00 2001 From: Kenneth Graunke Date: Thu, 18 Nov 2010 17:54:07 -0800 Subject: glsl: Combine many instruction lowering passes into one. This should save on the overhead of tree-walking and provide a convenient place to add more instruction lowering in the future. Signed-off-by: Ian Romanick --- src/glsl/Makefile | 5 +- src/glsl/SConscript | 5 +- src/glsl/glsl_parser_extras.cpp | 2 +- src/glsl/ir_optimization.h | 10 +- src/glsl/lower_div_to_mul_rcp.cpp | 114 --------------- src/glsl/lower_explog_to_explog2.cpp | 85 ------------ src/glsl/lower_instructions.cpp | 262 +++++++++++++++++++++++++++++++++++ src/glsl/lower_mod_to_fract.cpp | 90 ------------ src/glsl/lower_sub_to_add_neg.cpp | 76 ---------- src/mesa/drivers/dri/i965/brw_fs.cpp | 10 +- src/mesa/program/ir_to_mesa.cpp | 5 +- 11 files changed, 281 insertions(+), 383 deletions(-) delete mode 100644 src/glsl/lower_div_to_mul_rcp.cpp delete mode 100644 src/glsl/lower_explog_to_explog2.cpp create mode 100644 src/glsl/lower_instructions.cpp delete mode 100644 src/glsl/lower_mod_to_fract.cpp delete mode 100644 src/glsl/lower_sub_to_add_neg.cpp (limited to 'src/glsl/SConscript') diff --git a/src/glsl/Makefile b/src/glsl/Makefile index ea4d6be5a41..f5aadc347bd 100644 --- a/src/glsl/Makefile +++ b/src/glsl/Makefile @@ -52,14 +52,11 @@ CXX_SOURCES = \ loop_analysis.cpp \ loop_controls.cpp \ loop_unroll.cpp \ - lower_div_to_mul_rcp.cpp \ - lower_explog_to_explog2.cpp \ lower_if_to_cond_assign.cpp \ + lower_instructions.cpp \ lower_jumps.cpp \ lower_mat_op_to_vec.cpp \ - lower_mod_to_fract.cpp \ lower_noise.cpp \ - lower_sub_to_add_neg.cpp \ lower_texture_projection.cpp \ lower_variable_index_to_cond_assign.cpp \ lower_vec_index_to_cond_assign.cpp \ diff --git a/src/glsl/SConscript b/src/glsl/SConscript index 4cb2d8bcb36..f708635e53a 100644 --- a/src/glsl/SConscript +++ b/src/glsl/SConscript @@ -49,14 +49,11 @@ sources = [ 'loop_analysis.cpp', 'loop_controls.cpp', 'loop_unroll.cpp', - 'lower_div_to_mul_rcp.cpp', - 'lower_explog_to_explog2.cpp', 'lower_if_to_cond_assign.cpp', + 'lower_instructions.cpp', 'lower_jumps.cpp', 'lower_mat_op_to_vec.cpp', - 'lower_mod_to_fract.cpp', 'lower_noise.cpp', - 'lower_sub_to_add_neg.cpp', 'lower_variable_index_to_cond_assign.cpp', 'lower_vec_index_to_cond_assign.cpp', 'lower_vec_index_to_swizzle.cpp', diff --git a/src/glsl/glsl_parser_extras.cpp b/src/glsl/glsl_parser_extras.cpp index b2c378dd858..302cfbc5668 100644 --- a/src/glsl/glsl_parser_extras.cpp +++ b/src/glsl/glsl_parser_extras.cpp @@ -708,7 +708,7 @@ do_common_optimization(exec_list *ir, bool linked, unsigned max_unroll_iteration { GLboolean progress = GL_FALSE; - progress = do_sub_to_add_neg(ir) || progress; + progress = lower_instructions(ir, SUB_TO_ADD_NEG) || progress; if (linked) { progress = do_function_inlining(ir) || progress; diff --git a/src/glsl/ir_optimization.h b/src/glsl/ir_optimization.h index 1f8da16bcb2..fa497a45551 100644 --- a/src/glsl/ir_optimization.h +++ b/src/glsl/ir_optimization.h @@ -28,6 +28,13 @@ * Prototypes for optimization passes to be called by the compiler and drivers. */ +/* Operations for lower_instructions() */ +#define SUB_TO_ADD_NEG 0x01 +#define DIV_TO_MUL_RCP 0x02 +#define EXP_TO_EXP2 0x04 +#define LOG_TO_LOG2 0x08 +#define MOD_TO_FRACT 0x10 + bool do_common_optimization(exec_list *ir, bool linked, unsigned max_unroll_iterations); bool do_algebraic(exec_list *instructions); @@ -40,8 +47,6 @@ bool do_dead_code(exec_list *instructions); bool do_dead_code_local(exec_list *instructions); bool do_dead_code_unlinked(exec_list *instructions); bool do_dead_functions(exec_list *instructions); -bool do_div_to_mul_rcp(exec_list *instructions); -bool do_explog_to_explog2(exec_list *instructions); bool do_function_inlining(exec_list *instructions); bool do_lower_jumps(exec_list *instructions, bool pull_out_jumps = true, bool lower_sub_return = true, bool lower_main_return = false, bool lower_continue = false, bool lower_break = false); bool do_lower_texture_projection(exec_list *instructions); @@ -56,6 +61,7 @@ bool do_swizzle_swizzle(exec_list *instructions); bool do_tree_grafting(exec_list *instructions); bool do_vec_index_to_cond_assign(exec_list *instructions); bool do_vec_index_to_swizzle(exec_list *instructions); +bool lower_instructions(exec_list *instructions, unsigned what_to_lower); bool lower_noise(exec_list *instructions); bool lower_variable_index_to_cond_assign(exec_list *instructions, bool lower_input, bool lower_output, bool lower_temp, bool lower_uniform); diff --git a/src/glsl/lower_div_to_mul_rcp.cpp b/src/glsl/lower_div_to_mul_rcp.cpp deleted file mode 100644 index db9d6854cda..00000000000 --- a/src/glsl/lower_div_to_mul_rcp.cpp +++ /dev/null @@ -1,114 +0,0 @@ -/* - * Copyright © 2010 Intel Corporation - * - * 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_div_to_mul_rcp.cpp - * - * Breaks an ir_unop_div expression down to op0 * (rcp(op1)). - * - * Many GPUs don't have a divide instruction (945 and 965 included), - * but they do have an RCP instruction to compute an approximate - * reciprocal. By breaking the operation down, constant reciprocals - * can get constant folded. - */ - -#include "ir.h" -#include "glsl_types.h" - -class ir_div_to_mul_rcp_visitor : public ir_hierarchical_visitor { -public: - ir_div_to_mul_rcp_visitor() - { - this->made_progress = false; - } - - ir_visitor_status visit_leave(ir_expression *); - - bool made_progress; -}; - -bool -do_div_to_mul_rcp(exec_list *instructions) -{ - ir_div_to_mul_rcp_visitor v; - - visit_list_elements(&v, instructions); - return v.made_progress; -} - -ir_visitor_status -ir_div_to_mul_rcp_visitor::visit_leave(ir_expression *ir) -{ - if (ir->operation != ir_binop_div) - return visit_continue; - - if (!ir->operands[1]->type->is_integer()) { - /* New expression for the 1.0 / op1 */ - ir_rvalue *expr; - expr = new(ir) ir_expression(ir_unop_rcp, - ir->operands[1]->type, - ir->operands[1], - NULL); - - /* op0 / op1 -> op0 * (1.0 / op1) */ - ir->operation = ir_binop_mul; - ir->operands[1] = expr; - } else { - /* Be careful with integer division -- we need to do it as a - * float and re-truncate, since rcp(n > 1) of an integer would - * just be 0. - */ - ir_rvalue *op0, *op1; - const struct glsl_type *vec_type; - - vec_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, - ir->operands[1]->type->vector_elements, - ir->operands[1]->type->matrix_columns); - - if (ir->operands[1]->type->base_type == GLSL_TYPE_INT) - op1 = new(ir) ir_expression(ir_unop_i2f, vec_type, ir->operands[1], NULL); - else - op1 = new(ir) ir_expression(ir_unop_u2f, vec_type, ir->operands[1], NULL); - - op1 = new(ir) ir_expression(ir_unop_rcp, op1->type, op1, NULL); - - vec_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, - ir->operands[0]->type->vector_elements, - ir->operands[0]->type->matrix_columns); - - if (ir->operands[0]->type->base_type == GLSL_TYPE_INT) - op0 = new(ir) ir_expression(ir_unop_i2f, vec_type, ir->operands[0], NULL); - else - op0 = new(ir) ir_expression(ir_unop_u2f, vec_type, ir->operands[0], NULL); - - op0 = new(ir) ir_expression(ir_binop_mul, vec_type, op0, op1); - - ir->operation = ir_unop_f2i; - ir->operands[0] = op0; - ir->operands[1] = NULL; - } - - this->made_progress = true; - - return visit_continue; -} diff --git a/src/glsl/lower_explog_to_explog2.cpp b/src/glsl/lower_explog_to_explog2.cpp deleted file mode 100644 index 8ad02625443..00000000000 --- a/src/glsl/lower_explog_to_explog2.cpp +++ /dev/null @@ -1,85 +0,0 @@ -/* - * Copyright © 2010 Intel Corporation - * - * 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_explog_to_explog2.cpp - * - * Many GPUs don't have a base e log or exponent instruction, but they - * do have base 2 versions, so this pass converts exp and log to exp2 - * and log2 operations. - */ - -#include "main/core.h" /* for log2f on MSVC */ -#include "ir.h" -#include "glsl_types.h" - -class ir_explog_to_explog2_visitor : public ir_hierarchical_visitor { -public: - ir_explog_to_explog2_visitor() - { - this->progress = false; - } - - ir_visitor_status visit_leave(ir_expression *); - - bool progress; -}; - -bool -do_explog_to_explog2(exec_list *instructions) -{ - ir_explog_to_explog2_visitor v; - - visit_list_elements(&v, instructions); - return v.progress; -} - -ir_visitor_status -ir_explog_to_explog2_visitor::visit_leave(ir_expression *ir) -{ - if (ir->operation == ir_unop_exp) { - void *mem_ctx = talloc_parent(ir); - ir_constant *log2_e = new(mem_ctx) ir_constant(log2f(M_E)); - - ir->operation = ir_unop_exp2; - ir->operands[0] = new(mem_ctx) ir_expression(ir_binop_mul, - ir->operands[0]->type, - ir->operands[0], - log2_e); - this->progress = true; - } - - if (ir->operation == ir_unop_log) { - void *mem_ctx = talloc_parent(ir); - - ir->operation = ir_binop_mul; - ir->operands[0] = new(mem_ctx) ir_expression(ir_unop_log2, - ir->operands[0]->type, - ir->operands[0], - NULL); - ir->operands[1] = new(mem_ctx) ir_constant(1.0f / log2f(M_E)); - this->progress = true; - } - - return visit_continue; -} diff --git a/src/glsl/lower_instructions.cpp b/src/glsl/lower_instructions.cpp new file mode 100644 index 00000000000..d460ba1a97a --- /dev/null +++ b/src/glsl/lower_instructions.cpp @@ -0,0 +1,262 @@ +/* + * Copyright © 2010 Intel Corporation + * + * 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_instructions.cpp + * + * Many GPUs lack native instructions for certain expression operations, and + * must replace them with some other expression tree. This pass lowers some + * of the most common cases, allowing the lowering code to be implemented once + * rather than in each driver backend. + * + * Currently supported transformations: + * - SUB_TO_ADD_NEG + * - DIV_TO_MUL_RCP + * - EXP_TO_EXP2 + * - LOG_TO_LOG2 + * - MOD_TO_FRACT + * + * SUB_TO_ADD_NEG: + * --------------- + * Breaks an ir_binop_sub expression down to add(op0, neg(op1)) + * + * This simplifies expression reassociation, and for many backends + * there is no subtract operation separate from adding the negation. + * For backends with native subtract operations, they will probably + * want to recognize add(op0, neg(op1)) or the other way around to + * produce a subtract anyway. + * + * DIV_TO_MUL_RCP: + * --------------- + * Breaks an ir_unop_div expression down to op0 * (rcp(op1)). + * + * Many GPUs don't have a divide instruction (945 and 965 included), + * but they do have an RCP instruction to compute an approximate + * reciprocal. By breaking the operation down, constant reciprocals + * can get constant folded. + * + * EXP_TO_EXP2 and LOG_TO_LOG2: + * ---------------------------- + * Many GPUs don't have a base e log or exponent instruction, but they + * do have base 2 versions, so this pass converts exp and log to exp2 + * and log2 operations. + * + * MOD_TO_FRACT: + * ------------- + * Breaks an ir_unop_mod expression down to (op1 * fract(op0 / op1)) + * + * Many GPUs don't have a MOD instruction (945 and 965 included), and + * if we have to break it down like this anyway, it gives an + * opportunity to do things like constant fold the (1.0 / op1) easily. + */ + +#include "main/core.h" /* for M_E */ +#include "glsl_types.h" +#include "ir.h" +#include "ir_optimization.h" + +class lower_instructions_visitor : public ir_hierarchical_visitor { +public: + lower_instructions_visitor(unsigned lower) + : progress(false), lower(lower) { } + + ir_visitor_status visit_leave(ir_expression *); + + bool progress; + +private: + unsigned lower; /** Bitfield of which operations to lower */ + + void sub_to_add_neg(ir_expression *); + void div_to_mul_rcp(ir_expression *); + void mod_to_fract(ir_expression *); + void exp_to_exp2(ir_expression *); + void log_to_log2(ir_expression *); +}; + +/** + * Determine if a particular type of lowering should occur + */ +#define lowering(x) (this->lower & x) + +bool +lower_instructions(exec_list *instructions, unsigned what_to_lower) +{ + lower_instructions_visitor v(what_to_lower); + + visit_list_elements(&v, instructions); + return v.progress; +} + +void +lower_instructions_visitor::sub_to_add_neg(ir_expression *ir) +{ + ir->operation = ir_binop_add; + ir->operands[1] = new(ir) ir_expression(ir_unop_neg, ir->operands[1]->type, + ir->operands[1], NULL); + this->progress = true; +} + +void +lower_instructions_visitor::div_to_mul_rcp(ir_expression *ir) +{ + if (!ir->operands[1]->type->is_integer()) { + /* New expression for the 1.0 / op1 */ + ir_rvalue *expr; + expr = new(ir) ir_expression(ir_unop_rcp, + ir->operands[1]->type, + ir->operands[1], + NULL); + + /* op0 / op1 -> op0 * (1.0 / op1) */ + ir->operation = ir_binop_mul; + ir->operands[1] = expr; + } else { + /* Be careful with integer division -- we need to do it as a + * float and re-truncate, since rcp(n > 1) of an integer would + * just be 0. + */ + ir_rvalue *op0, *op1; + const struct glsl_type *vec_type; + + vec_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, + ir->operands[1]->type->vector_elements, + ir->operands[1]->type->matrix_columns); + + if (ir->operands[1]->type->base_type == GLSL_TYPE_INT) + op1 = new(ir) ir_expression(ir_unop_i2f, vec_type, ir->operands[1], NULL); + else + op1 = new(ir) ir_expression(ir_unop_u2f, vec_type, ir->operands[1], NULL); + + op1 = new(ir) ir_expression(ir_unop_rcp, op1->type, op1, NULL); + + vec_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, + ir->operands[0]->type->vector_elements, + ir->operands[0]->type->matrix_columns); + + if (ir->operands[0]->type->base_type == GLSL_TYPE_INT) + op0 = new(ir) ir_expression(ir_unop_i2f, vec_type, ir->operands[0], NULL); + else + op0 = new(ir) ir_expression(ir_unop_u2f, vec_type, ir->operands[0], NULL); + + op0 = new(ir) ir_expression(ir_binop_mul, vec_type, op0, op1); + + ir->operation = ir_unop_f2i; + ir->operands[0] = op0; + ir->operands[1] = NULL; + } + + this->progress = true; +} + +void +lower_instructions_visitor::exp_to_exp2(ir_expression *ir) +{ + ir_constant *log2_e = new(ir) ir_constant(log2f(M_E)); + + ir->operation = ir_unop_exp2; + ir->operands[0] = new(ir) ir_expression(ir_binop_mul, ir->operands[0]->type, + ir->operands[0], log2_e); + this->progress = true; +} + +void +lower_instructions_visitor::log_to_log2(ir_expression *ir) +{ + ir->operation = ir_binop_mul; + ir->operands[0] = new(ir) ir_expression(ir_unop_log2, ir->operands[0]->type, + ir->operands[0], NULL); + ir->operands[1] = new(ir) ir_constant(1.0f / log2f(M_E)); + this->progress = true; +} + +void +lower_instructions_visitor::mod_to_fract(ir_expression *ir) +{ + ir_variable *temp = new(ir) ir_variable(ir->operands[1]->type, "mod_b", + ir_var_temporary); + this->base_ir->insert_before(temp); + + ir_assignment *const assign = + new(ir) ir_assignment(new(ir) ir_dereference_variable(temp), + ir->operands[1], NULL); + + this->base_ir->insert_before(assign); + + ir_expression *const div_expr = + new(ir) ir_expression(ir_binop_div, ir->operands[0]->type, + ir->operands[0], + new(ir) ir_dereference_variable(temp)); + + /* Don't generate new IR that would need to be lowered in an additional + * pass. + */ + if (lowering(DIV_TO_MUL_RCP)) + div_to_mul_rcp(div_expr); + + ir_rvalue *expr = new(ir) ir_expression(ir_unop_fract, + ir->operands[0]->type, + div_expr, + NULL); + + ir->operation = ir_binop_mul; + ir->operands[0] = new(ir) ir_dereference_variable(temp); + ir->operands[1] = expr; + this->progress = true; +} + +ir_visitor_status +lower_instructions_visitor::visit_leave(ir_expression *ir) +{ + switch (ir->operation) { + case ir_binop_sub: + if (lowering(SUB_TO_ADD_NEG)) + sub_to_add_neg(ir); + break; + + case ir_binop_div: + if (lowering(DIV_TO_MUL_RCP)) + div_to_mul_rcp(ir); + break; + + case ir_unop_exp: + if (lowering(EXP_TO_EXP2)) + exp_to_exp2(ir); + break; + + case ir_unop_log: + if (lowering(LOG_TO_LOG2)) + log_to_log2(ir); + break; + + case ir_binop_mod: + if (lowering(MOD_TO_FRACT)) + mod_to_fract(ir); + break; + + default: + return visit_continue; + } + + return visit_continue; +} diff --git a/src/glsl/lower_mod_to_fract.cpp b/src/glsl/lower_mod_to_fract.cpp deleted file mode 100644 index ff907ae97cc..00000000000 --- a/src/glsl/lower_mod_to_fract.cpp +++ /dev/null @@ -1,90 +0,0 @@ -/* - * Copyright © 2010 Intel Corporation - * - * 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_mod_to_fract.cpp - * - * Breaks an ir_unop_mod expression down to (op1 * fract(op0 / op1)) - * - * Many GPUs don't have a MOD instruction (945 and 965 included), and - * if we have to break it down like this anyway, it gives an - * opportunity to do things like constant fold the (1.0 / op1) easily. - */ - -#include "ir.h" - -class ir_mod_to_fract_visitor : public ir_hierarchical_visitor { -public: - ir_mod_to_fract_visitor() - { - this->made_progress = false; - } - - ir_visitor_status visit_leave(ir_expression *); - - bool made_progress; -}; - -bool -do_mod_to_fract(exec_list *instructions) -{ - ir_mod_to_fract_visitor v; - - visit_list_elements(&v, instructions); - return v.made_progress; -} - -ir_visitor_status -ir_mod_to_fract_visitor::visit_leave(ir_expression *ir) -{ - if (ir->operation != ir_binop_mod) - return visit_continue; - - ir_variable *temp = new(ir) ir_variable(ir->operands[1]->type, "mod_b", - ir_var_temporary); - this->base_ir->insert_before(temp); - - ir_assignment *assign; - ir_rvalue *expr; - - assign = new(ir) ir_assignment(new(ir) ir_dereference_variable(temp), - ir->operands[1], NULL); - this->base_ir->insert_before(assign); - - expr = new(ir) ir_expression(ir_binop_div, - ir->operands[0]->type, - ir->operands[0], - new(ir) ir_dereference_variable(temp)); - - expr = new(ir) ir_expression(ir_unop_fract, - ir->operands[0]->type, - expr, - NULL); - - ir->operation = ir_binop_mul; - ir->operands[0] = new(ir) ir_dereference_variable(temp); - ir->operands[1] = expr; - this->made_progress = true; - - return visit_continue; -} diff --git a/src/glsl/lower_sub_to_add_neg.cpp b/src/glsl/lower_sub_to_add_neg.cpp deleted file mode 100644 index 9e4019709b4..00000000000 --- a/src/glsl/lower_sub_to_add_neg.cpp +++ /dev/null @@ -1,76 +0,0 @@ -/* - * Copyright © 2010 Intel Corporation - * - * 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_sub_to_add_neg.cpp - * - * Breaks an ir_binop_sub expression down to add(op0, neg(op1)) - * - * This simplifies expression reassociation, and for many backends - * there is no subtract operation separate from adding the negation. - * For backends with native subtract operations, they will probably - * want to recognize add(op0, neg(op1)) or the other way around to - * produce a subtract anyway. - */ - -#include "ir.h" - -class ir_sub_to_add_neg_visitor : public ir_hierarchical_visitor { -public: - ir_sub_to_add_neg_visitor() - { - this->progress = false; - } - - ir_visitor_status visit_leave(ir_expression *); - - bool progress; -}; - -bool -do_sub_to_add_neg(exec_list *instructions) -{ - ir_sub_to_add_neg_visitor v; - - visit_list_elements(&v, instructions); - return v.progress; -} - -ir_visitor_status -ir_sub_to_add_neg_visitor::visit_leave(ir_expression *ir) -{ - if (ir->operation != ir_binop_sub) - return visit_continue; - - void *mem_ctx = talloc_parent(ir); - - ir->operation = ir_binop_add; - ir->operands[1] = new(mem_ctx) ir_expression(ir_unop_neg, - ir->operands[1]->type, - ir->operands[1], - NULL); - - this->progress = true; - - return visit_continue; -} diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp index e1cb94452d4..610b7e3c3eb 100644 --- a/src/mesa/drivers/dri/i965/brw_fs.cpp +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp @@ -101,10 +101,12 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) clone_ir_list(mem_ctx, shader->ir, shader->base.ir); do_mat_op_to_vec(shader->ir); - do_mod_to_fract(shader->ir); - do_div_to_mul_rcp(shader->ir); - do_sub_to_add_neg(shader->ir); - do_explog_to_explog2(shader->ir); + lower_instructions(shader->ir, + MOD_TO_FRACT | + DIV_TO_MUL_RCP | + SUB_TO_ADD_NEG | + EXP_TO_EXP2 | + LOG_TO_LOG2); do_lower_texture_projection(shader->ir); brw_do_cubemap_normalize(shader->ir); diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp index 1cb81830426..c7a8594f729 100644 --- a/src/mesa/program/ir_to_mesa.cpp +++ b/src/mesa/program/ir_to_mesa.cpp @@ -2798,9 +2798,8 @@ _mesa_ir_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) /* Lowering */ do_mat_op_to_vec(ir); - do_mod_to_fract(ir); - do_div_to_mul_rcp(ir); - do_explog_to_explog2(ir); + lower_instructions(ir, MOD_TO_FRACT | DIV_TO_MUL_RCP | EXP_TO_EXP2 + | LOG_TO_LOG2); progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress; -- cgit v1.2.3 From 7aebe181f3eea51ae76f7abe200df239c7dd948a Mon Sep 17 00:00:00 2001 From: Vinson Lee Date: Fri, 19 Nov 2010 17:22:23 -0800 Subject: glsl: Add lower_vector.cpp to SConscript. --- src/glsl/SConscript | 1 + 1 file changed, 1 insertion(+) (limited to 'src/glsl/SConscript') diff --git a/src/glsl/SConscript b/src/glsl/SConscript index f708635e53a..fd22f668631 100644 --- a/src/glsl/SConscript +++ b/src/glsl/SConscript @@ -57,6 +57,7 @@ sources = [ 'lower_variable_index_to_cond_assign.cpp', 'lower_vec_index_to_cond_assign.cpp', 'lower_vec_index_to_swizzle.cpp', + 'lower_vector.cpp', 'opt_algebraic.cpp', 'opt_constant_folding.cpp', 'opt_constant_propagation.cpp', -- cgit v1.2.3