/* * 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_mat_op_to_vec.cpp * * Breaks matrix operation expressions down to a series of vector operations. * * Generally this is how we have to codegen matrix operations for a * GPU, so this gives us the chance to constant fold operations on a * column or row. */ #include "ir.h" #include "ir_expression_flattening.h" #include "glsl_types.h" class ir_mat_op_to_vec_visitor : public ir_hierarchical_visitor { public: ir_mat_op_to_vec_visitor() { this->made_progress = false; this->mem_ctx = NULL; } ir_visitor_status visit_leave(ir_assignment *); ir_dereference *get_column(ir_variable *var, int col); ir_rvalue *get_element(ir_variable *var, int col, int row); void do_mul_mat_mat(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var); void do_mul_mat_vec(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var); void do_mul_vec_mat(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var); void do_mul_mat_scalar(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var); void do_equal_mat_mat(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var, bool test_equal); void *mem_ctx; bool made_progress; }; static bool mat_op_to_vec_predicate(ir_instruction *ir) { ir_expression *expr = ir->as_expression(); unsigned int i; if (!expr) return false; for (i = 0; i < expr->get_num_operands(); i++) { if (expr->operands[i]->type->is_matrix()) return true; } return false; } bool do_mat_op_to_vec(exec_list *instructions) { ir_mat_op_to_vec_visitor v; /* Pull out any matrix expression to a separate assignment to a * temp. This will make our handling of the breakdown to * operations on the matrix's vector components much easier. */ do_expression_flattening(instructions, mat_op_to_vec_predicate); visit_list_elements(&v, instructions); return v.made_progress; } ir_rvalue * ir_mat_op_to_vec_visitor::get_element(ir_variable *var, int col, int row) { ir_dereference *deref; deref = new(mem_ctx) ir_dereference_variable(var); if (var->type->is_matrix()) { deref = new(mem_ctx) ir_dereference_array(var, new(mem_ctx) ir_constant(col)); } else { assert(col == 0); } return new(mem_ctx) ir_swizzle(deref, row, 0, 0, 0, 1); } ir_dereference * ir_mat_op_to_vec_visitor::get_column(ir_variable *var, int row) { ir_dereference *deref; if (!var->type->is_matrix()) { deref = new(mem_ctx) ir_dereference_variable(var); } else { deref = new(mem_ctx) ir_dereference_variable(var); deref = new(mem_ctx) ir_dereference_array(deref, new(mem_ctx) ir_constant(row)); } return deref; } void ir_mat_op_to_vec_visitor::do_mul_mat_mat(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var) { int b_col, i; ir_assignment *assign; ir_expression *expr; for (b_col = 0; b_col < b_var->type->matrix_columns; b_col++) { ir_rvalue *a = get_column(a_var, 0); ir_rvalue *b = get_element(b_var, b_col, 0); /* first column */ expr = new(mem_ctx) ir_expression(ir_binop_mul, a->type, a, b); /* following columns */ for (i = 1; i < a_var->type->matrix_columns; i++) { ir_expression *mul_expr; a = get_column(a_var, i); b = get_element(b_var, b_col, i); mul_expr = new(mem_ctx) ir_expression(ir_binop_mul, a->type, a, b); expr = new(mem_ctx) ir_expression(ir_binop_add, a->type, expr, mul_expr); } ir_rvalue *result = get_column(result_var, b_col); assign = new(mem_ctx) ir_assignment(result, expr, NULL); base_ir->insert_before(assign); } } void ir_mat_op_to_vec_visitor::do_mul_mat_vec(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var) { int i; ir_rvalue *a = get_column(a_var, 0); ir_rvalue *b = get_element(b_var, 0, 0); ir_assignment *assign; ir_expression *expr; /* first column */ expr = new(mem_ctx) ir_expression(ir_binop_mul, result_var->type, a, b); /* following columns */ for (i = 1; i < a_var->type->matrix_columns; i++) { ir_expression *mul_expr; a = get_column(a_var, i); b = get_element(b_var, 0, i); mul_expr = new(mem_ctx) ir_expression(ir_binop_mul, result_var->type, a, b); expr = new(mem_ctx) ir_expression(ir_binop_add, result_var->type, expr, mul_expr); } ir_rvalue *result = new(mem_ctx) ir_dereference_variable(result_var); assign = new(mem_ctx) ir_assignment(result, expr, NULL); base_ir->insert_before(assign); } void ir_mat_op_to_vec_visitor::do_mul_vec_mat(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var) { int i; for (i = 0; i < b_var->type->matrix_columns; i++) { ir_rvalue *a = new(mem_ctx) ir_dereference_variable(a_var); ir_rvalue *b = get_column(b_var, i); ir_rvalue *result; ir_expression *column_expr; ir_assignment *column_assign; result = new(mem_ctx) ir_dereference_variable(result_var); result = new(mem_ctx) ir_swizzle(result, i, 0, 0, 0, 1); column_expr = new(mem_ctx) ir_expression(ir_binop_dot, result->type, a, b); column_assign = new(mem_ctx) ir_assignment(result, column_expr, NULL); base_ir->insert_before(column_assign); } } void ir_mat_op_to_vec_visitor::do_mul_mat_scalar(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var) { int i; for (i = 0; i < a_var->type->matrix_columns; i++) { ir_rvalue *a = get_column(a_var, i); ir_rvalue *b = new(mem_ctx) ir_dereference_variable(b_var); ir_rvalue *result = get_column(result_var, i); ir_expression *column_expr; ir_assignment *column_assign; column_expr = new(mem_ctx) ir_expression(ir_binop_mul, result->type, a, b); column_assign = new(mem_ctx) ir_assignment(result, column_expr, NULL); base_ir->insert_before(column_assign); } } void ir_mat_op_to_vec_visitor::do_equal_mat_mat(ir_variable *result_var, ir_variable *a_var, ir_variable *b_var, bool test_equal) { /* This essentially implements the following GLSL: * * bool equal(mat4 a, mat4 b) * { * return !any(bvec4(a[0] != b[0], * a[1] != b[1], * a[2] != b[2], * a[3] != b[3]); * } * * bool nequal(mat4 a, mat4 b) * { * return any(bvec4(a[0] != b[0], * a[1] != b[1], * a[2] != b[2], * a[3] != b[3]); * } */ const unsigned columns = a_var->type->matrix_columns; const glsl_type *const bvec_type = glsl_type::get_instance(GLSL_TYPE_BOOL, columns, 1); ir_variable *const tmp_bvec = new(this->mem_ctx) ir_variable(bvec_type, "mat_cmp_bvec", ir_var_temporary); this->base_ir->insert_before(tmp_bvec); for (unsigned i = 0; i < columns; i++) { ir_dereference *const op0 = get_column(a_var, i); ir_dereference *const op1 = get_column(b_var, i); ir_expression *const cmp = new(this->mem_ctx) ir_expression(ir_binop_any_nequal, glsl_type::bool_type, op0, op1); ir_dereference *const lhs = new(this->mem_ctx) ir_dereference_variable(tmp_bvec); ir_assignment *const assign = new(this->mem_ctx) ir_assignment(lhs, cmp, NULL, (1U << i)); this->base_ir->insert_before(assign); } ir_rvalue *const val = new(this->mem_ctx) ir_dereference_variable(tmp_bvec); ir_expression *any = new(this->mem_ctx) ir_expression(ir_unop_any, glsl_type::bool_type, val, NULL); if (test_equal) any = new(this->mem_ctx) ir_expression(ir_unop_logic_not, glsl_type::bool_type, any, NULL); ir_rvalue *const result = new(this->mem_ctx) ir_dereference_variable(result_var); ir_assignment *const assign = new(mem_ctx) ir_assignment(result, any, NULL); base_ir->insert_before(assign); } static bool has_matrix_operand(const ir_expression *expr, unsigned &columns) { for (unsigned i = 0; i < expr->get_num_operands(); i++) { if (expr->operands[i]->type->is_matrix()) { columns = expr->operands[i]->type->matrix_columns; return true; } } return false; } ir_visitor_status ir_mat_op_to_vec_visitor::visit_leave(ir_assignment *orig_assign) { ir_expression *orig_expr = orig_assign->rhs->as_expression(); unsigned int i, matrix_columns = 1; ir_variable *op_var[2]; if (!orig_expr) return visit_continue; if (!has_matrix_operand(orig_expr, matrix_columns)) return visit_continue; assert(orig_expr->get_num_operands() <= 2); mem_ctx = hieralloc_parent(orig_assign); ir_dereference_variable *lhs_deref = orig_assign->lhs->as_dereference_variable(); assert(lhs_deref); ir_variable *result_var = lhs_deref->var; /* Store the expression operands in temps so we can use them * multiple times. */ for (i = 0; i < orig_expr->get_num_operands(); i++) { ir_assignment *assign; op_var[i] = new(mem_ctx) ir_variable(orig_expr->operands[i]->type, "mat_op_to_vec", ir_var_temporary); base_ir->insert_before(op_var[i]); lhs_deref = new(mem_ctx) ir_dereference_variable(op_var[i]); assign = new(mem_ctx) ir_assignment(lhs_deref, orig_expr->operands[i], NULL); base_ir->insert_before(assign); } /* OK, time to break down this matrix operation. */ switch (orig_expr->operation) { case ir_unop_neg: { const unsigned mask = (1U << result_var->type->vector_elements) - 1; /* Apply the operation to each column.*/ for (i = 0; i < matrix_columns; i++) { ir_rvalue *op0 = get_column(op_var[0], i); ir_dereference *result = get_column(result_var, i); ir_expression *column_expr; ir_assignment *column_assign; column_expr = new(mem_ctx) ir_expression(orig_expr->operation, result->type, op0, NULL); column_assign = new(mem_ctx) ir_assignment(result, column_expr, NULL, mask); assert(column_assign->write_mask != 0); base_ir->insert_before(column_assign); } break; } case ir_binop_add: case ir_binop_sub: case ir_binop_div: case ir_binop_mod: { const unsigned mask = (1U << result_var->type->vector_elements) - 1; /* For most operations, the matrix version is just going * column-wise through and applying the operation to each column * if available. */ for (i = 0; i < matrix_columns; i++) { ir_rvalue *op0 = get_column(op_var[0], i); ir_rvalue *op1 = get_column(op_var[1], i); ir_dereference *result = get_column(result_var, i); ir_expression *column_expr; ir_assignment *column_assign; column_expr = new(mem_ctx) ir_expression(orig_expr->operation, result->type, op0, op1); column_assign = new(mem_ctx) ir_assignment(result, column_expr, NULL, mask); assert(column_assign->write_mask != 0); base_ir->insert_before(column_assign); } break; } case ir_binop_mul: if (op_var[0]->type->is_matrix()) { if (op_var[1]->type->is_matrix()) { do_mul_mat_mat(result_var, op_var[0], op_var[1]); } else if (op_var[1]->type->is_vector()) { do_mul_mat_vec(result_var, op_var[0], op_var[1]); } else { assert(op_var[1]->type->is_scalar()); do_mul_mat_scalar(result_var, op_var[0], op_var[1]); } } else { assert(op_var[1]->type->is_matrix()); if (op_var[0]->type->is_vector()) { do_mul_vec_mat(result_var, op_var[0], op_var[1]); } else { assert(op_var[0]->type->is_scalar()); do_mul_mat_scalar(result_var, op_var[1], op_var[0]); } } break; case ir_binop_all_equal: case ir_binop_any_nequal: do_equal_mat_mat(result_var, op_var[1], op_var[0], (orig_expr->operation == ir_binop_all_equal)); break; default: printf("FINISHME: Handle matrix operation for %s\n", orig_expr->operator_string()); abort(); } orig_assign->remove(); this->made_progress = true; return visit_continue; }