From 11d6f1c69871d0b7edc28f639256460839fccd2d Mon Sep 17 00:00:00 2001 From: Ian Romanick Date: Tue, 16 Nov 2010 12:01:42 -0800 Subject: [PATCH] glsl: Add ir_quadop_vector expression The vector operator collects 2, 3, or 4 scalar components into a vector. Doing this has several advantages. First, it will make ud-chain tracking for components of vectors much easier. Second, a later optimization pass could collect scalars into vectors to allow generation of SWZ instructions (or similar as operands to other instructions on R200 and i915). It also enables an easy way to generate IR for SWZ instructions in the ARB_vertex_program assembler. --- src/glsl/Makefile | 1 + src/glsl/ir.cpp | 25 ++- src/glsl/ir.h | 14 +- src/glsl/ir_clone.cpp | 3 +- src/glsl/ir_constant_expression.cpp | 18 +++ src/glsl/ir_optimization.h | 1 + src/glsl/ir_validate.cpp | 45 ++++++ src/glsl/lower_vector.cpp | 224 +++++++++++++++++++++++++++ src/glsl/opt_algebraic.cpp | 2 +- src/mesa/drivers/dri/i965/brw_fs.cpp | 1 + src/mesa/program/ir_to_mesa.cpp | 132 ++++++++++++++++ 11 files changed, 460 insertions(+), 6 deletions(-) create mode 100644 src/glsl/lower_vector.cpp diff --git a/src/glsl/Makefile b/src/glsl/Makefile index 62984f81c0a..ea4d6be5a41 100644 --- a/src/glsl/Makefile +++ b/src/glsl/Makefile @@ -64,6 +64,7 @@ CXX_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 \ diff --git a/src/glsl/ir.cpp b/src/glsl/ir.cpp index 1f5e2ebdcb7..741e3cb1775 100644 --- a/src/glsl/ir.cpp +++ b/src/glsl/ir.cpp @@ -200,18 +200,35 @@ ir_expression::ir_expression(int op, const struct glsl_type *type, this->operation = ir_expression_operation(op); this->operands[0] = op0; this->operands[1] = NULL; + this->operands[2] = NULL; + this->operands[3] = NULL; } ir_expression::ir_expression(int op, const struct glsl_type *type, ir_rvalue *op0, ir_rvalue *op1) { - assert((op1 == NULL) && (get_num_operands(ir_expression_operation(op)) == 1) + assert(((op1 == NULL) && (get_num_operands(ir_expression_operation(op)) == 1)) || (get_num_operands(ir_expression_operation(op)) == 2)); this->ir_type = ir_type_expression; this->type = type; this->operation = ir_expression_operation(op); this->operands[0] = op0; this->operands[1] = op1; + this->operands[2] = NULL; + this->operands[3] = NULL; +} + +ir_expression::ir_expression(int op, const struct glsl_type *type, + ir_rvalue *op0, ir_rvalue *op1, + ir_rvalue *op2, ir_rvalue *op3) +{ + this->ir_type = ir_type_expression; + this->type = type; + this->operation = ir_expression_operation(op); + this->operands[0] = op0; + this->operands[1] = op1; + this->operands[2] = op2; + this->operands[3] = op3; } unsigned int @@ -225,6 +242,9 @@ ir_expression::get_num_operands(ir_expression_operation op) if (op <= ir_last_binop) return 2; + if (op == ir_quadop_vector) + return 4; + assert(false); return 0; } @@ -287,12 +307,13 @@ static const char *const operator_strs[] = { "min", "max", "pow", + "vector", }; const char *ir_expression::operator_string(ir_expression_operation op) { assert((unsigned int) op < Elements(operator_strs)); - assert(Elements(operator_strs) == (ir_binop_pow + 1)); + assert(Elements(operator_strs) == (ir_quadop_vector + 1)); return operator_strs[op]; } diff --git a/src/glsl/ir.h b/src/glsl/ir.h index 99fdaa3b09b..be0da07b3b6 100644 --- a/src/glsl/ir.h +++ b/src/glsl/ir.h @@ -33,6 +33,7 @@ extern "C" { #include } +#include "glsl_types.h" #include "list.h" #include "ir_visitor.h" #include "ir_hierarchical_visitor.h" @@ -824,6 +825,8 @@ enum ir_expression_operation { */ ir_last_binop = ir_binop_pow, + ir_quadop_vector, + /** * A sentinel marking the last of all operations. */ @@ -843,6 +846,12 @@ public: ir_expression(int op, const struct glsl_type *type, ir_rvalue *, ir_rvalue *); + /** + * Constructor for quad operator expressions + */ + ir_expression(int op, const struct glsl_type *type, + ir_rvalue *, ir_rvalue *, ir_rvalue *, ir_rvalue *); + virtual ir_expression *as_expression() { return this; @@ -868,7 +877,8 @@ public: */ unsigned int get_num_operands() const { - return get_num_operands(operation); + return (this->operation == ir_quadop_vector) + ? this->type->vector_elements : get_num_operands(operation); } /** @@ -895,7 +905,7 @@ public: virtual ir_visitor_status accept(ir_hierarchical_visitor *); ir_expression_operation operation; - ir_rvalue *operands[2]; + ir_rvalue *operands[4]; }; diff --git a/src/glsl/ir_clone.cpp b/src/glsl/ir_clone.cpp index 4032647c3a2..325f6066154 100644 --- a/src/glsl/ir_clone.cpp +++ b/src/glsl/ir_clone.cpp @@ -168,7 +168,8 @@ ir_expression::clone(void *mem_ctx, struct hash_table *ht) const op[i] = this->operands[i]->clone(mem_ctx, ht); } - return new(mem_ctx) ir_expression(this->operation, this->type, op[0], op[1]); + return new(mem_ctx) ir_expression(this->operation, this->type, + op[0], op[1], op[2], op[3]); } ir_dereference_variable * diff --git a/src/glsl/ir_constant_expression.cpp b/src/glsl/ir_constant_expression.cpp index 1fe15050470..4fd6d09a3af 100644 --- a/src/glsl/ir_constant_expression.cpp +++ b/src/glsl/ir_constant_expression.cpp @@ -788,6 +788,24 @@ ir_expression::constant_expression_value() } break; + case ir_quadop_vector: + for (unsigned c = 0; c < this->type->vector_elements; c++) { + switch (this->type->base_type) { + case GLSL_TYPE_INT: + data.i[c] = op[c]->value.i[0]; + break; + case GLSL_TYPE_UINT: + data.u[c] = op[c]->value.u[0]; + break; + case GLSL_TYPE_FLOAT: + data.f[c] = op[c]->value.f[0]; + break; + default: + assert(0); + } + } + break; + default: /* FINISHME: Should handle all expression types. */ return NULL; diff --git a/src/glsl/ir_optimization.h b/src/glsl/ir_optimization.h index ffdc66b9f75..1f8da16bcb2 100644 --- a/src/glsl/ir_optimization.h +++ b/src/glsl/ir_optimization.h @@ -59,4 +59,5 @@ bool do_vec_index_to_swizzle(exec_list *instructions); 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); +bool lower_quadop_vector(exec_list *instructions, bool dont_lower_swz); bool optimize_redundant_jumps(exec_list *instructions); diff --git a/src/glsl/ir_validate.cpp b/src/glsl/ir_validate.cpp index 2a066c1a277..5b055f64d38 100644 --- a/src/glsl/ir_validate.cpp +++ b/src/glsl/ir_validate.cpp @@ -374,6 +374,51 @@ ir_validate::visit_leave(ir_expression *ir) assert(ir->operands[0]->type->is_vector()); assert(ir->operands[0]->type == ir->operands[1]->type); break; + + case ir_quadop_vector: + /* The vector operator collects some number of scalars and generates a + * vector from them. + * + * - All of the operands must be scalar. + * - Number of operands must matche the size of the resulting vector. + * - Base type of the operands must match the base type of the result. + */ + assert(ir->type->is_vector()); + switch (ir->type->vector_elements) { + case 2: + assert(ir->operands[0]->type->is_scalar()); + assert(ir->operands[0]->type->base_type == ir->type->base_type); + assert(ir->operands[1]->type->is_scalar()); + assert(ir->operands[1]->type->base_type == ir->type->base_type); + assert(ir->operands[2] == NULL); + assert(ir->operands[3] == NULL); + break; + case 3: + assert(ir->operands[0]->type->is_scalar()); + assert(ir->operands[0]->type->base_type == ir->type->base_type); + assert(ir->operands[1]->type->is_scalar()); + assert(ir->operands[1]->type->base_type == ir->type->base_type); + assert(ir->operands[2]->type->is_scalar()); + assert(ir->operands[2]->type->base_type == ir->type->base_type); + assert(ir->operands[3] == NULL); + break; + case 4: + assert(ir->operands[0]->type->is_scalar()); + assert(ir->operands[0]->type->base_type == ir->type->base_type); + assert(ir->operands[1]->type->is_scalar()); + assert(ir->operands[1]->type->base_type == ir->type->base_type); + assert(ir->operands[2]->type->is_scalar()); + assert(ir->operands[2]->type->base_type == ir->type->base_type); + assert(ir->operands[3]->type->is_scalar()); + assert(ir->operands[3]->type->base_type == ir->type->base_type); + break; + default: + /* The is_vector assertion above should prevent execution from ever + * getting here. + */ + assert(!"Should not get here."); + break; + } } return visit_continue; diff --git a/src/glsl/lower_vector.cpp b/src/glsl/lower_vector.cpp new file mode 100644 index 00000000000..ae501201004 --- /dev/null +++ b/src/glsl/lower_vector.cpp @@ -0,0 +1,224 @@ +/* + * 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_vector.cpp + * IR lowering pass to remove some types of ir_quadop_vector + * + * \author Ian Romanick + */ + +#include "ir.h" +#include "ir_rvalue_visitor.h" + +class lower_vector_visitor : public ir_rvalue_visitor { +public: + lower_vector_visitor() : progress(false) + { + /* empty */ + } + + void handle_rvalue(ir_rvalue **rvalue); + + /** + * Should SWZ-like expressions be lowered? + */ + bool dont_lower_swz; + + bool progress; +}; + +/** + * Determine if an IR expression tree looks like an extended swizzle + * + * Extended swizzles consist of access of a single vector source (with possible + * per component negation) and the constants -1, 0, or 1. + */ +bool +is_extended_swizzle(ir_expression *ir) +{ + /* Track any variables that are accessed by this expression. + */ + ir_variable *var = NULL; + + assert(ir->operation == ir_quadop_vector); + + for (unsigned i = 0; i < ir->type->vector_elements; i++) { + ir_rvalue *op = ir->operands[i]; + + while (op != NULL) { + switch (op->ir_type) { + case ir_type_constant: { + const ir_constant *const c = op->as_constant(); + + if (!c->is_one() && !c->is_zero() && !c->is_negative_one()) + return false; + + op = NULL; + break; + } + + case ir_type_dereference_variable: { + ir_dereference_variable *const d = (ir_dereference_variable *) op; + + if ((var != NULL) && (var != d->var)) + return false; + + var = d->var; + op = NULL; + break; + } + + case ir_type_expression: { + ir_expression *const ex = (ir_expression *) op; + + if (ex->operation != ir_unop_neg) + return false; + + op = ex->operands[0]; + break; + } + + case ir_type_swizzle: + op = ((ir_swizzle *) op)->val; + break; + + default: + return false; + } + } + } + + return true; +} + +void +lower_vector_visitor::handle_rvalue(ir_rvalue **rvalue) +{ + if (!*rvalue) + return; + + ir_expression *expr = (*rvalue)->as_expression(); + if ((expr == NULL) || (expr->operation != ir_quadop_vector)) + return; + + if (this->dont_lower_swz && is_extended_swizzle(expr)) + return; + + /* FINISHME: Is this the right thing to use for the talloc context? + */ + void *const mem_ctx = expr; + + assert(expr->type->vector_elements == expr->get_num_operands()); + + /* Generate a temporary with the same type as the ir_quadop_operation. + */ + ir_variable *const temp = + new(mem_ctx) ir_variable(expr->type, "vecop_tmp", ir_var_temporary); + + this->base_ir->insert_before(temp); + + /* Counter of the number of components collected so far. + */ + unsigned assigned; + + /* Write-mask in the destination that receives counted by 'assigned'. + */ + unsigned write_mask; + + + /* Generate upto four assignments to that variable. Try to group component + * assignments together: + * + * - All constant components can be assigned at once. + * - All assigments of components from a single variable with the same + * unary operator can be assigned at once. + */ + ir_constant_data d = { { 0 } }; + + assigned = 0; + write_mask = 0; + for (unsigned i = 0; i < expr->type->vector_elements; i++) { + const ir_constant *const c = expr->operands[i]->as_constant(); + + if (c == NULL) + continue; + + switch (expr->type->base_type) { + case GLSL_TYPE_UINT: d.u[assigned] = c->value.u[0]; break; + case GLSL_TYPE_INT: d.i[assigned] = c->value.i[0]; break; + case GLSL_TYPE_FLOAT: d.f[assigned] = c->value.f[0]; break; + case GLSL_TYPE_BOOL: d.b[assigned] = c->value.b[0]; break; + defatul: assert(!"Should not get here."); break; + } + + write_mask |= (1U << i); + assigned++; + } + + assert((write_mask == 0) == (assigned == 0)); + + /* If there were constant values, generate an assignment. + */ + if (assigned > 0) { + ir_constant *const c = + new(mem_ctx) ir_constant(glsl_type::get_instance(expr->type->base_type, + assigned, 0), + &d); + ir_dereference *const lhs = new(mem_ctx) ir_dereference_variable(temp); + ir_assignment *const assign = + new(mem_ctx) ir_assignment(lhs, c, NULL, write_mask); + + this->base_ir->insert_before(assign); + } + + /* FINISHME: This should try to coalesce assignments. + */ + for (unsigned i = 0; i < expr->type->vector_elements; i++) { + if (expr->operands[i]->ir_type == ir_type_constant) + continue; + + ir_dereference *const lhs = new(mem_ctx) ir_dereference_variable(temp); + ir_assignment *const assign = + new(mem_ctx) ir_assignment(lhs, expr->operands[i], NULL, (1U << i)); + + this->base_ir->insert_before(assign); + assigned++; + } + + assert(assigned == expr->type->vector_elements); + + *rvalue = new(mem_ctx) ir_dereference_variable(temp); + this->progress = true; +} + +bool +lower_quadop_vector(exec_list *instructions, bool dont_lower_swz) +{ + lower_vector_visitor v; + + v.dont_lower_swz = dont_lower_swz; + visit_list_elements(&v, instructions); + + return v.progress; +} diff --git a/src/glsl/opt_algebraic.cpp b/src/glsl/opt_algebraic.cpp index 9a8080bff3d..3c9af85f312 100644 --- a/src/glsl/opt_algebraic.cpp +++ b/src/glsl/opt_algebraic.cpp @@ -394,7 +394,7 @@ ir_algebraic_visitor::handle_rvalue(ir_rvalue **rvalue) return; ir_expression *expr = (*rvalue)->as_expression(); - if (!expr) + if (!expr || expr->operation == ir_quadop_vector) return; *rvalue = handle_expression(expr); diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp index 105327c7fe7..e1cb94452d4 100644 --- a/src/mesa/drivers/dri/i965/brw_fs.cpp +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp @@ -130,6 +130,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) GL_TRUE, /* temp */ GL_TRUE /* uniform */ ) || progress; + progress = lower_quadop_vector(shader->ir, false) || progress; } while (progress); validate_ir_tree(shader->ir); diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp index 1b5337e92ee..1cb81830426 100644 --- a/src/mesa/program/ir_to_mesa.cpp +++ b/src/mesa/program/ir_to_mesa.cpp @@ -289,6 +289,8 @@ public: GLboolean try_emit_mad(ir_expression *ir, int mul_operand); + void emit_swz(ir_expression *ir); + bool process_move_condition(ir_rvalue *ir); void *mem_ctx; @@ -957,6 +959,123 @@ ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir, (*num_reladdr)--; } +void +ir_to_mesa_visitor::emit_swz(ir_expression *ir) +{ + /* Assume that the vector operator is in a form compatible with OPCODE_SWZ. + * This means that each of the operands is either an immediate value of -1, + * 0, or 1, or is a component from one source register (possibly with + * negation). + */ + uint8_t components[4] = { 0 }; + bool negate[4] = { false }; + ir_variable *var = NULL; + + for (unsigned i = 0; i < ir->type->vector_elements; i++) { + ir_rvalue *op = ir->operands[i]; + + assert(op->type->is_scalar()); + + while (op != NULL) { + switch (op->ir_type) { + case ir_type_constant: { + + assert(op->type->is_scalar()); + + const ir_constant *const c = op->as_constant(); + if (c->is_one()) { + components[i] = SWIZZLE_ONE; + } else if (c->is_zero()) { + components[i] = SWIZZLE_ZERO; + } else if (c->is_negative_one()) { + components[i] = SWIZZLE_ONE; + negate[i] = true; + } else { + assert(!"SWZ constant must be 0.0 or 1.0."); + } + + op = NULL; + break; + } + + case ir_type_dereference_variable: { + ir_dereference_variable *const deref = + (ir_dereference_variable *) op; + + assert((var == NULL) || (deref->var == var)); + components[i] = SWIZZLE_X; + var = deref->var; + op = NULL; + break; + } + + case ir_type_expression: { + ir_expression *const expr = (ir_expression *) op; + + assert(expr->operation == ir_unop_neg); + negate[i] = true; + + op = expr->operands[0]; + break; + } + + case ir_type_swizzle: { + ir_swizzle *const swiz = (ir_swizzle *) op; + + components[i] = swiz->mask.x; + op = swiz->val; + break; + } + + default: + assert(!"Should not get here."); + return; + } + } + } + + assert(var != NULL); + + ir_dereference_variable *const deref = + new(mem_ctx) ir_dereference_variable(var); + + this->result.file = PROGRAM_UNDEFINED; + deref->accept(this); + if (this->result.file == PROGRAM_UNDEFINED) { + ir_print_visitor v; + printf("Failed to get tree for expression operand:\n"); + deref->accept(&v); + exit(1); + } + + ir_to_mesa_src_reg src; + + src = this->result; + src.swizzle = MAKE_SWIZZLE4(components[0], + components[1], + components[2], + components[3]); + src.negate = ((unsigned(negate[0]) << 0) + | (unsigned(negate[1]) << 1) + | (unsigned(negate[2]) << 2) + | (unsigned(negate[3]) << 3)); + + /* Storage for our result. Ideally for an assignment we'd be using the + * actual storage for the result here, instead. + */ + const ir_to_mesa_src_reg result_src = get_temp(ir->type); + ir_to_mesa_dst_reg result_dst = ir_to_mesa_dst_reg_from_src(result_src); + + /* Limit writes to the channels that will be used by result_src later. + * This does limit this temp's use as a temporary for multi-instruction + * sequences. + */ + result_dst.writemask = (1 << ir->type->vector_elements) - 1; + + ir_to_mesa_emit_op1(ir, OPCODE_SWZ, result_dst, src); + this->result = result_src; +} + void ir_to_mesa_visitor::visit(ir_expression *ir) { @@ -974,6 +1093,11 @@ ir_to_mesa_visitor::visit(ir_expression *ir) return; } + if (ir->operation == ir_quadop_vector) { + this->emit_swz(ir); + return; + } + for (operand = 0; operand < ir->get_num_operands(); operand++) { this->result.file = PROGRAM_UNDEFINED; ir->operands[operand]->accept(this); @@ -1231,6 +1355,12 @@ ir_to_mesa_visitor::visit(ir_expression *ir) case ir_unop_round_even: assert(!"GLSL 1.30 features unsupported"); break; + + case ir_quadop_vector: + /* This operation should have already been handled. + */ + assert(!"Should not get here."); + break; } this->result = result_src; @@ -2676,6 +2806,8 @@ _mesa_ir_link_shader(struct gl_context *ctx, struct gl_shader_program *prog) progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress; + progress = lower_quadop_vector(ir, true) || progress; + if (options->EmitNoIfs) progress = do_if_to_cond_assign(ir) || progress; -- 2.30.2