From 1e1e2c35b405bbb537a6bbe35d9f0831111e272f Mon Sep 17 00:00:00 2001 From: marha Date: Wed, 26 Jan 2011 07:24:15 +0000 Subject: mesa xkeyboard-config git update 26 jan 2011 --- mesalib/src/glsl/ir_reader.cpp | 1995 ++++++++++----------- mesalib/src/glsl/link_functions.cpp | 528 +++--- mesalib/src/glsl/linker.cpp | 3348 +++++++++++++++++------------------ 3 files changed, 2941 insertions(+), 2930 deletions(-) (limited to 'mesalib/src/glsl') diff --git a/mesalib/src/glsl/ir_reader.cpp b/mesalib/src/glsl/ir_reader.cpp index 5b1eccb46..b78aa051b 100644 --- a/mesalib/src/glsl/ir_reader.cpp +++ b/mesalib/src/glsl/ir_reader.cpp @@ -1,996 +1,999 @@ -/* - * 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. - */ - -#include "ir_reader.h" -#include "glsl_parser_extras.h" -#include "glsl_types.h" -#include "s_expression.h" - -const static bool debug = false; - -class ir_reader { -public: - ir_reader(_mesa_glsl_parse_state *); - - void read(exec_list *instructions, const char *src, bool scan_for_protos); - -private: - void *mem_ctx; - _mesa_glsl_parse_state *state; - - void ir_read_error(s_expression *, const char *fmt, ...); - - const glsl_type *read_type(s_expression *); - - void scan_for_prototypes(exec_list *, s_expression *); - ir_function *read_function(s_expression *, bool skip_body); - void read_function_sig(ir_function *, s_expression *, bool skip_body); - - void read_instructions(exec_list *, s_expression *, ir_loop *); - ir_instruction *read_instruction(s_expression *, ir_loop *); - ir_variable *read_declaration(s_expression *); - ir_if *read_if(s_expression *, ir_loop *); - ir_loop *read_loop(s_expression *); - ir_return *read_return(s_expression *); - ir_rvalue *read_rvalue(s_expression *); - ir_assignment *read_assignment(s_expression *); - ir_expression *read_expression(s_expression *); - ir_call *read_call(s_expression *); - ir_swizzle *read_swizzle(s_expression *); - ir_constant *read_constant(s_expression *); - ir_texture *read_texture(s_expression *); - - ir_dereference *read_dereference(s_expression *); -}; - -ir_reader::ir_reader(_mesa_glsl_parse_state *state) : state(state) -{ - this->mem_ctx = state; -} - -void -_mesa_glsl_read_ir(_mesa_glsl_parse_state *state, exec_list *instructions, - const char *src, bool scan_for_protos) -{ - ir_reader r(state); - r.read(instructions, src, scan_for_protos); -} - -void -ir_reader::read(exec_list *instructions, const char *src, bool scan_for_protos) -{ - s_expression *expr = s_expression::read_expression(mem_ctx, src); - if (expr == NULL) { - ir_read_error(NULL, "couldn't parse S-Expression."); - return; - } - - if (scan_for_protos) { - scan_for_prototypes(instructions, expr); - if (state->error) - return; - } - - read_instructions(instructions, expr, NULL); - talloc_free(expr); - - if (debug) - validate_ir_tree(instructions); -} - -void -ir_reader::ir_read_error(s_expression *expr, const char *fmt, ...) -{ - va_list ap; - - state->error = true; - - if (state->current_function != NULL) - state->info_log = talloc_asprintf_append(state->info_log, - "In function %s:\n", - state->current_function->function_name()); - state->info_log = talloc_strdup_append(state->info_log, "error: "); - - va_start(ap, fmt); - state->info_log = talloc_vasprintf_append(state->info_log, fmt, ap); - va_end(ap); - state->info_log = talloc_strdup_append(state->info_log, "\n"); - - if (expr != NULL) { - state->info_log = talloc_strdup_append(state->info_log, - "...in this context:\n "); - expr->print(); - state->info_log = talloc_strdup_append(state->info_log, "\n\n"); - } -} - -const glsl_type * -ir_reader::read_type(s_expression *expr) -{ - s_expression *s_base_type; - s_int *s_size; - - s_pattern pat[] = { "array", s_base_type, s_size }; - if (MATCH(expr, pat)) { - const glsl_type *base_type = read_type(s_base_type); - if (base_type == NULL) { - ir_read_error(NULL, "when reading base type of array type"); - return NULL; - } - - return glsl_type::get_array_instance(base_type, s_size->value()); - } - - s_symbol *type_sym = SX_AS_SYMBOL(expr); - if (type_sym == NULL) { - ir_read_error(expr, "expected "); - return NULL; - } - - const glsl_type *type = state->symbols->get_type(type_sym->value()); - if (type == NULL) - ir_read_error(expr, "invalid type: %s", type_sym->value()); - - return type; -} - - -void -ir_reader::scan_for_prototypes(exec_list *instructions, s_expression *expr) -{ - s_list *list = SX_AS_LIST(expr); - if (list == NULL) { - ir_read_error(expr, "Expected ( ...); found an atom."); - return; - } - - foreach_iter(exec_list_iterator, it, list->subexpressions) { - s_list *sub = SX_AS_LIST(it.get()); - if (sub == NULL) - continue; // not a (function ...); ignore it. - - s_symbol *tag = SX_AS_SYMBOL(sub->subexpressions.get_head()); - if (tag == NULL || strcmp(tag->value(), "function") != 0) - continue; // not a (function ...); ignore it. - - ir_function *f = read_function(sub, true); - if (f == NULL) - return; - instructions->push_tail(f); - } -} - -ir_function * -ir_reader::read_function(s_expression *expr, bool skip_body) -{ - bool added = false; - s_symbol *name; - - s_pattern pat[] = { "function", name }; - if (!PARTIAL_MATCH(expr, pat)) { - ir_read_error(expr, "Expected (function (signature ...) ...)"); - return NULL; - } - - ir_function *f = state->symbols->get_function(name->value()); - if (f == NULL) { - f = new(mem_ctx) ir_function(name->value()); - added = state->symbols->add_function(f); - assert(added); - } - - exec_list_iterator it = ((s_list *) expr)->subexpressions.iterator(); - it.next(); // skip "function" tag - it.next(); // skip function name - for (/* nothing */; it.has_next(); it.next()) { - s_expression *s_sig = (s_expression *) it.get(); - read_function_sig(f, s_sig, skip_body); - } - return added ? f : NULL; -} - -void -ir_reader::read_function_sig(ir_function *f, s_expression *expr, bool skip_body) -{ - s_expression *type_expr; - s_list *paramlist; - s_list *body_list; - - s_pattern pat[] = { "signature", type_expr, paramlist, body_list }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "Expected (signature (parameters ...) " - "( ...))"); - return; - } - - const glsl_type *return_type = read_type(type_expr); - if (return_type == NULL) - return; - - s_symbol *paramtag = SX_AS_SYMBOL(paramlist->subexpressions.get_head()); - if (paramtag == NULL || strcmp(paramtag->value(), "parameters") != 0) { - ir_read_error(paramlist, "Expected (parameters ...)"); - return; - } - - // Read the parameters list into a temporary place. - exec_list hir_parameters; - state->symbols->push_scope(); - - exec_list_iterator it = paramlist->subexpressions.iterator(); - for (it.next() /* skip "parameters" */; it.has_next(); it.next()) { - ir_variable *var = read_declaration((s_expression *) it.get()); - if (var == NULL) - return; - - hir_parameters.push_tail(var); - } - - ir_function_signature *sig = f->exact_matching_signature(&hir_parameters); - if (sig == NULL && skip_body) { - /* If scanning for prototypes, generate a new signature. */ - sig = new(mem_ctx) ir_function_signature(return_type); - sig->is_builtin = true; - f->add_signature(sig); - } else if (sig != NULL) { - const char *badvar = sig->qualifiers_match(&hir_parameters); - if (badvar != NULL) { - ir_read_error(expr, "function `%s' parameter `%s' qualifiers " - "don't match prototype", f->name, badvar); - return; - } - - if (sig->return_type != return_type) { - ir_read_error(expr, "function `%s' return type doesn't " - "match prototype", f->name); - return; - } - } else { - /* No prototype for this body exists - skip it. */ - state->symbols->pop_scope(); - return; - } - assert(sig != NULL); - - sig->replace_parameters(&hir_parameters); - - if (!skip_body && !body_list->subexpressions.is_empty()) { - if (sig->is_defined) { - ir_read_error(expr, "function %s redefined", f->name); - return; - } - state->current_function = sig; - read_instructions(&sig->body, body_list, NULL); - state->current_function = NULL; - sig->is_defined = true; - } - - state->symbols->pop_scope(); -} - -void -ir_reader::read_instructions(exec_list *instructions, s_expression *expr, - ir_loop *loop_ctx) -{ - // Read in a list of instructions - s_list *list = SX_AS_LIST(expr); - if (list == NULL) { - ir_read_error(expr, "Expected ( ...); found an atom."); - return; - } - - foreach_iter(exec_list_iterator, it, list->subexpressions) { - s_expression *sub = (s_expression*) it.get(); - ir_instruction *ir = read_instruction(sub, loop_ctx); - if (ir != NULL) { - /* Global variable declarations should be moved to the top, before - * any functions that might use them. Functions are added to the - * instruction stream when scanning for prototypes, so without this - * hack, they always appear before variable declarations. - */ - if (state->current_function == NULL && ir->as_variable() != NULL) - instructions->push_head(ir); - else - instructions->push_tail(ir); - } - } -} - - -ir_instruction * -ir_reader::read_instruction(s_expression *expr, ir_loop *loop_ctx) -{ - s_symbol *symbol = SX_AS_SYMBOL(expr); - if (symbol != NULL) { - if (strcmp(symbol->value(), "break") == 0 && loop_ctx != NULL) - return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_break); - if (strcmp(symbol->value(), "continue") == 0 && loop_ctx != NULL) - return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_continue); - } - - s_list *list = SX_AS_LIST(expr); - if (list == NULL || list->subexpressions.is_empty()) { - ir_read_error(expr, "Invalid instruction.\n"); - return NULL; - } - - s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); - if (tag == NULL) { - ir_read_error(expr, "expected instruction tag"); - return NULL; - } - - ir_instruction *inst = NULL; - if (strcmp(tag->value(), "declare") == 0) { - inst = read_declaration(list); - } else if (strcmp(tag->value(), "assign") == 0) { - inst = read_assignment(list); - } else if (strcmp(tag->value(), "if") == 0) { - inst = read_if(list, loop_ctx); - } else if (strcmp(tag->value(), "loop") == 0) { - inst = read_loop(list); - } else if (strcmp(tag->value(), "return") == 0) { - inst = read_return(list); - } else if (strcmp(tag->value(), "function") == 0) { - inst = read_function(list, false); - } else { - inst = read_rvalue(list); - if (inst == NULL) - ir_read_error(NULL, "when reading instruction"); - } - return inst; -} - -ir_variable * -ir_reader::read_declaration(s_expression *expr) -{ - s_list *s_quals; - s_expression *s_type; - s_symbol *s_name; - - s_pattern pat[] = { "declare", s_quals, s_type, s_name }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (declare () )"); - return NULL; - } - - const glsl_type *type = read_type(s_type); - if (type == NULL) - return NULL; - - ir_variable *var = new(mem_ctx) ir_variable(type, s_name->value(), - ir_var_auto); - - foreach_iter(exec_list_iterator, it, s_quals->subexpressions) { - s_symbol *qualifier = SX_AS_SYMBOL(it.get()); - if (qualifier == NULL) { - ir_read_error(expr, "qualifier list must contain only symbols"); - return NULL; - } - - // FINISHME: Check for duplicate/conflicting qualifiers. - if (strcmp(qualifier->value(), "centroid") == 0) { - var->centroid = 1; - } else if (strcmp(qualifier->value(), "invariant") == 0) { - var->invariant = 1; - } else if (strcmp(qualifier->value(), "uniform") == 0) { - var->mode = ir_var_uniform; - } else if (strcmp(qualifier->value(), "auto") == 0) { - var->mode = ir_var_auto; - } else if (strcmp(qualifier->value(), "in") == 0) { - var->mode = ir_var_in; - } else if (strcmp(qualifier->value(), "out") == 0) { - var->mode = ir_var_out; - } else if (strcmp(qualifier->value(), "inout") == 0) { - var->mode = ir_var_inout; - } else if (strcmp(qualifier->value(), "smooth") == 0) { - var->interpolation = ir_var_smooth; - } else if (strcmp(qualifier->value(), "flat") == 0) { - var->interpolation = ir_var_flat; - } else if (strcmp(qualifier->value(), "noperspective") == 0) { - var->interpolation = ir_var_noperspective; - } else { - ir_read_error(expr, "unknown qualifier: %s", qualifier->value()); - return NULL; - } - } - - // Add the variable to the symbol table - state->symbols->add_variable(var); - - return var; -} - - -ir_if * -ir_reader::read_if(s_expression *expr, ir_loop *loop_ctx) -{ - s_expression *s_cond; - s_expression *s_then; - s_expression *s_else; - - s_pattern pat[] = { "if", s_cond, s_then, s_else }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (if (...) (...))"); - return NULL; - } - - ir_rvalue *condition = read_rvalue(s_cond); - if (condition == NULL) { - ir_read_error(NULL, "when reading condition of (if ...)"); - return NULL; - } - - ir_if *iff = new(mem_ctx) ir_if(condition); - - read_instructions(&iff->then_instructions, s_then, loop_ctx); - read_instructions(&iff->else_instructions, s_else, loop_ctx); - if (state->error) { - delete iff; - iff = NULL; - } - return iff; -} - - -ir_loop * -ir_reader::read_loop(s_expression *expr) -{ - s_expression *s_counter, *s_from, *s_to, *s_inc, *s_body; - - s_pattern pat[] = { "loop", s_counter, s_from, s_to, s_inc, s_body }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (loop " - " )"); - return NULL; - } - - // FINISHME: actually read the count/from/to fields. - - ir_loop *loop = new(mem_ctx) ir_loop; - read_instructions(&loop->body_instructions, s_body, loop); - if (state->error) { - delete loop; - loop = NULL; - } - return loop; -} - - -ir_return * -ir_reader::read_return(s_expression *expr) -{ - s_expression *s_retval; - - s_pattern pat[] = { "return", s_retval}; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (return )"); - return NULL; - } - - ir_rvalue *retval = read_rvalue(s_retval); - if (retval == NULL) { - ir_read_error(NULL, "when reading return value"); - return NULL; - } - - return new(mem_ctx) ir_return(retval); -} - - -ir_rvalue * -ir_reader::read_rvalue(s_expression *expr) -{ - s_list *list = SX_AS_LIST(expr); - if (list == NULL || list->subexpressions.is_empty()) - return NULL; - - s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); - if (tag == NULL) { - ir_read_error(expr, "expected rvalue tag"); - return NULL; - } - - ir_rvalue *rvalue = read_dereference(list); - if (rvalue != NULL || state->error) - return rvalue; - else if (strcmp(tag->value(), "swiz") == 0) { - rvalue = read_swizzle(list); - } else if (strcmp(tag->value(), "expression") == 0) { - rvalue = read_expression(list); - } else if (strcmp(tag->value(), "call") == 0) { - rvalue = read_call(list); - } else if (strcmp(tag->value(), "constant") == 0) { - rvalue = read_constant(list); - } else { - rvalue = read_texture(list); - if (rvalue == NULL && !state->error) - ir_read_error(expr, "unrecognized rvalue tag: %s", tag->value()); - } - - return rvalue; -} - -ir_assignment * -ir_reader::read_assignment(s_expression *expr) -{ - s_expression *cond_expr = NULL; - s_expression *lhs_expr, *rhs_expr; - s_list *mask_list; - - s_pattern pat4[] = { "assign", mask_list, lhs_expr, rhs_expr }; - s_pattern pat5[] = { "assign", cond_expr, mask_list, lhs_expr, rhs_expr }; - if (!MATCH(expr, pat4) && !MATCH(expr, pat5)) { - ir_read_error(expr, "expected (assign [] () " - " )"); - return NULL; - } - - ir_rvalue *condition = NULL; - if (cond_expr != NULL) { - condition = read_rvalue(cond_expr); - if (condition == NULL) { - ir_read_error(NULL, "when reading condition of assignment"); - return NULL; - } - } - - unsigned mask = 0; - - s_symbol *mask_symbol; - s_pattern mask_pat[] = { mask_symbol }; - if (MATCH(mask_list, mask_pat)) { - const char *mask_str = mask_symbol->value(); - unsigned mask_length = strlen(mask_str); - if (mask_length > 4) { - ir_read_error(expr, "invalid write mask: %s", mask_str); - return NULL; - } - - const unsigned idx_map[] = { 3, 0, 1, 2 }; /* w=bit 3, x=0, y=1, z=2 */ - - for (unsigned i = 0; i < mask_length; i++) { - if (mask_str[i] < 'w' || mask_str[i] > 'z') { - ir_read_error(expr, "write mask contains invalid character: %c", - mask_str[i]); - return NULL; - } - mask |= 1 << idx_map[mask_str[i] - 'w']; - } - } else if (!mask_list->subexpressions.is_empty()) { - ir_read_error(mask_list, "expected () or ()"); - return NULL; - } - - ir_dereference *lhs = read_dereference(lhs_expr); - if (lhs == NULL) { - ir_read_error(NULL, "when reading left-hand side of assignment"); - return NULL; - } - - ir_rvalue *rhs = read_rvalue(rhs_expr); - if (rhs == NULL) { - ir_read_error(NULL, "when reading right-hand side of assignment"); - return NULL; - } - - if (mask == 0 && (lhs->type->is_vector() || lhs->type->is_scalar())) { - ir_read_error(expr, "non-zero write mask required."); - return NULL; - } - - return new(mem_ctx) ir_assignment(lhs, rhs, condition, mask); -} - -ir_call * -ir_reader::read_call(s_expression *expr) -{ - s_symbol *name; - s_list *params; - - s_pattern pat[] = { "call", name, params }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (call ( ...))"); - return NULL; - } - - exec_list parameters; - - foreach_iter(exec_list_iterator, it, params->subexpressions) { - s_expression *expr = (s_expression*) it.get(); - ir_rvalue *param = read_rvalue(expr); - if (param == NULL) { - ir_read_error(expr, "when reading parameter to function call"); - return NULL; - } - parameters.push_tail(param); - } - - ir_function *f = state->symbols->get_function(name->value()); - if (f == NULL) { - ir_read_error(expr, "found call to undefined function %s", - name->value()); - return NULL; - } - - ir_function_signature *callee = f->matching_signature(¶meters); - if (callee == NULL) { - ir_read_error(expr, "couldn't find matching signature for function " - "%s", name->value()); - return NULL; - } - - return new(mem_ctx) ir_call(callee, ¶meters); -} - -ir_expression * -ir_reader::read_expression(s_expression *expr) -{ - s_expression *s_type; - s_symbol *s_op; - s_expression *s_arg1; - - s_pattern pat[] = { "expression", s_type, s_op, s_arg1 }; - if (!PARTIAL_MATCH(expr, pat)) { - ir_read_error(expr, "expected (expression " - " [])"); - return NULL; - } - s_expression *s_arg2 = (s_expression *) s_arg1->next; // may be tail sentinel - - const glsl_type *type = read_type(s_type); - if (type == NULL) - return NULL; - - /* Read the operator */ - ir_expression_operation op = ir_expression::get_operator(s_op->value()); - if (op == (ir_expression_operation) -1) { - ir_read_error(expr, "invalid operator: %s", s_op->value()); - return NULL; - } - - unsigned num_operands = ir_expression::get_num_operands(op); - if (num_operands == 1 && !s_arg1->next->is_tail_sentinel()) { - ir_read_error(expr, "expected (expression %s )", - s_op->value()); - return NULL; - } - - ir_rvalue *arg1 = read_rvalue(s_arg1); - ir_rvalue *arg2 = NULL; - if (arg1 == NULL) { - ir_read_error(NULL, "when reading first operand of %s", s_op->value()); - return NULL; - } - - if (num_operands == 2) { - if (s_arg2->is_tail_sentinel() || !s_arg2->next->is_tail_sentinel()) { - ir_read_error(expr, "expected (expression %s " - ")", s_op->value()); - return NULL; - } - arg2 = read_rvalue(s_arg2); - if (arg2 == NULL) { - ir_read_error(NULL, "when reading second operand of %s", - s_op->value()); - return NULL; - } - } - - return new(mem_ctx) ir_expression(op, type, arg1, arg2); -} - -ir_swizzle * -ir_reader::read_swizzle(s_expression *expr) -{ - s_symbol *swiz; - s_expression *sub; - - s_pattern pat[] = { "swiz", swiz, sub }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (swiz )"); - return NULL; - } - - if (strlen(swiz->value()) > 4) { - ir_read_error(expr, "expected a valid swizzle; found %s", swiz->value()); - return NULL; - } - - ir_rvalue *rvalue = read_rvalue(sub); - if (rvalue == NULL) - return NULL; - - ir_swizzle *ir = ir_swizzle::create(rvalue, swiz->value(), - rvalue->type->vector_elements); - if (ir == NULL) - ir_read_error(expr, "invalid swizzle"); - - return ir; -} - -ir_constant * -ir_reader::read_constant(s_expression *expr) -{ - s_expression *type_expr; - s_list *values; - - s_pattern pat[] = { "constant", type_expr, values }; - if (!MATCH(expr, pat)) { - ir_read_error(expr, "expected (constant (...))"); - return NULL; - } - - const glsl_type *type = read_type(type_expr); - if (type == NULL) - return NULL; - - if (values == NULL) { - ir_read_error(expr, "expected (constant (...))"); - return NULL; - } - - if (type->is_array()) { - unsigned elements_supplied = 0; - exec_list elements; - foreach_iter(exec_list_iterator, it, values->subexpressions) { - s_expression *elt = (s_expression *) it.get(); - ir_constant *ir_elt = read_constant(elt); - if (ir_elt == NULL) - return NULL; - elements.push_tail(ir_elt); - elements_supplied++; - } - - if (elements_supplied != type->length) { - ir_read_error(values, "expected exactly %u array elements, " - "given %u", type->length, elements_supplied); - return NULL; - } - return new(mem_ctx) ir_constant(type, &elements); - } - - const glsl_type *const base_type = type->get_base_type(); - - ir_constant_data data = { { 0 } }; - - // Read in list of values (at most 16). - int k = 0; - foreach_iter(exec_list_iterator, it, values->subexpressions) { - if (k >= 16) { - ir_read_error(values, "expected at most 16 numbers"); - return NULL; - } - - s_expression *expr = (s_expression*) it.get(); - - if (base_type->base_type == GLSL_TYPE_FLOAT) { - s_number *value = SX_AS_NUMBER(expr); - if (value == NULL) { - ir_read_error(values, "expected numbers"); - return NULL; - } - data.f[k] = value->fvalue(); - } else { - s_int *value = SX_AS_INT(expr); - if (value == NULL) { - ir_read_error(values, "expected integers"); - return NULL; - } - - switch (base_type->base_type) { - case GLSL_TYPE_UINT: { - data.u[k] = value->value(); - break; - } - case GLSL_TYPE_INT: { - data.i[k] = value->value(); - break; - } - case GLSL_TYPE_BOOL: { - data.b[k] = value->value(); - break; - } - default: - ir_read_error(values, "unsupported constant type"); - return NULL; - } - } - ++k; - } - - return new(mem_ctx) ir_constant(type, &data); -} - -ir_dereference * -ir_reader::read_dereference(s_expression *expr) -{ - s_symbol *s_var; - s_expression *s_subject; - s_expression *s_index; - s_symbol *s_field; - - s_pattern var_pat[] = { "var_ref", s_var }; - s_pattern array_pat[] = { "array_ref", s_subject, s_index }; - s_pattern record_pat[] = { "record_ref", s_subject, s_field }; - - if (MATCH(expr, var_pat)) { - ir_variable *var = state->symbols->get_variable(s_var->value()); - if (var == NULL) { - ir_read_error(expr, "undeclared variable: %s", s_var->value()); - return NULL; - } - return new(mem_ctx) ir_dereference_variable(var); - } else if (MATCH(expr, array_pat)) { - ir_rvalue *subject = read_rvalue(s_subject); - if (subject == NULL) { - ir_read_error(NULL, "when reading the subject of an array_ref"); - return NULL; - } - - ir_rvalue *idx = read_rvalue(s_index); - if (subject == NULL) { - ir_read_error(NULL, "when reading the index of an array_ref"); - return NULL; - } - return new(mem_ctx) ir_dereference_array(subject, idx); - } else if (MATCH(expr, record_pat)) { - ir_rvalue *subject = read_rvalue(s_subject); - if (subject == NULL) { - ir_read_error(NULL, "when reading the subject of a record_ref"); - return NULL; - } - return new(mem_ctx) ir_dereference_record(subject, s_field->value()); - } - return NULL; -} - -ir_texture * -ir_reader::read_texture(s_expression *expr) -{ - s_symbol *tag = NULL; - s_expression *s_sampler = NULL; - s_expression *s_coord = NULL; - s_list *s_offset = NULL; - s_expression *s_proj = NULL; - s_list *s_shadow = NULL; - s_expression *s_lod = NULL; - - ir_texture_opcode op; - - s_pattern tex_pattern[] = - { "tex", s_sampler, s_coord, s_offset, s_proj, s_shadow }; - s_pattern txf_pattern[] = - { "txf", s_sampler, s_coord, s_offset, s_lod }; - s_pattern other_pattern[] = - { tag, s_sampler, s_coord, s_offset, s_proj, s_shadow, s_lod }; - - if (MATCH(expr, tex_pattern)) { - op = ir_tex; - } else if (MATCH(expr, txf_pattern)) { - op = ir_txf; - } else if (MATCH(expr, other_pattern)) { - op = ir_texture::get_opcode(tag->value()); - if (op == -1) - return NULL; - } - - ir_texture *tex = new(mem_ctx) ir_texture(op); - - // Read sampler (must be a deref) - ir_dereference *sampler = read_dereference(s_sampler); - if (sampler == NULL) { - ir_read_error(NULL, "when reading sampler in (%s ...)", - tex->opcode_string()); - return NULL; - } - tex->set_sampler(sampler); - - // Read coordinate (any rvalue) - tex->coordinate = read_rvalue(s_coord); - if (tex->coordinate == NULL) { - ir_read_error(NULL, "when reading coordinate in (%s ...)", - tex->opcode_string()); - return NULL; - } - - // Read texel offset, i.e. (0 0 0) - s_int *offset_x; - s_int *offset_y; - s_int *offset_z; - s_pattern offset_pat[] = { offset_x, offset_y, offset_z }; - if (!MATCH(s_offset, offset_pat)) { - ir_read_error(s_offset, "expected ( )"); - return NULL; - } - tex->offsets[0] = offset_x->value(); - tex->offsets[1] = offset_y->value(); - tex->offsets[2] = offset_z->value(); - - if (op != ir_txf) { - s_int *proj_as_int = SX_AS_INT(s_proj); - if (proj_as_int && proj_as_int->value() == 1) { - tex->projector = NULL; - } else { - tex->projector = read_rvalue(s_proj); - if (tex->projector == NULL) { - ir_read_error(NULL, "when reading projective divide in (%s ..)", - tex->opcode_string()); - return NULL; - } - } - - if (s_shadow->subexpressions.is_empty()) { - tex->shadow_comparitor = NULL; - } else { - tex->shadow_comparitor = read_rvalue(s_shadow); - if (tex->shadow_comparitor == NULL) { - ir_read_error(NULL, "when reading shadow comparitor in (%s ..)", - tex->opcode_string()); - return NULL; - } - } - } - - switch (op) { - case ir_txb: - tex->lod_info.bias = read_rvalue(s_lod); - if (tex->lod_info.bias == NULL) { - ir_read_error(NULL, "when reading LOD bias in (txb ...)"); - return NULL; - } - break; - case ir_txl: - case ir_txf: - tex->lod_info.lod = read_rvalue(s_lod); - if (tex->lod_info.lod == NULL) { - ir_read_error(NULL, "when reading LOD in (%s ...)", - tex->opcode_string()); - return NULL; - } - break; - case ir_txd: { - s_expression *s_dx, *s_dy; - s_pattern dxdy_pat[] = { s_dx, s_dy }; - if (!MATCH(s_lod, dxdy_pat)) { - ir_read_error(s_lod, "expected (dPdx dPdy) in (txd ...)"); - return NULL; - } - tex->lod_info.grad.dPdx = read_rvalue(s_dx); - if (tex->lod_info.grad.dPdx == NULL) { - ir_read_error(NULL, "when reading dPdx in (txd ...)"); - return NULL; - } - tex->lod_info.grad.dPdy = read_rvalue(s_dy); - if (tex->lod_info.grad.dPdy == NULL) { - ir_read_error(NULL, "when reading dPdy in (txd ...)"); - return NULL; - } - break; - } - default: - // tex doesn't have any extra parameters. - break; - }; - return tex; -} +/* + * 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. + */ + +#include "ir_reader.h" +#include "glsl_parser_extras.h" +#include "glsl_types.h" +#include "s_expression.h" + +const static bool debug = false; + +class ir_reader { +public: + ir_reader(_mesa_glsl_parse_state *); + + void read(exec_list *instructions, const char *src, bool scan_for_protos); + +private: + void *mem_ctx; + _mesa_glsl_parse_state *state; + + void ir_read_error(s_expression *, const char *fmt, ...); + + const glsl_type *read_type(s_expression *); + + void scan_for_prototypes(exec_list *, s_expression *); + ir_function *read_function(s_expression *, bool skip_body); + void read_function_sig(ir_function *, s_expression *, bool skip_body); + + void read_instructions(exec_list *, s_expression *, ir_loop *); + ir_instruction *read_instruction(s_expression *, ir_loop *); + ir_variable *read_declaration(s_expression *); + ir_if *read_if(s_expression *, ir_loop *); + ir_loop *read_loop(s_expression *); + ir_return *read_return(s_expression *); + ir_rvalue *read_rvalue(s_expression *); + ir_assignment *read_assignment(s_expression *); + ir_expression *read_expression(s_expression *); + ir_call *read_call(s_expression *); + ir_swizzle *read_swizzle(s_expression *); + ir_constant *read_constant(s_expression *); + ir_texture *read_texture(s_expression *); + + ir_dereference *read_dereference(s_expression *); +}; + +ir_reader::ir_reader(_mesa_glsl_parse_state *state) : state(state) +{ + this->mem_ctx = state; +} + +void +_mesa_glsl_read_ir(_mesa_glsl_parse_state *state, exec_list *instructions, + const char *src, bool scan_for_protos) +{ + ir_reader r(state); + r.read(instructions, src, scan_for_protos); +} + +void +ir_reader::read(exec_list *instructions, const char *src, bool scan_for_protos) +{ + s_expression *expr = s_expression::read_expression(mem_ctx, src); + if (expr == NULL) { + ir_read_error(NULL, "couldn't parse S-Expression."); + return; + } + + if (scan_for_protos) { + scan_for_prototypes(instructions, expr); + if (state->error) + return; + } + + read_instructions(instructions, expr, NULL); + talloc_free(expr); + + if (debug) + validate_ir_tree(instructions); +} + +void +ir_reader::ir_read_error(s_expression *expr, const char *fmt, ...) +{ + va_list ap; + + state->error = true; + + if (state->current_function != NULL) + state->info_log = talloc_asprintf_append(state->info_log, + "In function %s:\n", + state->current_function->function_name()); + state->info_log = talloc_strdup_append(state->info_log, "error: "); + + va_start(ap, fmt); + state->info_log = talloc_vasprintf_append(state->info_log, fmt, ap); + va_end(ap); + state->info_log = talloc_strdup_append(state->info_log, "\n"); + + if (expr != NULL) { + state->info_log = talloc_strdup_append(state->info_log, + "...in this context:\n "); + expr->print(); + state->info_log = talloc_strdup_append(state->info_log, "\n\n"); + } +} + +const glsl_type * +ir_reader::read_type(s_expression *expr) +{ + s_expression *s_base_type; + s_int *s_size; + + s_pattern pat[] = { "array", s_base_type, s_size }; + if (MATCH(expr, pat)) { + const glsl_type *base_type = read_type(s_base_type); + if (base_type == NULL) { + ir_read_error(NULL, "when reading base type of array type"); + return NULL; + } + + return glsl_type::get_array_instance(base_type, s_size->value()); + } + + s_symbol *type_sym = SX_AS_SYMBOL(expr); + if (type_sym == NULL) { + ir_read_error(expr, "expected "); + return NULL; + } + + const glsl_type *type = state->symbols->get_type(type_sym->value()); + if (type == NULL) + ir_read_error(expr, "invalid type: %s", type_sym->value()); + + return type; +} + + +void +ir_reader::scan_for_prototypes(exec_list *instructions, s_expression *expr) +{ + s_list *list = SX_AS_LIST(expr); + if (list == NULL) { + ir_read_error(expr, "Expected ( ...); found an atom."); + return; + } + + foreach_iter(exec_list_iterator, it, list->subexpressions) { + s_list *sub = SX_AS_LIST(it.get()); + if (sub == NULL) + continue; // not a (function ...); ignore it. + + s_symbol *tag = SX_AS_SYMBOL(sub->subexpressions.get_head()); + if (tag == NULL || strcmp(tag->value(), "function") != 0) + continue; // not a (function ...); ignore it. + + ir_function *f = read_function(sub, true); + if (f == NULL) + return; + instructions->push_tail(f); + } +} + +ir_function * +ir_reader::read_function(s_expression *expr, bool skip_body) +{ + bool added = false; + s_symbol *name; + + s_pattern pat[] = { "function", name }; + if (!PARTIAL_MATCH(expr, pat)) { + ir_read_error(expr, "Expected (function (signature ...) ...)"); + return NULL; + } + + ir_function *f = state->symbols->get_function(name->value()); + if (f == NULL) { + f = new(mem_ctx) ir_function(name->value()); + added = state->symbols->add_function(f); + assert(added); + } + + exec_list_iterator it = ((s_list *) expr)->subexpressions.iterator(); + it.next(); // skip "function" tag + it.next(); // skip function name + for (/* nothing */; it.has_next(); it.next()) { + s_expression *s_sig = (s_expression *) it.get(); + read_function_sig(f, s_sig, skip_body); + } + return added ? f : NULL; +} + +void +ir_reader::read_function_sig(ir_function *f, s_expression *expr, bool skip_body) +{ + s_expression *type_expr; + s_list *paramlist; + s_list *body_list; + + s_pattern pat[] = { "signature", type_expr, paramlist, body_list }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "Expected (signature (parameters ...) " + "( ...))"); + return; + } + + const glsl_type *return_type = read_type(type_expr); + if (return_type == NULL) + return; + + s_symbol *paramtag = SX_AS_SYMBOL(paramlist->subexpressions.get_head()); + if (paramtag == NULL || strcmp(paramtag->value(), "parameters") != 0) { + ir_read_error(paramlist, "Expected (parameters ...)"); + return; + } + + // Read the parameters list into a temporary place. + exec_list hir_parameters; + state->symbols->push_scope(); + + exec_list_iterator it = paramlist->subexpressions.iterator(); + for (it.next() /* skip "parameters" */; it.has_next(); it.next()) { + ir_variable *var = read_declaration((s_expression *) it.get()); + if (var == NULL) + return; + + hir_parameters.push_tail(var); + } + + ir_function_signature *sig = f->exact_matching_signature(&hir_parameters); + if (sig == NULL && skip_body) { + /* If scanning for prototypes, generate a new signature. */ + sig = new(mem_ctx) ir_function_signature(return_type); + sig->is_builtin = true; + f->add_signature(sig); + } else if (sig != NULL) { + const char *badvar = sig->qualifiers_match(&hir_parameters); + if (badvar != NULL) { + ir_read_error(expr, "function `%s' parameter `%s' qualifiers " + "don't match prototype", f->name, badvar); + return; + } + + if (sig->return_type != return_type) { + ir_read_error(expr, "function `%s' return type doesn't " + "match prototype", f->name); + return; + } + } else { + /* No prototype for this body exists - skip it. */ + state->symbols->pop_scope(); + return; + } + assert(sig != NULL); + + sig->replace_parameters(&hir_parameters); + + if (!skip_body && !body_list->subexpressions.is_empty()) { + if (sig->is_defined) { + ir_read_error(expr, "function %s redefined", f->name); + return; + } + state->current_function = sig; + read_instructions(&sig->body, body_list, NULL); + state->current_function = NULL; + sig->is_defined = true; + } + + state->symbols->pop_scope(); +} + +void +ir_reader::read_instructions(exec_list *instructions, s_expression *expr, + ir_loop *loop_ctx) +{ + // Read in a list of instructions + s_list *list = SX_AS_LIST(expr); + if (list == NULL) { + ir_read_error(expr, "Expected ( ...); found an atom."); + return; + } + + foreach_iter(exec_list_iterator, it, list->subexpressions) { + s_expression *sub = (s_expression*) it.get(); + ir_instruction *ir = read_instruction(sub, loop_ctx); + if (ir != NULL) { + /* Global variable declarations should be moved to the top, before + * any functions that might use them. Functions are added to the + * instruction stream when scanning for prototypes, so without this + * hack, they always appear before variable declarations. + */ + if (state->current_function == NULL && ir->as_variable() != NULL) + instructions->push_head(ir); + else + instructions->push_tail(ir); + } + } +} + + +ir_instruction * +ir_reader::read_instruction(s_expression *expr, ir_loop *loop_ctx) +{ + s_symbol *symbol = SX_AS_SYMBOL(expr); + if (symbol != NULL) { + if (strcmp(symbol->value(), "break") == 0 && loop_ctx != NULL) + return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_break); + if (strcmp(symbol->value(), "continue") == 0 && loop_ctx != NULL) + return new(mem_ctx) ir_loop_jump(ir_loop_jump::jump_continue); + } + + s_list *list = SX_AS_LIST(expr); + if (list == NULL || list->subexpressions.is_empty()) { + ir_read_error(expr, "Invalid instruction.\n"); + return NULL; + } + + s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); + if (tag == NULL) { + ir_read_error(expr, "expected instruction tag"); + return NULL; + } + + ir_instruction *inst = NULL; + if (strcmp(tag->value(), "declare") == 0) { + inst = read_declaration(list); + } else if (strcmp(tag->value(), "assign") == 0) { + inst = read_assignment(list); + } else if (strcmp(tag->value(), "if") == 0) { + inst = read_if(list, loop_ctx); + } else if (strcmp(tag->value(), "loop") == 0) { + inst = read_loop(list); + } else if (strcmp(tag->value(), "return") == 0) { + inst = read_return(list); + } else if (strcmp(tag->value(), "function") == 0) { + inst = read_function(list, false); + } else { + inst = read_rvalue(list); + if (inst == NULL) + ir_read_error(NULL, "when reading instruction"); + } + return inst; +} + +ir_variable * +ir_reader::read_declaration(s_expression *expr) +{ + s_list *s_quals; + s_expression *s_type; + s_symbol *s_name; + + s_pattern pat[] = { "declare", s_quals, s_type, s_name }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (declare () )"); + return NULL; + } + + const glsl_type *type = read_type(s_type); + if (type == NULL) + return NULL; + + ir_variable *var = new(mem_ctx) ir_variable(type, s_name->value(), + ir_var_auto); + + foreach_iter(exec_list_iterator, it, s_quals->subexpressions) { + s_symbol *qualifier = SX_AS_SYMBOL(it.get()); + if (qualifier == NULL) { + ir_read_error(expr, "qualifier list must contain only symbols"); + return NULL; + } + + // FINISHME: Check for duplicate/conflicting qualifiers. + if (strcmp(qualifier->value(), "centroid") == 0) { + var->centroid = 1; + } else if (strcmp(qualifier->value(), "invariant") == 0) { + var->invariant = 1; + } else if (strcmp(qualifier->value(), "uniform") == 0) { + var->mode = ir_var_uniform; + } else if (strcmp(qualifier->value(), "auto") == 0) { + var->mode = ir_var_auto; + } else if (strcmp(qualifier->value(), "in") == 0) { + var->mode = ir_var_in; + } else if (strcmp(qualifier->value(), "out") == 0) { + var->mode = ir_var_out; + } else if (strcmp(qualifier->value(), "inout") == 0) { + var->mode = ir_var_inout; + } else if (strcmp(qualifier->value(), "smooth") == 0) { + var->interpolation = ir_var_smooth; + } else if (strcmp(qualifier->value(), "flat") == 0) { + var->interpolation = ir_var_flat; + } else if (strcmp(qualifier->value(), "noperspective") == 0) { + var->interpolation = ir_var_noperspective; + } else { + ir_read_error(expr, "unknown qualifier: %s", qualifier->value()); + return NULL; + } + } + + // Add the variable to the symbol table + state->symbols->add_variable(var); + + return var; +} + + +ir_if * +ir_reader::read_if(s_expression *expr, ir_loop *loop_ctx) +{ + s_expression *s_cond; + s_expression *s_then; + s_expression *s_else; + + s_pattern pat[] = { "if", s_cond, s_then, s_else }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (if (...) (...))"); + return NULL; + } + + ir_rvalue *condition = read_rvalue(s_cond); + if (condition == NULL) { + ir_read_error(NULL, "when reading condition of (if ...)"); + return NULL; + } + + ir_if *iff = new(mem_ctx) ir_if(condition); + + read_instructions(&iff->then_instructions, s_then, loop_ctx); + read_instructions(&iff->else_instructions, s_else, loop_ctx); + if (state->error) { + delete iff; + iff = NULL; + } + return iff; +} + + +ir_loop * +ir_reader::read_loop(s_expression *expr) +{ + s_expression *s_counter, *s_from, *s_to, *s_inc, *s_body; + + s_pattern pat[] = { "loop", s_counter, s_from, s_to, s_inc, s_body }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (loop " + " )"); + return NULL; + } + + // FINISHME: actually read the count/from/to fields. + + ir_loop *loop = new(mem_ctx) ir_loop; + read_instructions(&loop->body_instructions, s_body, loop); + if (state->error) { + delete loop; + loop = NULL; + } + return loop; +} + + +ir_return * +ir_reader::read_return(s_expression *expr) +{ + s_expression *s_retval; + + s_pattern pat[] = { "return", s_retval}; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (return )"); + return NULL; + } + + ir_rvalue *retval = read_rvalue(s_retval); + if (retval == NULL) { + ir_read_error(NULL, "when reading return value"); + return NULL; + } + + return new(mem_ctx) ir_return(retval); +} + + +ir_rvalue * +ir_reader::read_rvalue(s_expression *expr) +{ + s_list *list = SX_AS_LIST(expr); + if (list == NULL || list->subexpressions.is_empty()) + return NULL; + + s_symbol *tag = SX_AS_SYMBOL(list->subexpressions.get_head()); + if (tag == NULL) { + ir_read_error(expr, "expected rvalue tag"); + return NULL; + } + + ir_rvalue *rvalue = read_dereference(list); + if (rvalue != NULL || state->error) + return rvalue; + else if (strcmp(tag->value(), "swiz") == 0) { + rvalue = read_swizzle(list); + } else if (strcmp(tag->value(), "expression") == 0) { + rvalue = read_expression(list); + } else if (strcmp(tag->value(), "call") == 0) { + rvalue = read_call(list); + } else if (strcmp(tag->value(), "constant") == 0) { + rvalue = read_constant(list); + } else { + rvalue = read_texture(list); + if (rvalue == NULL && !state->error) + ir_read_error(expr, "unrecognized rvalue tag: %s", tag->value()); + } + + return rvalue; +} + +ir_assignment * +ir_reader::read_assignment(s_expression *expr) +{ + s_expression *cond_expr = NULL; + s_expression *lhs_expr, *rhs_expr; + s_list *mask_list; + + s_pattern pat4[] = { "assign", mask_list, lhs_expr, rhs_expr }; + s_pattern pat5[] = { "assign", cond_expr, mask_list, lhs_expr, rhs_expr }; + if (!MATCH(expr, pat4) && !MATCH(expr, pat5)) { + ir_read_error(expr, "expected (assign [] () " + " )"); + return NULL; + } + + ir_rvalue *condition = NULL; + if (cond_expr != NULL) { + condition = read_rvalue(cond_expr); + if (condition == NULL) { + ir_read_error(NULL, "when reading condition of assignment"); + return NULL; + } + } + + unsigned mask = 0; + + s_symbol *mask_symbol; + s_pattern mask_pat[] = { mask_symbol }; + if (MATCH(mask_list, mask_pat)) { + const char *mask_str = mask_symbol->value(); + unsigned mask_length = strlen(mask_str); + if (mask_length > 4) { + ir_read_error(expr, "invalid write mask: %s", mask_str); + return NULL; + } + + const unsigned idx_map[] = { 3, 0, 1, 2 }; /* w=bit 3, x=0, y=1, z=2 */ + + for (unsigned i = 0; i < mask_length; i++) { + if (mask_str[i] < 'w' || mask_str[i] > 'z') { + ir_read_error(expr, "write mask contains invalid character: %c", + mask_str[i]); + return NULL; + } + mask |= 1 << idx_map[mask_str[i] - 'w']; + } + } else if (!mask_list->subexpressions.is_empty()) { + ir_read_error(mask_list, "expected () or ()"); + return NULL; + } + + ir_dereference *lhs = read_dereference(lhs_expr); + if (lhs == NULL) { + ir_read_error(NULL, "when reading left-hand side of assignment"); + return NULL; + } + + ir_rvalue *rhs = read_rvalue(rhs_expr); + if (rhs == NULL) { + ir_read_error(NULL, "when reading right-hand side of assignment"); + return NULL; + } + + if (mask == 0 && (lhs->type->is_vector() || lhs->type->is_scalar())) { + ir_read_error(expr, "non-zero write mask required."); + return NULL; + } + + return new(mem_ctx) ir_assignment(lhs, rhs, condition, mask); +} + +ir_call * +ir_reader::read_call(s_expression *expr) +{ + s_symbol *name; + s_list *params; + + s_pattern pat[] = { "call", name, params }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (call ( ...))"); + return NULL; + } + + exec_list parameters; + + foreach_iter(exec_list_iterator, it, params->subexpressions) { + s_expression *expr = (s_expression*) it.get(); + ir_rvalue *param = read_rvalue(expr); + if (param == NULL) { + ir_read_error(expr, "when reading parameter to function call"); + return NULL; + } + parameters.push_tail(param); + } + + ir_function *f = state->symbols->get_function(name->value()); + if (f == NULL) { + ir_read_error(expr, "found call to undefined function %s", + name->value()); + return NULL; + } + + ir_function_signature *callee = f->matching_signature(¶meters); + if (callee == NULL) { + ir_read_error(expr, "couldn't find matching signature for function " + "%s", name->value()); + return NULL; + } + + return new(mem_ctx) ir_call(callee, ¶meters); +} + +ir_expression * +ir_reader::read_expression(s_expression *expr) +{ + s_expression *s_type; + s_symbol *s_op; + s_expression *s_arg1; + + s_pattern pat[] = { "expression", s_type, s_op, s_arg1 }; + if (!PARTIAL_MATCH(expr, pat)) { + ir_read_error(expr, "expected (expression " + " [])"); + return NULL; + } + s_expression *s_arg2 = (s_expression *) s_arg1->next; // may be tail sentinel + + const glsl_type *type = read_type(s_type); + if (type == NULL) + return NULL; + + /* Read the operator */ + ir_expression_operation op = ir_expression::get_operator(s_op->value()); + if (op == (ir_expression_operation) -1) { + ir_read_error(expr, "invalid operator: %s", s_op->value()); + return NULL; + } + + unsigned num_operands = ir_expression::get_num_operands(op); + if (num_operands == 1 && !s_arg1->next->is_tail_sentinel()) { + ir_read_error(expr, "expected (expression %s )", + s_op->value()); + return NULL; + } + + ir_rvalue *arg1 = read_rvalue(s_arg1); + ir_rvalue *arg2 = NULL; + if (arg1 == NULL) { + ir_read_error(NULL, "when reading first operand of %s", s_op->value()); + return NULL; + } + + if (num_operands == 2) { + if (s_arg2->is_tail_sentinel() || !s_arg2->next->is_tail_sentinel()) { + ir_read_error(expr, "expected (expression %s " + ")", s_op->value()); + return NULL; + } + arg2 = read_rvalue(s_arg2); + if (arg2 == NULL) { + ir_read_error(NULL, "when reading second operand of %s", + s_op->value()); + return NULL; + } + } + + return new(mem_ctx) ir_expression(op, type, arg1, arg2); +} + +ir_swizzle * +ir_reader::read_swizzle(s_expression *expr) +{ + s_symbol *swiz; + s_expression *sub; + + s_pattern pat[] = { "swiz", swiz, sub }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (swiz )"); + return NULL; + } + + if (strlen(swiz->value()) > 4) { + ir_read_error(expr, "expected a valid swizzle; found %s", swiz->value()); + return NULL; + } + + ir_rvalue *rvalue = read_rvalue(sub); + if (rvalue == NULL) + return NULL; + + ir_swizzle *ir = ir_swizzle::create(rvalue, swiz->value(), + rvalue->type->vector_elements); + if (ir == NULL) + ir_read_error(expr, "invalid swizzle"); + + return ir; +} + +ir_constant * +ir_reader::read_constant(s_expression *expr) +{ + s_expression *type_expr; + s_list *values; + + s_pattern pat[] = { "constant", type_expr, values }; + if (!MATCH(expr, pat)) { + ir_read_error(expr, "expected (constant (...))"); + return NULL; + } + + const glsl_type *type = read_type(type_expr); + if (type == NULL) + return NULL; + + if (values == NULL) { + ir_read_error(expr, "expected (constant (...))"); + return NULL; + } + + if (type->is_array()) { + unsigned elements_supplied = 0; + exec_list elements; + foreach_iter(exec_list_iterator, it, values->subexpressions) { + s_expression *elt = (s_expression *) it.get(); + ir_constant *ir_elt = read_constant(elt); + if (ir_elt == NULL) + return NULL; + elements.push_tail(ir_elt); + elements_supplied++; + } + + if (elements_supplied != type->length) { + ir_read_error(values, "expected exactly %u array elements, " + "given %u", type->length, elements_supplied); + return NULL; + } + return new(mem_ctx) ir_constant(type, &elements); + } + + const glsl_type *const base_type = type->get_base_type(); + + ir_constant_data data = { { 0 } }; + + // Read in list of values (at most 16). + int k = 0; + foreach_iter(exec_list_iterator, it, values->subexpressions) { + if (k >= 16) { + ir_read_error(values, "expected at most 16 numbers"); + return NULL; + } + + s_expression *expr = (s_expression*) it.get(); + + if (base_type->base_type == GLSL_TYPE_FLOAT) { + s_number *value = SX_AS_NUMBER(expr); + if (value == NULL) { + ir_read_error(values, "expected numbers"); + return NULL; + } + data.f[k] = value->fvalue(); + } else { + s_int *value = SX_AS_INT(expr); + if (value == NULL) { + ir_read_error(values, "expected integers"); + return NULL; + } + + switch (base_type->base_type) { + case GLSL_TYPE_UINT: { + data.u[k] = value->value(); + break; + } + case GLSL_TYPE_INT: { + data.i[k] = value->value(); + break; + } + case GLSL_TYPE_BOOL: { + data.b[k] = value->value(); + break; + } + default: + ir_read_error(values, "unsupported constant type"); + return NULL; + } + } + ++k; + } + + return new(mem_ctx) ir_constant(type, &data); +} + +ir_dereference * +ir_reader::read_dereference(s_expression *expr) +{ + s_symbol *s_var; + s_expression *s_subject; + s_expression *s_index; + s_symbol *s_field; + + s_pattern var_pat[] = { "var_ref", s_var }; + s_pattern array_pat[] = { "array_ref", s_subject, s_index }; + s_pattern record_pat[] = { "record_ref", s_subject, s_field }; + + if (MATCH(expr, var_pat)) { + ir_variable *var = state->symbols->get_variable(s_var->value()); + if (var == NULL) { + ir_read_error(expr, "undeclared variable: %s", s_var->value()); + return NULL; + } + return new(mem_ctx) ir_dereference_variable(var); + } else if (MATCH(expr, array_pat)) { + ir_rvalue *subject = read_rvalue(s_subject); + if (subject == NULL) { + ir_read_error(NULL, "when reading the subject of an array_ref"); + return NULL; + } + + ir_rvalue *idx = read_rvalue(s_index); + if (subject == NULL) { + ir_read_error(NULL, "when reading the index of an array_ref"); + return NULL; + } + return new(mem_ctx) ir_dereference_array(subject, idx); + } else if (MATCH(expr, record_pat)) { + ir_rvalue *subject = read_rvalue(s_subject); + if (subject == NULL) { + ir_read_error(NULL, "when reading the subject of a record_ref"); + return NULL; + } + return new(mem_ctx) ir_dereference_record(subject, s_field->value()); + } + return NULL; +} + +ir_texture * +ir_reader::read_texture(s_expression *expr) +{ + s_symbol *tag = NULL; + s_expression *s_sampler = NULL; + s_expression *s_coord = NULL; + s_list *s_offset = NULL; + s_expression *s_proj = NULL; + s_list *s_shadow = NULL; + s_expression *s_lod = NULL; + + ir_texture_opcode op = ir_tex; /* silence warning */ + + s_pattern tex_pattern[] = + { "tex", s_sampler, s_coord, s_offset, s_proj, s_shadow }; + s_pattern txf_pattern[] = + { "txf", s_sampler, s_coord, s_offset, s_lod }; + s_pattern other_pattern[] = + { tag, s_sampler, s_coord, s_offset, s_proj, s_shadow, s_lod }; + + if (MATCH(expr, tex_pattern)) { + op = ir_tex; + } else if (MATCH(expr, txf_pattern)) { + op = ir_txf; + } else if (MATCH(expr, other_pattern)) { + op = ir_texture::get_opcode(tag->value()); + if (op == -1) + return NULL; + } else { + ir_read_error(NULL, "unexpected texture pattern"); + return NULL; + } + + ir_texture *tex = new(mem_ctx) ir_texture(op); + + // Read sampler (must be a deref) + ir_dereference *sampler = read_dereference(s_sampler); + if (sampler == NULL) { + ir_read_error(NULL, "when reading sampler in (%s ...)", + tex->opcode_string()); + return NULL; + } + tex->set_sampler(sampler); + + // Read coordinate (any rvalue) + tex->coordinate = read_rvalue(s_coord); + if (tex->coordinate == NULL) { + ir_read_error(NULL, "when reading coordinate in (%s ...)", + tex->opcode_string()); + return NULL; + } + + // Read texel offset, i.e. (0 0 0) + s_int *offset_x; + s_int *offset_y; + s_int *offset_z; + s_pattern offset_pat[] = { offset_x, offset_y, offset_z }; + if (!MATCH(s_offset, offset_pat)) { + ir_read_error(s_offset, "expected ( )"); + return NULL; + } + tex->offsets[0] = offset_x->value(); + tex->offsets[1] = offset_y->value(); + tex->offsets[2] = offset_z->value(); + + if (op != ir_txf) { + s_int *proj_as_int = SX_AS_INT(s_proj); + if (proj_as_int && proj_as_int->value() == 1) { + tex->projector = NULL; + } else { + tex->projector = read_rvalue(s_proj); + if (tex->projector == NULL) { + ir_read_error(NULL, "when reading projective divide in (%s ..)", + tex->opcode_string()); + return NULL; + } + } + + if (s_shadow->subexpressions.is_empty()) { + tex->shadow_comparitor = NULL; + } else { + tex->shadow_comparitor = read_rvalue(s_shadow); + if (tex->shadow_comparitor == NULL) { + ir_read_error(NULL, "when reading shadow comparitor in (%s ..)", + tex->opcode_string()); + return NULL; + } + } + } + + switch (op) { + case ir_txb: + tex->lod_info.bias = read_rvalue(s_lod); + if (tex->lod_info.bias == NULL) { + ir_read_error(NULL, "when reading LOD bias in (txb ...)"); + return NULL; + } + break; + case ir_txl: + case ir_txf: + tex->lod_info.lod = read_rvalue(s_lod); + if (tex->lod_info.lod == NULL) { + ir_read_error(NULL, "when reading LOD in (%s ...)", + tex->opcode_string()); + return NULL; + } + break; + case ir_txd: { + s_expression *s_dx, *s_dy; + s_pattern dxdy_pat[] = { s_dx, s_dy }; + if (!MATCH(s_lod, dxdy_pat)) { + ir_read_error(s_lod, "expected (dPdx dPdy) in (txd ...)"); + return NULL; + } + tex->lod_info.grad.dPdx = read_rvalue(s_dx); + if (tex->lod_info.grad.dPdx == NULL) { + ir_read_error(NULL, "when reading dPdx in (txd ...)"); + return NULL; + } + tex->lod_info.grad.dPdy = read_rvalue(s_dy); + if (tex->lod_info.grad.dPdy == NULL) { + ir_read_error(NULL, "when reading dPdy in (txd ...)"); + return NULL; + } + break; + } + default: + // tex doesn't have any extra parameters. + break; + }; + return tex; +} diff --git a/mesalib/src/glsl/link_functions.cpp b/mesalib/src/glsl/link_functions.cpp index ea9129fe3..861fa39b5 100644 --- a/mesalib/src/glsl/link_functions.cpp +++ b/mesalib/src/glsl/link_functions.cpp @@ -1,258 +1,270 @@ -/* - * 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. - */ - -#include -#include -#include - -#include "main/core.h" -#include "glsl_symbol_table.h" -#include "glsl_parser_extras.h" -#include "ir.h" -#include "program.h" -#include "program/hash_table.h" -#include "linker.h" - -static ir_function_signature * -find_matching_signature(const char *name, const exec_list *actual_parameters, - gl_shader **shader_list, unsigned num_shaders); - -class call_link_visitor : public ir_hierarchical_visitor { -public: - call_link_visitor(gl_shader_program *prog, gl_shader *linked, - gl_shader **shader_list, unsigned num_shaders) - { - this->prog = prog; - this->shader_list = shader_list; - this->num_shaders = num_shaders; - this->success = true; - this->linked = linked; - - this->locals = hash_table_ctor(0, hash_table_pointer_hash, - hash_table_pointer_compare); - } - - ~call_link_visitor() - { - hash_table_dtor(this->locals); - } - - virtual ir_visitor_status visit(ir_variable *ir) - { - hash_table_insert(locals, ir, ir); - return visit_continue; - } - - virtual ir_visitor_status visit_enter(ir_call *ir) - { - /* If ir is an ir_call from a function that was imported from another - * shader callee will point to an ir_function_signature in the original - * shader. In this case the function signature MUST NOT BE MODIFIED. - * Doing so will modify the original shader. This may prevent that - * shader from being linkable in other programs. - */ - const ir_function_signature *const callee = ir->get_callee(); - assert(callee != NULL); - const char *const name = callee->function_name(); - - /* Determine if the requested function signature already exists in the - * final linked shader. If it does, use it as the target of the call. - */ - ir_function_signature *sig = - find_matching_signature(name, &callee->parameters, &linked, 1); - if (sig != NULL) { - ir->set_callee(sig); - return visit_continue; - } - - /* Try to find the signature in one of the other shaders that is being - * linked. If it's not found there, return an error. - */ - sig = find_matching_signature(name, &ir->actual_parameters, shader_list, - num_shaders); - if (sig == NULL) { - /* FINISHME: Log the full signature of unresolved function. - */ - linker_error_printf(this->prog, "unresolved reference to function " - "`%s'\n", name); - this->success = false; - return visit_stop; - } - - /* Find the prototype information in the linked shader. Generate any - * details that may be missing. - */ - ir_function *f = linked->symbols->get_function(name); - if (f == NULL) - f = new(linked) ir_function(name); - - ir_function_signature *linked_sig = - f->exact_matching_signature(&callee->parameters); - if (linked_sig == NULL) { - linked_sig = new(linked) ir_function_signature(callee->return_type); - f->add_signature(linked_sig); - } - - /* At this point linked_sig and called may be the same. If ir is an - * ir_call from linked then linked_sig and callee will be - * ir_function_signatures that have no definitions (is_defined is false). - */ - assert(!linked_sig->is_defined); - assert(linked_sig->body.is_empty()); - - /* Create an in-place clone of the function definition. This multistep - * process introduces some complexity here, but it has some advantages. - * The parameter list and the and function body are cloned separately. - * The clone of the parameter list is used to prime the hashtable used - * to replace variable references in the cloned body. - * - * The big advantage is that the ir_function_signature does not change. - * This means that we don't have to process the rest of the IR tree to - * patch ir_call nodes. In addition, there is no way to remove or - * replace signature stored in a function. One could easily be added, - * but this avoids the need. - */ - struct hash_table *ht = hash_table_ctor(0, hash_table_pointer_hash, - hash_table_pointer_compare); - exec_list formal_parameters; - foreach_list_const(node, &sig->parameters) { - const ir_instruction *const original = (ir_instruction *) node; - assert(const_cast(original)->as_variable()); - - ir_instruction *copy = original->clone(linked, ht); - formal_parameters.push_tail(copy); - } - - linked_sig->replace_parameters(&formal_parameters); - - foreach_list_const(node, &sig->body) { - const ir_instruction *const original = (ir_instruction *) node; - - ir_instruction *copy = original->clone(linked, ht); - linked_sig->body.push_tail(copy); - } - - linked_sig->is_defined = true; - hash_table_dtor(ht); - - /* Patch references inside the function to things outside the function - * (i.e., function calls and global variables). - */ - linked_sig->accept(this); - - ir->set_callee(linked_sig); - - return visit_continue; - } - - virtual ir_visitor_status visit(ir_dereference_variable *ir) - { - if (hash_table_find(locals, ir->var) == NULL) { - /* The non-function variable must be a global, so try to find the - * variable in the shader's symbol table. If the variable is not - * found, then it's a global that *MUST* be defined in the original - * shader. - */ - ir_variable *var = linked->symbols->get_variable(ir->var->name); - if (var == NULL) { - /* Clone the ir_variable that the dereference already has and add - * it to the linked shader. - */ - var = ir->var->clone(linked, NULL); - linked->symbols->add_variable(var); - linked->ir->push_head(var); - } - - ir->var = var; - } - - return visit_continue; - } - - /** Was function linking successful? */ - bool success; - -private: - /** - * Shader program being linked - * - * This is only used for logging error messages. - */ - gl_shader_program *prog; - - /** List of shaders available for linking. */ - gl_shader **shader_list; - - /** Number of shaders available for linking. */ - unsigned num_shaders; - - /** - * Final linked shader - * - * This is used two ways. It is used to find global variables in the - * linked shader that are accessed by the function. It is also used to add - * global variables from the shader where the function originated. - */ - gl_shader *linked; - - /** - * Table of variables local to the function. - */ - hash_table *locals; -}; - - -/** - * Searches a list of shaders for a particular function definition - */ -ir_function_signature * -find_matching_signature(const char *name, const exec_list *actual_parameters, - gl_shader **shader_list, unsigned num_shaders) -{ - for (unsigned i = 0; i < num_shaders; i++) { - ir_function *const f = shader_list[i]->symbols->get_function(name); - - if (f == NULL) - continue; - - ir_function_signature *sig = f->matching_signature(actual_parameters); - - if ((sig == NULL) || !sig->is_defined) - continue; - - return sig; - } - - return NULL; -} - - -bool -link_function_calls(gl_shader_program *prog, gl_shader *main, - gl_shader **shader_list, unsigned num_shaders) -{ - call_link_visitor v(prog, main, shader_list, num_shaders); - - v.run(main->ir); - return v.success; -} +/* + * 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. + */ + +#include +#include +#include + +#include "main/core.h" +#include "glsl_symbol_table.h" +#include "glsl_parser_extras.h" +#include "ir.h" +#include "program.h" +#include "program/hash_table.h" +#include "linker.h" + +static ir_function_signature * +find_matching_signature(const char *name, const exec_list *actual_parameters, + gl_shader **shader_list, unsigned num_shaders); + +class call_link_visitor : public ir_hierarchical_visitor { +public: + call_link_visitor(gl_shader_program *prog, gl_shader *linked, + gl_shader **shader_list, unsigned num_shaders) + { + this->prog = prog; + this->shader_list = shader_list; + this->num_shaders = num_shaders; + this->success = true; + this->linked = linked; + + this->locals = hash_table_ctor(0, hash_table_pointer_hash, + hash_table_pointer_compare); + } + + ~call_link_visitor() + { + hash_table_dtor(this->locals); + } + + virtual ir_visitor_status visit(ir_variable *ir) + { + hash_table_insert(locals, ir, ir); + return visit_continue; + } + + virtual ir_visitor_status visit_enter(ir_call *ir) + { + /* If ir is an ir_call from a function that was imported from another + * shader callee will point to an ir_function_signature in the original + * shader. In this case the function signature MUST NOT BE MODIFIED. + * Doing so will modify the original shader. This may prevent that + * shader from being linkable in other programs. + */ + const ir_function_signature *const callee = ir->get_callee(); + assert(callee != NULL); + const char *const name = callee->function_name(); + + /* Determine if the requested function signature already exists in the + * final linked shader. If it does, use it as the target of the call. + */ + ir_function_signature *sig = + find_matching_signature(name, &callee->parameters, &linked, 1); + if (sig != NULL) { + ir->set_callee(sig); + return visit_continue; + } + + /* Try to find the signature in one of the other shaders that is being + * linked. If it's not found there, return an error. + */ + sig = find_matching_signature(name, &ir->actual_parameters, shader_list, + num_shaders); + if (sig == NULL) { + /* FINISHME: Log the full signature of unresolved function. + */ + linker_error_printf(this->prog, "unresolved reference to function " + "`%s'\n", name); + this->success = false; + return visit_stop; + } + + /* Find the prototype information in the linked shader. Generate any + * details that may be missing. + */ + ir_function *f = linked->symbols->get_function(name); + if (f == NULL) + f = new(linked) ir_function(name); + + ir_function_signature *linked_sig = + f->exact_matching_signature(&callee->parameters); + if (linked_sig == NULL) { + linked_sig = new(linked) ir_function_signature(callee->return_type); + f->add_signature(linked_sig); + } + + /* At this point linked_sig and called may be the same. If ir is an + * ir_call from linked then linked_sig and callee will be + * ir_function_signatures that have no definitions (is_defined is false). + */ + assert(!linked_sig->is_defined); + assert(linked_sig->body.is_empty()); + + /* Create an in-place clone of the function definition. This multistep + * process introduces some complexity here, but it has some advantages. + * The parameter list and the and function body are cloned separately. + * The clone of the parameter list is used to prime the hashtable used + * to replace variable references in the cloned body. + * + * The big advantage is that the ir_function_signature does not change. + * This means that we don't have to process the rest of the IR tree to + * patch ir_call nodes. In addition, there is no way to remove or + * replace signature stored in a function. One could easily be added, + * but this avoids the need. + */ + struct hash_table *ht = hash_table_ctor(0, hash_table_pointer_hash, + hash_table_pointer_compare); + exec_list formal_parameters; + foreach_list_const(node, &sig->parameters) { + const ir_instruction *const original = (ir_instruction *) node; + assert(const_cast(original)->as_variable()); + + ir_instruction *copy = original->clone(linked, ht); + formal_parameters.push_tail(copy); + } + + linked_sig->replace_parameters(&formal_parameters); + + foreach_list_const(node, &sig->body) { + const ir_instruction *const original = (ir_instruction *) node; + + ir_instruction *copy = original->clone(linked, ht); + linked_sig->body.push_tail(copy); + } + + linked_sig->is_defined = true; + hash_table_dtor(ht); + + /* Patch references inside the function to things outside the function + * (i.e., function calls and global variables). + */ + linked_sig->accept(this); + + ir->set_callee(linked_sig); + + return visit_continue; + } + + virtual ir_visitor_status visit(ir_dereference_variable *ir) + { + if (hash_table_find(locals, ir->var) == NULL) { + /* The non-function variable must be a global, so try to find the + * variable in the shader's symbol table. If the variable is not + * found, then it's a global that *MUST* be defined in the original + * shader. + */ + ir_variable *var = linked->symbols->get_variable(ir->var->name); + if (var == NULL) { + /* Clone the ir_variable that the dereference already has and add + * it to the linked shader. + */ + var = ir->var->clone(linked, NULL); + linked->symbols->add_variable(var); + linked->ir->push_head(var); + } else if (var->type->is_array()) { + /* It is possible to have a global array declared in multiple + * shaders without a size. The array is implicitly sized by the + * maximal access to it in *any* shader. Because of this, we + * need to track the maximal access to the array as linking pulls + * more functions in that access the array. + */ + var->max_array_access = + MAX2(var->max_array_access, ir->var->max_array_access); + + if (var->type->length == 0 && ir->var->type->length != 0) + var->type = ir->var->type; + } + + ir->var = var; + } + + return visit_continue; + } + + /** Was function linking successful? */ + bool success; + +private: + /** + * Shader program being linked + * + * This is only used for logging error messages. + */ + gl_shader_program *prog; + + /** List of shaders available for linking. */ + gl_shader **shader_list; + + /** Number of shaders available for linking. */ + unsigned num_shaders; + + /** + * Final linked shader + * + * This is used two ways. It is used to find global variables in the + * linked shader that are accessed by the function. It is also used to add + * global variables from the shader where the function originated. + */ + gl_shader *linked; + + /** + * Table of variables local to the function. + */ + hash_table *locals; +}; + + +/** + * Searches a list of shaders for a particular function definition + */ +ir_function_signature * +find_matching_signature(const char *name, const exec_list *actual_parameters, + gl_shader **shader_list, unsigned num_shaders) +{ + for (unsigned i = 0; i < num_shaders; i++) { + ir_function *const f = shader_list[i]->symbols->get_function(name); + + if (f == NULL) + continue; + + ir_function_signature *sig = f->matching_signature(actual_parameters); + + if ((sig == NULL) || !sig->is_defined) + continue; + + return sig; + } + + return NULL; +} + + +bool +link_function_calls(gl_shader_program *prog, gl_shader *main, + gl_shader **shader_list, unsigned num_shaders) +{ + call_link_visitor v(prog, main, shader_list, num_shaders); + + v.run(main->ir); + return v.success; +} diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp index 9f4b4149c..bf7a56353 100644 --- a/mesalib/src/glsl/linker.cpp +++ b/mesalib/src/glsl/linker.cpp @@ -1,1676 +1,1672 @@ -/* - * 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 linker.cpp - * GLSL linker implementation - * - * Given a set of shaders that are to be linked to generate a final program, - * there are three distinct stages. - * - * In the first stage shaders are partitioned into groups based on the shader - * type. All shaders of a particular type (e.g., vertex shaders) are linked - * together. - * - * - Undefined references in each shader are resolve to definitions in - * another shader. - * - Types and qualifiers of uniforms, outputs, and global variables defined - * in multiple shaders with the same name are verified to be the same. - * - Initializers for uniforms and global variables defined - * in multiple shaders with the same name are verified to be the same. - * - * The result, in the terminology of the GLSL spec, is a set of shader - * executables for each processing unit. - * - * After the first stage is complete, a series of semantic checks are performed - * on each of the shader executables. - * - * - Each shader executable must define a \c main function. - * - Each vertex shader executable must write to \c gl_Position. - * - Each fragment shader executable must write to either \c gl_FragData or - * \c gl_FragColor. - * - * In the final stage individual shader executables are linked to create a - * complete exectuable. - * - * - Types of uniforms defined in multiple shader stages with the same name - * are verified to be the same. - * - Initializers for uniforms defined in multiple shader stages with the - * same name are verified to be the same. - * - Types and qualifiers of outputs defined in one stage are verified to - * be the same as the types and qualifiers of inputs defined with the same - * name in a later stage. - * - * \author Ian Romanick - */ -#include -#include -#include -#include - -#include "main/core.h" -#include "glsl_symbol_table.h" -#include "ir.h" -#include "program.h" -#include "program/hash_table.h" -#include "linker.h" -#include "ir_optimization.h" - -extern "C" { -#include "main/shaderobj.h" -} - -/** - * Visitor that determines whether or not a variable is ever written. - */ -class find_assignment_visitor : public ir_hierarchical_visitor { -public: - find_assignment_visitor(const char *name) - : name(name), found(false) - { - /* empty */ - } - - virtual ir_visitor_status visit_enter(ir_assignment *ir) - { - ir_variable *const var = ir->lhs->variable_referenced(); - - if (strcmp(name, var->name) == 0) { - found = true; - return visit_stop; - } - - return visit_continue_with_parent; - } - - virtual ir_visitor_status visit_enter(ir_call *ir) - { - exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator(); - foreach_iter(exec_list_iterator, iter, *ir) { - ir_rvalue *param_rval = (ir_rvalue *)iter.get(); - ir_variable *sig_param = (ir_variable *)sig_iter.get(); - - if (sig_param->mode == ir_var_out || - sig_param->mode == ir_var_inout) { - ir_variable *var = param_rval->variable_referenced(); - if (var && strcmp(name, var->name) == 0) { - found = true; - return visit_stop; - } - } - sig_iter.next(); - } - - return visit_continue_with_parent; - } - - bool variable_found() - { - return found; - } - -private: - const char *name; /**< Find writes to a variable with this name. */ - bool found; /**< Was a write to the variable found? */ -}; - - -/** - * Visitor that determines whether or not a variable is ever read. - */ -class find_deref_visitor : public ir_hierarchical_visitor { -public: - find_deref_visitor(const char *name) - : name(name), found(false) - { - /* empty */ - } - - virtual ir_visitor_status visit(ir_dereference_variable *ir) - { - if (strcmp(this->name, ir->var->name) == 0) { - this->found = true; - return visit_stop; - } - - return visit_continue; - } - - bool variable_found() const - { - return this->found; - } - -private: - const char *name; /**< Find writes to a variable with this name. */ - bool found; /**< Was a write to the variable found? */ -}; - - -void -linker_error_printf(gl_shader_program *prog, const char *fmt, ...) -{ - va_list ap; - - prog->InfoLog = talloc_strdup_append(prog->InfoLog, "error: "); - va_start(ap, fmt); - prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, ap); - va_end(ap); -} - - -void -invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode, - int generic_base) -{ - foreach_list(node, sh->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != (unsigned) mode)) - continue; - - /* Only assign locations for generic attributes / varyings / etc. - */ - if ((var->location >= generic_base) && !var->explicit_location) - var->location = -1; - } -} - - -/** - * Determine the number of attribute slots required for a particular type - * - * This code is here because it implements the language rules of a specific - * GLSL version. Since it's a property of the language and not a property of - * types in general, it doesn't really belong in glsl_type. - */ -unsigned -count_attribute_slots(const glsl_type *t) -{ - /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec: - * - * "A scalar input counts the same amount against this limit as a vec4, - * so applications may want to consider packing groups of four - * unrelated float inputs together into a vector to better utilize the - * capabilities of the underlying hardware. A matrix input will use up - * multiple locations. The number of locations used will equal the - * number of columns in the matrix." - * - * The spec does not explicitly say how arrays are counted. However, it - * should be safe to assume the total number of slots consumed by an array - * is the number of entries in the array multiplied by the number of slots - * consumed by a single element of the array. - */ - - if (t->is_array()) - return t->array_size() * count_attribute_slots(t->element_type()); - - if (t->is_matrix()) - return t->matrix_columns; - - return 1; -} - - -/** - * Verify that a vertex shader executable meets all semantic requirements - * - * \param shader Vertex shader executable to be verified - */ -bool -validate_vertex_shader_executable(struct gl_shader_program *prog, - struct gl_shader *shader) -{ - if (shader == NULL) - return true; - - find_assignment_visitor find("gl_Position"); - find.run(shader->ir); - if (!find.variable_found()) { - linker_error_printf(prog, - "vertex shader does not write to `gl_Position'\n"); - return false; - } - - return true; -} - - -/** - * Verify that a fragment shader executable meets all semantic requirements - * - * \param shader Fragment shader executable to be verified - */ -bool -validate_fragment_shader_executable(struct gl_shader_program *prog, - struct gl_shader *shader) -{ - if (shader == NULL) - return true; - - find_assignment_visitor frag_color("gl_FragColor"); - find_assignment_visitor frag_data("gl_FragData"); - - frag_color.run(shader->ir); - frag_data.run(shader->ir); - - if (frag_color.variable_found() && frag_data.variable_found()) { - linker_error_printf(prog, "fragment shader writes to both " - "`gl_FragColor' and `gl_FragData'\n"); - return false; - } - - return true; -} - - -/** - * Generate a string describing the mode of a variable - */ -static const char * -mode_string(const ir_variable *var) -{ - switch (var->mode) { - case ir_var_auto: - return (var->read_only) ? "global constant" : "global variable"; - - case ir_var_uniform: return "uniform"; - case ir_var_in: return "shader input"; - case ir_var_out: return "shader output"; - case ir_var_inout: return "shader inout"; - - case ir_var_temporary: - default: - assert(!"Should not get here."); - return "invalid variable"; - } -} - - -/** - * Perform validation of global variables used across multiple shaders - */ -bool -cross_validate_globals(struct gl_shader_program *prog, - struct gl_shader **shader_list, - unsigned num_shaders, - bool uniforms_only) -{ - /* Examine all of the uniforms in all of the shaders and cross validate - * them. - */ - glsl_symbol_table variables; - for (unsigned i = 0; i < num_shaders; i++) { - if (shader_list[i] == NULL) - continue; - - foreach_list(node, shader_list[i]->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if (var == NULL) - continue; - - if (uniforms_only && (var->mode != ir_var_uniform)) - continue; - - /* Don't cross validate temporaries that are at global scope. These - * will eventually get pulled into the shaders 'main'. - */ - if (var->mode == ir_var_temporary) - continue; - - /* If a global with this name has already been seen, verify that the - * new instance has the same type. In addition, if the globals have - * initializers, the values of the initializers must be the same. - */ - ir_variable *const existing = variables.get_variable(var->name); - if (existing != NULL) { - if (var->type != existing->type) { - /* Consider the types to be "the same" if both types are arrays - * of the same type and one of the arrays is implicitly sized. - * In addition, set the type of the linked variable to the - * explicitly sized array. - */ - if (var->type->is_array() - && existing->type->is_array() - && (var->type->fields.array == existing->type->fields.array) - && ((var->type->length == 0) - || (existing->type->length == 0))) { - if (existing->type->length == 0) { - existing->type = var->type; - existing->max_array_access = - MAX2(existing->max_array_access, - var->max_array_access); - } - } else { - linker_error_printf(prog, "%s `%s' declared as type " - "`%s' and type `%s'\n", - mode_string(var), - var->name, var->type->name, - existing->type->name); - return false; - } - } - - if (var->explicit_location) { - if (existing->explicit_location - && (var->location != existing->location)) { - linker_error_printf(prog, "explicit locations for %s " - "`%s' have differing values\n", - mode_string(var), var->name); - return false; - } - - existing->location = var->location; - existing->explicit_location = true; - } - - /* FINISHME: Handle non-constant initializers. - */ - if (var->constant_value != NULL) { - if (existing->constant_value != NULL) { - if (!var->constant_value->has_value(existing->constant_value)) { - linker_error_printf(prog, "initializers for %s " - "`%s' have differing values\n", - mode_string(var), var->name); - return false; - } - } else - /* If the first-seen instance of a particular uniform did not - * have an initializer but a later instance does, copy the - * initializer to the version stored in the symbol table. - */ - /* FINISHME: This is wrong. The constant_value field should - * FINISHME: not be modified! Imagine a case where a shader - * FINISHME: without an initializer is linked in two different - * FINISHME: programs with shaders that have differing - * FINISHME: initializers. Linking with the first will - * FINISHME: modify the shader, and linking with the second - * FINISHME: will fail. - */ - existing->constant_value = - var->constant_value->clone(talloc_parent(existing), NULL); - } - - if (existing->invariant != var->invariant) { - linker_error_printf(prog, "declarations for %s `%s' have " - "mismatching invariant qualifiers\n", - mode_string(var), var->name); - return false; - } - if (existing->centroid != var->centroid) { - linker_error_printf(prog, "declarations for %s `%s' have " - "mismatching centroid qualifiers\n", - mode_string(var), var->name); - return false; - } - } else - variables.add_variable(var); - } - } - - return true; -} - - -/** - * Perform validation of uniforms used across multiple shader stages - */ -bool -cross_validate_uniforms(struct gl_shader_program *prog) -{ - return cross_validate_globals(prog, prog->_LinkedShaders, - MESA_SHADER_TYPES, true); -} - - -/** - * Validate that outputs from one stage match inputs of another - */ -bool -cross_validate_outputs_to_inputs(struct gl_shader_program *prog, - gl_shader *producer, gl_shader *consumer) -{ - glsl_symbol_table parameters; - /* FINISHME: Figure these out dynamically. */ - const char *const producer_stage = "vertex"; - const char *const consumer_stage = "fragment"; - - /* Find all shader outputs in the "producer" stage. - */ - foreach_list(node, producer->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - /* FINISHME: For geometry shaders, this should also look for inout - * FINISHME: variables. - */ - if ((var == NULL) || (var->mode != ir_var_out)) - continue; - - parameters.add_variable(var); - } - - - /* Find all shader inputs in the "consumer" stage. Any variables that have - * matching outputs already in the symbol table must have the same type and - * qualifiers. - */ - foreach_list(node, consumer->ir) { - ir_variable *const input = ((ir_instruction *) node)->as_variable(); - - /* FINISHME: For geometry shaders, this should also look for inout - * FINISHME: variables. - */ - if ((input == NULL) || (input->mode != ir_var_in)) - continue; - - ir_variable *const output = parameters.get_variable(input->name); - if (output != NULL) { - /* Check that the types match between stages. - */ - if (input->type != output->type) { - /* There is a bit of a special case for gl_TexCoord. This - * built-in is unsized by default. Appliations that variable - * access it must redeclare it with a size. There is some - * language in the GLSL spec that implies the fragment shader - * and vertex shader do not have to agree on this size. Other - * driver behave this way, and one or two applications seem to - * rely on it. - * - * Neither declaration needs to be modified here because the array - * sizes are fixed later when update_array_sizes is called. - * - * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec: - * - * "Unlike user-defined varying variables, the built-in - * varying variables don't have a strict one-to-one - * correspondence between the vertex language and the - * fragment language." - */ - if (!output->type->is_array() - || (strncmp("gl_", output->name, 3) != 0)) { - linker_error_printf(prog, - "%s shader output `%s' declared as " - "type `%s', but %s shader input declared " - "as type `%s'\n", - producer_stage, output->name, - output->type->name, - consumer_stage, input->type->name); - return false; - } - } - - /* Check that all of the qualifiers match between stages. - */ - if (input->centroid != output->centroid) { - linker_error_printf(prog, - "%s shader output `%s' %s centroid qualifier, " - "but %s shader input %s centroid qualifier\n", - producer_stage, - output->name, - (output->centroid) ? "has" : "lacks", - consumer_stage, - (input->centroid) ? "has" : "lacks"); - return false; - } - - if (input->invariant != output->invariant) { - linker_error_printf(prog, - "%s shader output `%s' %s invariant qualifier, " - "but %s shader input %s invariant qualifier\n", - producer_stage, - output->name, - (output->invariant) ? "has" : "lacks", - consumer_stage, - (input->invariant) ? "has" : "lacks"); - return false; - } - - if (input->interpolation != output->interpolation) { - linker_error_printf(prog, - "%s shader output `%s' specifies %s " - "interpolation qualifier, " - "but %s shader input specifies %s " - "interpolation qualifier\n", - producer_stage, - output->name, - output->interpolation_string(), - consumer_stage, - input->interpolation_string()); - return false; - } - } - } - - return true; -} - - -/** - * Populates a shaders symbol table with all global declarations - */ -static void -populate_symbol_table(gl_shader *sh) -{ - sh->symbols = new(sh) glsl_symbol_table; - - foreach_list(node, sh->ir) { - ir_instruction *const inst = (ir_instruction *) node; - ir_variable *var; - ir_function *func; - - if ((func = inst->as_function()) != NULL) { - sh->symbols->add_function(func); - } else if ((var = inst->as_variable()) != NULL) { - sh->symbols->add_variable(var); - } - } -} - - -/** - * Remap variables referenced in an instruction tree - * - * This is used when instruction trees are cloned from one shader and placed in - * another. These trees will contain references to \c ir_variable nodes that - * do not exist in the target shader. This function finds these \c ir_variable - * references and replaces the references with matching variables in the target - * shader. - * - * If there is no matching variable in the target shader, a clone of the - * \c ir_variable is made and added to the target shader. The new variable is - * added to \b both the instruction stream and the symbol table. - * - * \param inst IR tree that is to be processed. - * \param symbols Symbol table containing global scope symbols in the - * linked shader. - * \param instructions Instruction stream where new variable declarations - * should be added. - */ -void -remap_variables(ir_instruction *inst, struct gl_shader *target, - hash_table *temps) -{ - class remap_visitor : public ir_hierarchical_visitor { - public: - remap_visitor(struct gl_shader *target, - hash_table *temps) - { - this->target = target; - this->symbols = target->symbols; - this->instructions = target->ir; - this->temps = temps; - } - - virtual ir_visitor_status visit(ir_dereference_variable *ir) - { - if (ir->var->mode == ir_var_temporary) { - ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var); - - assert(var != NULL); - ir->var = var; - return visit_continue; - } - - ir_variable *const existing = - this->symbols->get_variable(ir->var->name); - if (existing != NULL) - ir->var = existing; - else { - ir_variable *copy = ir->var->clone(this->target, NULL); - - this->symbols->add_variable(copy); - this->instructions->push_head(copy); - ir->var = copy; - } - - return visit_continue; - } - - private: - struct gl_shader *target; - glsl_symbol_table *symbols; - exec_list *instructions; - hash_table *temps; - }; - - remap_visitor v(target, temps); - - inst->accept(&v); -} - - -/** - * Move non-declarations from one instruction stream to another - * - * The intended usage pattern of this function is to pass the pointer to the - * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node - * pointer) for \c last and \c false for \c make_copies on the first - * call. Successive calls pass the return value of the previous call for - * \c last and \c true for \c make_copies. - * - * \param instructions Source instruction stream - * \param last Instruction after which new instructions should be - * inserted in the target instruction stream - * \param make_copies Flag selecting whether instructions in \c instructions - * should be copied (via \c ir_instruction::clone) into the - * target list or moved. - * - * \return - * The new "last" instruction in the target instruction stream. This pointer - * is suitable for use as the \c last parameter of a later call to this - * function. - */ -exec_node * -move_non_declarations(exec_list *instructions, exec_node *last, - bool make_copies, gl_shader *target) -{ - hash_table *temps = NULL; - - if (make_copies) - temps = hash_table_ctor(0, hash_table_pointer_hash, - hash_table_pointer_compare); - - foreach_list_safe(node, instructions) { - ir_instruction *inst = (ir_instruction *) node; - - if (inst->as_function()) - continue; - - ir_variable *var = inst->as_variable(); - if ((var != NULL) && (var->mode != ir_var_temporary)) - continue; - - assert(inst->as_assignment() - || ((var != NULL) && (var->mode == ir_var_temporary))); - - if (make_copies) { - inst = inst->clone(target, NULL); - - if (var != NULL) - hash_table_insert(temps, inst, var); - else - remap_variables(inst, target, temps); - } else { - inst->remove(); - } - - last->insert_after(inst); - last = inst; - } - - if (make_copies) - hash_table_dtor(temps); - - return last; -} - -/** - * Get the function signature for main from a shader - */ -static ir_function_signature * -get_main_function_signature(gl_shader *sh) -{ - ir_function *const f = sh->symbols->get_function("main"); - if (f != NULL) { - exec_list void_parameters; - - /* Look for the 'void main()' signature and ensure that it's defined. - * This keeps the linker from accidentally pick a shader that just - * contains a prototype for main. - * - * We don't have to check for multiple definitions of main (in multiple - * shaders) because that would have already been caught above. - */ - ir_function_signature *sig = f->matching_signature(&void_parameters); - if ((sig != NULL) && sig->is_defined) { - return sig; - } - } - - return NULL; -} - - -/** - * Combine a group of shaders for a single stage to generate a linked shader - * - * \note - * If this function is supplied a single shader, it is cloned, and the new - * shader is returned. - */ -static struct gl_shader * -link_intrastage_shaders(void *mem_ctx, - struct gl_context *ctx, - struct gl_shader_program *prog, - struct gl_shader **shader_list, - unsigned num_shaders) -{ - /* Check that global variables defined in multiple shaders are consistent. - */ - if (!cross_validate_globals(prog, shader_list, num_shaders, false)) - return NULL; - - /* Check that there is only a single definition of each function signature - * across all shaders. - */ - for (unsigned i = 0; i < (num_shaders - 1); i++) { - foreach_list(node, shader_list[i]->ir) { - ir_function *const f = ((ir_instruction *) node)->as_function(); - - if (f == NULL) - continue; - - for (unsigned j = i + 1; j < num_shaders; j++) { - ir_function *const other = - shader_list[j]->symbols->get_function(f->name); - - /* If the other shader has no function (and therefore no function - * signatures) with the same name, skip to the next shader. - */ - if (other == NULL) - continue; - - foreach_iter (exec_list_iterator, iter, *f) { - ir_function_signature *sig = - (ir_function_signature *) iter.get(); - - if (!sig->is_defined || sig->is_builtin) - continue; - - ir_function_signature *other_sig = - other->exact_matching_signature(& sig->parameters); - - if ((other_sig != NULL) && other_sig->is_defined - && !other_sig->is_builtin) { - linker_error_printf(prog, - "function `%s' is multiply defined", - f->name); - return NULL; - } - } - } - } - } - - /* Find the shader that defines main, and make a clone of it. - * - * Starting with the clone, search for undefined references. If one is - * found, find the shader that defines it. Clone the reference and add - * it to the shader. Repeat until there are no undefined references or - * until a reference cannot be resolved. - */ - gl_shader *main = NULL; - for (unsigned i = 0; i < num_shaders; i++) { - if (get_main_function_signature(shader_list[i]) != NULL) { - main = shader_list[i]; - break; - } - } - - if (main == NULL) { - linker_error_printf(prog, "%s shader lacks `main'\n", - (shader_list[0]->Type == GL_VERTEX_SHADER) - ? "vertex" : "fragment"); - return NULL; - } - - gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type); - linked->ir = new(linked) exec_list; - clone_ir_list(mem_ctx, linked->ir, main->ir); - - populate_symbol_table(linked); - - /* The a pointer to the main function in the final linked shader (i.e., the - * copy of the original shader that contained the main function). - */ - ir_function_signature *const main_sig = get_main_function_signature(linked); - - /* Move any instructions other than variable declarations or function - * declarations into main. - */ - exec_node *insertion_point = - move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false, - linked); - - for (unsigned i = 0; i < num_shaders; i++) { - if (shader_list[i] == main) - continue; - - insertion_point = move_non_declarations(shader_list[i]->ir, - insertion_point, true, linked); - } - - /* Resolve initializers for global variables in the linked shader. - */ - unsigned num_linking_shaders = num_shaders; - for (unsigned i = 0; i < num_shaders; i++) - num_linking_shaders += shader_list[i]->num_builtins_to_link; - - gl_shader **linking_shaders = - (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *)); - - memcpy(linking_shaders, shader_list, - sizeof(linking_shaders[0]) * num_shaders); - - unsigned idx = num_shaders; - for (unsigned i = 0; i < num_shaders; i++) { - memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link, - sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link); - idx += shader_list[i]->num_builtins_to_link; - } - - assert(idx == num_linking_shaders); - - if (!link_function_calls(prog, linked, linking_shaders, - num_linking_shaders)) { - ctx->Driver.DeleteShader(ctx, linked); - linked = NULL; - } - - free(linking_shaders); - - /* Make a pass over all global variables to ensure that arrays with - * unspecified sizes have a size specified. The size is inferred from the - * max_array_access field. - */ - if (linked != NULL) { - foreach_list(node, linked->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if (var == NULL) - continue; - - if ((var->mode != ir_var_auto) && (var->mode != ir_var_temporary)) - continue; - - if (!var->type->is_array() || (var->type->length != 0)) - continue; - - const glsl_type *type = - glsl_type::get_array_instance(var->type->fields.array, - var->max_array_access); - - assert(type != NULL); - var->type = type; - } - } - - return linked; -} - - -struct uniform_node { - exec_node link; - struct gl_uniform *u; - unsigned slots; -}; - -/** - * Update the sizes of linked shader uniform arrays to the maximum - * array index used. - * - * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec: - * - * If one or more elements of an array are active, - * GetActiveUniform will return the name of the array in name, - * subject to the restrictions listed above. The type of the array - * is returned in type. The size parameter contains the highest - * array element index used, plus one. The compiler or linker - * determines the highest index used. There will be only one - * active uniform reported by the GL per uniform array. - - */ -static void -update_array_sizes(struct gl_shader_program *prog) -{ - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - foreach_list(node, prog->_LinkedShaders[i]->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_uniform && - var->mode != ir_var_in && - var->mode != ir_var_out) || - !var->type->is_array()) - continue; - - unsigned int size = var->max_array_access; - for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) { - if (prog->_LinkedShaders[j] == NULL) - continue; - - foreach_list(node2, prog->_LinkedShaders[j]->ir) { - ir_variable *other_var = ((ir_instruction *) node2)->as_variable(); - if (!other_var) - continue; - - if (strcmp(var->name, other_var->name) == 0 && - other_var->max_array_access > size) { - size = other_var->max_array_access; - } - } - } - - if (size + 1 != var->type->fields.array->length) { - var->type = glsl_type::get_array_instance(var->type->fields.array, - size + 1); - /* FINISHME: We should update the types of array - * dereferences of this variable now. - */ - } - } - } -} - -static void -add_uniform(void *mem_ctx, exec_list *uniforms, struct hash_table *ht, - const char *name, const glsl_type *type, GLenum shader_type, - unsigned *next_shader_pos, unsigned *total_uniforms) -{ - if (type->is_record()) { - for (unsigned int i = 0; i < type->length; i++) { - const glsl_type *field_type = type->fields.structure[i].type; - char *field_name = talloc_asprintf(mem_ctx, "%s.%s", name, - type->fields.structure[i].name); - - add_uniform(mem_ctx, uniforms, ht, field_name, field_type, - shader_type, next_shader_pos, total_uniforms); - } - } else { - uniform_node *n = (uniform_node *) hash_table_find(ht, name); - unsigned int vec4_slots; - const glsl_type *array_elem_type = NULL; - - if (type->is_array()) { - array_elem_type = type->fields.array; - /* Array of structures. */ - if (array_elem_type->is_record()) { - for (unsigned int i = 0; i < type->length; i++) { - char *elem_name = talloc_asprintf(mem_ctx, "%s[%d]", name, i); - add_uniform(mem_ctx, uniforms, ht, elem_name, array_elem_type, - shader_type, next_shader_pos, total_uniforms); - } - return; - } - } - - /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out - * vectors to vec4 slots. - */ - if (type->is_array()) { - if (array_elem_type->is_sampler()) - vec4_slots = type->length; - else - vec4_slots = type->length * array_elem_type->matrix_columns; - } else if (type->is_sampler()) { - vec4_slots = 1; - } else { - vec4_slots = type->matrix_columns; - } - - if (n == NULL) { - n = (uniform_node *) calloc(1, sizeof(struct uniform_node)); - n->u = (gl_uniform *) calloc(1, sizeof(struct gl_uniform)); - n->slots = vec4_slots; - - n->u->Name = strdup(name); - n->u->Type = type; - n->u->VertPos = -1; - n->u->FragPos = -1; - n->u->GeomPos = -1; - (*total_uniforms)++; - - hash_table_insert(ht, n, name); - uniforms->push_tail(& n->link); - } - - switch (shader_type) { - case GL_VERTEX_SHADER: - n->u->VertPos = *next_shader_pos; - break; - case GL_FRAGMENT_SHADER: - n->u->FragPos = *next_shader_pos; - break; - case GL_GEOMETRY_SHADER: - n->u->GeomPos = *next_shader_pos; - break; - } - - (*next_shader_pos) += vec4_slots; - } -} - -void -assign_uniform_locations(struct gl_shader_program *prog) -{ - /* */ - exec_list uniforms; - unsigned total_uniforms = 0; - hash_table *ht = hash_table_ctor(32, hash_table_string_hash, - hash_table_string_compare); - void *mem_ctx = talloc_new(NULL); - - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - unsigned next_position = 0; - - foreach_list(node, prog->_LinkedShaders[i]->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_uniform)) - continue; - - if (strncmp(var->name, "gl_", 3) == 0) { - /* At the moment, we don't allocate uniform locations for - * builtin uniforms. It's permitted by spec, and we'll - * likely switch to doing that at some point, but not yet. - */ - continue; - } - - var->location = next_position; - add_uniform(mem_ctx, &uniforms, ht, var->name, var->type, - prog->_LinkedShaders[i]->Type, - &next_position, &total_uniforms); - } - } - - talloc_free(mem_ctx); - - gl_uniform_list *ul = (gl_uniform_list *) - calloc(1, sizeof(gl_uniform_list)); - - ul->Size = total_uniforms; - ul->NumUniforms = total_uniforms; - ul->Uniforms = (gl_uniform *) calloc(total_uniforms, sizeof(gl_uniform)); - - unsigned idx = 0; - uniform_node *next; - for (uniform_node *node = (uniform_node *) uniforms.head - ; node->link.next != NULL - ; node = next) { - next = (uniform_node *) node->link.next; - - node->link.remove(); - memcpy(&ul->Uniforms[idx], node->u, sizeof(gl_uniform)); - idx++; - - free(node->u); - free(node); - } - - hash_table_dtor(ht); - - prog->Uniforms = ul; -} - - -/** - * Find a contiguous set of available bits in a bitmask - * - * \param used_mask Bits representing used (1) and unused (0) locations - * \param needed_count Number of contiguous bits needed. - * - * \return - * Base location of the available bits on success or -1 on failure. - */ -int -find_available_slots(unsigned used_mask, unsigned needed_count) -{ - unsigned needed_mask = (1 << needed_count) - 1; - const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count; - - /* The comparison to 32 is redundant, but without it GCC emits "warning: - * cannot optimize possibly infinite loops" for the loop below. - */ - if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32)) - return -1; - - for (int i = 0; i <= max_bit_to_test; i++) { - if ((needed_mask & ~used_mask) == needed_mask) - return i; - - needed_mask <<= 1; - } - - return -1; -} - - -bool -assign_attribute_locations(gl_shader_program *prog, unsigned max_attribute_index) -{ - /* Mark invalid attribute locations as being used. - */ - unsigned used_locations = (max_attribute_index >= 32) - ? ~0 : ~((1 << max_attribute_index) - 1); - - gl_shader *const sh = prog->_LinkedShaders[0]; - assert(sh->Type == GL_VERTEX_SHADER); - - /* Operate in a total of four passes. - * - * 1. Invalidate the location assignments for all vertex shader inputs. - * - * 2. Assign locations for inputs that have user-defined (via - * glBindVertexAttribLocation) locatoins. - * - * 3. Sort the attributes without assigned locations by number of slots - * required in decreasing order. Fragmentation caused by attribute - * locations assigned by the application may prevent large attributes - * from having enough contiguous space. - * - * 4. Assign locations to any inputs without assigned locations. - */ - - invalidate_variable_locations(sh, ir_var_in, VERT_ATTRIB_GENERIC0); - - if (prog->Attributes != NULL) { - for (unsigned i = 0; i < prog->Attributes->NumParameters; i++) { - ir_variable *const var = - sh->symbols->get_variable(prog->Attributes->Parameters[i].Name); - - /* Note: attributes that occupy multiple slots, such as arrays or - * matrices, may appear in the attrib array multiple times. - */ - if ((var == NULL) || (var->location != -1)) - continue; - - /* From page 61 of the OpenGL 4.0 spec: - * - * "LinkProgram will fail if the attribute bindings assigned by - * BindAttribLocation do not leave not enough space to assign a - * location for an active matrix attribute or an active attribute - * array, both of which require multiple contiguous generic - * attributes." - * - * Previous versions of the spec contain similar language but omit the - * bit about attribute arrays. - * - * Page 61 of the OpenGL 4.0 spec also says: - * - * "It is possible for an application to bind more than one - * attribute name to the same location. This is referred to as - * aliasing. This will only work if only one of the aliased - * attributes is active in the executable program, or if no path - * through the shader consumes more than one attribute of a set - * of attributes aliased to the same location. A link error can - * occur if the linker determines that every path through the - * shader consumes multiple aliased attributes, but - * implementations are not required to generate an error in this - * case." - * - * These two paragraphs are either somewhat contradictory, or I don't - * fully understand one or both of them. - */ - /* FINISHME: The code as currently written does not support attribute - * FINISHME: location aliasing (see comment above). - */ - const int attr = prog->Attributes->Parameters[i].StateIndexes[0]; - const unsigned slots = count_attribute_slots(var->type); - - /* Mask representing the contiguous slots that will be used by this - * attribute. - */ - const unsigned use_mask = (1 << slots) - 1; - - /* Generate a link error if the set of bits requested for this - * attribute overlaps any previously allocated bits. - */ - if ((~(use_mask << attr) & used_locations) != used_locations) { - linker_error_printf(prog, - "insufficient contiguous attribute locations " - "available for vertex shader input `%s'", - var->name); - return false; - } - - var->location = VERT_ATTRIB_GENERIC0 + attr; - used_locations |= (use_mask << attr); - } - } - - /* Temporary storage for the set of attributes that need locations assigned. - */ - struct temp_attr { - unsigned slots; - ir_variable *var; - - /* Used below in the call to qsort. */ - static int compare(const void *a, const void *b) - { - const temp_attr *const l = (const temp_attr *) a; - const temp_attr *const r = (const temp_attr *) b; - - /* Reversed because we want a descending order sort below. */ - return r->slots - l->slots; - } - } to_assign[16]; - - unsigned num_attr = 0; - - foreach_list(node, sh->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_in)) - continue; - - if (var->explicit_location) { - const unsigned slots = count_attribute_slots(var->type); - const unsigned use_mask = (1 << slots) - 1; - const int attr = var->location - VERT_ATTRIB_GENERIC0; - - if ((var->location >= (int)(max_attribute_index + VERT_ATTRIB_GENERIC0)) - || (var->location < 0)) { - linker_error_printf(prog, - "invalid explicit location %d specified for " - "`%s'\n", - (var->location < 0) ? var->location : attr, - var->name); - return false; - } else if (var->location >= VERT_ATTRIB_GENERIC0) { - used_locations |= (use_mask << attr); - } - } - - /* The location was explicitly assigned, nothing to do here. - */ - if (var->location != -1) - continue; - - to_assign[num_attr].slots = count_attribute_slots(var->type); - to_assign[num_attr].var = var; - num_attr++; - } - - /* If all of the attributes were assigned locations by the application (or - * are built-in attributes with fixed locations), return early. This should - * be the common case. - */ - if (num_attr == 0) - return true; - - qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare); - - /* VERT_ATTRIB_GENERIC0 is a psdueo-alias for VERT_ATTRIB_POS. It can only - * be explicitly assigned by via glBindAttribLocation. Mark it as reserved - * to prevent it from being automatically allocated below. - */ - find_deref_visitor find("gl_Vertex"); - find.run(sh->ir); - if (find.variable_found()) - used_locations |= (1 << 0); - - for (unsigned i = 0; i < num_attr; i++) { - /* Mask representing the contiguous slots that will be used by this - * attribute. - */ - const unsigned use_mask = (1 << to_assign[i].slots) - 1; - - int location = find_available_slots(used_locations, to_assign[i].slots); - - if (location < 0) { - linker_error_printf(prog, - "insufficient contiguous attribute locations " - "available for vertex shader input `%s'", - to_assign[i].var->name); - return false; - } - - to_assign[i].var->location = VERT_ATTRIB_GENERIC0 + location; - used_locations |= (use_mask << location); - } - - return true; -} - - -/** - * Demote shader inputs and outputs that are not used in other stages - */ -void -demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode) -{ - foreach_list(node, sh->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != int(mode))) - continue; - - /* A shader 'in' or 'out' variable is only really an input or output if - * its value is used by other shader stages. This will cause the variable - * to have a location assigned. - */ - if (var->location == -1) { - var->mode = ir_var_auto; - } - } -} - - -void -assign_varying_locations(struct gl_shader_program *prog, - gl_shader *producer, gl_shader *consumer) -{ - /* FINISHME: Set dynamically when geometry shader support is added. */ - unsigned output_index = VERT_RESULT_VAR0; - unsigned input_index = FRAG_ATTRIB_VAR0; - - /* Operate in a total of three passes. - * - * 1. Assign locations for any matching inputs and outputs. - * - * 2. Mark output variables in the producer that do not have locations as - * not being outputs. This lets the optimizer eliminate them. - * - * 3. Mark input variables in the consumer that do not have locations as - * not being inputs. This lets the optimizer eliminate them. - */ - - invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0); - invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0); - - foreach_list(node, producer->ir) { - ir_variable *const output_var = ((ir_instruction *) node)->as_variable(); - - if ((output_var == NULL) || (output_var->mode != ir_var_out) - || (output_var->location != -1)) - continue; - - ir_variable *const input_var = - consumer->symbols->get_variable(output_var->name); - - if ((input_var == NULL) || (input_var->mode != ir_var_in)) - continue; - - assert(input_var->location == -1); - - output_var->location = output_index; - input_var->location = input_index; - - /* FINISHME: Support for "varying" records in GLSL 1.50. */ - assert(!output_var->type->is_record()); - - if (output_var->type->is_array()) { - const unsigned slots = output_var->type->length - * output_var->type->fields.array->matrix_columns; - - output_index += slots; - input_index += slots; - } else { - const unsigned slots = output_var->type->matrix_columns; - - output_index += slots; - input_index += slots; - } - } - - foreach_list(node, consumer->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_in)) - continue; - - if (var->location == -1) { - if (prog->Version <= 120) { - /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec: - * - * Only those varying variables used (i.e. read) in - * the fragment shader executable must be written to - * by the vertex shader executable; declaring - * superfluous varying variables in a vertex shader is - * permissible. - * - * We interpret this text as meaning that the VS must - * write the variable for the FS to read it. See - * "glsl1-varying read but not written" in piglit. - */ - - linker_error_printf(prog, "fragment shader varying %s not written " - "by vertex shader\n.", var->name); - prog->LinkStatus = false; - } - - /* An 'in' variable is only really a shader input if its - * value is written by the previous stage. - */ - var->mode = ir_var_auto; - } - } -} - - -void -link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) -{ - void *mem_ctx = talloc_init("temporary linker context"); - - prog->LinkStatus = false; - prog->Validated = false; - prog->_Used = false; - - if (prog->InfoLog != NULL) - talloc_free(prog->InfoLog); - - prog->InfoLog = talloc_strdup(NULL, ""); - - /* Separate the shaders into groups based on their type. - */ - struct gl_shader **vert_shader_list; - unsigned num_vert_shaders = 0; - struct gl_shader **frag_shader_list; - unsigned num_frag_shaders = 0; - - vert_shader_list = (struct gl_shader **) - calloc(2 * prog->NumShaders, sizeof(struct gl_shader *)); - frag_shader_list = &vert_shader_list[prog->NumShaders]; - - unsigned min_version = UINT_MAX; - unsigned max_version = 0; - for (unsigned i = 0; i < prog->NumShaders; i++) { - min_version = MIN2(min_version, prog->Shaders[i]->Version); - max_version = MAX2(max_version, prog->Shaders[i]->Version); - - switch (prog->Shaders[i]->Type) { - case GL_VERTEX_SHADER: - vert_shader_list[num_vert_shaders] = prog->Shaders[i]; - num_vert_shaders++; - break; - case GL_FRAGMENT_SHADER: - frag_shader_list[num_frag_shaders] = prog->Shaders[i]; - num_frag_shaders++; - break; - case GL_GEOMETRY_SHADER: - /* FINISHME: Support geometry shaders. */ - assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER); - break; - } - } - - /* Previous to GLSL version 1.30, different compilation units could mix and - * match shading language versions. With GLSL 1.30 and later, the versions - * of all shaders must match. - */ - assert(min_version >= 100); - assert(max_version <= 130); - if ((max_version >= 130 || min_version == 100) - && min_version != max_version) { - linker_error_printf(prog, "all shaders must use same shading " - "language version\n"); - goto done; - } - - prog->Version = max_version; - - for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] != NULL) - ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]); - - prog->_LinkedShaders[i] = NULL; - } - - /* Link all shaders for a particular stage and validate the result. - */ - if (num_vert_shaders > 0) { - gl_shader *const sh = - link_intrastage_shaders(mem_ctx, ctx, prog, vert_shader_list, - num_vert_shaders); - - if (sh == NULL) - goto done; - - if (!validate_vertex_shader_executable(prog, sh)) - goto done; - - _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX], - sh); - } - - if (num_frag_shaders > 0) { - gl_shader *const sh = - link_intrastage_shaders(mem_ctx, ctx, prog, frag_shader_list, - num_frag_shaders); - - if (sh == NULL) - goto done; - - if (!validate_fragment_shader_executable(prog, sh)) - goto done; - - _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT], - sh); - } - - /* Here begins the inter-stage linking phase. Some initial validation is - * performed, then locations are assigned for uniforms, attributes, and - * varyings. - */ - if (cross_validate_uniforms(prog)) { - unsigned prev; - - for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { - if (prog->_LinkedShaders[prev] != NULL) - break; - } - - /* Validate the inputs of each stage with the output of the preceeding - * stage. - */ - for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - if (!cross_validate_outputs_to_inputs(prog, - prog->_LinkedShaders[prev], - prog->_LinkedShaders[i])) - goto done; - - prev = i; - } - - prog->LinkStatus = true; - } - - /* Do common optimization before assigning storage for attributes, - * uniforms, and varyings. Later optimization could possibly make - * some of that unused. - */ - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32)) - ; - } - - update_array_sizes(prog); - - assign_uniform_locations(prog); - - if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { - /* FINISHME: The value of the max_attribute_index parameter is - * FINISHME: implementation dependent based on the value of - * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be - * FINISHME: at least 16, so hardcode 16 for now. - */ - if (!assign_attribute_locations(prog, 16)) { - prog->LinkStatus = false; - goto done; - } - } - - unsigned prev; - for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { - if (prog->_LinkedShaders[prev] != NULL) - break; - } - - for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - assign_varying_locations(prog, - prog->_LinkedShaders[prev], - prog->_LinkedShaders[i]); - prev = i; - } - - if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { - demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX], - ir_var_out); - } - - if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) { - gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY]; - - demote_shader_inputs_and_outputs(sh, ir_var_in); - demote_shader_inputs_and_outputs(sh, ir_var_inout); - demote_shader_inputs_and_outputs(sh, ir_var_out); - } - - if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) { - gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT]; - - demote_shader_inputs_and_outputs(sh, ir_var_in); - } - - /* FINISHME: Assign fragment shader output locations. */ - -done: - free(vert_shader_list); - - for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { - if (prog->_LinkedShaders[i] == NULL) - continue; - - /* Retain any live IR, but trash the rest. */ - reparent_ir(prog->_LinkedShaders[i]->ir, prog->_LinkedShaders[i]->ir); - } - - talloc_free(mem_ctx); -} +/* + * 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 linker.cpp + * GLSL linker implementation + * + * Given a set of shaders that are to be linked to generate a final program, + * there are three distinct stages. + * + * In the first stage shaders are partitioned into groups based on the shader + * type. All shaders of a particular type (e.g., vertex shaders) are linked + * together. + * + * - Undefined references in each shader are resolve to definitions in + * another shader. + * - Types and qualifiers of uniforms, outputs, and global variables defined + * in multiple shaders with the same name are verified to be the same. + * - Initializers for uniforms and global variables defined + * in multiple shaders with the same name are verified to be the same. + * + * The result, in the terminology of the GLSL spec, is a set of shader + * executables for each processing unit. + * + * After the first stage is complete, a series of semantic checks are performed + * on each of the shader executables. + * + * - Each shader executable must define a \c main function. + * - Each vertex shader executable must write to \c gl_Position. + * - Each fragment shader executable must write to either \c gl_FragData or + * \c gl_FragColor. + * + * In the final stage individual shader executables are linked to create a + * complete exectuable. + * + * - Types of uniforms defined in multiple shader stages with the same name + * are verified to be the same. + * - Initializers for uniforms defined in multiple shader stages with the + * same name are verified to be the same. + * - Types and qualifiers of outputs defined in one stage are verified to + * be the same as the types and qualifiers of inputs defined with the same + * name in a later stage. + * + * \author Ian Romanick + */ +#include +#include +#include +#include + +#include "main/core.h" +#include "glsl_symbol_table.h" +#include "ir.h" +#include "program.h" +#include "program/hash_table.h" +#include "linker.h" +#include "ir_optimization.h" + +extern "C" { +#include "main/shaderobj.h" +} + +/** + * Visitor that determines whether or not a variable is ever written. + */ +class find_assignment_visitor : public ir_hierarchical_visitor { +public: + find_assignment_visitor(const char *name) + : name(name), found(false) + { + /* empty */ + } + + virtual ir_visitor_status visit_enter(ir_assignment *ir) + { + ir_variable *const var = ir->lhs->variable_referenced(); + + if (strcmp(name, var->name) == 0) { + found = true; + return visit_stop; + } + + return visit_continue_with_parent; + } + + virtual ir_visitor_status visit_enter(ir_call *ir) + { + exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator(); + foreach_iter(exec_list_iterator, iter, *ir) { + ir_rvalue *param_rval = (ir_rvalue *)iter.get(); + ir_variable *sig_param = (ir_variable *)sig_iter.get(); + + if (sig_param->mode == ir_var_out || + sig_param->mode == ir_var_inout) { + ir_variable *var = param_rval->variable_referenced(); + if (var && strcmp(name, var->name) == 0) { + found = true; + return visit_stop; + } + } + sig_iter.next(); + } + + return visit_continue_with_parent; + } + + bool variable_found() + { + return found; + } + +private: + const char *name; /**< Find writes to a variable with this name. */ + bool found; /**< Was a write to the variable found? */ +}; + + +/** + * Visitor that determines whether or not a variable is ever read. + */ +class find_deref_visitor : public ir_hierarchical_visitor { +public: + find_deref_visitor(const char *name) + : name(name), found(false) + { + /* empty */ + } + + virtual ir_visitor_status visit(ir_dereference_variable *ir) + { + if (strcmp(this->name, ir->var->name) == 0) { + this->found = true; + return visit_stop; + } + + return visit_continue; + } + + bool variable_found() const + { + return this->found; + } + +private: + const char *name; /**< Find writes to a variable with this name. */ + bool found; /**< Was a write to the variable found? */ +}; + + +void +linker_error_printf(gl_shader_program *prog, const char *fmt, ...) +{ + va_list ap; + + prog->InfoLog = talloc_strdup_append(prog->InfoLog, "error: "); + va_start(ap, fmt); + prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, ap); + va_end(ap); +} + + +void +invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode, + int generic_base) +{ + foreach_list(node, sh->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != (unsigned) mode)) + continue; + + /* Only assign locations for generic attributes / varyings / etc. + */ + if ((var->location >= generic_base) && !var->explicit_location) + var->location = -1; + } +} + + +/** + * Determine the number of attribute slots required for a particular type + * + * This code is here because it implements the language rules of a specific + * GLSL version. Since it's a property of the language and not a property of + * types in general, it doesn't really belong in glsl_type. + */ +unsigned +count_attribute_slots(const glsl_type *t) +{ + /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec: + * + * "A scalar input counts the same amount against this limit as a vec4, + * so applications may want to consider packing groups of four + * unrelated float inputs together into a vector to better utilize the + * capabilities of the underlying hardware. A matrix input will use up + * multiple locations. The number of locations used will equal the + * number of columns in the matrix." + * + * The spec does not explicitly say how arrays are counted. However, it + * should be safe to assume the total number of slots consumed by an array + * is the number of entries in the array multiplied by the number of slots + * consumed by a single element of the array. + */ + + if (t->is_array()) + return t->array_size() * count_attribute_slots(t->element_type()); + + if (t->is_matrix()) + return t->matrix_columns; + + return 1; +} + + +/** + * Verify that a vertex shader executable meets all semantic requirements + * + * \param shader Vertex shader executable to be verified + */ +bool +validate_vertex_shader_executable(struct gl_shader_program *prog, + struct gl_shader *shader) +{ + if (shader == NULL) + return true; + + find_assignment_visitor find("gl_Position"); + find.run(shader->ir); + if (!find.variable_found()) { + linker_error_printf(prog, + "vertex shader does not write to `gl_Position'\n"); + return false; + } + + return true; +} + + +/** + * Verify that a fragment shader executable meets all semantic requirements + * + * \param shader Fragment shader executable to be verified + */ +bool +validate_fragment_shader_executable(struct gl_shader_program *prog, + struct gl_shader *shader) +{ + if (shader == NULL) + return true; + + find_assignment_visitor frag_color("gl_FragColor"); + find_assignment_visitor frag_data("gl_FragData"); + + frag_color.run(shader->ir); + frag_data.run(shader->ir); + + if (frag_color.variable_found() && frag_data.variable_found()) { + linker_error_printf(prog, "fragment shader writes to both " + "`gl_FragColor' and `gl_FragData'\n"); + return false; + } + + return true; +} + + +/** + * Generate a string describing the mode of a variable + */ +static const char * +mode_string(const ir_variable *var) +{ + switch (var->mode) { + case ir_var_auto: + return (var->read_only) ? "global constant" : "global variable"; + + case ir_var_uniform: return "uniform"; + case ir_var_in: return "shader input"; + case ir_var_out: return "shader output"; + case ir_var_inout: return "shader inout"; + + case ir_var_temporary: + default: + assert(!"Should not get here."); + return "invalid variable"; + } +} + + +/** + * Perform validation of global variables used across multiple shaders + */ +bool +cross_validate_globals(struct gl_shader_program *prog, + struct gl_shader **shader_list, + unsigned num_shaders, + bool uniforms_only) +{ + /* Examine all of the uniforms in all of the shaders and cross validate + * them. + */ + glsl_symbol_table variables; + for (unsigned i = 0; i < num_shaders; i++) { + if (shader_list[i] == NULL) + continue; + + foreach_list(node, shader_list[i]->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if (var == NULL) + continue; + + if (uniforms_only && (var->mode != ir_var_uniform)) + continue; + + /* Don't cross validate temporaries that are at global scope. These + * will eventually get pulled into the shaders 'main'. + */ + if (var->mode == ir_var_temporary) + continue; + + /* If a global with this name has already been seen, verify that the + * new instance has the same type. In addition, if the globals have + * initializers, the values of the initializers must be the same. + */ + ir_variable *const existing = variables.get_variable(var->name); + if (existing != NULL) { + if (var->type != existing->type) { + /* Consider the types to be "the same" if both types are arrays + * of the same type and one of the arrays is implicitly sized. + * In addition, set the type of the linked variable to the + * explicitly sized array. + */ + if (var->type->is_array() + && existing->type->is_array() + && (var->type->fields.array == existing->type->fields.array) + && ((var->type->length == 0) + || (existing->type->length == 0))) { + if (var->type->length != 0) { + existing->type = var->type; + } + } else { + linker_error_printf(prog, "%s `%s' declared as type " + "`%s' and type `%s'\n", + mode_string(var), + var->name, var->type->name, + existing->type->name); + return false; + } + } + + if (var->explicit_location) { + if (existing->explicit_location + && (var->location != existing->location)) { + linker_error_printf(prog, "explicit locations for %s " + "`%s' have differing values\n", + mode_string(var), var->name); + return false; + } + + existing->location = var->location; + existing->explicit_location = true; + } + + /* FINISHME: Handle non-constant initializers. + */ + if (var->constant_value != NULL) { + if (existing->constant_value != NULL) { + if (!var->constant_value->has_value(existing->constant_value)) { + linker_error_printf(prog, "initializers for %s " + "`%s' have differing values\n", + mode_string(var), var->name); + return false; + } + } else + /* If the first-seen instance of a particular uniform did not + * have an initializer but a later instance does, copy the + * initializer to the version stored in the symbol table. + */ + /* FINISHME: This is wrong. The constant_value field should + * FINISHME: not be modified! Imagine a case where a shader + * FINISHME: without an initializer is linked in two different + * FINISHME: programs with shaders that have differing + * FINISHME: initializers. Linking with the first will + * FINISHME: modify the shader, and linking with the second + * FINISHME: will fail. + */ + existing->constant_value = + var->constant_value->clone(talloc_parent(existing), NULL); + } + + if (existing->invariant != var->invariant) { + linker_error_printf(prog, "declarations for %s `%s' have " + "mismatching invariant qualifiers\n", + mode_string(var), var->name); + return false; + } + if (existing->centroid != var->centroid) { + linker_error_printf(prog, "declarations for %s `%s' have " + "mismatching centroid qualifiers\n", + mode_string(var), var->name); + return false; + } + } else + variables.add_variable(var); + } + } + + return true; +} + + +/** + * Perform validation of uniforms used across multiple shader stages + */ +bool +cross_validate_uniforms(struct gl_shader_program *prog) +{ + return cross_validate_globals(prog, prog->_LinkedShaders, + MESA_SHADER_TYPES, true); +} + + +/** + * Validate that outputs from one stage match inputs of another + */ +bool +cross_validate_outputs_to_inputs(struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer) +{ + glsl_symbol_table parameters; + /* FINISHME: Figure these out dynamically. */ + const char *const producer_stage = "vertex"; + const char *const consumer_stage = "fragment"; + + /* Find all shader outputs in the "producer" stage. + */ + foreach_list(node, producer->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + /* FINISHME: For geometry shaders, this should also look for inout + * FINISHME: variables. + */ + if ((var == NULL) || (var->mode != ir_var_out)) + continue; + + parameters.add_variable(var); + } + + + /* Find all shader inputs in the "consumer" stage. Any variables that have + * matching outputs already in the symbol table must have the same type and + * qualifiers. + */ + foreach_list(node, consumer->ir) { + ir_variable *const input = ((ir_instruction *) node)->as_variable(); + + /* FINISHME: For geometry shaders, this should also look for inout + * FINISHME: variables. + */ + if ((input == NULL) || (input->mode != ir_var_in)) + continue; + + ir_variable *const output = parameters.get_variable(input->name); + if (output != NULL) { + /* Check that the types match between stages. + */ + if (input->type != output->type) { + /* There is a bit of a special case for gl_TexCoord. This + * built-in is unsized by default. Appliations that variable + * access it must redeclare it with a size. There is some + * language in the GLSL spec that implies the fragment shader + * and vertex shader do not have to agree on this size. Other + * driver behave this way, and one or two applications seem to + * rely on it. + * + * Neither declaration needs to be modified here because the array + * sizes are fixed later when update_array_sizes is called. + * + * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec: + * + * "Unlike user-defined varying variables, the built-in + * varying variables don't have a strict one-to-one + * correspondence between the vertex language and the + * fragment language." + */ + if (!output->type->is_array() + || (strncmp("gl_", output->name, 3) != 0)) { + linker_error_printf(prog, + "%s shader output `%s' declared as " + "type `%s', but %s shader input declared " + "as type `%s'\n", + producer_stage, output->name, + output->type->name, + consumer_stage, input->type->name); + return false; + } + } + + /* Check that all of the qualifiers match between stages. + */ + if (input->centroid != output->centroid) { + linker_error_printf(prog, + "%s shader output `%s' %s centroid qualifier, " + "but %s shader input %s centroid qualifier\n", + producer_stage, + output->name, + (output->centroid) ? "has" : "lacks", + consumer_stage, + (input->centroid) ? "has" : "lacks"); + return false; + } + + if (input->invariant != output->invariant) { + linker_error_printf(prog, + "%s shader output `%s' %s invariant qualifier, " + "but %s shader input %s invariant qualifier\n", + producer_stage, + output->name, + (output->invariant) ? "has" : "lacks", + consumer_stage, + (input->invariant) ? "has" : "lacks"); + return false; + } + + if (input->interpolation != output->interpolation) { + linker_error_printf(prog, + "%s shader output `%s' specifies %s " + "interpolation qualifier, " + "but %s shader input specifies %s " + "interpolation qualifier\n", + producer_stage, + output->name, + output->interpolation_string(), + consumer_stage, + input->interpolation_string()); + return false; + } + } + } + + return true; +} + + +/** + * Populates a shaders symbol table with all global declarations + */ +static void +populate_symbol_table(gl_shader *sh) +{ + sh->symbols = new(sh) glsl_symbol_table; + + foreach_list(node, sh->ir) { + ir_instruction *const inst = (ir_instruction *) node; + ir_variable *var; + ir_function *func; + + if ((func = inst->as_function()) != NULL) { + sh->symbols->add_function(func); + } else if ((var = inst->as_variable()) != NULL) { + sh->symbols->add_variable(var); + } + } +} + + +/** + * Remap variables referenced in an instruction tree + * + * This is used when instruction trees are cloned from one shader and placed in + * another. These trees will contain references to \c ir_variable nodes that + * do not exist in the target shader. This function finds these \c ir_variable + * references and replaces the references with matching variables in the target + * shader. + * + * If there is no matching variable in the target shader, a clone of the + * \c ir_variable is made and added to the target shader. The new variable is + * added to \b both the instruction stream and the symbol table. + * + * \param inst IR tree that is to be processed. + * \param symbols Symbol table containing global scope symbols in the + * linked shader. + * \param instructions Instruction stream where new variable declarations + * should be added. + */ +void +remap_variables(ir_instruction *inst, struct gl_shader *target, + hash_table *temps) +{ + class remap_visitor : public ir_hierarchical_visitor { + public: + remap_visitor(struct gl_shader *target, + hash_table *temps) + { + this->target = target; + this->symbols = target->symbols; + this->instructions = target->ir; + this->temps = temps; + } + + virtual ir_visitor_status visit(ir_dereference_variable *ir) + { + if (ir->var->mode == ir_var_temporary) { + ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var); + + assert(var != NULL); + ir->var = var; + return visit_continue; + } + + ir_variable *const existing = + this->symbols->get_variable(ir->var->name); + if (existing != NULL) + ir->var = existing; + else { + ir_variable *copy = ir->var->clone(this->target, NULL); + + this->symbols->add_variable(copy); + this->instructions->push_head(copy); + ir->var = copy; + } + + return visit_continue; + } + + private: + struct gl_shader *target; + glsl_symbol_table *symbols; + exec_list *instructions; + hash_table *temps; + }; + + remap_visitor v(target, temps); + + inst->accept(&v); +} + + +/** + * Move non-declarations from one instruction stream to another + * + * The intended usage pattern of this function is to pass the pointer to the + * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node + * pointer) for \c last and \c false for \c make_copies on the first + * call. Successive calls pass the return value of the previous call for + * \c last and \c true for \c make_copies. + * + * \param instructions Source instruction stream + * \param last Instruction after which new instructions should be + * inserted in the target instruction stream + * \param make_copies Flag selecting whether instructions in \c instructions + * should be copied (via \c ir_instruction::clone) into the + * target list or moved. + * + * \return + * The new "last" instruction in the target instruction stream. This pointer + * is suitable for use as the \c last parameter of a later call to this + * function. + */ +exec_node * +move_non_declarations(exec_list *instructions, exec_node *last, + bool make_copies, gl_shader *target) +{ + hash_table *temps = NULL; + + if (make_copies) + temps = hash_table_ctor(0, hash_table_pointer_hash, + hash_table_pointer_compare); + + foreach_list_safe(node, instructions) { + ir_instruction *inst = (ir_instruction *) node; + + if (inst->as_function()) + continue; + + ir_variable *var = inst->as_variable(); + if ((var != NULL) && (var->mode != ir_var_temporary)) + continue; + + assert(inst->as_assignment() + || ((var != NULL) && (var->mode == ir_var_temporary))); + + if (make_copies) { + inst = inst->clone(target, NULL); + + if (var != NULL) + hash_table_insert(temps, inst, var); + else + remap_variables(inst, target, temps); + } else { + inst->remove(); + } + + last->insert_after(inst); + last = inst; + } + + if (make_copies) + hash_table_dtor(temps); + + return last; +} + +/** + * Get the function signature for main from a shader + */ +static ir_function_signature * +get_main_function_signature(gl_shader *sh) +{ + ir_function *const f = sh->symbols->get_function("main"); + if (f != NULL) { + exec_list void_parameters; + + /* Look for the 'void main()' signature and ensure that it's defined. + * This keeps the linker from accidentally pick a shader that just + * contains a prototype for main. + * + * We don't have to check for multiple definitions of main (in multiple + * shaders) because that would have already been caught above. + */ + ir_function_signature *sig = f->matching_signature(&void_parameters); + if ((sig != NULL) && sig->is_defined) { + return sig; + } + } + + return NULL; +} + + +/** + * Combine a group of shaders for a single stage to generate a linked shader + * + * \note + * If this function is supplied a single shader, it is cloned, and the new + * shader is returned. + */ +static struct gl_shader * +link_intrastage_shaders(void *mem_ctx, + struct gl_context *ctx, + struct gl_shader_program *prog, + struct gl_shader **shader_list, + unsigned num_shaders) +{ + /* Check that global variables defined in multiple shaders are consistent. + */ + if (!cross_validate_globals(prog, shader_list, num_shaders, false)) + return NULL; + + /* Check that there is only a single definition of each function signature + * across all shaders. + */ + for (unsigned i = 0; i < (num_shaders - 1); i++) { + foreach_list(node, shader_list[i]->ir) { + ir_function *const f = ((ir_instruction *) node)->as_function(); + + if (f == NULL) + continue; + + for (unsigned j = i + 1; j < num_shaders; j++) { + ir_function *const other = + shader_list[j]->symbols->get_function(f->name); + + /* If the other shader has no function (and therefore no function + * signatures) with the same name, skip to the next shader. + */ + if (other == NULL) + continue; + + foreach_iter (exec_list_iterator, iter, *f) { + ir_function_signature *sig = + (ir_function_signature *) iter.get(); + + if (!sig->is_defined || sig->is_builtin) + continue; + + ir_function_signature *other_sig = + other->exact_matching_signature(& sig->parameters); + + if ((other_sig != NULL) && other_sig->is_defined + && !other_sig->is_builtin) { + linker_error_printf(prog, + "function `%s' is multiply defined", + f->name); + return NULL; + } + } + } + } + } + + /* Find the shader that defines main, and make a clone of it. + * + * Starting with the clone, search for undefined references. If one is + * found, find the shader that defines it. Clone the reference and add + * it to the shader. Repeat until there are no undefined references or + * until a reference cannot be resolved. + */ + gl_shader *main = NULL; + for (unsigned i = 0; i < num_shaders; i++) { + if (get_main_function_signature(shader_list[i]) != NULL) { + main = shader_list[i]; + break; + } + } + + if (main == NULL) { + linker_error_printf(prog, "%s shader lacks `main'\n", + (shader_list[0]->Type == GL_VERTEX_SHADER) + ? "vertex" : "fragment"); + return NULL; + } + + gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type); + linked->ir = new(linked) exec_list; + clone_ir_list(mem_ctx, linked->ir, main->ir); + + populate_symbol_table(linked); + + /* The a pointer to the main function in the final linked shader (i.e., the + * copy of the original shader that contained the main function). + */ + ir_function_signature *const main_sig = get_main_function_signature(linked); + + /* Move any instructions other than variable declarations or function + * declarations into main. + */ + exec_node *insertion_point = + move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false, + linked); + + for (unsigned i = 0; i < num_shaders; i++) { + if (shader_list[i] == main) + continue; + + insertion_point = move_non_declarations(shader_list[i]->ir, + insertion_point, true, linked); + } + + /* Resolve initializers for global variables in the linked shader. + */ + unsigned num_linking_shaders = num_shaders; + for (unsigned i = 0; i < num_shaders; i++) + num_linking_shaders += shader_list[i]->num_builtins_to_link; + + gl_shader **linking_shaders = + (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *)); + + memcpy(linking_shaders, shader_list, + sizeof(linking_shaders[0]) * num_shaders); + + unsigned idx = num_shaders; + for (unsigned i = 0; i < num_shaders; i++) { + memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link, + sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link); + idx += shader_list[i]->num_builtins_to_link; + } + + assert(idx == num_linking_shaders); + + if (!link_function_calls(prog, linked, linking_shaders, + num_linking_shaders)) { + ctx->Driver.DeleteShader(ctx, linked); + linked = NULL; + } + + free(linking_shaders); + + /* Make a pass over all variable declarations to ensure that arrays with + * unspecified sizes have a size specified. The size is inferred from the + * max_array_access field. + */ + if (linked != NULL) { + class array_sizing_visitor : public ir_hierarchical_visitor { + public: + virtual ir_visitor_status visit(ir_variable *var) + { + if (var->type->is_array() && (var->type->length == 0)) { + const glsl_type *type = + glsl_type::get_array_instance(var->type->fields.array, + var->max_array_access); + + assert(type != NULL); + var->type = type; + } + + return visit_continue; + } + } v; + + v.run(linked->ir); + } + + return linked; +} + + +struct uniform_node { + exec_node link; + struct gl_uniform *u; + unsigned slots; +}; + +/** + * Update the sizes of linked shader uniform arrays to the maximum + * array index used. + * + * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec: + * + * If one or more elements of an array are active, + * GetActiveUniform will return the name of the array in name, + * subject to the restrictions listed above. The type of the array + * is returned in type. The size parameter contains the highest + * array element index used, plus one. The compiler or linker + * determines the highest index used. There will be only one + * active uniform reported by the GL per uniform array. + + */ +static void +update_array_sizes(struct gl_shader_program *prog) +{ + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + foreach_list(node, prog->_LinkedShaders[i]->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_uniform && + var->mode != ir_var_in && + var->mode != ir_var_out) || + !var->type->is_array()) + continue; + + unsigned int size = var->max_array_access; + for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) { + if (prog->_LinkedShaders[j] == NULL) + continue; + + foreach_list(node2, prog->_LinkedShaders[j]->ir) { + ir_variable *other_var = ((ir_instruction *) node2)->as_variable(); + if (!other_var) + continue; + + if (strcmp(var->name, other_var->name) == 0 && + other_var->max_array_access > size) { + size = other_var->max_array_access; + } + } + } + + if (size + 1 != var->type->fields.array->length) { + var->type = glsl_type::get_array_instance(var->type->fields.array, + size + 1); + /* FINISHME: We should update the types of array + * dereferences of this variable now. + */ + } + } + } +} + +static void +add_uniform(void *mem_ctx, exec_list *uniforms, struct hash_table *ht, + const char *name, const glsl_type *type, GLenum shader_type, + unsigned *next_shader_pos, unsigned *total_uniforms) +{ + if (type->is_record()) { + for (unsigned int i = 0; i < type->length; i++) { + const glsl_type *field_type = type->fields.structure[i].type; + char *field_name = talloc_asprintf(mem_ctx, "%s.%s", name, + type->fields.structure[i].name); + + add_uniform(mem_ctx, uniforms, ht, field_name, field_type, + shader_type, next_shader_pos, total_uniforms); + } + } else { + uniform_node *n = (uniform_node *) hash_table_find(ht, name); + unsigned int vec4_slots; + const glsl_type *array_elem_type = NULL; + + if (type->is_array()) { + array_elem_type = type->fields.array; + /* Array of structures. */ + if (array_elem_type->is_record()) { + for (unsigned int i = 0; i < type->length; i++) { + char *elem_name = talloc_asprintf(mem_ctx, "%s[%d]", name, i); + add_uniform(mem_ctx, uniforms, ht, elem_name, array_elem_type, + shader_type, next_shader_pos, total_uniforms); + } + return; + } + } + + /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out + * vectors to vec4 slots. + */ + if (type->is_array()) { + if (array_elem_type->is_sampler()) + vec4_slots = type->length; + else + vec4_slots = type->length * array_elem_type->matrix_columns; + } else if (type->is_sampler()) { + vec4_slots = 1; + } else { + vec4_slots = type->matrix_columns; + } + + if (n == NULL) { + n = (uniform_node *) calloc(1, sizeof(struct uniform_node)); + n->u = (gl_uniform *) calloc(1, sizeof(struct gl_uniform)); + n->slots = vec4_slots; + + n->u->Name = strdup(name); + n->u->Type = type; + n->u->VertPos = -1; + n->u->FragPos = -1; + n->u->GeomPos = -1; + (*total_uniforms)++; + + hash_table_insert(ht, n, name); + uniforms->push_tail(& n->link); + } + + switch (shader_type) { + case GL_VERTEX_SHADER: + n->u->VertPos = *next_shader_pos; + break; + case GL_FRAGMENT_SHADER: + n->u->FragPos = *next_shader_pos; + break; + case GL_GEOMETRY_SHADER: + n->u->GeomPos = *next_shader_pos; + break; + } + + (*next_shader_pos) += vec4_slots; + } +} + +void +assign_uniform_locations(struct gl_shader_program *prog) +{ + /* */ + exec_list uniforms; + unsigned total_uniforms = 0; + hash_table *ht = hash_table_ctor(32, hash_table_string_hash, + hash_table_string_compare); + void *mem_ctx = talloc_new(NULL); + + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + unsigned next_position = 0; + + foreach_list(node, prog->_LinkedShaders[i]->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_uniform)) + continue; + + if (strncmp(var->name, "gl_", 3) == 0) { + /* At the moment, we don't allocate uniform locations for + * builtin uniforms. It's permitted by spec, and we'll + * likely switch to doing that at some point, but not yet. + */ + continue; + } + + var->location = next_position; + add_uniform(mem_ctx, &uniforms, ht, var->name, var->type, + prog->_LinkedShaders[i]->Type, + &next_position, &total_uniforms); + } + } + + talloc_free(mem_ctx); + + gl_uniform_list *ul = (gl_uniform_list *) + calloc(1, sizeof(gl_uniform_list)); + + ul->Size = total_uniforms; + ul->NumUniforms = total_uniforms; + ul->Uniforms = (gl_uniform *) calloc(total_uniforms, sizeof(gl_uniform)); + + unsigned idx = 0; + uniform_node *next; + for (uniform_node *node = (uniform_node *) uniforms.head + ; node->link.next != NULL + ; node = next) { + next = (uniform_node *) node->link.next; + + node->link.remove(); + memcpy(&ul->Uniforms[idx], node->u, sizeof(gl_uniform)); + idx++; + + free(node->u); + free(node); + } + + hash_table_dtor(ht); + + prog->Uniforms = ul; +} + + +/** + * Find a contiguous set of available bits in a bitmask + * + * \param used_mask Bits representing used (1) and unused (0) locations + * \param needed_count Number of contiguous bits needed. + * + * \return + * Base location of the available bits on success or -1 on failure. + */ +int +find_available_slots(unsigned used_mask, unsigned needed_count) +{ + unsigned needed_mask = (1 << needed_count) - 1; + const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count; + + /* The comparison to 32 is redundant, but without it GCC emits "warning: + * cannot optimize possibly infinite loops" for the loop below. + */ + if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32)) + return -1; + + for (int i = 0; i <= max_bit_to_test; i++) { + if ((needed_mask & ~used_mask) == needed_mask) + return i; + + needed_mask <<= 1; + } + + return -1; +} + + +bool +assign_attribute_locations(gl_shader_program *prog, unsigned max_attribute_index) +{ + /* Mark invalid attribute locations as being used. + */ + unsigned used_locations = (max_attribute_index >= 32) + ? ~0 : ~((1 << max_attribute_index) - 1); + + gl_shader *const sh = prog->_LinkedShaders[0]; + assert(sh->Type == GL_VERTEX_SHADER); + + /* Operate in a total of four passes. + * + * 1. Invalidate the location assignments for all vertex shader inputs. + * + * 2. Assign locations for inputs that have user-defined (via + * glBindVertexAttribLocation) locatoins. + * + * 3. Sort the attributes without assigned locations by number of slots + * required in decreasing order. Fragmentation caused by attribute + * locations assigned by the application may prevent large attributes + * from having enough contiguous space. + * + * 4. Assign locations to any inputs without assigned locations. + */ + + invalidate_variable_locations(sh, ir_var_in, VERT_ATTRIB_GENERIC0); + + if (prog->Attributes != NULL) { + for (unsigned i = 0; i < prog->Attributes->NumParameters; i++) { + ir_variable *const var = + sh->symbols->get_variable(prog->Attributes->Parameters[i].Name); + + /* Note: attributes that occupy multiple slots, such as arrays or + * matrices, may appear in the attrib array multiple times. + */ + if ((var == NULL) || (var->location != -1)) + continue; + + /* From page 61 of the OpenGL 4.0 spec: + * + * "LinkProgram will fail if the attribute bindings assigned by + * BindAttribLocation do not leave not enough space to assign a + * location for an active matrix attribute or an active attribute + * array, both of which require multiple contiguous generic + * attributes." + * + * Previous versions of the spec contain similar language but omit the + * bit about attribute arrays. + * + * Page 61 of the OpenGL 4.0 spec also says: + * + * "It is possible for an application to bind more than one + * attribute name to the same location. This is referred to as + * aliasing. This will only work if only one of the aliased + * attributes is active in the executable program, or if no path + * through the shader consumes more than one attribute of a set + * of attributes aliased to the same location. A link error can + * occur if the linker determines that every path through the + * shader consumes multiple aliased attributes, but + * implementations are not required to generate an error in this + * case." + * + * These two paragraphs are either somewhat contradictory, or I don't + * fully understand one or both of them. + */ + /* FINISHME: The code as currently written does not support attribute + * FINISHME: location aliasing (see comment above). + */ + const int attr = prog->Attributes->Parameters[i].StateIndexes[0]; + const unsigned slots = count_attribute_slots(var->type); + + /* Mask representing the contiguous slots that will be used by this + * attribute. + */ + const unsigned use_mask = (1 << slots) - 1; + + /* Generate a link error if the set of bits requested for this + * attribute overlaps any previously allocated bits. + */ + if ((~(use_mask << attr) & used_locations) != used_locations) { + linker_error_printf(prog, + "insufficient contiguous attribute locations " + "available for vertex shader input `%s'", + var->name); + return false; + } + + var->location = VERT_ATTRIB_GENERIC0 + attr; + used_locations |= (use_mask << attr); + } + } + + /* Temporary storage for the set of attributes that need locations assigned. + */ + struct temp_attr { + unsigned slots; + ir_variable *var; + + /* Used below in the call to qsort. */ + static int compare(const void *a, const void *b) + { + const temp_attr *const l = (const temp_attr *) a; + const temp_attr *const r = (const temp_attr *) b; + + /* Reversed because we want a descending order sort below. */ + return r->slots - l->slots; + } + } to_assign[16]; + + unsigned num_attr = 0; + + foreach_list(node, sh->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_in)) + continue; + + if (var->explicit_location) { + const unsigned slots = count_attribute_slots(var->type); + const unsigned use_mask = (1 << slots) - 1; + const int attr = var->location - VERT_ATTRIB_GENERIC0; + + if ((var->location >= (int)(max_attribute_index + VERT_ATTRIB_GENERIC0)) + || (var->location < 0)) { + linker_error_printf(prog, + "invalid explicit location %d specified for " + "`%s'\n", + (var->location < 0) ? var->location : attr, + var->name); + return false; + } else if (var->location >= VERT_ATTRIB_GENERIC0) { + used_locations |= (use_mask << attr); + } + } + + /* The location was explicitly assigned, nothing to do here. + */ + if (var->location != -1) + continue; + + to_assign[num_attr].slots = count_attribute_slots(var->type); + to_assign[num_attr].var = var; + num_attr++; + } + + /* If all of the attributes were assigned locations by the application (or + * are built-in attributes with fixed locations), return early. This should + * be the common case. + */ + if (num_attr == 0) + return true; + + qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare); + + /* VERT_ATTRIB_GENERIC0 is a psdueo-alias for VERT_ATTRIB_POS. It can only + * be explicitly assigned by via glBindAttribLocation. Mark it as reserved + * to prevent it from being automatically allocated below. + */ + find_deref_visitor find("gl_Vertex"); + find.run(sh->ir); + if (find.variable_found()) + used_locations |= (1 << 0); + + for (unsigned i = 0; i < num_attr; i++) { + /* Mask representing the contiguous slots that will be used by this + * attribute. + */ + const unsigned use_mask = (1 << to_assign[i].slots) - 1; + + int location = find_available_slots(used_locations, to_assign[i].slots); + + if (location < 0) { + linker_error_printf(prog, + "insufficient contiguous attribute locations " + "available for vertex shader input `%s'", + to_assign[i].var->name); + return false; + } + + to_assign[i].var->location = VERT_ATTRIB_GENERIC0 + location; + used_locations |= (use_mask << location); + } + + return true; +} + + +/** + * Demote shader inputs and outputs that are not used in other stages + */ +void +demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode) +{ + foreach_list(node, sh->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != int(mode))) + continue; + + /* A shader 'in' or 'out' variable is only really an input or output if + * its value is used by other shader stages. This will cause the variable + * to have a location assigned. + */ + if (var->location == -1) { + var->mode = ir_var_auto; + } + } +} + + +void +assign_varying_locations(struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer) +{ + /* FINISHME: Set dynamically when geometry shader support is added. */ + unsigned output_index = VERT_RESULT_VAR0; + unsigned input_index = FRAG_ATTRIB_VAR0; + + /* Operate in a total of three passes. + * + * 1. Assign locations for any matching inputs and outputs. + * + * 2. Mark output variables in the producer that do not have locations as + * not being outputs. This lets the optimizer eliminate them. + * + * 3. Mark input variables in the consumer that do not have locations as + * not being inputs. This lets the optimizer eliminate them. + */ + + invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0); + invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0); + + foreach_list(node, producer->ir) { + ir_variable *const output_var = ((ir_instruction *) node)->as_variable(); + + if ((output_var == NULL) || (output_var->mode != ir_var_out) + || (output_var->location != -1)) + continue; + + ir_variable *const input_var = + consumer->symbols->get_variable(output_var->name); + + if ((input_var == NULL) || (input_var->mode != ir_var_in)) + continue; + + assert(input_var->location == -1); + + output_var->location = output_index; + input_var->location = input_index; + + /* FINISHME: Support for "varying" records in GLSL 1.50. */ + assert(!output_var->type->is_record()); + + if (output_var->type->is_array()) { + const unsigned slots = output_var->type->length + * output_var->type->fields.array->matrix_columns; + + output_index += slots; + input_index += slots; + } else { + const unsigned slots = output_var->type->matrix_columns; + + output_index += slots; + input_index += slots; + } + } + + foreach_list(node, consumer->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_in)) + continue; + + if (var->location == -1) { + if (prog->Version <= 120) { + /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec: + * + * Only those varying variables used (i.e. read) in + * the fragment shader executable must be written to + * by the vertex shader executable; declaring + * superfluous varying variables in a vertex shader is + * permissible. + * + * We interpret this text as meaning that the VS must + * write the variable for the FS to read it. See + * "glsl1-varying read but not written" in piglit. + */ + + linker_error_printf(prog, "fragment shader varying %s not written " + "by vertex shader\n.", var->name); + prog->LinkStatus = false; + } + + /* An 'in' variable is only really a shader input if its + * value is written by the previous stage. + */ + var->mode = ir_var_auto; + } + } +} + + +void +link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) +{ + void *mem_ctx = talloc_init("temporary linker context"); + + prog->LinkStatus = false; + prog->Validated = false; + prog->_Used = false; + + if (prog->InfoLog != NULL) + talloc_free(prog->InfoLog); + + prog->InfoLog = talloc_strdup(NULL, ""); + + /* Separate the shaders into groups based on their type. + */ + struct gl_shader **vert_shader_list; + unsigned num_vert_shaders = 0; + struct gl_shader **frag_shader_list; + unsigned num_frag_shaders = 0; + + vert_shader_list = (struct gl_shader **) + calloc(2 * prog->NumShaders, sizeof(struct gl_shader *)); + frag_shader_list = &vert_shader_list[prog->NumShaders]; + + unsigned min_version = UINT_MAX; + unsigned max_version = 0; + for (unsigned i = 0; i < prog->NumShaders; i++) { + min_version = MIN2(min_version, prog->Shaders[i]->Version); + max_version = MAX2(max_version, prog->Shaders[i]->Version); + + switch (prog->Shaders[i]->Type) { + case GL_VERTEX_SHADER: + vert_shader_list[num_vert_shaders] = prog->Shaders[i]; + num_vert_shaders++; + break; + case GL_FRAGMENT_SHADER: + frag_shader_list[num_frag_shaders] = prog->Shaders[i]; + num_frag_shaders++; + break; + case GL_GEOMETRY_SHADER: + /* FINISHME: Support geometry shaders. */ + assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER); + break; + } + } + + /* Previous to GLSL version 1.30, different compilation units could mix and + * match shading language versions. With GLSL 1.30 and later, the versions + * of all shaders must match. + */ + assert(min_version >= 100); + assert(max_version <= 130); + if ((max_version >= 130 || min_version == 100) + && min_version != max_version) { + linker_error_printf(prog, "all shaders must use same shading " + "language version\n"); + goto done; + } + + prog->Version = max_version; + + for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] != NULL) + ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]); + + prog->_LinkedShaders[i] = NULL; + } + + /* Link all shaders for a particular stage and validate the result. + */ + if (num_vert_shaders > 0) { + gl_shader *const sh = + link_intrastage_shaders(mem_ctx, ctx, prog, vert_shader_list, + num_vert_shaders); + + if (sh == NULL) + goto done; + + if (!validate_vertex_shader_executable(prog, sh)) + goto done; + + _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX], + sh); + } + + if (num_frag_shaders > 0) { + gl_shader *const sh = + link_intrastage_shaders(mem_ctx, ctx, prog, frag_shader_list, + num_frag_shaders); + + if (sh == NULL) + goto done; + + if (!validate_fragment_shader_executable(prog, sh)) + goto done; + + _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT], + sh); + } + + /* Here begins the inter-stage linking phase. Some initial validation is + * performed, then locations are assigned for uniforms, attributes, and + * varyings. + */ + if (cross_validate_uniforms(prog)) { + unsigned prev; + + for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { + if (prog->_LinkedShaders[prev] != NULL) + break; + } + + /* Validate the inputs of each stage with the output of the preceeding + * stage. + */ + for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + if (!cross_validate_outputs_to_inputs(prog, + prog->_LinkedShaders[prev], + prog->_LinkedShaders[i])) + goto done; + + prev = i; + } + + prog->LinkStatus = true; + } + + /* Do common optimization before assigning storage for attributes, + * uniforms, and varyings. Later optimization could possibly make + * some of that unused. + */ + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32)) + ; + } + + update_array_sizes(prog); + + assign_uniform_locations(prog); + + if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { + /* FINISHME: The value of the max_attribute_index parameter is + * FINISHME: implementation dependent based on the value of + * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be + * FINISHME: at least 16, so hardcode 16 for now. + */ + if (!assign_attribute_locations(prog, 16)) { + prog->LinkStatus = false; + goto done; + } + } + + unsigned prev; + for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { + if (prog->_LinkedShaders[prev] != NULL) + break; + } + + for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + assign_varying_locations(prog, + prog->_LinkedShaders[prev], + prog->_LinkedShaders[i]); + prev = i; + } + + if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { + demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX], + ir_var_out); + } + + if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) { + gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY]; + + demote_shader_inputs_and_outputs(sh, ir_var_in); + demote_shader_inputs_and_outputs(sh, ir_var_inout); + demote_shader_inputs_and_outputs(sh, ir_var_out); + } + + if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) { + gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT]; + + demote_shader_inputs_and_outputs(sh, ir_var_in); + } + + /* FINISHME: Assign fragment shader output locations. */ + +done: + free(vert_shader_list); + + for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] == NULL) + continue; + + /* Retain any live IR, but trash the rest. */ + reparent_ir(prog->_LinkedShaders[i]->ir, prog->_LinkedShaders[i]->ir); + } + + talloc_free(mem_ctx); +} -- cgit v1.2.3