SPIRV-Tools/source/validate_cfg.cpp
David Neto dadd5161bb Check strict domination of merge block
If a merge block is reachable, then it must be *strictly* dominated
by its header.  Until now we've allowed the header and the merge
block to be the same.

Fixes https://github.com/KhronosGroup/SPIRV-Tools/issues/551

Also: Use dominates and postdominates methods on BasicBlock to
improve readability.
2017-02-10 12:26:10 -05:00

537 lines
19 KiB
C++

// Copyright (c) 2015-2016 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "validate.h"
#include <algorithm>
#include <cassert>
#include <functional>
#include <iostream>
#include <map>
#include <string>
#include <tuple>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#include "val/basic_block.h"
#include "val/construct.h"
#include "val/function.h"
#include "val/validation_state.h"
using std::find;
using std::function;
using std::get;
using std::ignore;
using std::make_pair;
using std::make_tuple;
using std::numeric_limits;
using std::pair;
using std::string;
using std::tie;
using std::transform;
using std::tuple;
using std::unordered_map;
using std::unordered_set;
using std::vector;
using libspirv::BasicBlock;
namespace libspirv {
namespace {
using bb_ptr = BasicBlock*;
using cbb_ptr = const BasicBlock*;
using bb_iter = vector<BasicBlock*>::const_iterator;
struct block_info {
cbb_ptr block; ///< pointer to the block
bb_iter iter; ///< Iterator to the current child node being processed
};
/// Returns true if a block with @p id is found in the @p work_list vector
///
/// @param[in] work_list Set of blocks visited in the the depth first traversal
/// of the CFG
/// @param[in] id The ID of the block being checked
///
/// @return true if the edge work_list.back().block->id() => id is a back-edge
bool FindInWorkList(const vector<block_info>& work_list, uint32_t id) {
for (const auto b : work_list) {
if (b.block->id() == id) return true;
}
return false;
}
} // namespace
void DepthFirstTraversal(const BasicBlock* entry,
get_blocks_func successor_func,
function<void(cbb_ptr)> preorder,
function<void(cbb_ptr)> postorder,
function<void(cbb_ptr, cbb_ptr)> backedge) {
unordered_set<uint32_t> processed;
/// NOTE: work_list is the sequence of nodes from the root node to the node
/// being processed in the traversal
vector<block_info> work_list;
work_list.reserve(10);
work_list.push_back({entry, begin(*successor_func(entry))});
preorder(entry);
processed.insert(entry->id());
while (!work_list.empty()) {
block_info& top = work_list.back();
if (top.iter == end(*successor_func(top.block))) {
postorder(top.block);
work_list.pop_back();
} else {
BasicBlock* child = *top.iter;
top.iter++;
if (FindInWorkList(work_list, child->id())) {
backedge(top.block, child);
}
if (processed.count(child->id()) == 0) {
preorder(child);
work_list.emplace_back(
block_info{child, begin(*successor_func(child))});
processed.insert(child->id());
}
}
}
}
vector<pair<BasicBlock*, BasicBlock*>> CalculateDominators(
const vector<cbb_ptr>& postorder, get_blocks_func predecessor_func) {
struct block_detail {
size_t dominator; ///< The index of blocks's dominator in post order array
size_t postorder_index; ///< The index of the block in the post order array
};
const size_t undefined_dom = postorder.size();
unordered_map<cbb_ptr, block_detail> idoms;
for (size_t i = 0; i < postorder.size(); i++) {
idoms[postorder[i]] = {undefined_dom, i};
}
idoms[postorder.back()].dominator = idoms[postorder.back()].postorder_index;
bool changed = true;
while (changed) {
changed = false;
for (auto b = postorder.rbegin() + 1; b != postorder.rend(); ++b) {
const vector<BasicBlock*>& predecessors = *predecessor_func(*b);
// Find the first processed/reachable predecessor that is reachable
// in the forward traversal.
auto res = find_if(begin(predecessors), end(predecessors),
[&idoms, undefined_dom](BasicBlock* pred) {
return idoms.count(pred) &&
idoms[pred].dominator != undefined_dom;
});
if (res == end(predecessors)) continue;
const BasicBlock* idom = *res;
size_t idom_idx = idoms[idom].postorder_index;
// all other predecessors
for (const auto* p : predecessors) {
if (idom == p) continue;
// Only consider nodes reachable in the forward traversal.
// Otherwise the intersection doesn't make sense and will never
// terminate.
if (!idoms.count(p)) continue;
if (idoms[p].dominator != undefined_dom) {
size_t finger1 = idoms[p].postorder_index;
size_t finger2 = idom_idx;
while (finger1 != finger2) {
while (finger1 < finger2) {
finger1 = idoms[postorder[finger1]].dominator;
}
while (finger2 < finger1) {
finger2 = idoms[postorder[finger2]].dominator;
}
}
idom_idx = finger1;
}
}
if (idoms[*b].dominator != idom_idx) {
idoms[*b].dominator = idom_idx;
changed = true;
}
}
}
vector<pair<bb_ptr, bb_ptr>> out;
for (auto idom : idoms) {
// NOTE: performing a const cast for convenient usage with
// UpdateImmediateDominators
out.push_back({const_cast<BasicBlock*>(get<0>(idom)),
const_cast<BasicBlock*>(postorder[get<1>(idom).dominator])});
}
return out;
}
void printDominatorList(const BasicBlock& b) {
std::cout << b.id() << " is dominated by: ";
const BasicBlock* bb = &b;
while (bb->immediate_dominator() != bb) {
bb = bb->immediate_dominator();
std::cout << bb->id() << " ";
}
}
#define CFG_ASSERT(ASSERT_FUNC, TARGET) \
if (spv_result_t rcode = ASSERT_FUNC(_, TARGET)) return rcode
spv_result_t FirstBlockAssert(ValidationState_t& _, uint32_t target) {
if (_.current_function().IsFirstBlock(target)) {
return _.diag(SPV_ERROR_INVALID_CFG)
<< "First block " << _.getIdName(target) << " of function "
<< _.getIdName(_.current_function().id()) << " is targeted by block "
<< _.getIdName(_.current_function().current_block()->id());
}
return SPV_SUCCESS;
}
spv_result_t MergeBlockAssert(ValidationState_t& _, uint32_t merge_block) {
if (_.current_function().IsBlockType(merge_block, kBlockTypeMerge)) {
return _.diag(SPV_ERROR_INVALID_CFG)
<< "Block " << _.getIdName(merge_block)
<< " is already a merge block for another header";
}
return SPV_SUCCESS;
}
/// Update the continue construct's exit blocks once the backedge blocks are
/// identified in the CFG.
void UpdateContinueConstructExitBlocks(
Function& function, const vector<pair<uint32_t, uint32_t>>& back_edges) {
auto& constructs = function.constructs();
// TODO(umar): Think of a faster way to do this
for (auto& edge : back_edges) {
uint32_t back_edge_block_id;
uint32_t loop_header_block_id;
tie(back_edge_block_id, loop_header_block_id) = edge;
auto is_this_header = [=](Construct& c) {
return c.type() == ConstructType::kLoop &&
c.entry_block()->id() == loop_header_block_id;
};
for (auto construct : constructs) {
if (is_this_header(construct)) {
Construct* continue_construct =
construct.corresponding_constructs().back();
assert(continue_construct->type() == ConstructType::kContinue);
BasicBlock* back_edge_block;
tie(back_edge_block, ignore) = function.GetBlock(back_edge_block_id);
continue_construct->set_exit(back_edge_block);
}
}
}
}
tuple<string, string, string> ConstructNames(ConstructType type) {
string construct_name, header_name, exit_name;
switch (type) {
case ConstructType::kSelection:
construct_name = "selection";
header_name = "selection header";
exit_name = "merge block";
break;
case ConstructType::kLoop:
construct_name = "loop";
header_name = "loop header";
exit_name = "merge block";
break;
case ConstructType::kContinue:
construct_name = "continue";
header_name = "continue target";
exit_name = "back-edge block";
break;
case ConstructType::kCase:
construct_name = "case";
header_name = "case entry block";
exit_name = "case exit block";
break;
default:
assert(1 == 0 && "Not defined type");
}
return make_tuple(construct_name, header_name, exit_name);
}
/// Constructs an error message for construct validation errors
string ConstructErrorString(const Construct& construct,
const string& header_string,
const string& exit_string,
const string& dominate_text) {
string construct_name, header_name, exit_name;
tie(construct_name, header_name, exit_name) =
ConstructNames(construct.type());
// TODO(umar): Add header block for continue constructs to error message
return "The " + construct_name + " construct with the " + header_name + " " +
header_string + " " + dominate_text + " the " + exit_name + " " +
exit_string;
}
spv_result_t StructuredControlFlowChecks(
const ValidationState_t& _, const Function& function,
const vector<pair<uint32_t, uint32_t>>& back_edges) {
/// Check all backedges target only loop headers and have exactly one
/// back-edge branching to it
// Map a loop header to blocks with back-edges to the loop header.
std::map<uint32_t, std::unordered_set<uint32_t>> loop_latch_blocks;
for (auto back_edge : back_edges) {
uint32_t back_edge_block;
uint32_t header_block;
tie(back_edge_block, header_block) = back_edge;
if (!function.IsBlockType(header_block, kBlockTypeLoop)) {
return _.diag(SPV_ERROR_INVALID_CFG)
<< "Back-edges (" << _.getIdName(back_edge_block) << " -> "
<< _.getIdName(header_block)
<< ") can only be formed between a block and a loop header.";
}
loop_latch_blocks[header_block].insert(back_edge_block);
}
// Check the loop headers have exactly one back-edge branching to it
for (BasicBlock* loop_header : function.ordered_blocks()) {
if (!loop_header->reachable()) continue;
if (!loop_header->is_type(kBlockTypeLoop)) continue;
auto loop_header_id = loop_header->id();
auto num_latch_blocks = loop_latch_blocks[loop_header_id].size();
if (num_latch_blocks != 1) {
return _.diag(SPV_ERROR_INVALID_CFG)
<< "Loop header " << _.getIdName(loop_header_id)
<< " is targeted by " << num_latch_blocks
<< " back-edge blocks but the standard requires exactly one";
}
}
// Check construct rules
for (const Construct& construct : function.constructs()) {
auto header = construct.entry_block();
auto merge = construct.exit_block();
if (header->reachable() && !merge) {
string construct_name, header_name, exit_name;
tie(construct_name, header_name, exit_name) =
ConstructNames(construct.type());
return _.diag(SPV_ERROR_INTERNAL)
<< "Construct " + construct_name + " with " + header_name + " " +
_.getIdName(header->id()) + " does not have a " +
exit_name + ". This may be a bug in the validator.";
}
// If the exit block is reachable then it's dominated by the
// header.
if (merge && merge->reachable()) {
if (!header->dominates(*merge)) {
return _.diag(SPV_ERROR_INVALID_CFG) << ConstructErrorString(
construct, _.getIdName(header->id()),
_.getIdName(merge->id()), "does not dominate");
}
// If it's really a merge block for a selection or loop, then it must be
// *strictly* dominated by the header.
if (construct.ExitBlockIsMergeBlock() && (header == merge)) {
return _.diag(SPV_ERROR_INVALID_CFG) << ConstructErrorString(
construct, _.getIdName(header->id()),
_.getIdName(merge->id()), "does not strictly dominate");
}
}
// Check post-dominance for continue constructs. But dominance and
// post-dominance only make sense when the construct is reachable.
if (header->reachable() && construct.type() == ConstructType::kContinue) {
if (!merge->postdominates(*header)) {
return _.diag(SPV_ERROR_INVALID_CFG) << ConstructErrorString(
construct, _.getIdName(header->id()),
_.getIdName(merge->id()), "is not post dominated by");
}
}
// TODO(umar): an OpSwitch block dominates all its defined case
// constructs
// TODO(umar): each case construct has at most one branch to another
// case construct
// TODO(umar): each case construct is branched to by at most one other
// case construct
// TODO(umar): if Target T1 branches to Target T2, or if Target T1
// branches to the Default and the Default branches to Target T2, then
// T1 must immediately precede T2 in the list of the OpSwitch Target
// operands
}
return SPV_SUCCESS;
}
spv_result_t PerformCfgChecks(ValidationState_t& _) {
for (auto& function : _.functions()) {
// Check all referenced blocks are defined within a function
if (function.undefined_block_count() != 0) {
string undef_blocks("{");
for (auto undefined_block : function.undefined_blocks()) {
undef_blocks += _.getIdName(undefined_block) + " ";
}
return _.diag(SPV_ERROR_INVALID_CFG)
<< "Block(s) " << undef_blocks << "\b}"
<< " are referenced but not defined in function "
<< _.getIdName(function.id());
}
// Set each block's immediate dominator and immediate postdominator,
// and find all back-edges.
//
// We want to analyze all the blocks in the function, even in degenerate
// control flow cases including unreachable blocks. So use the augmented
// CFG to ensure we cover all the blocks.
vector<const BasicBlock*> postorder;
vector<const BasicBlock*> postdom_postorder;
vector<pair<uint32_t, uint32_t>> back_edges;
auto ignore_block = [](cbb_ptr) {};
auto ignore_edge = [](cbb_ptr, cbb_ptr) {};
if (!function.ordered_blocks().empty()) {
/// calculate dominators
DepthFirstTraversal(
function.first_block(), function.AugmentedCFGSuccessorsFunction(),
ignore_block, [&](cbb_ptr b) { postorder.push_back(b); },
ignore_edge);
auto edges = libspirv::CalculateDominators(
postorder, function.AugmentedCFGPredecessorsFunction());
for (auto edge : edges) {
edge.first->SetImmediateDominator(edge.second);
}
/// calculate post dominators
DepthFirstTraversal(
function.pseudo_exit_block(),
function.AugmentedCFGPredecessorsFunction(), ignore_block,
[&](cbb_ptr b) { postdom_postorder.push_back(b); }, ignore_edge);
auto postdom_edges = libspirv::CalculateDominators(
postdom_postorder, function.AugmentedCFGSuccessorsFunction());
for (auto edge : postdom_edges) {
edge.first->SetImmediatePostDominator(edge.second);
}
/// calculate back edges.
DepthFirstTraversal(
function.pseudo_entry_block(),
function
.AugmentedCFGSuccessorsFunctionIncludingHeaderToContinueEdge(),
ignore_block, ignore_block, [&](cbb_ptr from, cbb_ptr to) {
back_edges.emplace_back(from->id(), to->id());
});
}
UpdateContinueConstructExitBlocks(function, back_edges);
auto& blocks = function.ordered_blocks();
if (!blocks.empty()) {
// Check if the order of blocks in the binary appear before the blocks
// they dominate
for (auto block = begin(blocks) + 1; block != end(blocks); ++block) {
if (auto idom = (*block)->immediate_dominator()) {
if (idom != function.pseudo_entry_block() &&
block == std::find(begin(blocks), block, idom)) {
return _.diag(SPV_ERROR_INVALID_CFG)
<< "Block " << _.getIdName((*block)->id())
<< " appears in the binary before its dominator "
<< _.getIdName(idom->id());
}
}
}
// If we have structed control flow, check that no block has a control
// flow nesting depth larger than the limit.
if (_.HasCapability(SpvCapabilityShader)) {
const int control_flow_nesting_depth_limit = 1023;
for (auto block = begin(blocks); block != end(blocks); ++block) {
if (function.GetBlockDepth(*block) >
control_flow_nesting_depth_limit) {
return _.diag(SPV_ERROR_INVALID_CFG)
<< "Maximum Control Flow nesting depth exceeded.";
}
}
}
}
/// Structured control flow checks are only required for shader capabilities
if (_.HasCapability(SpvCapabilityShader)) {
if (auto error = StructuredControlFlowChecks(_, function, back_edges))
return error;
}
}
return SPV_SUCCESS;
}
spv_result_t CfgPass(ValidationState_t& _,
const spv_parsed_instruction_t* inst) {
SpvOp opcode = static_cast<SpvOp>(inst->opcode);
switch (opcode) {
case SpvOpLabel:
if (auto error = _.current_function().RegisterBlock(inst->result_id))
return error;
break;
case SpvOpLoopMerge: {
uint32_t merge_block = inst->words[inst->operands[0].offset];
uint32_t continue_block = inst->words[inst->operands[1].offset];
CFG_ASSERT(MergeBlockAssert, merge_block);
if (auto error = _.current_function().RegisterLoopMerge(merge_block,
continue_block))
return error;
} break;
case SpvOpSelectionMerge: {
uint32_t merge_block = inst->words[inst->operands[0].offset];
CFG_ASSERT(MergeBlockAssert, merge_block);
if (auto error = _.current_function().RegisterSelectionMerge(merge_block))
return error;
} break;
case SpvOpBranch: {
uint32_t target = inst->words[inst->operands[0].offset];
CFG_ASSERT(FirstBlockAssert, target);
_.current_function().RegisterBlockEnd({target}, opcode);
} break;
case SpvOpBranchConditional: {
uint32_t tlabel = inst->words[inst->operands[1].offset];
uint32_t flabel = inst->words[inst->operands[2].offset];
CFG_ASSERT(FirstBlockAssert, tlabel);
CFG_ASSERT(FirstBlockAssert, flabel);
_.current_function().RegisterBlockEnd({tlabel, flabel}, opcode);
} break;
case SpvOpSwitch: {
vector<uint32_t> cases;
for (int i = 1; i < inst->num_operands; i += 2) {
uint32_t target = inst->words[inst->operands[i].offset];
CFG_ASSERT(FirstBlockAssert, target);
cases.push_back(target);
}
_.current_function().RegisterBlockEnd({cases}, opcode);
} break;
case SpvOpKill:
case SpvOpReturn:
case SpvOpReturnValue:
case SpvOpUnreachable:
_.current_function().RegisterBlockEnd(vector<uint32_t>(), opcode);
break;
default:
break;
}
return SPV_SUCCESS;
}
} // namespace libspirv