AuroraMiddleware/SPIRV-Tools
1
0
Fork 0
mirror of https://github.com/KhronosGroup/SPIRV-Tools synced 2024-10-19 11:30:15 +00:00
Code Issues Packages Projects Releases Wiki Activity
c4dc046399
SPIRV-Tools/source/opt/cfg.cpp

311 lines
11 KiB
C++
Raw Normal View History

Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
// Copyright (c) 2017 Google 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 "cfg.h"
Move class CFG from namespace opt to namespace ir. It makes more sense to have the CFG inside the ir name space, as it is descriptive of the representation.
2017-11-02 15:51:07 +00:00
#include "cfa.h"
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
#include "ir_builder.h"
Store all enabled capabilities in the feature manger. In order to keep track of all of the implicit capabilities as well as the explicit ones, we will add them all to the feature manager. That is the object that needs to be queried when checking if a capability is enabled. The name of the "HasCapability" function in the module was changed to make it more obvious that it does not check for implied capabilities. Keep an spv_context and AssemblyGrammar in IRContext
2017-12-19 19:18:13 +00:00
#include "ir_context.h"
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
#include "module.h"
namespace spvtools {
Move class CFG from namespace opt to namespace ir. It makes more sense to have the CFG inside the ir name space, as it is descriptive of the representation.
2017-11-02 15:51:07 +00:00
namespace ir {
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
namespace {
// Universal Limit of ResultID + 1
const int kInvalidId = 0x400000;
} // namespace
CFG::CFG(ir::Module* module)
: module_(module),
pseudo_entry_block_(std::unique_ptr<ir::Instruction>(
Adding an unique id to Instruction generated by IRContext Each instruction is given an unique id that can be used for ordering purposes. The ids are generated via the IRContext. Major changes: * Instructions now contain a uint32_t for unique id and a cached context pointer * Most constructors have been modified to take a context as input * unfortunately I cannot remove the default and copy constructors, but developers should avoid these * Added accessors to parents of basic block and function * Removed the copy constructors for BasicBlock and Function and replaced them with Clone functions * Reworked BuildModule to return an IRContext owning the built module * Since all instructions require a context, the context now becomes the basic unit for IR * Added a constructor to context to create an owned module internally * Replaced uses of Instruction's copy constructor with Clone whereever I found them * Reworked the linker functionality to perform clones into a different context instead of moves * Updated many tests to be consistent with the above changes * Still need to add new tests to cover added functionality * Added comparison operators to Instruction * Added an internal option to LinkerOptions to verify merged ids are unique * Added a test for the linker to verify merged ids are unique * Updated MergeReturnPass to supply a context * Updated DecorationManager to supply a context for cloned decorations * Reworked several portions of the def use tests in anticipation of next set of changes
2017-11-14 19:11:50 +00:00
new ir::Instruction(module->context(), SpvOpLabel, 0, 0, {}))),
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
pseudo_exit_block_(std::unique_ptr<ir::Instruction>(new ir::Instruction(
module->context(), SpvOpLabel, 0, kInvalidId, {}))) {
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
for (auto& fn : *module) {
for (auto& blk : fn) {
Add LoopUtils class to gather some loop transformation support. This patch adds LoopUtils class to handle some loop related transformations. For now it has 2 transformations that simplifies other transformations such as loop unroll or unswitch: - Dedicate exit blocks: this ensure that all exit basic block (out-of-loop basic blocks that have a predecessor in the loop) have all their predecessors in the loop; - Loop Closed SSA (LCSSA): this ensure that all definitions in a loop are used inside the loop or in a phi instruction in an exit basic block. It also adds the following capabilities: - Loop::IsLCSSA to test if the loop is in a LCSSA form - Loop::GetOrCreatePreHeaderBlock that can build a loop preheader if required; - New methods to allow on the fly updates of the loop descriptors. - New methods to allow on the fly updates of the CFG analysis. - Instruction::SetOperand to allow expression of the index relative to Instruction::NumOperands (to be compatible with the index returned by DefUseManager::ForEachUse)
2018-01-26 12:07:10 +00:00
RegisterBlock(&blk);
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
}
}
}
Disambiguate between const and nonconst ForEachSuccessorLabel This helps VisualStudio 2013 compile the code. Contributes to #1262
2018-02-02 22:17:42 +00:00
void CFG::AddEdges(ir::BasicBlock* blk) {
uint32_t blk_id = blk->id();
// Force the creation of an entry, not all basic block have predecessors
// (such as the entry blocks and some unreachables).
label2preds_[blk_id];
const auto* const_blk = blk;
const_blk->ForEachSuccessorLabel(
[blk_id, this](const uint32_t succ_id) { AddEdge(blk_id, succ_id); });
}
Add LoopUtils class to gather some loop transformation support. This patch adds LoopUtils class to handle some loop related transformations. For now it has 2 transformations that simplifies other transformations such as loop unroll or unswitch: - Dedicate exit blocks: this ensure that all exit basic block (out-of-loop basic blocks that have a predecessor in the loop) have all their predecessors in the loop; - Loop Closed SSA (LCSSA): this ensure that all definitions in a loop are used inside the loop or in a phi instruction in an exit basic block. It also adds the following capabilities: - Loop::IsLCSSA to test if the loop is in a LCSSA form - Loop::GetOrCreatePreHeaderBlock that can build a loop preheader if required; - New methods to allow on the fly updates of the loop descriptors. - New methods to allow on the fly updates of the CFG analysis. - Instruction::SetOperand to allow expression of the index relative to Instruction::NumOperands (to be compatible with the index returned by DefUseManager::ForEachUse)
2018-01-26 12:07:10 +00:00
void CFG::RemoveNonExistingEdges(uint32_t blk_id) {
std::vector<uint32_t> updated_pred_list;
for (uint32_t id : preds(blk_id)) {
Disambiguate between const and nonconst ForEachSuccessorLabel This helps VisualStudio 2013 compile the code. Contributes to #1262
2018-02-02 22:17:42 +00:00
const ir::BasicBlock* pred_blk = block(id);
Add LoopUtils class to gather some loop transformation support. This patch adds LoopUtils class to handle some loop related transformations. For now it has 2 transformations that simplifies other transformations such as loop unroll or unswitch: - Dedicate exit blocks: this ensure that all exit basic block (out-of-loop basic blocks that have a predecessor in the loop) have all their predecessors in the loop; - Loop Closed SSA (LCSSA): this ensure that all definitions in a loop are used inside the loop or in a phi instruction in an exit basic block. It also adds the following capabilities: - Loop::IsLCSSA to test if the loop is in a LCSSA form - Loop::GetOrCreatePreHeaderBlock that can build a loop preheader if required; - New methods to allow on the fly updates of the loop descriptors. - New methods to allow on the fly updates of the CFG analysis. - Instruction::SetOperand to allow expression of the index relative to Instruction::NumOperands (to be compatible with the index returned by DefUseManager::ForEachUse)
2018-01-26 12:07:10 +00:00
bool has_branch = false;
pred_blk->ForEachSuccessorLabel([&has_branch, blk_id](uint32_t succ) {
Loop invariant code motion initial implementation
2018-01-29 10:39:55 +00:00
if (succ == blk_id) {
has_branch = true;
}
Add LoopUtils class to gather some loop transformation support. This patch adds LoopUtils class to handle some loop related transformations. For now it has 2 transformations that simplifies other transformations such as loop unroll or unswitch: - Dedicate exit blocks: this ensure that all exit basic block (out-of-loop basic blocks that have a predecessor in the loop) have all their predecessors in the loop; - Loop Closed SSA (LCSSA): this ensure that all definitions in a loop are used inside the loop or in a phi instruction in an exit basic block. It also adds the following capabilities: - Loop::IsLCSSA to test if the loop is in a LCSSA form - Loop::GetOrCreatePreHeaderBlock that can build a loop preheader if required; - New methods to allow on the fly updates of the loop descriptors. - New methods to allow on the fly updates of the CFG analysis. - Instruction::SetOperand to allow expression of the index relative to Instruction::NumOperands (to be compatible with the index returned by DefUseManager::ForEachUse)
2018-01-26 12:07:10 +00:00
});
if (has_branch) updated_pred_list.push_back(id);
}
label2preds_.at(blk_id) = std::move(updated_pred_list);
}
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
void CFG::ComputeStructuredOrder(ir::Function* func, ir::BasicBlock* root,
Move class CFG from namespace opt to namespace ir. It makes more sense to have the CFG inside the ir name space, as it is descriptive of the representation.
2017-11-02 15:51:07 +00:00
std::list<ir::BasicBlock*>* order) {
Store all enabled capabilities in the feature manger. In order to keep track of all of the implicit capabilities as well as the explicit ones, we will add them all to the feature manager. That is the object that needs to be queried when checking if a capability is enabled. The name of the "HasCapability" function in the module was changed to make it more obvious that it does not check for implied capabilities. Keep an spv_context and AssemblyGrammar in IRContext
2017-12-19 19:18:13 +00:00
assert(module_->context()->get_feature_mgr()->HasCapability(
SpvCapabilityShader) &&
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
"This only works on structured control flow");
// Compute structured successors and do DFS.
ComputeStructuredSuccessors(func);
auto ignore_block = [](cbb_ptr) {};
auto ignore_edge = [](cbb_ptr, cbb_ptr) {};
auto get_structured_successors = [this](const ir::BasicBlock* b) {
return &(block2structured_succs_[b]);
};
// TODO(greg-lunarg): Get rid of const_cast by making moving const
// out of the cfa.h prototypes and into the invoking code.
auto post_order = [&](cbb_ptr b) {
order->push_front(const_cast<ir::BasicBlock*>(b));
};
spvtools::CFA<ir::BasicBlock>::DepthFirstTraversal(
Move class CFG from namespace opt to namespace ir. It makes more sense to have the CFG inside the ir name space, as it is descriptive of the representation.
2017-11-02 15:51:07 +00:00
root, get_structured_successors, ignore_block, post_order, ignore_edge);
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
}
Make use of the instruction folder. Implementation of the simplification pass. - Create pass that calls the instruction folder on each instruction and propagate instructions that fold to a copy. This will do copy propagation as well. - Did not use the propagator engine because I want to modify the instruction as we go along. - Change folding to not allocate new instructions, but make changes in place. This change had a big impact on compile time. - Add simplification pass to the legalization passes in place of insert-extract elimination. - Added test cases for new folding rules. - Added tests for the simplification pass - Added a method to the CFG to apply a function to the basic blocks in reverse post order. Contributes to #1164.
2018-02-02 16:55:05 +00:00
void CFG::ForEachBlockInReversePostOrder(
BasicBlock* bb, const std::function<void(BasicBlock*)>& f) {
std::vector<BasicBlock*> po;
std::unordered_set<BasicBlock*> seen;
ComputePostOrderTraversal(bb, &po, &seen);
for (auto current_bb = po.rbegin(); current_bb != po.rend(); ++current_bb) {
if (!IsPseudoExitBlock(*current_bb) && !IsPseudoEntryBlock(*current_bb)) {
f(*current_bb);
}
}
}
Move class CFG from namespace opt to namespace ir. It makes more sense to have the CFG inside the ir name space, as it is descriptive of the representation.
2017-11-02 15:51:07 +00:00
void CFG::ComputeStructuredSuccessors(ir::Function* func) {
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
block2structured_succs_.clear();
for (auto& blk : *func) {
// If no predecessors in function, make successor to pseudo entry.
if (label2preds_[blk.id()].size() == 0)
block2structured_succs_[&pseudo_entry_block_].push_back(&blk);
// If header, make merge block first successor and continue block second
// successor if there is one.
uint32_t mbid = blk.MergeBlockIdIfAny();
if (mbid != 0) {
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
block2structured_succs_[&blk].push_back(block(mbid));
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
uint32_t cbid = blk.ContinueBlockIdIfAny();
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
if (cbid != 0) {
block2structured_succs_[&blk].push_back(block(cbid));
}
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
}
// Add true successors.
Disambiguate between const and nonconst ForEachSuccessorLabel This helps VisualStudio 2013 compile the code. Contributes to #1262
2018-02-02 22:17:42 +00:00
const auto& const_blk = blk;
const_blk.ForEachSuccessorLabel([&blk, this](const uint32_t sbid) {
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
block2structured_succs_[&blk].push_back(block(sbid));
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
});
}
}
Make use of the instruction folder. Implementation of the simplification pass. - Create pass that calls the instruction folder on each instruction and propagate instructions that fold to a copy. This will do copy propagation as well. - Did not use the propagator engine because I want to modify the instruction as we go along. - Change folding to not allocate new instructions, but make changes in place. This change had a big impact on compile time. - Add simplification pass to the legalization passes in place of insert-extract elimination. - Added test cases for new folding rules. - Added tests for the simplification pass - Added a method to the CFG to apply a function to the basic blocks in reverse post order. Contributes to #1164.
2018-02-02 16:55:05 +00:00
void CFG::ComputePostOrderTraversal(BasicBlock* bb, vector<BasicBlock*>* order,
unordered_set<BasicBlock*>* seen) {
seen->insert(bb);
static_cast<const BasicBlock*>(bb)->ForEachSuccessorLabel(
[&order, &seen, this](const uint32_t sbid) {
BasicBlock* succ_bb = id2block_[sbid];
if (!seen->count(succ_bb)) {
ComputePostOrderTraversal(succ_bb, order, seen);
}
});
order->push_back(bb);
}
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
BasicBlock* CFG::SplitLoopHeader(ir::BasicBlock* bb) {
assert(bb->GetLoopMergeInst() && "Expecting bb to be the header of a loop.");
Function* fn = bb->GetParent();
IRContext* context = fn->context();
// Find the insertion point for the new bb.
Function::iterator header_it = std::find_if(
fn->begin(), fn->end(),
[bb](BasicBlock& block_in_func) { return &block_in_func == bb; });
assert(header_it != fn->end());
Fixed remaining review comments from #1380
2018-03-21 16:52:13 +00:00
const std::vector<uint32_t>& pred = preds(bb->id());
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
// Find the back edge
ir::BasicBlock* latch_block = nullptr;
Fixed remaining review comments from #1380
2018-03-21 16:52:13 +00:00
Function::iterator latch_block_iter = header_it;
while (++latch_block_iter != fn->end()) {
// If blocks are in the proper order, then the only branch that appears
// after the header is the latch.
if (std::find(pred.begin(), pred.end(), latch_block_iter->id()) !=
pred.end()) {
break;
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
}
}
Fixed remaining review comments from #1380
2018-03-21 16:52:13 +00:00
assert(latch_block_iter != fn->end() && "Could not find the latch.");
latch_block = &*latch_block_iter;
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
RemoveSuccessorEdges(bb);
// Create the new header bb basic bb.
// Leave the phi instructions behind.
auto iter = bb->begin();
while (iter->opcode() == SpvOpPhi) {
++iter;
}
std::unique_ptr<ir::BasicBlock> newBlock(
bb->SplitBasicBlock(context, context->TakeNextId(), iter));
// Insert the new bb in the correct position
auto insert_pos = header_it;
++insert_pos;
ir::BasicBlock* new_header = &*insert_pos.InsertBefore(std::move(newBlock));
new_header->SetParent(fn);
uint32_t new_header_id = new_header->id();
context->AnalyzeDefUse(new_header->GetLabelInst());
// Update cfg
RegisterBlock(new_header);
// Update bb mappings.
context->set_instr_block(new_header->GetLabelInst(), new_header);
new_header->ForEachInst([new_header, context](ir::Instruction* inst) {
context->set_instr_block(inst, new_header);
});
// Adjust the OpPhi instructions as needed.
bb->ForEachPhiInst([latch_block, bb, new_header, context](Instruction* phi) {
std::vector<uint32_t> preheader_phi_ops;
std::vector<Operand> header_phi_ops;
Fixed remaining review comments from #1380
2018-03-21 16:52:13 +00:00
// Identify where the original inputs to original OpPhi belong: header or
Change merge return pass to handle structured cfg. We are seeing shaders that have multiple returns in a functions. These functions must get inlined for legalization purposes; however, the inliner does not know how to inline functions that have multiple returns. The solution we will go with it to improve the merge return pass to handle structured control flow. Note that the merge return pass will assume the cfg has been cleanedup by dead branch elimination. Fixes #857.
2018-03-06 16:20:28 +00:00
// preheader.
for (uint32_t i = 0; i < phi->NumInOperands(); i += 2) {
uint32_t def_id = phi->GetSingleWordInOperand(i);
uint32_t branch_id = phi->GetSingleWordInOperand(i + 1);
if (branch_id == latch_block->id()) {
header_phi_ops.push_back({SPV_OPERAND_TYPE_ID, {def_id}});
header_phi_ops.push_back({SPV_OPERAND_TYPE_ID, {branch_id}});
} else {
preheader_phi_ops.push_back(def_id);
preheader_phi_ops.push_back(branch_id);
}
}
// Create a phi instruction if and only if the preheader_phi_ops has more
// than one pair.
if (preheader_phi_ops.size() > 2) {
opt::InstructionBuilder builder(
context, &*bb->begin(),
ir::IRContext::kAnalysisDefUse |
ir::IRContext::kAnalysisInstrToBlockMapping);
ir::Instruction* new_phi =
builder.AddPhi(phi->type_id(), preheader_phi_ops);
// Add the OpPhi to the header bb.
header_phi_ops.push_back({SPV_OPERAND_TYPE_ID, {new_phi->result_id()}});
header_phi_ops.push_back({SPV_OPERAND_TYPE_ID, {bb->id()}});
} else {
// An OpPhi with a single entry is just a copy. In this case use the same
// instruction in the new header.
header_phi_ops.push_back({SPV_OPERAND_TYPE_ID, {preheader_phi_ops[0]}});
header_phi_ops.push_back({SPV_OPERAND_TYPE_ID, {bb->id()}});
}
phi->RemoveFromList();
std::unique_ptr<ir::Instruction> phi_owner(phi);
phi->SetInOperands(std::move(header_phi_ops));
new_header->begin()->InsertBefore(std::move(phi_owner));
context->set_instr_block(phi, new_header);
context->AnalyzeUses(phi);
});
// Add a branch to the new header.
opt::InstructionBuilder branch_builder(
context, bb,
ir::IRContext::kAnalysisDefUse |
ir::IRContext::kAnalysisInstrToBlockMapping);
bb->AddInstruction(MakeUnique<ir::Instruction>(
context, SpvOpBranch, 0, 0,
std::initializer_list<ir::Operand>{
{SPV_OPERAND_TYPE_ID, {new_header->id()}}}));
context->AnalyzeUses(bb->terminator());
context->set_instr_block(bb->terminator(), bb);
label2preds_[new_header->id()].push_back(bb->id());
// Update the latch to branch to the new header.
latch_block->ForEachSuccessorLabel([bb, new_header_id](uint32_t* id) {
if (*id == bb->id()) {
*id = new_header_id;
}
});
ir::Instruction* latch_branch = latch_block->terminator();
context->AnalyzeUses(latch_branch);
label2preds_[new_header->id()].push_back(latch_block->id());
auto& block_preds = label2preds_[bb->id()];
auto latch_pos =
std::find(block_preds.begin(), block_preds.end(), latch_block->id());
assert(latch_pos != block_preds.end() && "The cfg was invalid.");
block_preds.erase(latch_pos);
// Update the loop descriptors
if (context->AreAnalysesValid(ir::IRContext::kAnalysisLoopAnalysis)) {
LoopDescriptor* loop_desc = context->GetLoopDescriptor(bb->GetParent());
Loop* loop = (*loop_desc)[bb->id()];
loop->AddBasicBlock(new_header_id);
loop->SetHeaderBlock(new_header);
loop_desc->SetBasicBlockToLoop(new_header_id, loop);
loop->RemoveBasicBlock(bb->id());
loop->SetPreHeaderBlock(bb);
Loop* parent_loop = loop->GetParent();
if (parent_loop != nullptr) {
parent_loop->AddBasicBlock(bb->id());
loop_desc->SetBasicBlockToLoop(bb->id(), parent_loop);
} else {
loop_desc->SetBasicBlockToLoop(bb->id(), nullptr);
}
}
return new_header;
}
unordered_set<BasicBlock*> CFG::FindReachableBlocks(BasicBlock* start) {
std::unordered_set<BasicBlock*> reachable_blocks;
ForEachBlockInReversePostOrder(start, [&reachable_blocks](BasicBlock* bb) {
reachable_blocks.insert(bb);
});
return reachable_blocks;
}
Re-format source tree - NFC. Re-formatted the source tree with the command: $ /usr/bin/clang-format -style=file -i \ $(find include source tools test utils -name '*.cpp' -or -name '*.h') This required a fix to source/val/decoration.h. It was not including spirv.h, which broke builds when the #include headers were re-ordered by clang-format.
2017-11-27 15:16:41 +00:00
} // namespace ir
Add a new class opt::CFG to represent the CFG for the module. This class moves some of the CFG-related functionality into a new class opt::CFG. There is some other code related to the CFG in the inliner and in opt::LocalSingleStoreElimPass that should also be moved, but that require more changes than this pure restructuring. I will move those bits in a follow-up PR. Currently, the CFG is computed every time a pass is instantiated, but this should be later moved to the new IRContext class that @s-perron is working on. Other re-factoring: - Add BasicBlock::ContinueBlockIdIfAny. Re-factored out of MergeBlockIdIfAny - Rewrite IsLoopHeader in terms of GetLoopMergeInst. - Run clang-format on some files.
2017-10-30 21:42:26 +00:00
} // namespace spvtools
Reference in New Issue Copy Permalink