This allows to add facts even if no corresponding ICmp instruction
exists in the IR.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D158837
Remove compares after replacing all uses. Cleaning dead compares can
enable additional simplifications when adjusting the position of the
pass slightly. In particular, it seems like the additional dead
instructions may prevent SimplifyCFG performing some folds.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D158760
This patch represents a condition with `Pred + LHS + RHS` in ReproducerEntry instead of `CmpInst + Not`.
It avoids creating temporary ICmpInsts in D155412.
Reviewed By: nikic, fhahn
Differential Revision: https://reviews.llvm.org/D155782
Simplification of non-equality predicates for solving constraint
systems is now supported by checking the validity of related
inequalities and equalities.
Differential Revision: https://reviews.llvm.org/D152684
Delegate checking of the constraint & its preconditions to the existing
::isValid. This reduces duplication and allows additional optimizations
together with D152730.
This patch extends the existing logic to handle cases where we have
branch conditions of the form (AND icmp, icmp) where the first icmp
implies the second. This can improve results in some cases, e.g. if
SimplifyCFG folded conditions from multiple branches to an AND.
The implementation handles this by adding a new type of check
(AndImpliedCheck), which are queued before conditional facts for the same
block.
When encountering AndImpliedChecks during solving, the first condition
is optimistically added to the constraint system, then we check if the
second icmp can be simplified, and finally the newly added entries are
removed.
The reason for doing things this way is to avoid clashes with signed
<-> unsigned condition transfer, which require us to re-order facts to
increase effectiveness.
Reviewed By: nikic, antoniofrighetto
Differential Revision: https://reviews.llvm.org/D151799
Simplification of equality predicates is now supported by
transferring equalities into inequalities. This is achieved
by separately checking that both `isConditionImplied(A >= B)`
and `isConditionImplied(A <= B)` hold.
Differential Revision: https://reviews.llvm.org/D152067
Handling `true` and `false` constant replacements is now abstracted
out into a single lambda function `ReplaceCmpWithConstant`, so as to
reduce code duplication.
If there is an optimisation opportunity and the function argument hasn’t been added to constraint
system through previous facts we fail to optimise it.
It might be a good idea to start the constraint system with all the function arguments added to the system
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D144879
This reverts commit 2a2a6bfcfe8e62886542cb673ac8df349cf26499.
This causes build failures:
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp: In member function ‘llvm::SmallVector<std::__cxx11::basic_string<char> > llvm::ConstraintSystem::getVarNamesList() const’:
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:118:10: error: invalid use of incomplete type ‘class llvm::Value’
118 | if (V->getName().empty())
| ^~
In file included from /home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:9:
/home/npopov/repos/llvm-project/llvm/include/llvm/Analysis/ConstraintSystem.h:21:7: note: forward declaration of ‘class llvm::Value’
21 | class Value;
| ^~~~~
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:119:22: error: invalid use of incomplete type ‘class llvm::Value’
119 | OperandName = V->getNameOrAsOperand();
| ^~
/home/npopov/repos/llvm-project/llvm/include/llvm/Analysis/ConstraintSystem.h:21:7: note: forward declaration of ‘class llvm::Value’
21 | class Value;
| ^~~~~
/home/npopov/repos/llvm-project/llvm/lib/Analysis/ConstraintSystem.cpp:121:41: error: invalid use of incomplete type ‘class llvm::Value’
121 | OperandName = std::string("%") + V->getName().str();
| ^~
/home/npopov/repos/llvm-project/llvm/include/llvm/Analysis/ConstraintSystem.h:21:7: note: forward declaration of ‘class llvm::Value’
21 | class Value;
| ^~~~~
Previously when constraint system outputs the rows in the system the variables used are x1,2...n making it hard to infer which ones they relate to in the IR
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D142618
This patch adds an optimization remark for each performed optimization
containing a module that can be used to reproduce the transformation.
The reproducer function contains a series of @llvm.assume calls, one for
each condition currently in scope. For each condition, the operand
instruction are cloned until we reach operands that have an entry in the
constraint system. Those will then be added as function arguments.
The reproducer functions are minimal, that is, they only contain the
conditions required for a given simplification. The resulting IR is very
compact and can be used to verify each transformation individually.
It also provides a python script to extract the IR from the remarks and
create LLVM IR files from it.
Reviewed By: paquette
Differential Revision: https://reviews.llvm.org/D143323
This reverts commit 40ffe9c167395256b43846733ab69eec17eead78.
Reverted because some comments where missed in the review https://reviews.llvm.org/D142647
Once the constraint system grows too large in terms of number of rows,
queries can become very slow. This patch adds a new option to limit the
number of rows tracked.
The python script below can be used to generate worst-case IR with a
chain of conditional branches with N branches.
With this limit, we get the following runtimes:
* python3 generate.py 100: 0.1s
* python3 generate.py 1000: 2s
* python3 generate.py 10000: 4s
Without the limit, the case with 1000 chained conditions takes 20+
seconds.
generate.py:
import sys
N = int(sys.argv[1])
args = []
checks = []
for i in range(0, N):
args.append('i32 %l{}'.format(i))
checks.append("""
bb{0}:
%c{0} = icmp uge i32 %l{0}, 100
br i1 %c{0}, label %bb{1}, label %exit
""".format(i, i+1))
print("""
define i1 @foo({0}) {{
{1}
bb{2}:
%c{2} = icmp uge i32 %l0, 100
ret i1 %c{2}
exit:
ret i1 false
}}
""".format(' ,'.join(args), '\n'.join(checks), N))
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D140926
Update the sort order to make sure that comesBefore is never used from
conditional facts, which are instructions but may use DFS numbers from
different blocks.
This fixes a crash in the added test on some platforms.
This allows interleaving facts and checks in a single block. In
particular this enables using facts from assumes for conditions in the
same block that come after the assume.
This could be extended to only try to simplify checks at the point where
a condition is used.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D138452
This slightly simplifies the code and addresses a correctness issue
where the index scaling for the precondition was not considered
properly.
Thanks to @nikic for pointing that out in D137840.
Use collectOffset to collect scaled indices and constant offset for GEP
instead of custom code. This simplifies the logic in decomposeGEP and
allows to handle all cases supported by the generic helper.
Replace the vector of DecompEntry with a struct that stores the
constant offset separately. I think this is cleaner than giving the
first element special handling.
This probably also fixes some potential ubsan errors by more
consistently using addWithOverflow/multiplyWithOverflow.
decompose() currently returns a mix of {} and 0 + 1*V on failure.
This changes it to always return the 0 + 1*V form, thus making
decompose() infallible.
This makes the code marginally more powerful, e.g. we now fold
sub_decomp_i80 by treating the constant as a symbolic value.
Differential Revision: https://reviews.llvm.org/D137847
