This is a case that is not picked up by integer range inference and
suggests a weakness with integer range inference on the index dialect.
The problem is that when `[1, SMAX_64]` is truncated to 32 bits, the
resulting range could be `[SMIN_32, SMAX_32]`, making the subsequent
comparison worthless. This is because integer range inference doesn't
know that the result of the max/min inference also changes based on the
bitwidth, and doing the truncation locally at the input of the
comparison op loses that information.
This also was a pattern that frequently showed up in our code, so adding
it as a folder allows dead code to be pruned more frequently.
Depends on D153731
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D153732
As discussed in https://github.com/llvm/llvm-project/issues/59714#issuecomment-1369518768, the folder for the remainder operations should be resillient when the rhs is 0.
The file `IndexOps.cpp` was already checking for multiple divisions by zero, so I tried to stick to the code style from those checks.
Fixes#59714.
As a side note, is it correct that remainder operations are never optimized away? I would expect that the following code
```
func.func @remu_test() -> index {
%c3 = index.constant 2
%c0 = index.constant 1
%0 = index.remu %c3, %c0
return %0 : index
}
```
would be optimized to
```
func.func @remu_test() -> index {
return index.constant 0 : index
}
```
when called with `mlir-opt --convert-scf-to-openmp temp.mlir`, but maybe I'm misunderstanding something.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D151476
Re-land D140899 to fix a missing dependency in the index dialect's
CMakeLists.txt.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D142147
This reverts commit 455305624884cf9237143e2ba0635fcc5ba5206a.
Linker error, unbreak build while I work out how to fix it.
Differential Revision: https://reviews.llvm.org/D142142
Implement InferIntRangeInterface for all operations in the Index dialect. The
inference implementation, unlike the one for Arith, accounts for the
fact that Index can be either 64 or 32 bits long by evaluating both
cases. Bounds are stored as if index were i64, but when inferring new
bounds, we compute both f(...) and f(trunc(...)). We then compare
trunc(f(...)) to f(trunc(...)). If they are equal in the relevant
range components, we use the 64-bit range computation, otherwise we
give the range ext(f(trunc(...))) union f(...).
Note that this can cause surprising behavior as seen in the tests,
where, for example, the order of min and max operations impacts the
behavior of the inference. The inference could perhaps be made more
precise in the future (ex. by tracking 32 and 64-bit results
separately and having them influence each other somehow) butt, since
my project targets an index=i32 platform and doesn't see index-valued
values > uint32_max, I'm not too concerned about it.
Depends on https://reviews.llvm.org/D141299
Depends on https://reviews.llvm.org/D141296
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D140899
Signed and unsigned minimum operations were missing from the Index
dialect and are needed to test integer range inference.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D141299
ConstantOp uses `%idx<value>` and BoolConstantOp uses true/false, which
is similar to the printing for arith::ConstantOp.
Differential Revision: https://reviews.llvm.org/D139175
This patch adds the and, or, and xor bitwise operations to
the index dialects with folders and LLVM lowerings.
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D138590
This patch adds the left shift, signed right shift, and unsigned right
shift operations to the index dialects with folders and LLVM lowerings.
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D137349
This patch adds folders for `index` dialect ops. Ths folders are
careful to ensure that fold results are valid on both 32-bit and 64-bit
targets.
Depends on D135689
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D135694
This patch adds the definitions for the operations and attributes (just
one enum attribute) for the `index` dialect.
Depends on D135688
Reviewed By: rriddle, jpienaar
Differential Revision: https://reviews.llvm.org/D135689
