[all-commits] [llvm/llvm-project] ff3749: [NFC] SimplifyCFGOpt::simplifyUnreachable(): pacif...

[Roman Lebedev via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: ff3749fc7933a6f45b77739380e434060ca7693d
      https://github.com/llvm/llvm-project/commit/ff3749fc7933a6f45b77739380e434060ca7693d
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/lib/Transforms/Utils/SimplifyCFG.cpp

  Log Message:
  -----------
  [NFC] SimplifyCFGOpt::simplifyUnreachable(): pacify unused variable warning

Thanks to Luke Benes for pointing it out.


  Commit: b3021a72a6d2fea8702b75f34f9f3317ae923d92
      https://github.com/llvm/llvm-project/commit/b3021a72a6d2fea8702b75f34f9f3317ae923d92
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/include/llvm/IR/PatternMatch.h
    M llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
    M llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
    M llvm/lib/Transforms/InstCombine/InstCombineNegator.cpp
    M llvm/lib/Transforms/InstCombine/InstructionCombining.cpp

  Log Message:
  -----------
  [IR][InstCombine] Add m_ImmConstant(), that matches on non-ConstantExpr constants, and use it

A pattern to ignore ConstantExpr's is quite common, since they frequently
lead into infinite combine loops, so let's make writing it easier.


  Commit: da4c7e15df33aa659d160b645c595b977c81ad02
      https://github.com/llvm/llvm-project/commit/da4c7e15df33aa659d160b645c595b977c81ad02
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/test/Transforms/InstCombine/vec_shuffle.ll

  Log Message:
  -----------
  [NFC][InstCombine] Autogenerate check lines in vec_shuffle.ll test


  Commit: 1fda23367d46955fcb6e605f4114d47e499f0901
      https://github.com/llvm/llvm-project/commit/1fda23367d46955fcb6e605f4114d47e499f0901
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/test/Transforms/InstCombine/and-xor-or.ll

  Log Message:
  -----------
  [NFC][InstCombine] Add test for `a & ~(a ^ b)` pattern

... which is a variation of `a & (a ^ ~b)` --> a & b`.
A follow-up patch exposes this missing fold, so we need to fix it first.


  Commit: 5b78303433c0778a839e89d20daa57fbc037d0c7
      https://github.com/llvm/llvm-project/commit/5b78303433c0778a839e89d20daa57fbc037d0c7
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
    M llvm/test/Transforms/InstCombine/and-xor-or.ll

  Log Message:
  -----------
  [InstCombine] Fold `a & ~(a ^ b)` to `x & y`

```
----------------------------------------
define i32 @and_xor_not_common_op(i32 %a, i32 %b) {
%0:
  %b2 = xor i32 %b, 4294967295
  %t2 = xor i32 %a, %b2
  %t4 = and i32 %t2, %a
  ret i32 %t4
}
=>
define i32 @and_xor_not_common_op(i32 %a, i32 %b) {
%0:
  %t4 = and i32 %a, %b
  ret i32 %t4
}
Transformation seems to be correct!
```


  Commit: 8001dcbd50ce501e75e06886c5e225298e9eea2b
      https://github.com/llvm/llvm-project/commit/8001dcbd50ce501e75e06886c5e225298e9eea2b
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    A llvm/test/Transforms/InstCombine/hoist-xor-by-constant-from-xor-by-value.ll

  Log Message:
  -----------
  [NFC][InstCombine] Add test coverage for `(x ^ C) ^ y` pattern


  Commit: d9ebaeeb468d6a8f29eb479f18d2790f7efb8565
      https://github.com/llvm/llvm-project/commit/d9ebaeeb468d6a8f29eb479f18d2790f7efb8565
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
    M llvm/test/Transforms/InstCombine/hoist-xor-by-constant-from-xor-by-value.ll
    M llvm/test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-scalar.ll
    M llvm/test/Transforms/InstCombine/invert-variable-mask-in-masked-merge-vector.ll
    M llvm/test/Transforms/InstCombine/or-xor.ll
    M llvm/test/Transforms/InstCombine/unfold-masked-merge-with-const-mask-scalar.ll
    M llvm/test/Transforms/InstCombine/unfold-masked-merge-with-const-mask-vector.ll
    M llvm/test/Transforms/InstCombine/vec_shuffle-inseltpoison.ll
    M llvm/test/Transforms/InstCombine/vec_shuffle.ll
    M llvm/test/Transforms/InstCombine/xor2.ll

  Log Message:
  -----------
  [InstCombine] Hoist xor-by-constant from xor-by-value

This is one of the deficiencies that can be observed in
https://godbolt.org/z/YPczsG after D91038 patch set.

This exposed two missing folds, one was fixed by the previous commit,
another one is `(A ^ B) | ~(A ^ B) --> -1` / `(A ^ B) & ~(A ^ B) --> 0`.

`-early-cse` will catch it: https://godbolt.org/z/4n1T1v,
but isn't meaningful to fix it in InstCombine,
because we'd need to essentially do our own CSE,
and we can't even rely on `Instruction::isIdenticalTo()`,
because there are no guarantees that the order of operands matches.
So let's just accept it as a loss.


  Commit: 6e074a8324d86af93930001e4bdcaff8ebc9890e
      https://github.com/llvm/llvm-project/commit/6e074a8324d86af93930001e4bdcaff8ebc9890e
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/test/Transforms/LoopIdiom/X86/left-shift-until-bittest.ll

  Log Message:
  -----------
  [NFC][LoopIdiom] Improve test coverage for 'left-shift-until-bittest' pattern

In particular, add tests with no-wrap flags on shift,
a test where %x is not `1`, and ensure that tests where %bit
is a constant bitwidth-1, or is not a constant bitwidth-1
test both liveout values.


  Commit: 25aebe2ccfb4622b17494c5cfdb2b422c93cee4d
      https://github.com/llvm/llvm-project/commit/25aebe2ccfb4622b17494c5cfdb2b422c93cee4d
  Author: Roman Lebedev <lebedev.ri at gmail.com>
  Date:   2020-12-24 (Thu, 24 Dec 2020)

  Changed paths:
    M llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
    M llvm/test/Transforms/LoopIdiom/X86/left-shift-until-bittest.ll

  Log Message:
  -----------
  [LoopIdiom] 'left-shift-until-bittest': keep no-wrap flags on shift, fix edge-case miscompilation for %x.next

While `%x.curr` is always safe to compute, because `LoopBackedgeTakenCount`
will always be smaller than `bitwidth(X)`, i.e. we never get poison,
rewriting `%x.next` is more complicated, however, because `X << LoopTripCount`
will be poison iff `LoopTripCount == bitwidth(X)` (which will happen
iff `BitPos` is `bitwidth(x) - 1` and `X` is `1`).

So unless we know that isn't the case (as alive2 notes, we know it's safe
to do iff shift had no-wrap flags, or bitpos does not indicate signbit,
or we know that %x is never `1`), we'll need to emit an alternative,
safe IR, by either just shifting the `%x.curr`, or conditionally selecting
between the computed `%x.next` and `0`..
Former IR looks better so let's do that.

While there, ensure that we don't drop no-wrap flags from said shift.


Compare: https://github.com/llvm/llvm-project/compare/e457896a6ef0...25aebe2ccfb4