[PATCH] D145299: [X86] Generate better code for std::bit_ceil

[Kazu Hirata via Phabricator via llvm-commits]

kazu created this revision.
Herald added subscribers: pengfei, hiraditya.
Herald added a project: All.
kazu requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Without this patch, std::bit_ceil is compiled as:

  8d 47 ff                   lea    -0x1(%rdi),%eax
  f3 0f bd c0                lzcnt  %eax,%eax
  f6 d8                      neg    %al
  b9 01 00 00 00             mov    $0x1,%ecx
  c4 e2 79 f7 c1             shlx   %eax,%ecx,%eax
  83 ff 02                   cmp    $0x2,%edi
  0f 42 c1                   cmovb  %ecx,%eax

Note cmp and cmovb at the end above.  Actually, the sequence produces
correct results without the cmp and cmovb at the end even for inputs 0
and 1:

  input   0    1    2
  -------------------
  lea    -1    0    1
  lzcnt   0   32   31
  neg     0  -32  -31
  &0x1f   0    0    1
  shlx    1    1    2

This patch recognizes the specific DAG and drops the conditional move,
saving 6 bytes at least.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D145299

Files:
  llvm/lib/Target/X86/X86ISelLowering.cpp
  llvm/test/CodeGen/X86/bit_ceil.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D145299.502358.patch
Type: text/x-patch
Size: 4506 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230304/aaea40eb/attachment.bin>