LLVM backend for 6502
Go to file
Duncan P. N. Exon Smith cc1e1707b8 blockfreq: Rewrite BlockFrequencyInfoImpl
Rewrite the shared implementation of BlockFrequencyInfo and
MachineBlockFrequencyInfo entirely.

The old implementation had a fundamental flaw:  precision losses from
nested loops (or very wide branches) compounded past loop exits (and
convergence points).

The @nested_loops testcase at the end of
test/Analysis/BlockFrequencyAnalysis/basic.ll is motivating.  This
function has three nested loops, with branch weights in the loop headers
of 1:4000 (exit:continue).  The old analysis gives non-sensical results:

    Printing analysis 'Block Frequency Analysis' for function 'nested_loops':
    ---- Block Freqs ----
     entry = 1.0
     for.cond1.preheader = 1.00103
     for.cond4.preheader = 5.5222
     for.body6 = 18095.19995
     for.inc8 = 4.52264
     for.inc11 = 0.00109
     for.end13 = 0.0

The new analysis gives correct results:

    Printing analysis 'Block Frequency Analysis' for function 'nested_loops':
    block-frequency-info: nested_loops
     - entry: float = 1.0, int = 8
     - for.cond1.preheader: float = 4001.0, int = 32007
     - for.cond4.preheader: float = 16008001.0, int = 128064007
     - for.body6: float = 64048012001.0, int = 512384096007
     - for.inc8: float = 16008001.0, int = 128064007
     - for.inc11: float = 4001.0, int = 32007
     - for.end13: float = 1.0, int = 8

Most importantly, the frequency leaving each loop matches the frequency
entering it.

The new algorithm leverages BlockMass and PositiveFloat to maintain
precision, separates "probability mass distribution" from "loop
scaling", and uses dithering to eliminate probability mass loss.  I have
unit tests for these types out of tree, but it was decided in the review
to make the classes private to BlockFrequencyInfoImpl, and try to shrink
them (or remove them entirely) in follow-up commits.

The new algorithm should generally have a complexity advantage over the
old.  The previous algorithm was quadratic in the worst case.  The new
algorithm is still worst-case quadratic in the presence of irreducible
control flow, but it's linear without it.

The key difference between the old algorithm and the new is that control
flow within a loop is evaluated separately from control flow outside,
limiting propagation of precision problems and allowing loop scale to be
calculated independently of mass distribution.  Loops are visited
bottom-up, their loop scales are calculated, and they are replaced by
pseudo-nodes.  Mass is then distributed through the function, which is
now a DAG.  Finally, loops are revisited top-down to multiply through
the loop scales and the masses distributed to pseudo nodes.

There are some remaining flaws.

  - Irreducible control flow isn't modelled correctly.  LoopInfo and
    MachineLoopInfo ignore irreducible edges, so this algorithm will
    fail to scale accordingly.  There's a note in the class
    documentation about how to get closer.  See also the comments in
    test/Analysis/BlockFrequencyInfo/irreducible.ll.

  - Loop scale is limited to 4096 per loop (2^12) to avoid exhausting
    the 64-bit integer precision used downstream.

  - The "bias" calculation proposed on llvmdev is *not* incorporated
    here.  This will be added in a follow-up commit, once comments from
    this review have been handled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206548 91177308-0d34-0410-b5e6-96231b3b80d8
2014-04-18 01:57:45 +00:00
autoconf ARM64: initial backend import 2014-03-29 10:18:08 +00:00
bindings [python-bindings] Remove some cruft that snuck in. 2014-02-23 04:43:26 +00:00
cmake Teach LLVMConfigVersion.cmake to behave as find_package() expects. 2014-04-16 23:15:31 +00:00
docs Fix example for VS2012. 2014-04-18 01:20:08 +00:00
examples Fix warnings about an variable only used in asserts. 2014-03-06 06:35:46 +00:00
include blockfreq: Rewrite BlockFrequencyInfoImpl 2014-04-18 01:57:45 +00:00
lib blockfreq: Rewrite BlockFrequencyInfoImpl 2014-04-18 01:57:45 +00:00
projects Remove projects/sample. 2014-03-12 22:40:22 +00:00
test blockfreq: Rewrite BlockFrequencyInfoImpl 2014-04-18 01:57:45 +00:00
tools Atomics: promote ARM's IR-based atomics pass to CodeGen. 2014-04-17 18:22:47 +00:00
unittests Make the User::value_op_iterator a random access iterator. I had written 2014-04-17 09:07:50 +00:00
utils remove some dead code 2014-04-17 22:26:44 +00:00
.arcconfig Updated phabricator server. 2014-04-07 03:57:04 +00:00
.clang-format Test commit. 2014-03-02 13:08:46 +00:00
.gitignore Remove projects/sample. 2014-03-12 22:40:22 +00:00
CMakeLists.txt [CMake] Reformat, if(MSVC)...else()...endif() 2014-04-14 21:58:19 +00:00
CODE_OWNERS.TXT ARM Linux support 2014-04-02 23:03:28 +00:00
configure ARM64: initial backend import 2014-03-29 10:18:08 +00:00
CREDITS.TXT Test commit. 2014-04-10 22:25:51 +00:00
LICENSE.TXT Remove projects/sample. 2014-03-12 22:40:22 +00:00
llvm.spec.in
LLVMBuild.txt
Makefile [configure/make] Propagate names of build host tools when making BuildTools 2014-03-25 21:45:41 +00:00
Makefile.common
Makefile.config.in Add a --enable-clang-plugin-support option to configure. 2014-03-10 16:58:35 +00:00
Makefile.rules Use -std=gnu++11 on cygwin and mingw. 2014-03-12 20:01:15 +00:00
README.txt

Low Level Virtual Machine (LLVM)
================================

This directory and its subdirectories contain source code for the Low Level
Virtual Machine, a toolkit for the construction of highly optimized compilers,
optimizers, and runtime environments.

LLVM is open source software. You may freely distribute it under the terms of
the license agreement found in LICENSE.txt.

Please see the documentation provided in docs/ for further
assistance with LLVM, and in particular docs/GettingStarted.rst for getting
started with LLVM and docs/README.txt for an overview of LLVM's
documentation setup.

If you're writing a package for LLVM, see docs/Packaging.rst for our
suggestions.