https://github.com/david-xl updated
https://github.com/llvm/llvm-project/pull/73845
>From 15dd2029a68a6fdbbc4eee5764ce966e0533880f Mon Sep 17 00:00:00 2001
From: David Li <davi...@google.com>
Date: Wed, 29 Nov 2023 11:56:31 -0800
Subject: [PATCH] Fix PGO documentation in user manual
---
clang/docs/UsersManual.rst | 51 ++++++++++++++++++++++++++------------
1 file changed, 35 insertions(+), 16 deletions(-)
diff --git a/clang/docs/UsersManual.rst b/clang/docs/UsersManual.rst
index 2e658557b0e310c..af10a73285ce6cf 100644
--- a/clang/docs/UsersManual.rst
+++ b/clang/docs/UsersManual.rst
@@ -2348,9 +2348,10 @@ differences between the two:
1. Profile data generated with one cannot be used by the other, and there is no
conversion tool that can convert one to the other. So, a profile generated
- via ``-fprofile-instr-generate`` must be used with ``-fprofile-instr-use``.
- Similarly, sampling profiles generated by external profilers must be
- converted and used with ``-fprofile-sample-use``.
+ via ``-fprofile-generate`` or ``-fprofile-instr-generate`` must be used with
+ ``-fprofile-use`` or ``-fprofile-instr-use``. Similarly, sampling profiles
+ generated by external profilers must be converted and used with
``-fprofile-sample-use``
+ or ``-fauto-profile``.
2. Instrumentation profile data can be used for code coverage analysis and
optimization.
@@ -2598,6 +2599,8 @@ Of those, 31,977 were spent inside the body of ``bar``.
The last line
of the profile (``2: 0``) corresponds to line 2 inside ``main``. No
samples were collected there.
+.. _prof_instr:
+
Profiling with Instrumentation
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -2607,11 +2610,25 @@ overhead during the profiling, but it provides more
detailed results than a
sampling profiler. It also provides reproducible results, at least to the
extent that the code behaves consistently across runs.
+Clang supports two types of instrumentation: frontend-based and IR-based.
+Frontend-based instrumentation can be enabled with the option
``-fprofile-instr-generate``,
+and IR-based instrumentation can be enabled with the option
``-fprofile-generate``.
+For best performance with PGO, IR-based instrumentation should be used. It has
+the benefits of lower instrumentation overhead, smaller raw profile size, and
+better runtime performance. Frontend-based instrumentation, on the other hand,
+has better source correlation, so it should be used with source line-based
+coverage testing.
+
+The flag ``-fcs-profile-generate`` also instruments programs using the same
+instrumentation method as ``-fprofile-generate``. However, it performs a
+post-inline late instrumentation and can produce context-sensitive profiles.
+
+
Here are the steps for using profile guided optimization with
instrumentation:
1. Build an instrumented version of the code by compiling and linking with the
- ``-fprofile-instr-generate`` option.
+ ``-fprofile-generate`` or ``-fprofile-instr-generate`` option.
.. code-block:: console
@@ -2674,8 +2691,8 @@ instrumentation:
Note that this step is necessary even when there is only one "raw" profile,
since the merge operation also changes the file format.
-4. Build the code again using the ``-fprofile-instr-use`` option to specify the
- collected profile data.
+4. Build the code again using the ``-fprofile-use`` or ``-fprofile-instr-use``
+ option to specify the collected profile data.
.. code-block:: console
@@ -2685,13 +2702,10 @@ instrumentation:
profile. As you make changes to your code, clang may no longer be able to
use the profile data. It will warn you when this happens.
-Profile generation using an alternative instrumentation method can be
-controlled by the GCC-compatible flags ``-fprofile-generate`` and
-``-fprofile-use``. Although these flags are semantically equivalent to
-their GCC counterparts, they *do not* handle GCC-compatible profiles.
-They are only meant to implement GCC's semantics with respect to
-profile creation and use. Flag ``-fcs-profile-generate`` also instruments
-programs using the same instrumentation method as ``-fprofile-generate``.
+Note that ``-fprofile-use`` option is semantically equivalent to
+its GCC counterpart, it *does not* handle profile formats produced by GCC.
+Both ``-fprofile-use`` and ``-fprofile-instr-use`` accept profiles in the
+indexed format, regardeless whether it is produced by frontend or the IR pass.
.. option:: -fprofile-generate[=<dirname>]
@@ -4401,13 +4415,18 @@ Execute ``clang-cl /?`` to see a list of supported
options:
Instrument only functions from files where names
don't match all the regexes separated by a semi-colon
-fprofile-filter-files=<value>
Instrument only functions from files where names
match any regex separated by a semi-colon
- -fprofile-instr-generate=<file>
- Generate instrumented code to collect execution
counts into <file>
+ -fprofile-generate[=<dirname>]
+ Generate instrumented code to collect execution
counts into a raw profile file in directory specified by the argument. The
filename has %m format by default. See :ref:`Profiling With Instrumentation
<prof_instr>` section for details.
+ (overridden by LLVM_PROFILE_FILE env var)
+ -fprofile-instr-generate=<file_name_pattern>
+ Generate instrumented code to collect execution
counts into the file whose name pattern is specified as the argument.
(overridden by LLVM_PROFILE_FILE env var)
-fprofile-instr-generate
Generate instrumented code to collect execution
counts into default.profraw file
(overridden by '=' form of option or
LLVM_PROFILE_FILE env var)
-fprofile-instr-use=<value>
+ Use instrumentation data for coverage testing or
profile-guided optimization
+ -fprofile-use=<value>
Use instrumentation data for profile-guided
optimization
-fprofile-remapping-file=<file>
Use the remappings described in <file> to match
the profile data against names in the program
@@ -4569,7 +4588,7 @@ clang-cl supports several features that require runtime
library support:
- Address Sanitizer (ASan): ``-fsanitize=address``
- Undefined Behavior Sanitizer (UBSan): ``-fsanitize=undefined``
- Code coverage: ``-fprofile-instr-generate -fcoverage-mapping``
-- Profile Guided Optimization (PGO): ``-fprofile-instr-generate``
+- Profile Guided Optimization (PGO): ``-fprofile-generate``
- Certain math operations (int128 division) require the builtins library
In order to use these features, the user must link the right runtime libraries
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
