Add profile summary information to GCOV by dc03-work · Pull Request #251 · google/autofdo

dc03-work · 2025-10-09T13:16:44Z

This patch adds supports for writing out profile summaries and histogram-based percentile information for samples to the GCOV profile, in a similar manner to how this information is written out for LLVM profiles. It also bumps the GCOV version to 3. This is part of the work to add support for the same to GCC.

erozenfeld · 2025-10-20T20:51:59Z

gcov.cc

+    if (absl::GetFlag(FLAGS_gcov_version) == 2 ||
+        absl::GetFlag(FLAGS_gcov_version) == 3) {


Suggested change

if (absl::GetFlag(FLAGS_gcov_version) == 2 ||

absl::GetFlag(FLAGS_gcov_version) == 3) {

if (absl::GetFlag(FLAGS_gcov_version) >= 2) {

erozenfeld · 2025-10-20T20:52:43Z

gcov.cc

+  if (absl::GetFlag(FLAGS_gcov_version) == 2 ||
+      absl::GetFlag(FLAGS_gcov_version) == 3) {


Suggested change

if (absl::GetFlag(FLAGS_gcov_version) == 2 ||

absl::GetFlag(FLAGS_gcov_version) == 3) {

if (absl::GetFlag(FLAGS_gcov_version) >= 2) {

profile_writer.cc

erozenfeld · 2025-11-04T02:02:06Z

@dc03-work I saw that you made some changes to your GCC patch. Does that require corresponding changes in this PR?

dc03-work · 2025-11-04T03:06:00Z

@dc03-work I saw that you made some changes to your GCC patch. Does that require corresponding changes in this PR?

No, but I will be rewriting the histogram calculation as a set of standalone routines.

dc03-work · 2025-11-13T04:47:03Z

@erozenfeld Hi, I have implemented the calculation as a standalone visitor and verified that it produces the same values as LLVM's implementation. Does this look OK now?

dc03-work · 2025-12-23T05:40:03Z

@erozenfeld Similar to #244, the GCC patch corresponding to this has landed as r16-6349-gdbe8e0efb7b029. Can we please land this if there are no immediate issues as well? This is also required for the AutoFDO flow to work.

This patch aims to implement summary support in auto-profile, similar to LLVM. The summary information stores various information about the profile being read such as the number of functions, the maximum sample count, the total number of samples and so on. It also adds a section called the "detailed summary" which contains a histogram-based calculation of the minimum execution count for a sample needed to belong to a specific percentile of samples. This is used to decide the hot count threshold (which can be controlled with a command line parameter). The default is any sample belonging to the 99th percentile being marked as hot. This patch requires the changes from google/autofdo#251 to work correctly. Signed-off-by: Dhruv Chawla <dhruvc@nvidia.com> gcc/ChangeLog: * auto-profile.cc (struct summary_info): New struct. (summary_info::read): New function. (summary_info::get_threshold_count): Likewise. (function_instance::read_function_instance): Read afdo_profile_info->sum_max directly from summary info. (autofdo_source_profile::read): Set afdo_hot_bb_threshold from param_hot_bb_count_ws_permille. (read_profile): Call summary_info->read. (end_auto_profile): Free afdo_summary_info. * gcov-io.h (GCOV_TAG_AFDO_SUMMARY): New define.

erozenfeld

Looks good, just a couple of minor suggestions.

erozenfeld · 2026-01-06T03:31:02Z

profile_reader.cc

+    for (unsigned i = 0; i < num; i++) {
+      gcov_read_unsigned(); // Cutoff
+      gcov_read_counter();  // Min count
+      gcov_read_counter();  // Num cutoffs


Should the comment say
// Num counts
?

erozenfeld · 2026-01-06T03:37:50Z

profile_writer.cc


+  ProfileSummaryInformation info = ProfileSummaryComputer::Compute(
+      *symbol_map_, {std::begin(ProfileSummaryInformation::default_cutoffs),
+                     std::end(ProfileSummaryInformation::default_cutoffs)});


The call to ProfileSummaryComputer::Compute can be moved to under
if (absl::GetFlag(FLAGS_gcov_version) >= 3)
check below.

snehasish

Is it possible to add a test for this?

profile_writer.cc

dc03-work · 2026-01-19T09:02:02Z

Is it possible to add a test for this?

@snehasish I've added the test case. I tried to make it work with the existing binaries / perf data in testdata/ but it seems like there's no binaries with mmap events + build id + debug info, so I hardcoded the values. I hope that is okay.

CMakeLists.txt

profile_reader.h

profile_writer.cc

profile_writer.h

snehasish · 2026-01-21T03:46:08Z

summary_calculation_test.cc

+  };
+  symbol_map.AddSourceCount("boo", boo_stack3, 150, 2);
+
+  ProfileSummaryInformation info = ProfileSummaryComputer::Compute(


Can we test the summary reading and writing by serializing it to a simple profile and then reading it back in?

Sure, I will add that.

snehasish

lgtm, just a couple of optional suggestions. Let me know when you are ready to merge.

snehasish · 2026-01-22T17:15:54Z

profile_writer.h

  StringIndexMap *map_;
 };

+class ProfileSummaryInformation {


If everything is public maybe this is better off as a struct?

snehasish · 2026-01-22T17:17:46Z

summary_calculation_test.cc

+
+  // Make sure the summary that was written out is the same as that which is
+  // read back in.
+  EXPECT_EQ(info_1.total_count_, info_2->total_count_);


If you define operator== for ProfileSummaryInformation and DetailedSummary then I think you can simplify all of these expectations to just EXPECT_THAT(info_1, Eq(info_2)).

This patch adds supports for writing out profile summaries and histogram-based percentile information for samples to the GCOV profile, in a similar manner to how this information is written out for LLVM profiles. It also bumps the GCOV version to 3.

dc03-work · 2026-01-23T05:21:16Z

Addressed comments and rebased on master. Feel free to merge :) Thanks for the reviews!

dc03-work · 2026-01-27T14:45:25Z

profile_writer.cc


+  // Write out the GCOV_TAG_AFDO_SUMMARY section.
+  if (absl::GetFlag(FLAGS_gcov_version) >= 3) {
+    assert(summary != nullptr);


Looks like I accidentally left an assert in here from the first commit that doesn't work anymore. I think it didn't fail to compile because the assert get preprocessed out - the summary variable doesn't exist anywhere anymore.

Send a PR to cleanup?

PR put up as #261.

I forgot to remove this assert as it was not compiled in Release mode.

erozenfeld suggested changes Oct 20, 2025

View reviewed changes

dc03-work force-pushed the gcov-summary-info branch from 72fa47a to 9627fd2 Compare November 13, 2025 04:44

erozenfeld suggested changes Jan 6, 2026

View reviewed changes

erozenfeld approved these changes Jan 7, 2026

View reviewed changes

dc03-work mentioned this pull request Jan 14, 2026

Write out file names to GCOV file #244

Merged

snehasish reviewed Jan 14, 2026

View reviewed changes

profile_writer.cc Show resolved Hide resolved

dc03-work changed the title ~~[RFC] Add profile summary information to GCOV~~ Add profile summary information to GCOV Jan 19, 2026

snehasish reviewed Jan 21, 2026

View reviewed changes

snehasish approved these changes Jan 22, 2026

View reviewed changes

dc03-work added 7 commits January 22, 2026 21:17

Simplify version comparison

96c7bf8

Implement summary calculation as standalone visitor

d123f70

Address review comments

da84bf2

Add testcase for summary calculation

9931071

Address review comments

d25dcf4

Address review comments 2

b9e980d

dc03-work force-pushed the gcov-summary-info branch from 35b9b9b to b9e980d Compare January 23, 2026 05:20

snehasish merged commit e954df8 into google:master Jan 23, 2026
2 of 3 checks passed

dc03-work commented Jan 27, 2026

View reviewed changes

dc03-work added a commit to dc03-work/autofdo that referenced this pull request Jan 30, 2026

Fix incorrect assertion from google#251

db9defb

I forgot to remove this assert as it was not compiled in Release mode.

dc03-work added a commit to dc03-work/autofdo that referenced this pull request Feb 5, 2026

Fix incorrect assertion from google#251

48144af

I forgot to remove this assert as it was not compiled in Release mode.

snehasish pushed a commit that referenced this pull request Feb 11, 2026

Fix incorrect assertion from #251 (#261)

16437b6

I forgot to remove this assert as it was not compiled in Release mode.

		if (absl::GetFlag(FLAGS_gcov_version) == 2 \|\|
		absl::GetFlag(FLAGS_gcov_version) == 3) {

	if (absl::GetFlag(FLAGS_gcov_version) == 2 \|\|
	absl::GetFlag(FLAGS_gcov_version) == 3) {
	if (absl::GetFlag(FLAGS_gcov_version) >= 2) {

Conversation

dc03-work commented Oct 9, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

erozenfeld commented Nov 4, 2025

Uh oh!

dc03-work commented Nov 4, 2025

Uh oh!

dc03-work commented Nov 13, 2025

Uh oh!

dc03-work commented Dec 23, 2025

Uh oh!

erozenfeld left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

erozenfeld Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

snehasish left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

dc03-work commented Jan 19, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

snehasish left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dc03-work commented Jan 23, 2026

Uh oh!

Uh oh!

dc03-work Jan 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

erozenfeld Jan 6, 2026 •

edited

Loading

dc03-work Jan 27, 2026 •

edited

Loading