fix(hipblaslt): use portable formatting for 64-bit values to fix warnings by ScottTodd · Pull Request #12326 · ROCm/rocm-libraries

ScottTodd · 2026-09-18T21:22:43Z

Motivation

This fixes ~500 warnings when building on Windows (see ROCm/TheRock#47), saving 1-2MB of log size per GPU family build.

For example:

3.4	In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/src/msgpack/MessagePack.cpp:27:
3.4	In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/msgpack/MessagePack.hpp:31:
3.4	In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/ContractionLibrary.hpp:34:
3.4	In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/GranularitySelectionLibrary.hpp:38:
3.4	C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/PropertyMatching.hpp:1106:67: warning: format specifies type 'long' but the argument has type 'long long' [-Wformat]
3.4	 1106 |                         printf("[ %ld ,%ld,%ld,%ld]\n", e.key[0], e.key[1], e.key[2], e.key[3]);
3.4	      |                                        ~~~                        ^~~~~~~~
3.4	      |                                        %lld
3.4	C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/PropertyMatching.hpp:1106:77: warning: format specifies type 'long' but the argument has type 'long long' [-Wformat]
3.4	 1106 |                         printf("[ %ld ,%ld,%ld,%ld]\n", e.key[0], e.key[1], e.key[2], e.key[3]);
3.4	      |                                            ~~~                              ^~~~~~~~
3.4	      |                                            %lld
...
3.4	C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/src/msgpack/MessagePack.cpp:413:9: note: in instantiation of function template specialization 'TensileLite::MessagePackLoadLibraryFile<TensileLite::ContractionProblemGemm, TensileLite::ContractionSolution>' requested here
3.4	  413 |         MessagePackLoadLibraryFile<ContractionProblemGemm, ContractionSolution>(
3.4	      |         ^

Technical Details

The %ld format specifier expects long, while the values are int64_t.

On 64-bit Linux, long is typically 64 bits
On 64-bit Windows, long is 32 bits while int64_t is always 64 bits (long long)

See https://en.cppreference.com/cpp/types/integer#Format_macro_constants, PRId64 is recommended for int64_t.

Note

The tensilelite/include/Tensile/PropertyMatching.hpp fix was already applied to Tensile in ROCm/Tensile#1623 / 37278db, including switching from printf to std::cout.

rocm-libraries/shared/tensile/Tensile/Source/lib/include/Tensile/PropertyMatching.hpp

Lines 698 to 701 in ab0a574

    
           auto compM = [&count, Debug](Entry const& e, long const M) { 
        
               if(Debug) 
        
                   std::cout << "[" << e.key[0] << "," << e.key[1] << "," << e.key[2] << "]" 
        
                             << std::endl;

Test Plan

Check that CI logs no longer contain these warnings.

Run details	Logs link	file size
baseline	35375442872-windows/...hipBLASLt_build.log	1247KB
this PR	35397683840-windows/...hipBLASLt_build.log	87KB

Submission Checklist

Look over the contributing guidelines at https://github.com/ROCm/TheRock/blob/main/GOVERNANCE.md#pull-requests.

Use stream output for templated matching keys and PRId64 for the grouped GEMM benchmark index. Avoid assuming that long matches the key or int64_t type on Windows. Comparing the saved September 15 local Windows build log with the September 18 hipBLASLt+dist build, format warnings fall from 536 to 0 and total warnings from 562 to 26. The remaining diagnostics are 18 deprecations, 7 C++17 extensions, and 1 potentially evaluated expression. The latest build completed successfully; git diff --cached --check passed. Generated with Codex

therock-pr-bot · 2026-09-18T21:27:02Z

✅ All Policy Checks Passed

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	⚠️ Warning	Error: Source/code files changed without an accompanying unit test. Expected: add at least one test file named like `test_<name>.py` / `test_<name>.cpp` (or `<name>_test.`). Current:* code file(s) changed: `projects/hipblaslt/clients/bench/src/client_groupedgemm_fixed_mk.cpp`, `projects/hipblaslt/tensilelite/include/Tensile/PropertyMatching.hpp`; no test file found
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

🎉 All policy checks passed!

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

therock-pr-bot · 2026-09-18T21:27:04Z

🎉 All checks passed! This PR is ready for review.

codecov · 2026-09-18T23:12:56Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop   #12326      +/-   ##
===========================================
+ Coverage    70.78%   70.94%   +0.16%     
===========================================
  Files         2842     2844       +2     
  Lines       465020   466180    +1160     
  Branches     68499    68770     +271     
===========================================
+ Hits        329152   330715    +1563     
+ Misses      112174   111583     -591     
- Partials     23694    23882     +188

Flag	Coverage Δ		*Carryforward flag
TensileLite-CPP	`48.75% <ø> (+2.35%)`	⬆️
TensileLite-Unit	`76.46% <ø> (+0.36%)`	⬆️
hipBLAS	`90.62% <ø> (ø)`		Carriedforward from 26b228b
hipBLASLt	`35.24% <ø> (ø)`		Carriedforward from 26b228b
hipCUB	`82.68% <ø> (ø)`		Carriedforward from 26b228b
hipDNN	`87.01% <ø> (ø)`		Carriedforward from 26b228b
hipFFT	`43.25% <ø> (ø)`		Carriedforward from 26b228b
hipRAND	`76.12% <ø> (ø)`		Carriedforward from 26b228b
hipSOLVER	`68.96% <ø> (ø)`		Carriedforward from 26b228b
hipSPARSE	`86.99% <ø> (ø)`		Carriedforward from 26b228b
rocBLAS	`48.31% <ø> (ø)`		Carriedforward from 26b228b
rocFFT	`51.59% <ø> (ø)`		Carriedforward from 26b228b
rocRAND	`57.42% <ø> (ø)`		Carriedforward from 26b228b
rocSOLVER	`76.83% <ø> (ø)`		Carriedforward from 26b228b
rocSPARSE	`74.61% <ø> (ø)`		Carriedforward from 26b228b
rocThrust	`93.15% <ø> (ø)`		Carriedforward from 26b228b

*This pull request uses carry forward flags. Click here to find out more.
see 34 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

newling

LGTM

tony-davis

Approving. Nice cleanup, and I appreciate that you linked both logs instead of just asserting the improvement. I pulled them down and counted myself: -Wformat goes 536 to 0, total warnings 562 to 26, and both builds completed with the same 2537 kernels, so it's a real apples-to-apples comparison.

I also checked the types on both sides. Key resolves to std::array<int64_t, N> (via the mappingKey<std::array<int64_t, ...>> dispatch in Serialization/MatchingLibrary.hpp), and in the client ldd and stride_d are std::vector<int64_t>, so PRId64 is right there. Adding <iostream> also closes a latent gap, since the file was already streaming std::cout around line 1060 without including it.

One non-blocking note, take it or leave it. The comparator lambdas at lines 1105, 1113, and 1121 still take long const M / long const N / long const B, and the values passed in are int64_t (e.g. std::min(key[0], (table.end() - 1)->key[0])). So on Windows those get truncated to 32 bits before the comparison, which is the same assumption this PR is removing from the format strings. You'd need a dimension above 2^31 to actually notice, so it's latent and I don't think it should hold the PR. But it's a one-word change to int64_t on lines you're already touching, so I'd probably just grab it while you're in there. Upstream Tensile has the same thing, so there's no parity reason to keep it.

Tiny thing: the Technical Details section says PPRId64 where it means PRId64.

ScottTodd · 2026-09-23T17:51:26Z

Tiny thing: the Technical Details section says PPRId64 where it means PRId64.

Thanks, fixed (that's what I get for typing descriptions by hand 😅)

ScottTodd · 2026-09-23T18:00:33Z

One non-blocking note, take it or leave it. The comparator lambdas at lines 1105, 1113, and 1121 still take long const M / long const N / long const B, and the values passed in are int64_t (e.g. std::min(key[0], (table.end() - 1)->key[0])). So on Windows those get truncated to 32 bits before the comparison, which is the same assumption this PR is removing from the format strings. You'd need a dimension above 2^31 to actually notice, so it's latent and I don't think it should hold the PR. But it's a one-word change to int64_t on lines you're already touching, so I'd probably just grab it while you're in there. Upstream Tensile has the same thing, so there's no parity reason to keep it.

I'd prefer to fix that across both/all code (shared/tensile/Tensile/Source/lib/include/Tensile/PropertyMatching.hpp and projects/hipblaslt/tensilelite/include/Tensile/PropertyMatching.hpp) in a follow-up PR.

fix(hipblaslt): use portable formatting for 64-bit values to fix warnings (#12326) ## Motivation This fixes ~500 warnings when building on Windows (see ROCm/TheRock#47), saving 1-2MB of log size per GPU family build. For example: ``` 3.4 In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/src/msgpack/MessagePack.cpp:27: 3.4 In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/msgpack/MessagePack.hpp:31: 3.4 In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/ContractionLibrary.hpp:34: 3.4 In file included from C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/GranularitySelectionLibrary.hpp:38: 3.4 C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/PropertyMatching.hpp:1106:67: warning: format specifies type 'long' but the argument has type 'long long' [-Wformat] 3.4 1106 | printf("[ %ld ,%ld,%ld,%ld]\n", e.key[0], e.key[1], e.key[2], e.key[3]); 3.4 | ~~~ ^~~~~~~~ 3.4 | %lld 3.4 C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/include\Tensile/PropertyMatching.hpp:1106:77: warning: format specifies type 'long' but the argument has type 'long long' [-Wformat] 3.4 1106 | printf("[ %ld ,%ld,%ld,%ld]\n", e.key[0], e.key[1], e.key[2], e.key[3]); 3.4 | ~~~ ^~~~~~~~ 3.4 | %lld ... 3.4 C:/home/runner/_work/TheRock/TheRock/rocm-libraries/projects/hipblaslt/tensilelite/src/msgpack/MessagePack.cpp:413:9: note: in instantiation of function template specialization 'TensileLite::MessagePackLoadLibraryFile<TensileLite::ContractionProblemGemm, TensileLite::ContractionSolution>' requested here 3.4 413 | MessagePackLoadLibraryFile<ContractionProblemGemm, ContractionSolution>( 3.4 | ^ ``` ## Technical Details The `%ld` format specifier expects `long`, while the values are `int64_t`. * On 64-bit Linux, `long` is typically 64 bits * On 64-bit Windows, `long` is 32 bits while `int64_t` is always 64 bits (`long long`) See https://en.cppreference.com/cpp/types/integer#Format_macro_constants, `PRId64` is recommended for `int64_t`. > [!NOTE] > The `tensilelite/include/Tensile/PropertyMatching.hpp` fix was already applied to _Tensile_ in ROCm/Tensile#1623 / ROCm/rocm-libraries@37278db, including switching from `printf` to `std::cout`. > https://github.com/ROCm/rocm-libraries/blob/ab0a5742e4480ade753a606fab6d6945fd2c0a51/shared/tensile/Tensile/Source/lib/include/Tensile/PropertyMatching.hpp#L698-L701 ## Test Plan Check that CI logs no longer contain these warnings. Run details | Logs link | file size -- | -- | --: baseline | [35375442872-windows/...hipBLASLt_build.log](https://therock-ci-artifacts-external.s3.amazonaws.com/ROCm-rocm-libraries/35375442872-windows/logs/math-libs/gfx1151/hipBLASLt_build.log) | 1247KB this PR | [35397683840-windows/...hipBLASLt_build.log](https://therock-ci-artifacts-external.s3.amazonaws.com/ROCm-rocm-libraries/35397683840-windows/logs/math-libs/gfx1151/hipBLASLt_build.log) | 87KB ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/TheRock/blob/main/GOVERNANCE.md#pull-requests.

assistant-librarian · 2026-09-25T12:04:17Z

TheRock Submodule Bump Activity

Newest first

2026-09-25 — Included in TheRock via ROCm/TheRock#8472.

github-actions Bot added project: hipblaslt project: hipsparselt ci:hipsparselt-fast labels Sep 18, 2026

assistant-librarian Bot added the organization: ROCm label Sep 18, 2026

ScottTodd marked this pull request as ready for review September 18, 2026 23:23

ScottTodd requested a review from a team as a code owner September 18, 2026 23:23

ScottTodd requested a review from TorreZuk September 18, 2026 23:24

talumbau requested review from talumbau and tony-davis September 23, 2026 15:05

newling approved these changes Sep 23, 2026

View reviewed changes

tony-davis approved these changes Sep 23, 2026

View reviewed changes

ScottTodd merged commit d9eafe9 into develop Sep 23, 2026
237 of 256 checks passed

ScottTodd deleted the users/scotttodd/hipblaslt-format-warnings branch September 23, 2026 18:01

Navigation Menu

Repository navigation

fix(hipblaslt): use portable formatting for 64-bit values to fix warnings - #12326

fix(hipblaslt): use portable formatting for 64-bit values to fix warnings#12326
ScottTodd merged 1 commit into
developfrom
users/scotttodd/hipblaslt-format-warnings

ScottTodd commented Sep 18, 2026 •

edited

Loading

Uh oh!

therock-pr-bot Bot commented Sep 18, 2026 •

edited

Loading

Uh oh!

therock-pr-bot Bot commented Sep 18, 2026

Uh oh!

codecov Bot commented Sep 18, 2026 •

edited

Loading

Uh oh!

newling left a comment

Uh oh!

tony-davis left a comment

Uh oh!

ScottTodd commented Sep 23, 2026

Uh oh!

ScottTodd commented Sep 23, 2026

Uh oh!

Uh oh!

assistant-librarian Bot commented Sep 25, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	auto compM = [&count, Debug](Entry const& e, long const M) {
	if(Debug)
	std::cout << "[" << e.key[0] << "," << e.key[1] << "," << e.key[2] << "]"
	<< std::endl;

Conversation

ScottTodd commented Sep 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Technical Details

Test Plan

Submission Checklist

Uh oh!

therock-pr-bot Bot commented Sep 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ All Policy Checks Passed

Uh oh!

therock-pr-bot Bot commented Sep 18, 2026

Uh oh!

codecov Bot commented Sep 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

newling left a comment

Choose a reason for hiding this comment

Uh oh!

tony-davis left a comment

Choose a reason for hiding this comment

Uh oh!

ScottTodd commented Sep 23, 2026

Uh oh!

ScottTodd commented Sep 23, 2026

Uh oh!

Uh oh!

assistant-librarian Bot commented Sep 25, 2026

TheRock Submodule Bump Activity

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ScottTodd commented Sep 18, 2026 •

edited

Loading

therock-pr-bot Bot commented Sep 18, 2026 •

edited

Loading

codecov Bot commented Sep 18, 2026 •

edited

Loading