CI: Build check with various optimization levels #546

ChinYikMing · 2025-01-25T13:32:18Z

The submodules are built locally, which may lead to build failures. Specifically, the dtc submodule fails to build when using GCC versions 11.4.0 or 12.3.0 with optimization levels set to -O0 or with debug flag -g. This is a critical issue, as it can disable the debug mode. To address this, optimization levels and debug level build are introduced to stabilize the pull request changes.

The compiled error:
/usr/bin/ld: /tmp/cc2eaHdM.ltrans0.ltrans.o: in function fdt_del_node': :(.text+0x279f0): undefined reference to fdt_node_end_offset_' collect2: error: ld returned 1 exit status

Note: I have only used gcc 11.4.0 and gcc 12.3.0 to test this out. Github runners should test other compilers.

After bisecting, the original cause was found after #534.

ChinYikMing · 2025-01-25T13:36:27Z

See action for more information.

Clang has no problem with various optimization build. Interesting.

RinHizakura · 2025-01-25T13:54:58Z

This is not the right way to fix it. The issue could happen because I only compile the object that I think we need, but maybe more objects are required for -g + -O0. However, this can't explain the different results on Clang and GCC that you mentioned. I also have no idea here.

Can you check if you can also fix this with the following change?

--- a/mk/system.mk
+++ b/mk/system.mk
@@ -28,7 +28,7 @@ DEV_OBJS := $(patsubst $(DEV_SRC)/%.c, $(DEV_OUT)/%.o, $(wildcard $(DEV_SRC)/*.c
 deps := $(DEV_OBJS:%.o=%.o.d)

 OBJS_EXT += system.o
-OBJS_EXT += dtc/libfdt/fdt.o dtc/libfdt/fdt_ro.o dtc/libfdt/fdt_rw.o
+OBJS_EXT += dtc/libfdt/fdt.o dtc/libfdt/fdt_ro.o dtc/libfdt/fdt_rw.o dtc/libfdt/fdt_wip.o

RinHizakura · 2025-01-25T14:01:11Z

Oh sorry, not aware that you are just trying to robust the CI/CD here instead of fixing anything. We can accept the test first and fix it in another PR.

RinHizakura · 2025-01-25T14:05:15Z

.github/workflows/main.yml

+        CC: ${{ steps.install_cc.outputs.cc }}
+      run: |
+            make distclean
+            make OPT_LEVEL=-Ofast ENABLE_SYSTEM=1 -j$(nproc)


Not sure if we need to take care of -Os or even just retain the presentative -O level only, otherwise the possible options may be too many. Will let @jserv decide this.

One more -Oz can be the candidate. It is related to size just like -Os. Thus, not embracing them at first.

ChinYikMing · 2025-01-25T14:08:07Z

Oh sorry, not aware that you are just trying to robust the CI/CD here instead of fixing anything. We can accept the test first and fix it in another PR.

Agree. Thanks! Since with default -O2 build, the problem does not exist. I revealed this problem when trying to do some debugging.

ChinYikMing · 2025-01-25T14:13:05Z

This is not the right way to fix it. The issue could happen because I only compile the object that I think we need, but maybe more objects are required for -g + -O0. However, this can't explain the different results on Clang and GCC that you mentioned. I also have no idea here.

Can you check if you can also fix this with the following change?
--- a/mk/system.mk
+++ b/mk/system.mk
@@ -28,7 +28,7 @@ DEV_OBJS := $(patsubst $(DEV_SRC)/%.c, $(DEV_OUT)/%.o, $(wildcard $(DEV_SRC)/*.c
 deps := $(DEV_OBJS:%.o=%.o.d)

 OBJS_EXT += system.o
-OBJS_EXT += dtc/libfdt/fdt.o dtc/libfdt/fdt_ro.o dtc/libfdt/fdt_rw.o
+OBJS_EXT += dtc/libfdt/fdt.o dtc/libfdt/fdt_ro.o dtc/libfdt/fdt_rw.o dtc/libfdt/fdt_wip.o

Tested it out, fixed it! Another commit will be push later.

ChinYikMing · 2025-01-25T14:14:48Z

@RinHizakura One more thing, I noticed the dtc submodule has integrated the dtc compiler, shall we use it? So that, the new user does not need to install the dtc on their own.

ChinYikMing · 2025-01-25T14:20:24Z

The action that passed the various optimization level build after apply commit 139af81: link.

otteryc · 2025-01-26T03:04:51Z

Link-Time Optimization (LTO) may be the culprit when building with -O0 flag. GCC worked fine after adding the -fno-lto flag.

ChinYikMing · 2025-01-26T06:45:50Z

Link-Time Optimization (LTO) may be the culprit when building with -O0 flag. GCC worked fine after adding the -fno-lto flag.

Without modifying the lto configuration, adding this object dtc/libfdt/fdt_wip.o solves it.

RinHizakura · 2025-01-27T04:39:32Z

One more thing, I noticed the dtc submodule has integrated the dtc compiler, shall we use it? So that, the new user does not need to install the dtc on their own.

I think we don't need to introduce such complications because building dtc may require other dependencies.

jserv

Rebase the latest master branch to resolve Arm64 host breakage on CI.

jserv

Benchmarks

Benchmark suite	Current: `1a2bcc9`	Previous: `64ea138`	Ratio
`Dhrystone`	`1313` Average DMIPS over 10 runs	`1324` Average DMIPS over 10 runs	`1.01`
`Coremark`	`962.001` Average iterations/sec over 10 runs	`969.943` Average iterations/sec over 10 runs	`1.01`

This comment was automatically generated by workflow using github-action-benchmark.

The submodules are built locally, which may lead to build failures. Specifically, the dtc submodule fails to build when using GCC versions 11.4.0 or 12.3.0 with optimization levels set to -O0 or with debug flag -g. This is a critical issue, as it can disable the debug mode. To address this, optimization levels and debug level build are introduced to stabilize the pull request changes. The compiled error: /usr/bin/ld: /tmp/cc2eaHdM.ltrans0.ltrans.o: in function fdt_del_node': <artificial>:(.text+0x279f0): undefined reference to fdt_node_end_offset_' collect2: error: ld returned 1 exit status Note: I have only used gcc 11.4.0 and gcc 12.3.0 to test this out. Github runners should test other compilers. After bisecting, the original cause was found after sysprog21#534.

The build error: /usr/bin/ld: /tmp/cc2eaHdM.ltrans0.ltrans.o: in function fdt_del_node': <artificial>:(.text+0x279f0): undefined reference to fdt_node_end_offset_' collect2: error: ld returned 1 exit status

ChinYikMing requested a review from RinHizakura January 25, 2025 13:32

RinHizakura reviewed Jan 25, 2025

View reviewed changes

jserv requested changes Jan 27, 2025

View reviewed changes

ChinYikMing force-pushed the ci-opt-level-build branch from 139af81 to 1a2bcc9 Compare January 27, 2025 17:51

jserv reviewed Jan 27, 2025

View reviewed changes

ChinYikMing added 2 commits January 28, 2025 02:00

Fix dtc submodule build error when using gcc -g or -O0

0ffcdf6

The build error: /usr/bin/ld: /tmp/cc2eaHdM.ltrans0.ltrans.o: in function fdt_del_node': <artificial>:(.text+0x279f0): undefined reference to fdt_node_end_offset_' collect2: error: ld returned 1 exit status

ChinYikMing force-pushed the ci-opt-level-build branch from 1a2bcc9 to 0ffcdf6 Compare January 27, 2025 18:00

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

CI: Build check with various optimization levels #546

CI: Build check with various optimization levels #546

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025 •

edited

Loading

RinHizakura commented Jan 25, 2025 •

edited

Loading

RinHizakura commented Jan 25, 2025 •

edited

Loading

RinHizakura Jan 25, 2025 •

edited

Loading

ChinYikMing Jan 25, 2025 •

edited

Loading

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025 •

edited

Loading

otteryc commented Jan 26, 2025 •

edited

Loading

ChinYikMing commented Jan 26, 2025

RinHizakura commented Jan 27, 2025

jserv left a comment

jserv left a comment

CI: Build check with various optimization levels #546

Are you sure you want to change the base?

CI: Build check with various optimization levels #546

Conversation

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025 • edited Loading

RinHizakura commented Jan 25, 2025 • edited Loading

RinHizakura commented Jan 25, 2025 • edited Loading

RinHizakura Jan 25, 2025 • edited Loading

Choose a reason for hiding this comment

ChinYikMing Jan 25, 2025 • edited Loading

Choose a reason for hiding this comment

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025

ChinYikMing commented Jan 25, 2025 • edited Loading

otteryc commented Jan 26, 2025 • edited Loading

ChinYikMing commented Jan 26, 2025

RinHizakura commented Jan 27, 2025

jserv left a comment

Choose a reason for hiding this comment

jserv left a comment

Choose a reason for hiding this comment

Benchmarks

ChinYikMing commented Jan 25, 2025 •

edited

Loading

RinHizakura commented Jan 25, 2025 •

edited

Loading

RinHizakura commented Jan 25, 2025 •

edited

Loading

RinHizakura Jan 25, 2025 •

edited

Loading

ChinYikMing Jan 25, 2025 •

edited

Loading

ChinYikMing commented Jan 25, 2025 •

edited

Loading

otteryc commented Jan 26, 2025 •

edited

Loading