CI: Integrate macOS/arm64 #569
Conversation
Code Review Agent Run #288b18Actionable Suggestions - 0Review Details
AI Code Review powered by |
Quote from the issue. The idea is similar to 7af92d2. |
|
The JIT test fails as mention in the issue. So that this changes can be postponed to be merged until the JIT warning error is handled. See CI. |
Changelist by BitoThis pull request implements the following key changes.
|
There was a problem hiding this comment.
Benchmarks
| Benchmark suite | Current: ad16cd8 | Previous: f6af403 | Ratio |
|---|---|---|---|
Dhrystone |
1239 Average DMIPS over 10 runs |
1331 Average DMIPS over 10 runs |
1.07 |
Coremark |
937.092 Average iterations/sec over 10 runs |
946.757 Average iterations/sec over 10 runs |
1.01 |
This comment was automatically generated by workflow using github-action-benchmark.
|
The macOS host CI completes in approximately 25 minutes, exceeding the Linux host by 7 minutes but still noticeably faster than Linux/arm64. This seems acceptable for adopting the macOS host CI. |
| strategy: | ||
| fail-fast: false | ||
| matrix: | ||
| compiler: [gcc-14, clang] |
There was a problem hiding this comment.
I am not sure if it makes sense to validate gcc-14 builds on macOS.
There was a problem hiding this comment.
I am not sure if it makes sense to validate gcc-14 builds on macOS.
gcc is still available via brew install gcc (specifically as gcc-14, the latest version). Embracing gcc builds in CI on macOS could benefit gcc enthusiasts? Additionally, building with gcc on macOS uncovered some potential build errors or warnings, which have been addressed in these changes.
There was a problem hiding this comment.
I agree with the use of gcc. Let's skip the tests on different optimization flags since we already check the general optimizations on Ubuntu Linux prior to macOS builds.
There was a problem hiding this comment.
I agree with the use of gcc. Let's skip the tests on different optimization flags since we already check the general optimizations on Ubuntu Linux prior to macOS builds.
But, the softfloat warning is revealed with gcc-14 + O1 flag on macOS, see the commit message of 1d497b1. Shall we keep them?
There was a problem hiding this comment.
But, the softfloat warning is revealed with
gcc-14+ O1 flag on macOS, see the commit message of 1d497b1. Shall we keep them?
The default CFLAGS includes -O2, and it should be sufficient to make the compilation optimization issues verbose.
There was a problem hiding this comment.
But, the softfloat warning is revealed with
gcc-14+ O1 flag on macOS, see the commit message of 1d497b1. Shall we keep them?The default
CFLAGSincludes-O2, and it should be sufficient to make the compilation optimization issues verbose.
Got it, but I will remain the '-Wno-uninitialized' CFLAGS for softfloat.
ChinYikMing
force-pushed
the
ci-macOS
branch
from
7af92d2 to
60e6807
Compare
Code Review Agent Run Status
|
ChinYikMing
force-pushed
the
ci-macOS
branch
from
60e6807 to
7a032db
Compare
Code Review Agent Run Status
|
ChinYikMing
force-pushed
the
ci-macOS
branch
from
7a032db to
2d9f2f1
Compare
Code Review Agent Run Status
|
ChinYikMing
force-pushed
the
ci-macOS
branch
2 times, most recently
from
8172b84 to
cd7c0e1
Compare
Code Review Agent Run Status
|
ChinYikMing
force-pushed
the
ci-macOS
branch
from
cd7c0e1 to
1482689
Compare
Code Review Agent Run Status
|
|
TODO: add emcc build in CI. #576 Edit: done. |
ChinYikMing
force-pushed
the
ci-macOS
branch
from
1482689 to
4d27818
Compare
Code Review Agent Run Status
|
ChinYikMing
force-pushed
the
ci-macOS
branch
from
4d27818 to
295097a
Compare
Code Review Agent Run Status
|
ChinYikMing
force-pushed
the
ci-macOS
branch
from
295097a to
eccf1a4
Compare
Code Review Agent Run Status
|
ChinYikMing
force-pushed
the
ci-macOS
branch
from
eccf1a4 to
da51ade
Compare
Code Review Agent Run Status
|
|
One remaining issue to be fixed is that fetching failure of the latest release tag of rv32emu-prebuilt. on macOS/arm64, the failure can be happened very often. The commit da51ade eliminates calling the Github API then avoid the issue. Trying to investigate the root cause. |
Error:
Undefined symbols for architecture arm64:
"_ebreak_handler", referenced from:
lC0 in riscv.o
"_ecall_handler", referenced from:
lC0 in riscv.o
"_log_impl", referenced from:
_rv_create in riscv.o
_rv_create in riscv.o
_rv_create in riscv.o
_rv_profile in riscv.o
_rv_profile in riscv.o
_rv_run in riscv.o
"_log_level_string", referenced from:
_rv_create in riscv.o
"_log_set_level", referenced from:
_rv_create in riscv.o
"_log_set_stdout_stream", referenced from:
_rv_remap_stdstream in riscv.o
"_memcpy_handler", referenced from:
lC0 in riscv.o
"_memory_delete", referenced from:
_rv_delete in riscv.o
_rv_create in riscv.o
"_memory_ifetch", referenced from:
_on_mem_ifetch in riscv.o
"_memory_new", referenced from:
_rv_create in riscv.o
"_memory_read_b", referenced from:
_on_mem_read_b in riscv.o
"_memory_read_s", referenced from:
_on_mem_read_s in riscv.o
"_memory_read_w", referenced from:
_on_mem_read_w in riscv.o
"_memory_write_b", referenced from:
_on_mem_write_b in riscv.o
"_memory_write_s", referenced from:
_on_mem_write_s in riscv.o
"_memory_write_w", referenced from:
_on_mem_write_w in riscv.o
"_memset_handler", referenced from:
lC0 in riscv.o
"_rv", referenced from:
_rv_async_block_clear in riscv.o
"_rv_step", referenced from:
_rv_run in riscv.o
_rv_run in riscv.o
"_trap_handler", referenced from:
lC0 in riscv.o
ld: symbol(s) not found for architecture arm64
collect2: error: ld returned 1 exit status
make: *** [build/rv_histogram] Error 1
Add emulate.o, syscall.o, syscall_sdl.o, io.o, and log.o objects and
conditional build rv_async_block_clear() fixes the error and passes the
build.
Specify using gcc-14 instead of gcc (which typically points to the
latest version of gcc) due to a macOS limitation in the
rlalik/setup-cpp-compiler@master action. According to [1], each gcc must
have specified version.
Softfloat build warning error using gcc-14 with optimization flag O1:
In file included from src/softfloat/source/include/internals.h:42,
from src/softfloat/source/s_mulAddF64.c:40:
In function 'softfloat_sub128',
inlined from 'softfloat_mulAddF64' at
src/softfloat/source/s_mulAddF64.c:185:17:
src/softfloat/source/include/primitives.h:526:17: error: 'sig128C.v64'
may be used uninitialized [-Werror=maybe-uninitialized]
526 | z.v64 = a64 - b64;
| ~~~~^~~~~
src/softfloat/source/s_mulAddF64.c: In function 'softfloat_mulAddF64':
src/softfloat/source/s_mulAddF64.c:66:20: note: 'sig128C.v64' was
declared here
66 | struct uint128 sig128C;
| ^~~~~~~
In function 'softfloat_sub128',
inlined from 'softfloat_mulAddF64' at
src/softfloat/source/s_mulAddF64.c:185:17:
src/softfloat/source/include/primitives.h:527:18: error: 'sig128C.v0'
may be used uninitialized [-Werror=maybe-uninitialized]
527 | z.v64 -= (a0 < b0);
| ~~~~^~~~~
src/softfloat/source/s_mulAddF64.c: In function 'softfloat_mulAddF64':
src/softfloat/source/s_mulAddF64.c:66:20: note: 'sig128C.v0' was
declared here
66 | struct uint128 sig128C;
Add -Wno-initialized to surpress them.
Drop the undefined behavior test on macOS/arm64 using gcc-14, as its
libsanitizer's configure.txt does not yet support it, check [2].
[1] https://github.com/rlalik/setup-cpp-compiler
[2] https://github.com/iains/gcc-darwin-arm64/blob/master-wip-apple-si/
libsanitizer/configure.tgt
Close sysprog21#519
unused function 'emit_dataproc_3source' [-Werror,-Wunused-function]
Refer to [1] and [2], leveraging ${{ secrets.GITHUB_TOKEN }} to be
authenticated which increases the Github API request rate limit.
[1] https://docs.github.com/en/rest/using-the-rest-api/rate-limits-for-the-rest-api?apiVersion=2022-11-28#primary-rate-limit-for-github_token-in-github-actions
[2] https://docs.github.com/en/rest/using-the-rest-api/rate-limits-for-the-rest-api?apiVersion=2022-11-28#getting-a-higher-rate-limit
Code Review Agent Run Status
|
Summary by Bito
This PR enhances the CI pipeline support for macOS/arm64 builds by updating build configurations and managing macOS-specific dependencies. Key improvements include restructuring prebuilt artifact handling, implementing conditional compilation guards for RISC-V instructions, and overall optimizations for cross-platform compatibility.Unit tests added: False
Estimated effort to review (1-5, lower is better): 2