Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

RVV 1.0 integration testing on real hardware #1184

Open
wants to merge 32 commits into
base: master
Choose a base branch
from

Conversation

alitariq4589
Copy link

This PR contains a riscv64-cross.txt file for cross compiling for riscv64 architecture on meson, a cloud-v-pipeline file for CI flow and a Makefile for running tests on RISC-V architecture (since meson fails when binaries are transferred from one platform to the other).

The Makefile contains a very simple bash script in it which executes are the executables in the build/test directory recursively and prints out the passed and failed tests' count along with the total number of counts.

Currently the passing criteria is that all the binaries should be executed with a return code of zero. Else the test will fail.

A complete CI build of this is present at: https://dash.cloud-v.co/blue/organizations/jenkins/simde-fork/detail/simde-fork/48/pipeline/51/

Copy link
Collaborator

@mr-c mr-c left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you @alitariq4589 , I left you some comments

riscv64-cross.txt Outdated Show resolved Hide resolved
cloud-v-pipeline Outdated Show resolved Hide resolved
cloud-v-pipeline Outdated Show resolved Hide resolved
Makefile Outdated Show resolved Hide resolved
@alitariq4589
Copy link
Author

@mr-c I made all the changes you requested. Also, previously, the compile was for riscv64 default target (which was rv64gc). Now I have added -march=rv64gcv to compile for vector extension. For rv64gc, all 2002 tests pass but for rv64gcv, tests are failing. I think it is because of some porting issue in the source code macros.

@mr-c
Copy link
Collaborator

mr-c commented May 27, 2024

Thank you @alitariq4589 ; this is exactly the sort of issue we want to catch using CI!

@zengdage @howjmay @eric900115 Can you look into this compile errors when using -march=rv64gcv and also why they weren't caught with our existing testing?

Details

../test/arm/neon/../../../simde/arm/neon/types.h:54:5: error: unknown type name 'fixed_vint8m1_t'
   54 |     fixed_vint8m1_t sv64;
      |     ^~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:67:5: error: unknown type name 'fixed_vint16m1_t'
   67 |     fixed_vint16m1_t sv64;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:80:5: error: unknown type name 'fixed_vint32m1_t'
   80 |     fixed_vint32m1_t sv64;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:93:5: error: unknown type name 'fixed_vint64m1_t'
   93 |     fixed_vint64m1_t sv64;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:106:5: error: unknown type name 'fixed_vuint8m1_t'
  106 |     fixed_vuint8m1_t sv64;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:119:5: error: unknown type name 'fixed_vuint16m1_t'
  119 |     fixed_vuint16m1_t sv64;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:132:5: error: unknown type name 'fixed_vuint32m1_t'
  132 |     fixed_vuint32m1_t sv64;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:145:5: error: unknown type name 'fixed_vuint64m1_t'
  145 |     fixed_vuint64m1_t sv64;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:175:5: error: unknown type name 'fixed_vfloat32m1_t'
  175 |     fixed_vfloat32m1_t sv64;
      |     ^~~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:188:5: error: unknown type name 'fixed_vfloat64m1_t'
  188 |     fixed_vfloat64m1_t sv64;
      |     ^~~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:196:5: error: unknown type name 'fixed_vuint8m1_t'
  196 |     fixed_vuint8m1_t sv64;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:203:5: error: unknown type name 'fixed_vuint16m1_t'
  203 |     fixed_vuint16m1_t sv64;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:210:5: error: unknown type name 'fixed_vuint64m1_t'
  210 |     fixed_vuint64m1_t sv64;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:230:5: error: unknown type name 'fixed_vint8m1_t'
  230 |     fixed_vint8m1_t sv128;
      |     ^~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:251:5: error: unknown type name 'fixed_vint16m1_t'
  251 |     fixed_vint16m1_t sv128;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:276:5: error: unknown type name 'fixed_vint32m1_t'
  276 |     fixed_vint32m1_t sv128;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:297:5: error: unknown type name 'fixed_vint64m1_t'
  297 |     fixed_vint64m1_t sv128;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:318:5: error: unknown type name 'fixed_vuint8m1_t'
  318 |     fixed_vuint8m1_t sv128;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:339:5: error: unknown type name 'fixed_vuint16m1_t'
  339 |     fixed_vuint16m1_t sv128;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:360:5: error: unknown type name 'fixed_vuint32m1_t'
  360 |     fixed_vuint32m1_t sv128;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:381:5: error: unknown type name 'fixed_vuint64m1_t'
  381 |     fixed_vuint64m1_t sv128;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:427:5: error: unknown type name 'fixed_vfloat32m1_t'
  427 |     fixed_vfloat32m1_t sv128;
      |     ^~~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:448:5: error: unknown type name 'fixed_vfloat64m1_t'
  448 |     fixed_vfloat64m1_t sv128;
      |     ^~~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:456:5: error: unknown type name 'fixed_vuint8m1_t'
  456 |     fixed_vuint8m1_t sv128;
      |     ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:463:5: error: unknown type name 'fixed_vuint16m1_t'
  463 |     fixed_vuint16m1_t sv128;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:470:5: error: unknown type name 'fixed_vuint64m1_t'
  470 |     fixed_vuint64m1_t sv128;
      |     ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:781:11: error: unknown type name 'fixed_vint8m1_t'
  781 |   typedef fixed_vint8m1_t    simde_int8x8_t;
      |           ^~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:782:11: error: unknown type name 'fixed_vint16m1_t'
  782 |   typedef fixed_vint16m1_t   simde_int16x4_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:783:11: error: unknown type name 'fixed_vint32m1_t'
  783 |   typedef fixed_vint32m1_t   simde_int32x2_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:784:11: error: unknown type name 'fixed_vint64m1_t'
  784 |   typedef fixed_vint64m1_t   simde_int64x1_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:785:11: error: unknown type name 'fixed_vuint8m1_t'
  785 |   typedef fixed_vuint8m1_t   simde_uint8x8_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:786:11: error: unknown type name 'fixed_vuint16m1_t'
  786 |   typedef fixed_vuint16m1_t  simde_uint16x4_t;
      |           ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:787:11: error: unknown type name 'fixed_vuint32m1_t'
  787 |   typedef fixed_vuint32m1_t  simde_uint32x2_t;
      |           ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:788:11: error: unknown type name 'fixed_vuint64m1_t'
  788 |   typedef fixed_vuint64m1_t  simde_uint64x1_t;
      |           ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:789:11: error: unknown type name 'fixed_vfloat32m1_t'
  789 |   typedef fixed_vfloat32m1_t simde_float32x2_t;
      |           ^~~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:790:11: error: unknown type name 'fixed_vfloat64m1_t'
  790 |   typedef fixed_vfloat64m1_t simde_float64x1_t;
      |           ^~~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:792:11: error: unknown type name 'fixed_vint8m1_t'
  792 |   typedef fixed_vint8m1_t    simde_int8x16_t;
      |           ^~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:793:11: error: unknown type name 'fixed_vint16m1_t'
  793 |   typedef fixed_vint16m1_t   simde_int16x8_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:794:11: error: unknown type name 'fixed_vint32m1_t'
  794 |   typedef fixed_vint32m1_t   simde_int32x4_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:795:11: error: unknown type name 'fixed_vint64m1_t'
  795 |   typedef fixed_vint64m1_t   simde_int64x2_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:796:11: error: unknown type name 'fixed_vuint8m1_t'
  796 |   typedef fixed_vuint8m1_t   simde_uint8x16_t;
      |           ^~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:797:11: error: unknown type name 'fixed_vuint16m1_t'
  797 |   typedef fixed_vuint16m1_t  simde_uint16x8_t;
      |           ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:798:11: error: unknown type name 'fixed_vuint32m1_t'
  798 |   typedef fixed_vuint32m1_t  simde_uint32x4_t;
      |           ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:799:11: error: unknown type name 'fixed_vuint64m1_t'
  799 |   typedef fixed_vuint64m1_t  simde_uint64x2_t;
      |           ^~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:800:11: error: unknown type name 'fixed_vfloat32m1_t'
  800 |   typedef fixed_vfloat32m1_t simde_float32x4_t;
      |           ^~~~~~~~~~~~~~~~~~
../test/arm/neon/../../../simde/arm/neon/types.h:801:11: error: unknown type name 'fixed_vfloat64m1_t'
  801 |   typedef fixed_vfloat64m1_t simde_float64x2_t;
      |           ^~~~~~~~~~~~~~~~~~

Full log: https://dash.cloud-v.co/job/simde/15/consoleFull

@zengdage
Copy link
Contributor

@alitariq4589 @mr-c Hi, I think the compiler errors are caused by fixed_vint8m1_t and it's variants. In gcc/clang, it has be supported by the following patches. Maybe you need to add some flags like-mrvv-vector-bits=128 to build it.

https://gcc.gnu.org/pipermail/gcc-patches/2024-March/648204.html
https://reviews.llvm.org/D142144

f6b10ed6-794c-43f5-8222-7364ac903181

@alitariq4589
Copy link
Author

@zengdage Thanks for providing me the link to the patch changelog. I added the compilation flag which you requested. However, it is unrecognized in the RISC-V 64 GNU GCC cross-compiler. Also, I am using RISC-V 64-bit GNU cross compiler (nightly build pre-release version April 12, 2024 which is the latest version in pre-release list) on x86 because compiling the source code on K230 kendryte is very time-consuming because of the lack of compute resources on it.

Also, I have tried searching for fixed_vint8m1_t string but could not find it in the https://gcc.gnu.org/pipermail/gcc-patches/2024-March/648204.html link. Maybe they did not add it yet?

@zengdage
Copy link
Contributor

@alitariq4589 please look at that, https://godbolt.org/z/r7Ksbo1oP , I think it's the usage of fixed_vint8m1_t in the latest gcc.

@alitariq4589
Copy link
Author

@zengdage Thanks for referring me to the link. The compiler version in the environment is 13.2.0. I will update the compiler to 14.1.0 and will try to compile again.

@alitariq4589
Copy link
Author

I have compiled the master branch from source and the compiled binary includes gcc version 13.2.0. I think latest version 14.1.0 is not yet upstreamed in the RISC-V GNU Toolchain cross compilers which will cause us to wait till they do so.

@alitariq4589
Copy link
Author

I have updated the riscv gnu toolchain version to 14.1.1 and also updated the CI files. Previous errors are resolved but there are some new errors in the log (full log available at: https://dash.cloud-v.co/view/SIMDe/job/simde/25/console).

@alitariq4589
Copy link
Author

@mr-c @zengdage bumping this PR.

@mr-c
Copy link
Collaborator

mr-c commented Jun 11, 2024

@zengdage Any thoughts? See https://dash.cloud-v.co/view/SIMDe/job/simde/25/console

riscv64-unknown-linux-gnu-gcc -march=rv64gcv_zvl256b -mrvv-vector-bits=zvl -mabi=lp64d -Ofast -Itest/arm/neon/addv-emul-c.p -Itest/arm/neon -I../test/arm/neon -I. -I.. -fdiagnostics-color=always -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -std=c99 -g -fopenmp-simd -DSIMDE_CONSTRAINED_COMPILATION -DSIMDE_ENABLE_OPENMP -Wno-psabi -DSIMDE_TEST_BARE -DSIMDE_NO_NATIVE -MD -MQ test/arm/neon/addv-emul-c.p/addv.c.o -MF test/arm/neon/addv-emul-c.p/addv.c.o.d -o test/arm/neon/addv-emul-c.p/addv.c.o -c ../test/arm/neon/addv.c
/tmp/ccPu0NMf.s: Assembler messages:
/tmp/ccPu0NMf.s:1253: Error: instruction vsetvli requires absolute expression
/tmp/ccPu0NMf.s:1253: Error: illegal operands `vsetvli t6,t6,e0,m1,ta,ma'
/tmp/ccPu0NMf.s:1535: Error: instruction vsetvli requires absolute expression
/tmp/ccPu0NMf.s:1535: Error: illegal operands `vsetvli a2,a2,e0,m1,ta,ma'
/tmp/ccPu0NMf.s:2863: Error: instruction vsetvli requires absolute expression
/tmp/ccPu0NMf.s:2863: Error: illegal operands `vsetvli t6,t6,e0,m1,ta,ma'
/tmp/ccPu0NMf.s:3146: Error: instruction vsetvli requires absolute expression
/tmp/ccPu0NMf.s:3146: Error: illegal operands `vsetvli a2,a2,e0,m1,ta,ma'

@zengdage
Copy link
Contributor

instruction vsetvli requires absolute expression

@alitariq4589 Sorry for the late reply, and I don't know why it get these errors. I try to use the gcc 14.1 in Compiler Explorer, https://godbolt.org/z/4qzrdPs7v , and it can compile addv.c successfully.

According to the google result,
https://inbox.sourceware.org/binutils/2022121416064372037428@eswincomputing.com/T/
riscv-software-src/riscv-tests#9

I think the issue have been fixed in gcc 14.1, so may you check that is something mismatched in your configurations.

module load riscv64-gnu-glibc/14.1.1
riscv64-unknown-linux-gnu-gcc --version
meson setup build --cross-file=./docker/cross-files/riscv64-gcc-13-unknown.cross -Db_lundef=false
'''
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does it need to be modified here? 13->14 ?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@mr-c mr-c changed the title RVV 1.0 integration RVV 1.0 integration testing on real hardware Jun 21, 2024
Comment on lines 2 to 3
c = ['riscv64-unknown-linux-gnu-gcc', '-march=rv64gcv_zvl256b', '-mrvv-vector-bits=zvl', '-mabi=lp64d', '-Ofast']
cpp = ['riscv64-unknown-linux-gnu-g++', '-march=rv64gcv_zvl256b', '-mrvv-vector-bits=zvl', '-mabi=lp64d', '-Ofast']
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Normally we don't build with -Ofast, maybe that matters for your cross compiler?

Suggested change
c = ['riscv64-unknown-linux-gnu-gcc', '-march=rv64gcv_zvl256b', '-mrvv-vector-bits=zvl', '-mabi=lp64d', '-Ofast']
cpp = ['riscv64-unknown-linux-gnu-g++', '-march=rv64gcv_zvl256b', '-mrvv-vector-bits=zvl', '-mabi=lp64d', '-Ofast']
c = ['riscv64-unknown-linux-gnu-gcc', '-march=rv64gcv_zvl256b', '-mrvv-vector-bits=zvl', '-mabi=lp64d', '-O3']
cpp = ['riscv64-unknown-linux-gnu-g++', '-march=rv64gcv_zvl256b', '-mrvv-vector-bits=zvl', '-mabi=lp64d', '-O3']

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the suggestion. I will try rebuilding it without -Ofast flag

@alitariq4589
Copy link
Author

@mr-c @zengdage I have run the build with -O3 and without any optimization at all.

  • The build with -O3 failed during compile stage. The complete log is available at this link

  • The build without any optimization flags passed at the compile stage but the tests failed during execution in the kendryte k230 RISC-V compute instance. The complete list of failed tests are available in test.mak.log. The complete log is available at this link

I have not yet configured the pipeline to publish the test.mak.log file so following is the test.mak.log file.
test.mak.log

@alitariq4589
Copy link
Author

Update: I have updated the pipeline file to archive artifacts of builds.

The latest build is #39 where you can check the test.mak.log file in the Build Artifacts section.

@mr-c
Copy link
Collaborator

mr-c commented Jul 1, 2024

Update: I have updated the pipeline file to archive artifacts of builds.

The latest build is #39 where you can check the test.mak.log file in the Build Artifacts section.

Huh, that is a lot of errors. I guess they need reporting upstream. Do you get the same errors without -march=rv64gcv_zvl256b -mrvv-vector-bits=zvl -mabi=lp64d ?

@alitariq4589
Copy link
Author

Update: I have updated the pipeline file to archive artifacts of builds.
The latest build is #39 where you can check the test.mak.log file in the Build Artifacts section.

Huh, that is a lot of errors. I guess they need reporting upstream. Do you get the same errors without -march=rv64gcv_zvl256b -mrvv-vector-bits=zvl -mabi=lp64d ?

@mr-c I tried building and running this just for RV64GC in the past (i.e. without vector extension flags) and it ran successfully. The build log is available at private CI pipeline here

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants