Can't compile tensorflow 2.19.1 on riscv #102159
Description
Issue type
Build/Install
Have you reproduced the bug with TensorFlow Nightly?
No
Source
source
TensorFlow version
tf 2.19.1
Custom code
Yes
OS platform and distribution
EulixOS 12.3.1-33.eos30
Mobile device
No response
Python version
3.11.6
Bazel version
2.19.1
GCC/compiler version
gcc (GCC) 12.3.1 (EulixOS 12.3.1-33.eos30)
CUDA/cuDNN version
No
GPU model and memory
No
Current behavior?
I am trying to compile tf 2.19.1 on riscv64. to add support and let bazel build tf with the python on riscv, I create a fold :third_party/python_riscv_reposity which contains BUILD.bazel ,WORKSPACE and riscv64_py_cc_toolchain.bzl. The detail comments is droped below.And I register them in WORKSPACE
local_repository(
name ="python_riscv64-unknown-linux-gnu",
path = "third_party/python_riscv64_toolchain",
)
register_toolchains(
"@python_riscv64-unknown-linux-gnu//:python_riscv64_toolchain",
"@python_riscv64-unknown-linux-gnu//:riscv64_py_cc_toolchain",
)
Now I meet:
ERROR: /root/.cache/bazel/_bazel_root/7b20b9b6ea80f6f214c7e8b22473e299/external/rules_python/python/cc/BUILD.bazel:15:22: in current_py_cc_headers rule @rules_python//python/cc:current_py_cc_headers:
Traceback (most recent call last):
File "/root/.cache/bazel/_bazel_root/7b20b9b6ea80f6f214c7e8b22473e299/external/rules_python/python/private/current_py_cc_headers.bzl", line 21, column 27, in _current_py_cc_headers_impl
return py_cc_toolchain.headers.providers_map.values()
Error: <target @python_riscv64-unknown-linux-gnu//:python_headers> (rule 'cc_library') doesn't have provider 'headers'
ERROR: /root/.cache/bazel/_bazel_root/7b20b9b6ea80f6f214c7e8b22473e299/external/rules_python/python/cc/BUILD.bazel:15:22: Analysis of target '@rules_python//python/cc:current_py_cc_headers' failed
ERROR: Analysis of target '//tensorflow/tools/pip_package:wheel' failed; build aborted:
How can I solve it?
Standalone code to reproduce the issue
WORKSPACE
workspace(name = "python_riscv64_toolchain_repo")
BUILD.bazel
load("@rules_python//python:defs.bzl", "py_runtime", "py_runtime_pair")
load("riscv64_py_cc_toolchain.bzl", "riscv64_py_cc_toolchain_rule")
package(default_visibility = ["//visibility:public"])
cc_library(
name = "python_headers",
srcs=[],
hdrs = glob(["python3.11/*.h"]),
includes = ["python3.11"],
visibility = ["//visibility:public"],
)
filegroup(
name = "libpython",
srcs = ["libpython3.11.so"],
)
filegroup(
name = "python3",
srcs = ["bin/python"],
)
py_runtime(
name = "riscv64_py_runtime",
interpreter_path = "/AI/zjg/python/venv11/bin/python",
python_version = "PY3",
visibility = ["//visibility:public"],
)
py_runtime_pair(
name = "riscv64_py_runtime_pair",
py3_runtime = ":riscv64_py_runtime",
visibility = ["//visibility:public"],
)
toolchain(
name = "python_riscv64_toolchain",
toolchain_type = "@rules_python//python:toolchain_type",
toolchain = ":riscv64_py_runtime_pair",
visibility = ["//visibility:public"],
)
riscv64_py_cc_toolchain_rule(
name = "riscv64_py_cc_impl",
python_includes = ["python3.11"],
python_headers = [":python_headers"],
)
toolchain(
name = "riscv64_py_cc_toolchain",
toolchain_type = "@rules_python//python/cc:toolchain_type",
toolchain = ":riscv64_py_cc_impl",
visibility = ["//visibility:public"],
)
riscv64_py_cc_toolchain.bzl
def _riscv64_py_cc_toolchain_impl(ctx):
py_cc = ctx.attr.python_headers[0]
return platform_common.ToolchainInfo(
# py_cc_toolchain = ctx.attr.python_headers,
py_cc_toolchain = py_cc,
python_headers = ctx.attr.python_headers,
python_includes = ctx.attr.python_includes,
)
riscv64_py_cc_toolchain_rule = rule(
implementation = _riscv64_py_cc_toolchain_impl,
attrs = {
"python_includes": attr.string_list(),
"python_headers": attr.label_list(allow_files=False),
},
)Relevant log output
No