Source: blender
Version: 4.0.2+dfsg-1
Severity: important
Tags: FTBFS, patch
User: debian-ri...@lists.debian.org
Usertags: riscv64
X-Debbugs-Cc: debian-ri...@lists.debian.org
Dear llvmlite Maintainer,
The blender compile failed on riscv64 caused by losing precision when cast
from ‘void*’ to ‘CUdeviceptr’.The crucial buildd log below:
```
/<<PKGBUILDDIR>>/intern/cycles/device/cuda/device_impl.cpp:528:25:
error: cast from ‘void*’ to ‘CUdeviceptr’ {aka ‘unsigned int’} loses
precision [-fpermissive]
528 | cuda_assert(cuMemFree((CUdeviceptr)device_pointer));
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~
/<<PKGBUILDDIR>>/intern/cycles/device/../device/cuda/util.h:32:25:
note: in definition of macro ‘cuda_device_assert’
32 | CUresult result = stmt; \
| ^~~~
/<<PKGBUILDDIR>>/intern/cycles/device/cuda/device_impl.cpp:528:3:
note: in expansion of macro ‘cuda_assert’
528 | cuda_assert(cuMemFree((CUdeviceptr)device_pointer));
| ^~~~~~~~~~~
/<<PKGBUILDDIR>>/intern/cycles/device/cuda/device_impl.cpp: In member
function ‘virtual void ccl::CUDADevice::copy_host_to_device(void*,
void*, size_t)’:
/<<PKGBUILDDIR>>/intern/cycles/device/cuda/device_impl.cpp:558:28:
error: cast from ‘void*’ to ‘CUdeviceptr’ {aka ‘unsigned int’} loses
precision [-fpermissive]
558 | cuda_assert(cuMemcpyHtoD((CUdeviceptr)device_pointer,
host_pointer, size));
```
The full buildd log is here:
https://buildd.debian.org/status/fetch.php?pkg=blender&arch=riscv64&ver=4.0.2%2Bdfsg-1&stamp=1704028470&raw=0
My solution to this issue:
Convert the problematic conversion code to a uintptr_t intermediate type.I
have tested this modification in local ,and it works well.Please let me
know wheather this patch can be accepted. The patch is in the attachment.
Gui-Yue
Best Regards
