Blender 3.1 beta crashes when trying to render on the GPU with Metal #95477

New Issue

cotooleimages@gmail.com · 2022-02-03T17:42:55+01:00

2022-02-03 17:42:55 +01:00

System Information
Operating system: macOS-12.1-arm64-arm-64bit 64 Bits
Graphics card: Apple M1 Max Apple 4.1 Metal - 76.3

Blender Version
Broken: version: 3.1.0 Beta, branch: master, commit date: 2022-02-02 15:25, hash: 43b0ff3054
Worked: 3.1 beta (on CPU)

Short description of error
Whenever I try to render this scene on the GPU with Metal, Blender crashes (it renders fine on the CPU).

Exact steps for others to reproduce the error
Just render the scene on the GPU with Metal

Interstellar try1.blend

**System Information** Operating system: macOS-12.1-arm64-arm-64bit 64 Bits Graphics card: Apple M1 Max Apple 4.1 Metal - 76.3 **Blender Version** Broken: version: 3.1.0 Beta, branch: master, commit date: 2022-02-02 15:25, hash: `43b0ff3054` Worked: 3.1 beta (on CPU) **Short description of error** Whenever I try to render this scene on the GPU with Metal, Blender crashes (it renders fine on the CPU). **Exact steps for others to reproduce the error** Just render the scene on the GPU with Metal [Interstellar try1.blend](https://archive.blender.org/developer/F12844033/Interstellar_try1.blend)

cotooleimages@gmail.com commented

2022-02-03 17:42:55 +01:00

Added subscriber: @BlenderChris

Pratik Borhade commented

2022-02-04 02:24:33 +01:00

Added subscribers: @mano-wii, @PratikPB2123

Pratik Borhade commented

2022-02-04 02:24:33 +01:00

@mano-wii , can you check?

Germano Cavalcante commented

2022-02-04 15:17:13 +01:00

Changed status from 'Needs Triage' to: 'Confirmed'

Germano Cavalcante commented

2022-02-04 15:17:13 +01:00

Added subscriber: @Michael-Jones

Germano Cavalcante commented

2022-02-04 15:17:13 +01:00

I can confirm a crash in ccl::MetalDevice::tex_alloc:

__pthread_kill (@__pthread_kill:5)
pthread_kill (@pthread_kill:75)
abort (@abort:44)
__assert_rtn (@__assert_rtn:56)
MTLReportFailure.cold.1 (@bool MTLGetEnvCase<MTLErrorModeType>(char const*, MTLErrorModeType&, std::__1::vector<std::__1::pair<char const*, MTLErrorModeType>, std::__1::allocator<std::__1::pair<char const*, MTLErrorModeType> > > const&) (.cold.1):3)
MTLReportFailure (@MTLReportFailure:122)
_MTLMessageContextEnd (@_MTLMessageContextEnd:270)
-[MTLTextureDescriptorInternal validateWithDevice:] (@-[MTLTextureDescriptorInternal validateWithDevice:]:646)
___lldb_unnamed_symbol5130 (@___lldb_unnamed_symbol5130:20)
ccl::MetalDevice::tex_alloc(ccl::device_texture&) (/Users/germanocavalcante/blender-git/blender/intern/cycles/device/metal/device_impl.mm:890)
ccl::MultiDevice::mem_copy_to(ccl::device_memory&) (/Users/germanocavalcante/blender-git/blender/intern/cycles/device/multi/device.cpp:307)
ccl::ImageManager::device_load_image(ccl::Device*, ccl::Scene*, int, ccl::Progress*) (/Users/germanocavalcante/blender-git/blender/intern/cycles/scene/image.cpp:777)
std::__1::__function::__value_func<void ()>::operator()() const (/Users/germanocavalcante/Applications/Xcode-beta.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.0.sdk/usr/include/c++/v1/functional:1885)
std::__1::function<void ()>::operator()() const (/Users/germanocavalcante/Applications/Xcode-beta.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.0.sdk/usr/include/c++/v1/functional:2560)
tbb::internal::function_task<std::__1::function<void ()> >::execute() (/Users/germanocavalcante/blender-git/lib/darwin_arm64/tbb/include/tbb/task.h:1059)
tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::process_bypass_loop(tbb::internal::context_guard_helper<false>&, tbb::task*, long) (@tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::process_bypass_loop(tbb::internal::context_guard_helper<false>&, tbb::task*, long):113)
tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::local_wait_for_all(tbb::task&, tbb::task*) (@tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::local_wait_for_all(tbb::task&, tbb::task*):49)
tbb::internal::arena::process(tbb::internal::generic_scheduler&) (@tbb::internal::arena::process(tbb::internal::generic_scheduler&):66)
tbb::internal::market::process(rml::job&) (@tbb::internal::market::process(rml::job&):13)
tbb::internal::rml::private_worker::run() (@tbb::internal::rml::private_worker::run():75)
tbb::internal::rml::private_worker::thread_routine(void*) (@tbb::internal::rml::private_worker::thread_routine(void*):6)
_pthread_start (@_pthread_start:40)

I can confirm a crash in `ccl::MetalDevice::tex_alloc`: ``` __pthread_kill (@__pthread_kill:5) pthread_kill (@pthread_kill:75) abort (@abort:44) __assert_rtn (@__assert_rtn:56) MTLReportFailure.cold.1 (@bool MTLGetEnvCase<MTLErrorModeType>(char const*, MTLErrorModeType&, std::__1::vector<std::__1::pair<char const*, MTLErrorModeType>, std::__1::allocator<std::__1::pair<char const*, MTLErrorModeType> > > const&) (.cold.1):3) MTLReportFailure (@MTLReportFailure:122) _MTLMessageContextEnd (@_MTLMessageContextEnd:270) -[MTLTextureDescriptorInternal validateWithDevice:] (@-[MTLTextureDescriptorInternal validateWithDevice:]:646) ___lldb_unnamed_symbol5130 (@___lldb_unnamed_symbol5130:20) ccl::MetalDevice::tex_alloc(ccl::device_texture&) (/Users/germanocavalcante/blender-git/blender/intern/cycles/device/metal/device_impl.mm:890) ccl::MultiDevice::mem_copy_to(ccl::device_memory&) (/Users/germanocavalcante/blender-git/blender/intern/cycles/device/multi/device.cpp:307) ccl::ImageManager::device_load_image(ccl::Device*, ccl::Scene*, int, ccl::Progress*) (/Users/germanocavalcante/blender-git/blender/intern/cycles/scene/image.cpp:777) std::__1::__function::__value_func<void ()>::operator()() const (/Users/germanocavalcante/Applications/Xcode-beta.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.0.sdk/usr/include/c++/v1/functional:1885) std::__1::function<void ()>::operator()() const (/Users/germanocavalcante/Applications/Xcode-beta.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.0.sdk/usr/include/c++/v1/functional:2560) tbb::internal::function_task<std::__1::function<void ()> >::execute() (/Users/germanocavalcante/blender-git/lib/darwin_arm64/tbb/include/tbb/task.h:1059) tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::process_bypass_loop(tbb::internal::context_guard_helper<false>&, tbb::task*, long) (@tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::process_bypass_loop(tbb::internal::context_guard_helper<false>&, tbb::task*, long):113) tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::local_wait_for_all(tbb::task&, tbb::task*) (@tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::local_wait_for_all(tbb::task&, tbb::task*):49) tbb::internal::arena::process(tbb::internal::generic_scheduler&) (@tbb::internal::arena::process(tbb::internal::generic_scheduler&):66) tbb::internal::market::process(rml::job&) (@tbb::internal::market::process(rml::job&):13) tbb::internal::rml::private_worker::run() (@tbb::internal::rml::private_worker::run():75) tbb::internal::rml::private_worker::thread_routine(void*) (@tbb::internal::rml::private_worker::thread_routine(void*):6) _pthread_start (@_pthread_start:40) ```

Michael Jones (Apple) commented

2022-02-04 18:06:43 +01:00

The allocation is failing due to an individual texture exceeding MTLTexture dimension limits:

MTLTextureDescriptor has width (21600) greater than the maximum allowed size of 16384.

The allocation is failing due to an individual texture exceeding `MTLTexture` dimension limits: `MTLTextureDescriptor has width (21600) greater than the maximum allowed size of 16384.`

Brecht Van Lommel commented

2022-02-04 18:24:34 +01:00

Added subscriber: @brecht

Brecht Van Lommel commented

2022-02-04 18:24:34 +01:00

Maybe in the future we can solve that with a tiled texture cache system, but for now it would be good to show an error to the user instead of crashing.

cotooleimages@gmail.com commented

2022-02-04 18:40:28 +01:00

In #95477#1301802, @Michael-Jones wrote:
The allocation is failing due to an individual texture exceeding MTLTexture dimension limits:

MTLTextureDescriptor has width (21600) greater than the maximum allowed size of 16384.

Is there a fix for this, or is it just an area where I'll need to cope with rendering with the CPU for now?

> In #95477#1301802, @Michael-Jones wrote: > The allocation is failing due to an individual texture exceeding `MTLTexture` dimension limits: > > `MTLTextureDescriptor has width (21600) greater than the maximum allowed size of 16384.` Is there a fix for this, or is it just an area where I'll need to cope with rendering with the CPU for now?

Michael Jones (Apple) commented

2022-02-07 11:40:22 +01:00

In #95477#1301828, @brecht wrote:
Maybe in the future we can solve that with a tiled texture cache system, but for now it would be good to show an error to the user instead of crashing.

Is Device::set_error(...) the preferred method of error reporting for cases like this?

> In #95477#1301828, @brecht wrote: > Maybe in the future we can solve that with a tiled texture cache system, but for now it would be good to show an error to the user instead of crashing. Is `Device::set_error(...)` the preferred method of error reporting for cases like this?

Brecht Van Lommel commented

2022-02-07 15:15:53 +01:00

Yes.