Crash when enabling Open Shading Language with Cycles on macOS and Apple Silicon M1 #86530

New Issue

Phil Hargett · 2021-03-12T16:30:09+01:00

Phil Hargett commented

2021-03-12 16:30:09 +01:00

System Information
Operating system: macOS Big Sur 11.2.3
CPU: Apple M1
Graphics card: Apple M1

Blender Version
Broken: 2.93 alpha on master, built locally, currently at commit b1ef55abdb (although older commits also have this issue)
Worked: 2.92 Blender Intel version running on Apple M1 via Rosetta 2 does not crash

Short description of error

Enabling Open Shading Language on the Render Properties panel with Render Engine set to Cycles causes an immediate crash.

First noticed this on my own scenes, but can reproduce by just opening blender and using the default scene.

Exact steps for others to reproduce the error

Open Blender, just use default scene
Set Viewport shading to Rendered Preview (so that Cycles results can be previewed in real-time)
On Render Properties panel, set Render Engine to Cycles (leave CPU selected)
On Render Properties panel, check the "Open Shading Language" checkbox
Crash

Details gathered so far

I've been building and using Blender successfully after progress was made on #78710 and the Building Blender instructions for Mac were known to work again. My current projects don't need OSL, but was just experimenting when I discovered the crash.

After building an Xcode project and attaching Xcode to my locally built Blender app (again, this is all running on the new Apple M1/ARM64 chip, not Intel), I did find that the crashing thread seems to be in OSL library code:

- 0	0x0000000108b53f00 in llvm::CallInst::Create(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::ArrayRef<llvm::OperandBundleDefT<llvm::Value*> >, llvm::Twine const&, llvm::Instruction*) ()
- 1	0x0000000108b495d8 in llvm::IRBuilderBase::CreateCall(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::Twine const&, llvm::MDNode*) ()
- 2	0x0000000104a2cee0 in OSL_v1_11::pvt::LLVM_Util::call_function(llvm::Value*, OpenImageIO_v2_1::span<llvm::Value* const, -1l>) ()
- 3	0x00000001049ff1bc in OSL_v1_11::pvt::llvm_gen_closure(OSL_v1_11::pvt::BackendLLVM&, int) ()
- 4	0x0000000104a084c4 in OSL_v1_11::pvt::BackendLLVM::build_llvm_code(int, int, llvm::BasicBlock*) ()
- 5	0x0000000104a0a860 in OSL_v1_11::pvt::BackendLLVM::build_llvm_instance(bool) ()
- 6	0x0000000104a20cd4 in OSL_v1_11::pvt::BackendLLVM::run() ()
- 7	0x00000001048cacec in OSL_v1_11::pvt::ShadingSystemImpl::optimize_group(OSL_v1_11::ShaderGroup&, OSL_v1_11::ShadingContext*, bool) ()
- 8	0x00000001048c943c in OSL_v1_11::pvt::ShadingSystemImpl::optimize_all_groups(int, int, int, bool) ()
- 9	0x00000001048dfbac in void* std::__1::__thread_proxy<std::__1::tuple<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct> >, void (*)(OSL_v1_11::pvt::ShadingSystemImpl*, int, int, bool), OSL_v1_11::pvt::ShadingSystemImpl*, int, int, bool> >(void*) ()
#10	0x0000000181f2e06c in _pthread_start ()

After reviewing other threads to see if they were doing anything interesting (as opposed to just looping on semaphores or event loops), there is this thread:

- 0	0x0000000181ef6acc in __ulock_wait ()
- 1	0x0000000181f2fc14 in _pthread_join ()
- 2	0x0000000181ec644c in std::__1::thread::join() ()
- 3	0x00000001048c92f0 in OSL_v1_11::pvt::ShadingSystemImpl::optimize_all_groups(int, int, int, bool) ()
- 4	0x0000000105f4ae8c in ccl::OSLShaderManager::device_update(ccl::Device*, ccl::DeviceScene*, ccl::Scene*, ccl::Progress&) ()
- 5	0x0000000105f54718 in ccl::Scene::device_update(ccl::Device*, ccl::Progress&) ()
- 6	0x0000000105f55e1c in ccl::Scene::update(ccl::Progress&, bool&) ()
- 7	0x0000000105f5bfb0 in ccl::Session::update_scene() ()
- 8	0x0000000105f5d904 in ccl::Session::run_cpu() ()
- 9	0x0000000105f5afe4 in ccl::Session::run() ()
- 10	0x0000000108df87c0 in ccl::thread::run(void*) ()
- 11	0x0000000181f2e06c in _pthread_start ()

This is pretty much the last line of method ccl::OSLShaderManager::device_update.

**System Information** Operating system: macOS Big Sur 11.2.3 CPU: Apple M1 Graphics card: Apple M1 **Blender Version** Broken: 2.93 alpha on master, built locally, currently at commit b1ef55abdbb (although older commits also have this issue) Worked: 2.92 Blender Intel version running on Apple M1 via Rosetta 2 does not crash **Short description of error** Enabling Open Shading Language on the Render Properties panel with Render Engine set to Cycles causes an immediate crash. First noticed this on my own scenes, but can reproduce by just opening blender and using the default scene. **Exact steps for others to reproduce the error** 1. Open Blender, just use default scene 2. Set Viewport shading to Rendered Preview (so that Cycles results can be previewed in real-time) 3. On Render Properties panel, set Render Engine to Cycles (leave CPU selected) 4. On Render Properties panel, check the "Open Shading Language" checkbox 5. Crash **Details gathered so far** I've been building and using Blender successfully after progress was made on #78710 and the Building Blender instructions for Mac were known to work again. My current projects don't need OSL, but was just experimenting when I discovered the crash. After building an Xcode project and attaching Xcode to my locally built Blender app (again, this is all running on the new Apple M1/ARM64 chip, not Intel), I did find that the crashing thread seems to be in OSL library code: ``` - 0 0x0000000108b53f00 in llvm::CallInst::Create(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::ArrayRef<llvm::OperandBundleDefT<llvm::Value*> >, llvm::Twine const&, llvm::Instruction*) () - 1 0x0000000108b495d8 in llvm::IRBuilderBase::CreateCall(llvm::FunctionType*, llvm::Value*, llvm::ArrayRef<llvm::Value*>, llvm::Twine const&, llvm::MDNode*) () - 2 0x0000000104a2cee0 in OSL_v1_11::pvt::LLVM_Util::call_function(llvm::Value*, OpenImageIO_v2_1::span<llvm::Value* const, -1l>) () - 3 0x00000001049ff1bc in OSL_v1_11::pvt::llvm_gen_closure(OSL_v1_11::pvt::BackendLLVM&, int) () - 4 0x0000000104a084c4 in OSL_v1_11::pvt::BackendLLVM::build_llvm_code(int, int, llvm::BasicBlock*) () - 5 0x0000000104a0a860 in OSL_v1_11::pvt::BackendLLVM::build_llvm_instance(bool) () - 6 0x0000000104a20cd4 in OSL_v1_11::pvt::BackendLLVM::run() () - 7 0x00000001048cacec in OSL_v1_11::pvt::ShadingSystemImpl::optimize_group(OSL_v1_11::ShaderGroup&, OSL_v1_11::ShadingContext*, bool) () - 8 0x00000001048c943c in OSL_v1_11::pvt::ShadingSystemImpl::optimize_all_groups(int, int, int, bool) () - 9 0x00000001048dfbac in void* std::__1::__thread_proxy<std::__1::tuple<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct> >, void (*)(OSL_v1_11::pvt::ShadingSystemImpl*, int, int, bool), OSL_v1_11::pvt::ShadingSystemImpl*, int, int, bool> >(void*) () #10 0x0000000181f2e06c in _pthread_start () ``` After reviewing other threads to see if they were doing anything interesting (as opposed to just looping on semaphores or event loops), there is this thread: ``` - 0 0x0000000181ef6acc in __ulock_wait () - 1 0x0000000181f2fc14 in _pthread_join () - 2 0x0000000181ec644c in std::__1::thread::join() () - 3 0x00000001048c92f0 in OSL_v1_11::pvt::ShadingSystemImpl::optimize_all_groups(int, int, int, bool) () - 4 0x0000000105f4ae8c in ccl::OSLShaderManager::device_update(ccl::Device*, ccl::DeviceScene*, ccl::Scene*, ccl::Progress&) () - 5 0x0000000105f54718 in ccl::Scene::device_update(ccl::Device*, ccl::Progress&) () - 6 0x0000000105f55e1c in ccl::Scene::update(ccl::Progress&, bool&) () - 7 0x0000000105f5bfb0 in ccl::Session::update_scene() () - 8 0x0000000105f5d904 in ccl::Session::run_cpu() () - 9 0x0000000105f5afe4 in ccl::Session::run() () - 10 0x0000000108df87c0 in ccl::thread::run(void*) () - 11 0x0000000181f2e06c in _pthread_start () ``` This is pretty much the last line of method `ccl::OSLShaderManager::device_update`.

Phil Hargett commented

2021-03-12 16:30:09 +01:00

Added subscriber: @hargettp

Sebastián Barschkis commented

2021-03-12 16:38:06 +01:00

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

hujin commented

2021-03-31 18:57:37 +02:00

Added subscriber: @2046411367

Phil Hargett commented

2021-04-08 23:44:56 +02:00

So, for what it's worth, I changed this line: https://git.blender.org/gitweb/gitweb.cgi/blender.git/blob/HEAD:/intern/cycles/render/osl.cpp#l171

from this call:

ss->optimize_all_groups();

to this call:

ss->optimize_all_groups(1, true);

This basically makes the optimize_all_groups synchronous on the calling thread, rather than spread out over multiple threads. Such a change should eliminate thread safety in OSL as a cause.

Repeating the test still caused a crash in OSL. This suggests the bug might be in OSL on M1--although I can't rule out we didn't pass a bad data structure or pointer to OSL before this.

So, for what it's worth, I changed this line: https://git.blender.org/gitweb/gitweb.cgi/blender.git/blob/HEAD:/intern/cycles/render/osl.cpp#l171 from this call: `ss->optimize_all_groups();` to this call: `ss->optimize_all_groups(1, true);` This basically makes the `optimize_all_groups` synchronous on the calling thread, rather than spread out over multiple threads. Such a change should eliminate thread safety in OSL as a cause. Repeating the test still caused a crash in OSL. This suggests the bug might be in OSL on M1--although I can't rule out we didn't pass a bad data structure or pointer to OSL before this.

Timothy Cook commented

2021-04-14 16:10:35 +02:00

Added subscriber: @CookItOff

Timothy Cook commented

2021-04-14 16:10:35 +02:00

I can confirm the crash on M1 with the Arm 64 official build.

Phil Hargett commented

2021-04-15 02:37:14 +02:00

FYI, filed this ticket for OSL: Crash in optimize_all_groups running in pre-release Blender on Apple Silicon . Not 100% sure its OSL, but hoping for guidance.

FYI, filed this ticket for OSL: [Crash in optimize_all_groups running in pre-release Blender on Apple Silicon ](https://github.com/AcademySoftwareFoundation/OpenShadingLanguage/issues/1359). Not 100% sure its OSL, but hoping for guidance.

blender-admin commented

2021-04-19 14:41:14 +02:00

This issue was referenced by 33f218fa3f

This issue was referenced by 33f218fa3f93a3d574d81d65628188921d6f100c

Brecht Van Lommel commented

2021-04-19 15:05:21 +02:00

Changed status from 'Confirmed' to: 'Resolved'

Brecht Van Lommel closed this issue

2021-04-19 15:05:21 +02:00

Brecht Van Lommel self-assigned this 2021-04-19 15:05:21 +02:00

blender-admin commented

2021-04-19 15:06:10 +02:00