kvark
Crash at start on Snapdragon X elite
Vulkan sdk is installed, vkcube is working fine.
C:\Development\blade>cargo run --example bunnymark
warning: `C:\Development\blade\.cargo\config` is deprecated in favor of `config.toml`
note: if you need to support cargo 1.38 or earlier, you can symlink `config` to `config.toml`
warning: use of deprecated method `winit::event_loop::EventLoop::<T>::create_window`: use `ActiveEventLoop::create_window` instead
--> examples\bunnymark\main.rs:371:29
|
371 | let window = event_loop.create_window(window_attributes).unwrap();
| ^^^^^^^^^^^^^
|
= note: `#[warn(deprecated)]` on by default
warning: use of deprecated method `winit::event_loop::EventLoop::<T>::run`: use `EventLoop::run_app` instead
--> examples\bunnymark\main.rs:400:10
|
400 | .run(|event, target| {
| ^^^
warning: `blade` (example "bunnymark") generated 2 warnings
Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.24s
Running `target\debug\examples\bunnymark.exe`
DeviceInformation { is_software_emulated: false, device_name: "Qualcomm(R) Adreno(TM) X1-85 GPU", driver_name: "Qualcomm Technologies Inc. Adreno Vulkan Driver", driver_info: "Driver Build: 36a15f63c3, , 1727332442\nDate: 09/26/2024\nCompiler Version: E031.47.15.00\nDriver Branch: \n" }
thread 'main' panicked at blade-graphics\src\vulkan\pipeline.rs:601:18:
called `Result::unwrap()` on an `Err` value: ([0x0], ERROR_UNKNOWN)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
error: process didn't exit successfully: `target\debug\examples\bunnymark.exe` (exit code: 101)
Interesting, thank you for the report! Could you run with validation enabled? I.e. have the validation layers installed and do "validation: true" in the context desc.
I changed the validation to true then false in bunnymark/main.rs:82
I still get the same output. Note, I'm not a rust developer
It looks like it fails at vkCreateGraphicsPipelines(). Most examples (except init, mini) fails at this specific call.
Thread 0, Frame 0:
vkCreateGraphicsPipelines(device, pipelineCache, createInfoCount, pCreateInfos, pAllocator, pPipelines) returns VkResult VK_ERROR_UNKNOWN (-13):
device: VkDevice = 000001EFE0839220
pipelineCache: VkPipelineCache = 0000000000000000
createInfoCount: uint32_t = 1
pCreateInfos: const VkGraphicsPipelineCreateInfo* = 0000009416741790
pCreateInfos[0]: const VkGraphicsPipelineCreateInfo = 0000009416741790:
sType: VkStructureType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO (28)
pNext: const void* = VkPipelineRenderingCreateInfo
flags: VkPipelineCreateFlags = 0
stageCount: uint32_t = 2
pStages: const VkPipelineShaderStageCreateInfo* = 00000094167401E0
pStages[0]: const VkPipelineShaderStageCreateInfo = 00000094167401E0:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO (18)
pNext: const void* = NULL
flags: VkPipelineShaderStageCreateFlags = 0
stage: VkShaderStageFlagBits = 1 (VK_SHADER_STAGE_VERTEX_BIT)
module: VkShaderModule = E7E6D0000000000F
pName: const char* = "vs_main"
pSpecializationInfo: const VkSpecializationInfo* = NULL
pStages[1]: const VkPipelineShaderStageCreateInfo = 0000009416740210:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO (18)
pNext: const void* = NULL
flags: VkPipelineShaderStageCreateFlags = 0
stage: VkShaderStageFlagBits = 16 (VK_SHADER_STAGE_FRAGMENT_BIT)
module: VkShaderModule = D5B26F0000000010
pName: const char* = "fs_main"
pSpecializationInfo: const VkSpecializationInfo* = NULL
pVertexInputState: const VkPipelineVertexInputStateCreateInfo* = 00000094167403C0:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO (19)
pNext: const void* = NULL
flags: VkPipelineVertexInputStateCreateFlags = 0
vertexBindingDescriptionCount: uint32_t = 0
pVertexBindingDescriptions: const VkVertexInputBindingDescription* = 0000000000000004
vertexAttributeDescriptionCount: uint32_t = 0
pVertexAttributeDescriptions: const VkVertexInputAttributeDescription* = 0000000000000004
pInputAssemblyState: const VkPipelineInputAssemblyStateCreateInfo* = 0000009416740450:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO (20)
pNext: const void* = NULL
flags: VkPipelineInputAssemblyStateCreateFlags = 0
topology: VkPrimitiveTopology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST (3)
primitiveRestartEnable: VkBool32 = 0
pTessellationState: const VkPipelineTessellationStateCreateInfo* = NULL
pViewportState: const VkPipelineViewportStateCreateInfo* = 00000094167406C0:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO (22)
pNext: const void* = NULL
flags: VkPipelineViewportStateCreateFlags = 0
viewportCount: uint32_t = 1
pViewports: const VkViewport* = UNUSED
scissorCount: uint32_t = 1
pScissors: const VkRect2D* = UNUSED
pRasterizationState: const VkPipelineRasterizationStateCreateInfo* = 00000094167404B0:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO (23)
pNext: const void* = NULL
flags: VkPipelineRasterizationStateCreateFlags = 0
depthClampEnable: VkBool32 = 0
rasterizerDiscardEnable: VkBool32 = 0
polygonMode: VkPolygonMode = VK_POLYGON_MODE_FILL (0)
cullMode: VkCullModeFlags = 0 (VK_CULL_MODE_NONE)
frontFace: VkFrontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE (0)
depthBiasEnable: VkBool32 = 0
depthBiasConstantFactor: float = 0
depthBiasClamp: float = 0
depthBiasSlopeFactor: float = 0
lineWidth: float = 1
pMultisampleState: const VkPipelineMultisampleStateCreateInfo* = 0000009416740788:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO (24)
pNext: const void* = NULL
flags: VkPipelineMultisampleStateCreateFlags = 0
rasterizationSamples: VkSampleCountFlagBits = 1 (VK_SAMPLE_COUNT_1_BIT)
sampleShadingEnable: VkBool32 = 0
minSampleShading: float = 0
pSampleMask: const VkSampleMask* = 0000009416740780
pSampleMask[0]: const VkSampleMask = 4294967295
alphaToCoverageEnable: VkBool32 = 0
alphaToOneEnable: VkBool32 = 0
pDepthStencilState: const VkPipelineDepthStencilStateCreateInfo* = 0000009416740850:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO (25)
pNext: const void* = NULL
flags: VkPipelineDepthStencilStateCreateFlags = 0
depthTestEnable: VkBool32 = 0
depthWriteEnable: VkBool32 = 0
depthCompareOp: VkCompareOp = VK_COMPARE_OP_NEVER (0)
depthBoundsTestEnable: VkBool32 = 0
stencilTestEnable: VkBool32 = 0
front: VkStencilOpState = 0000009416740878:
failOp: VkStencilOp = VK_STENCIL_OP_KEEP (0)
passOp: VkStencilOp = VK_STENCIL_OP_KEEP (0)
depthFailOp: VkStencilOp = VK_STENCIL_OP_KEEP (0)
compareOp: VkCompareOp = VK_COMPARE_OP_NEVER (0)
compareMask: uint32_t = 0
writeMask: uint32_t = 0
reference: uint32_t = 0
back: VkStencilOpState = 0000009416740894:
failOp: VkStencilOp = VK_STENCIL_OP_KEEP (0)
passOp: VkStencilOp = VK_STENCIL_OP_KEEP (0)
depthFailOp: VkStencilOp = VK_STENCIL_OP_KEEP (0)
compareOp: VkCompareOp = VK_COMPARE_OP_NEVER (0)
compareMask: uint32_t = 0
writeMask: uint32_t = 0
reference: uint32_t = 0
minDepthBounds: float = 0
maxDepthBounds: float = 0
pColorBlendState: const VkPipelineColorBlendStateCreateInfo* = 0000009416740F80:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO (26)
pNext: const void* = NULL
flags: VkPipelineColorBlendStateCreateFlags = 0
logicOpEnable: VkBool32 = 0
logicOp: VkLogicOp = VK_LOGIC_OP_CLEAR (0)
attachmentCount: uint32_t = 1
pAttachments: const VkPipelineColorBlendAttachmentState* = 000001EFE0E89EE0
pAttachments[0]: const VkPipelineColorBlendAttachmentState = 000001EFE0E89EE0:
blendEnable: VkBool32 = 1
srcColorBlendFactor: VkBlendFactor = VK_BLEND_FACTOR_ONE (1)
dstColorBlendFactor: VkBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA (7)
colorBlendOp: VkBlendOp = VK_BLEND_OP_ADD (0)
srcAlphaBlendFactor: VkBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_DST_ALPHA (9)
dstAlphaBlendFactor: VkBlendFactor = VK_BLEND_FACTOR_ONE (1)
alphaBlendOp: VkBlendOp = VK_BLEND_OP_ADD (0)
colorWriteMask: VkColorComponentFlags = 15 (VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT)
blendConstants: float[4] = 0000009416740FA8
blendConstants[0]: float = 0
blendConstants[1]: float = 0
blendConstants[2]: float = 0
blendConstants[3]: float = 0
pDynamicState: const VkPipelineDynamicStateCreateInfo* = 0000009416740680:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO (27)
pNext: const void* = NULL
flags: VkPipelineDynamicStateCreateFlags = 0
dynamicStateCount: uint32_t = 4
pDynamicStates: const VkDynamicState* = 0000009416740670
pDynamicStates[0]: const VkDynamicState = VK_DYNAMIC_STATE_VIEWPORT (0)
pDynamicStates[1]: const VkDynamicState = VK_DYNAMIC_STATE_SCISSOR (1)
pDynamicStates[2]: const VkDynamicState = VK_DYNAMIC_STATE_BLEND_CONSTANTS (4)
pDynamicStates[3]: const VkDynamicState = VK_DYNAMIC_STATE_STENCIL_REFERENCE (8)
layout: VkPipelineLayout = DD3A8A0000000015
renderPass: VkRenderPass = 0000000000000000
subpass: uint32_t = 0
basePipelineHandle: VkPipeline = 0000000000000000
basePipelineIndex: int32_t = 0
pNext: VkPipelineRenderingCreateInfo = 0000009416740FF0:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO (1000044002)
pNext: const void* = NULL
viewMask: uint32_t = 0
colorAttachmentCount: uint32_t = 1
pColorAttachmentFormats: const VkFormat* = 000001EFE01F3AB0
pColorAttachmentFormats[0]: const VkFormat = VK_FORMAT_B8G8R8A8_SRGB (50)
depthAttachmentFormat: VkFormat = VK_FORMAT_D32_SFLOAT_S8_UINT (130)
stencilAttachmentFormat: VkFormat = VK_FORMAT_D32_SFLOAT_S8_UINT (130)
pAllocator: const VkAllocationCallbacks* = NULL
pPipelines: VkPipeline* = 000001EFE01F3B30
pPipelines[0]: VkPipeline = 0000000000000000
There's no any errors spotted by the validation layer. I guess it's something with quirks in Qualcomm's Vulkan driver on Windows. Didn't have this problem on Linux (mesa/turnip).
It's really hard to say what part of the shader the Qualcomm's Vulkan driver doesn't like, but it looks like a driver bug to me.
It's really hard to say what part of the shader the Qualcomm's Vulkan driver doesn't like, but it looks like a driver bug to me.
I'm an experienced Rust dev w/ an X1E80100 and willing to help debug this issue further if you're interested, although I'll need guidance as I've done very little graphics development.
I'd first change the shader into something trivia, i.e. vertex doing nothing and fragment outputting a constant color. I assume that pipeline would be created successfully. And then I'd manually bisect the shader code (between the failing point and the successful point) to see which exact construct is making the driver to freak out.
Once we find the construct, we can work with current Naga developers to put a workaround in place in the shading translation.
After I changed the fragment part in bunnymark, it doesn't crash, and it renders a white box.
Its seems like the issue is textureSample function call
@fragment fn fs_main(vertex: VertexOutput) -> @location(0) vec4<f32> { return vec4<f32>(1.0, 1.0, 1.0, 1.0); }
After a bit of testing it seems like the crash occurs when you try to access vertex properties in fs_main
I can repro @HgGamer's test:
Acessing vertex.position in the shader works, but vertex.color or vertex.tex_coords causes the application to crash.
DeviceInformation {
is_software_emulated: false,
device_name: "Qualcomm(R) Adreno(TM) X1-85 GPU"
driver_name: "Qualcomm Technologies Inc. Adreno Vulkan Driver",
driver_info: "Driver Build: b1f52b5da7, Ibad1ce48fe, 1723618625\nDate: 08/14/2024\nCompiler Version: E031.47.09.00\nDriver Branch: \n"
}
Also, I tried multiple texture formats because I read somewhere that the GPU doesn't support every format.
There's a workaround implemented in Blender a couple months ago, which could be related and points to a driver issue with optimized SPIR-V. The workaround for Blender was to disable some optimizations (this is beyond my knowledge, so just a guess).
Thank you both for the investigation! The Blender thread appears to be very relevant but unfortunately lacking details. It would be useful to read the Qualcomm driver devs take on this. Shaderc optimizations is basically spirv-opt from SPIR-V tools. Blade doesn't use it - we have a pure-Rust shader pipeline, and we aren't running the shader through spirv-opt since we assume the driver already doing this anyway at pipeline creation.
There appear to be multiple other threads about Adreno GPU issues with SPIR-V shaders:
- https://github.com/devmiyax/yabause/issues/998
- https://github.com/o3de/o3de-extras/issues/625
- https://community.khronos.org/t/adreno-506-vkcreatecomputepipelines-has-returned-vk-incomplete/104071/13
On snapdragon devices there are two vulkan drivers, one is native, and the other is emulated with DX12 (Dozen). Its maybe possible to test it with each drivers.
Originally posted by @evgri243 in #19739 I assume it is more valuable here.
Just as an intro. I tried all possible combinations yesterday.
Build for: "x86_64-pc-windows-msvc" and "aarch64-pc-windows-msvc", Executed it with "Force Physical Device Only Preset" in Portability Loader Configuration of kvconfig:
- Qualcomm(R) Adreno (TM) X1-85 GPU with driver: "C:\WINDOWS\System32\DriverStore\FileRepository\qcdx8380.inf_arm64_1b5d57031d8d3198.\qcvkarm64xum.dll": fails with the error above for both build architectures:
Result::unwrap() on an Err value: ([0x0], ERROR_UNKNOWN)inblade-graphics\\src\\vulkan\\pipeline.rs:614.- Microsoft Direct3D12 (Qualcomm(R) Adreno(TM) X1-85 GPU) with Vulkan 1.2.295 driver: fails with for both architectures with
Unable to init GPU context: NoSupportedDeviceFoundincrates\gpui\src\platform\windows\platform.rs:111.- Microsoft Direct3D12 (Qualcomm(R) Adreno(TM) X1-85 GPU) with Vulkan 1.2.295 driver: fails with for both architectures with
Unable to init GPU context: NoSupportedDeviceFoundincrates\gpui\src\platform\windows\platform.rs:111.So far, no working configuration.
Direct3D version: 1.611.1-81528511 DXCore version: 10.0.26100.1-240331-1435.ge-release Windows version: 10.0.26120.3576 Graphics driver: the latest v31.0.96.0 beta OpenCL™, OpenGL®, and Vulkan® Compatibility Pack is installed, hence the Microsoft 3D12 devices. Tested with the latest master by the date Blade with minimal changes for obj2 library compatibility.
vkcubeworks with every forced device.
I can reproduce this with building it myself with MSVC 2022 on Windows on ARM and also with a CLANGARM64 build in MSYS2. Both fail in the exact same place:
[2025-04-03T20:50:58-06:00 ERROR client] Element not found. (0x80070490)
[2025-04-03T20:50:58-06:00 ERROR blade_graphics::hal::surface] No composite alpha flag for transparency: OPAQUE
[2025-04-03T20:50:59-06:00 ERROR zed::reliability] {
"thread": "main",
"payload": "called `Result::unwrap()` on an `Err` value: ([0x0], ERROR_UNKNOWN)",
"location_data": {
"file": "C:\\Users\\Hernan\\.cargo\\git\\checkouts\\blade-26bd638d88226de3\\b16f5c7\\blade-graphics\\src\\vulkan\\pipeline.rs",
"line": 614
},
"backtrace": [
"backtrace::backtrace::dbghelp64::trace+191492748",
"backtrace::backtrace::trace_unsynchronized<backtrace::capture::impl$4::create::closure_env$0>+191492748",
"backtrace::backtrace::trace<backtrace::capture::impl$4::create::closure_env$0>+191492484",
"backtrace::capture::Backtrace::create+191468952",
"backtrace::capture::Backtrace::new+191468836",
"zed::reliability::init_panic_hook::closure$0+1470324",
"alloc::boxed::impl$30::call+194289040",
"std::panicking::rust_panic_with_hook+194289040",
"std::panicking::begin_panic_handler::closure$0+194288284",
"std::sys::backtrace::__rust_end_short_backtrace<std::panicking::begin_panic_handler::closure_env$0,never$>+194280700"
],
"app_version": "0.180.2",
"app_commit_sha": "3775496b84cfc9ebf27376119742f3484d4997f2",
"release_channel": "stable",
"target": "aarch64-pc-windows-msvc",
"os_name": "Windows",
"os_version": "10.0.26100",
"architecture": "aarch64",
"panicked_on": 1743735059182,
"system_id": "f70ef1ce-d154-4628-bc79-66887ee17028",
"installation_id": "ed8b4376-2159-4e7d-bfd1-300540a4f679",
"session_id": "c901de50-bbf8-4e47-9abc-081f57784185"
}
I'm trying to get zed running on Windows with lavapipe, at least for the time being while the Snapdragon X Vulkan driver is in this broken state, but.
I can't even get the blade-graphics vehicle example to work with lavapipe, but the list of required Vulkan extensions in https://crates.io/crates/blade-graphics does not seem to be comprehensive (see https://github.com/kvark/blade/issues/255), and when I try to run the example I get a KHR_external_memory_win32 not implemented error, so I added this request: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13017 .
Lavapipe already supports ``KHR_external_memory_fd` so blade-graphics seem to work on Linux with lavapipe.
Update: lavapipe now works, see #255 As for the actual issue with Snapdragon, we need to find a way to get more feedback from the driver, if possible. Generally for this configuration, I'd expect it to be an internal failure of the driver not behaving up to specification.
@kvark Do you mean zed with #255 now works? I still have to give it a try.