Vulkano: Unexplainable seg-fault on compute pipeline creation
Hi,
I am currently trying to write a real time ray traced shadows (the hype is real) via vulkano.
So far everything works out great, but currently I am experiencing a strange bug.
When does it happen?
When Vulkano tries to create compute pipeline a seg-fault happens. RUST_BACKTRACE=1 returns nothing on that matter.
#
Since no Spirv compilation error happens I tried to find the "wrong" code by removing everything in the shader and then inserting lines till it crashes.
My compute pipeline creation code looks like this:
let dispatch_dims = [
(image_dims[0] / SHADOWTHREADSIZE[0]) + 1,
(image_dims[1] / SHADOWTHREADSIZE[1]) + 1,
1
];
//Now start the point and spot light compute shader
let point_desc = self.descriptor_pool_else.next()
.add_sampled_image(
position_img, self.gbuffer_sampler.clone()
).expect("failed to add position image to shadow pass")
.add_sampled_image(
normal_img, self.gbuffer_sampler.clone()
).expect("failed to add normal image to shadow pass")
.add_image(shadow_img)
.expect("failed to add target shadow image to shadow pass")
.add_buffer(bvh_buffer)
.expect("failed to add bvh buffer to compute pass")
.add_buffer(vertex_buffer)
.expect("failed to add vertex buffer to compute pass")
.add_image(debug_img)
.expect("failed to add debug image!")
.build().expect("failed to build bvh buffer");
let point_light_desc = light_sys.get_light_descriptorset(1, self.pipeline_else.clone());
//Finally dispatch the last shadow pass
let final_cb = command_buffer.dispatch(
dispatch_dims, self.pipeline_else.clone(),
(point_desc, point_light_desc), ()
).expect("failed to start point light shadows compute shader");
The whole module can be found here, WARNING spaghetti code :)
The whole shader can be found here: https://gitlab.com/Siebencorgie/jakar-engine/blob/master/data/shader/r_shadow_else.comp
However, its mostly a copy of the r_shadow_dir.comp shader, which works perfectly fine so far.
The only changes are that I ray trace not "AnyHit" but "ClosestHit" (since it is for point lights and later spot lights).
Anyways the seg-fault only happens when I write to the current_hit:
///Traces a ray through the scene, returns if there was any hit.
TriangleHit ClosestHit(Ray ray){
uint top_bvh_buffer_length = u_bvh.nodes.length();
TriangleHit current_hit = TriangleHit(vec2(0.0,0.0), infinity, 0);
//Start tracing at root
int idx = 0;
while(idx != -1){
if (idx == -1 || idx > (top_bvh_buffer_length-1)){
break;
}
int current_idx = idx;
idx++;
BvhNode node = u_bvh.nodes[current_idx];
//Check if we intersect
if (aabb_intersects(node.aabb_min.xyz, node.aabb_max.xyz, ray)){
if (node.isLeaf == 1){
TriangleHit leafhit = ray_intersect_mesh(ray, node.left, node.right);
if (leafhit.is_hit == 1 && leafhit.tvec > MINHITDISTANCE && leafhit.tvec < current_hit.tvec){
current_hit = leafhit; //<- If I remove this it works! --------------------------------------------------
}
}
}else{
//We can advance to the escape idx
idx = node.esc_idx;
}
}
return current_hit;
}
Question
My Question now is, has anyone an idea why this happens? Could this be some "over optimizing" by the glsl -> SPIRV compiler?
If you are reading this and you have a graphics card other than AMD vega 56 on Linux/Windows and some time on your hand.
Could you please:
- download the full source code from here: https://gitlab.com/Siebencorgie/jakar-engine/tree/master
- Download some simple gltf model (for instance this one: https://github.com/KhronosGroup/glTF-Sample-Models/tree/master/2.0/Lantern)
- put it into
examples/simple_scene/ - correct the path on line 86 in
examples/simple_scene/main.rs - run the example with
cargo run --example simple --release
If I can provide any more detail I'd be happy to assist. I tried everything I could think of, including changing every return of a struct to a inout value like here and deleting the whole function and rewriting it.
-Siebencorgie
SiebenCorgie
All 11 comments
I havent looked into your code at all, but maybe it's a bug in the shader compiler? Try using nlordell's branch switching to shaderc-rs https://github.com/vulkano-rs/vulkano/pull/947
rukai
on 15 Sep 2018
Hi,
I tried using his shaderc branch, but the segfault still happens.
I tried to get a stack trace via valgrid which gives me the following events:
==10646== Invalid read of size 4
==10646== at 0x8095DC3: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x8093D58: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x939461D: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x8079D57: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x8066493: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x806743E: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x805D4FC: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x80374E5: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x27BBBB: <vulkano::pipeline::compute_pipeline::ComputePipeline<()>>::new (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== by 0x22B233: jakar_engine::render::raytracing::shadow::ShadowRayTracing::new (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== by 0x361CE4: jakar_engine::render::raytracing::RayTracing::new (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== by 0x284CC0: <jakar_engine::render::render_builder::RenderBuilder as jakar_engine::render::renderer::BuildRender>::build (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== Address 0x20 is not stack'd, malloc'd or (recently) free'd
==10646==
==10646==
==10646== Process terminating with default action of signal 11 (SIGSEGV): dumping core
==10646== Access not within mapped region at address 0x20
==10646== at 0x8095DC3: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x8093D58: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x939461D: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x8079D57: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x8066493: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x806743E: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x805D4FC: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x80374E5: ??? (in /usr/lib/amdvlk64.so)
==10646== by 0x27BBBB: <vulkano::pipeline::compute_pipeline::ComputePipeline<()>>::new (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== by 0x22B233: jakar_engine::render::raytracing::shadow::ShadowRayTracing::new (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== by 0x361CE4: jakar_engine::render::raytracing::RayTracing::new (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== by 0x284CC0: <jakar_engine::render::render_builder::RenderBuilder as jakar_engine::render::renderer::BuildRender>::build (in /home/user/Code/Rust/engine/jakar-engine/simple)
==10646== If you believe this happened as a result of a stack
==10646== overflow in your program's main thread (unlikely but
==10646== possible), you can try to increase the size of the
==10646== main thread stack using the --main-stacksize= flag.
==10646== The main thread stack size used in this run was 8388608.
==10646==
==10646== HEAP SUMMARY:
==10646== in use at exit: 73,827,915 bytes in 393,837 blocks
==10646== total heap usage: 2,088,175 allocs, 1,694,349 frees, 648,476,882 bytes allocated
==10646==
==10646== 28 bytes in 1 blocks are possibly lost in loss record 577 of 6,498
==10646== at 0x483777F: malloc (vg_replace_malloc.c:299)
==10646== by 0x63C92B1: ??? (in /usr/lib/libxcb-dri3.so.0.0.0)
==10646== by 0x63C94E4: ??? (in /usr/lib/libxcb-dri3.so.0.0.0)
==10646== by 0x57E9C48: ???
==10646== by 0x4010549: call_init.part.0 (in /usr/lib/ld-2.28.so)
==10646== by 0x4010649: _dl_init (in /usr/lib/ld-2.28.so)
==10646== by 0x4014532: dl_open_worker (in /usr/lib/ld-2.28.so)
==10646== by 0x4A13E76: _dl_catch_exception (in /usr/lib/libc-2.28.so)
==10646== by 0x4013DFE: _dl_open (in /usr/lib/ld-2.28.so)
==10646== by 0x4893159: ??? (in /usr/lib/libdl-2.28.so)
==10646== by 0x4A13E76: _dl_catch_exception (in /usr/lib/libc-2.28.so)
==10646== by 0x4A13F12: _dl_catch_error (in /usr/lib/libc-2.28.so)
I also doubled the stack size but no changes.
I wont have access to my code for the next week, so It will be difficult to try new stuff, but after that I try to find a computer with a Nvidia card and test my code there. Maybe its a driver problem. But if so it would be strange because I already changed from RADV to AMDVLK.
SiebenCorgie
on 15 Sep 2018
Have you tried enabling validation layers?
nicokoch
on 15 Sep 2018
👍2
Oh, no I totally forgot about those. I'll do as soon as I come home!
SiebenCorgie
on 16 Sep 2018
Okay so I activated the lunarG layers and this is the output:
Thread 0, Frame 0:
vkCreateDescriptorSetLayout(device, pCreateInfo, pAllocator, pSetLayout) returns VkResult VK_SUCCESS (0):
device: VkDevice = 0x7f811c6cfb40
pCreateInfo: const VkDescriptorSetLayoutCreateInfo* = 0x7ffe6343a6b0:
sType: VkStructureType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO (32)
pNext: const void* = NULL
flags: VkDescriptorSetLayoutCreateFlags = 0
bindingCount: uint32_t = 6
pBindings: const VkDescriptorSetLayoutBinding* = 0x7ffe6343a728
pBindings[0]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a728:
binding: uint32_t = 0
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER (1)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = NULL
pBindings[1]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a740:
binding: uint32_t = 1
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER (1)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = NULL
pBindings[2]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a758:
binding: uint32_t = 2
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE (3)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[3]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a770:
binding: uint32_t = 3
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[4]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a788:
binding: uint32_t = 4
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[5]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a7a0:
binding: uint32_t = 5
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE (3)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pAllocator: const VkAllocationCallbacks* = NULL
pSetLayout: VkDescriptorSetLayout* = 0xcd
Thread 0, Frame 0:
vkCreateDescriptorSetLayout(device, pCreateInfo, pAllocator, pSetLayout) returns VkResult VK_SUCCESS (0):
device: VkDevice = 0x7f811c6cfb40
pCreateInfo: const VkDescriptorSetLayoutCreateInfo* = 0x7ffe6343a6b0:
sType: VkStructureType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO (32)
pNext: const void* = NULL
flags: VkDescriptorSetLayoutCreateFlags = 0
bindingCount: uint32_t = 4
pBindings: const VkDescriptorSetLayoutBinding* = 0x7ffe6343a728
pBindings[0]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a728:
binding: uint32_t = 0
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[1]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a740:
binding: uint32_t = 1
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[2]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a758:
binding: uint32_t = 2
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[3]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a770:
binding: uint32_t = 3
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER (6)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pAllocator: const VkAllocationCallbacks* = NULL
pSetLayout: VkDescriptorSetLayout* = 0xce
Thread 0, Frame 0:
vkCreatePipelineLayout(device, pCreateInfo, pAllocator, pPipelineLayout) returns VkResult VK_SUCCESS (0):
device: VkDevice = 0x7f811c6cfb40
pCreateInfo: const VkPipelineLayoutCreateInfo* = 0x7ffe6343aa90:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO (30)
pNext: const void* = NULL
flags: VkPipelineLayoutCreateFlags = 0
setLayoutCount: uint32_t = 2
pSetLayouts: const VkDescriptorSetLayout* = 0x7ffe6343ac00
pSetLayouts[0]: const VkDescriptorSetLayout = 0xcd
pSetLayouts[1]: const VkDescriptorSetLayout = 0xce
pushConstantRangeCount: uint32_t = 0
pPushConstantRanges: const VkPushConstantRange* = 0x7ffe6343aaec
pAllocator: const VkAllocationCallbacks* = NULL
pPipelineLayout: VkPipelineLayout* = 0xcf
Thread 0, Frame 0:
vkCreateComputePipelines(device, pipelineCache, createInfoCount, pCreateInfos, pAllocator, pPipelines) returns VkResult VK_SUCCESS (0):
device: VkDevice = 0x7f811c6cfb40
pipelineCache: VkPipelineCache = 0
createInfoCount: uint32_t = 1
pCreateInfos: const VkComputePipelineCreateInfo* = 0x7ffe6343ab50
pCreateInfos[0]: const VkComputePipelineCreateInfo = 0x7ffe6343ab50:
sType: VkStructureType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO (29)
pNext: const void* = NULL
flags: VkPipelineCreateFlags = 0
stage: VkPipelineShaderStageCreateInfo = 0x7ffe6343ab68:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO (18)
pNext: const void* = NULL
flags: VkPipelineShaderStageCreateFlags = 0
stage: VkShaderStageFlagBits = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
module: VkShaderModule = 0xcb
pName: const char* = "main"
pSpecializationInfo: const VkSpecializationInfo* = NULL
layout: VkPipelineLayout = 0xcf
basePipelineHandle: VkPipeline = 0
basePipelineIndex: int32_t = 0
pAllocator: const VkAllocationCallbacks* = NULL
pPipelines: VkPipeline* = 0x7ffe6343aa50
pPipelines[0]: VkPipeline = 0xd0
Thread 0, Frame 0:
vkCreateDescriptorSetLayout(device, pCreateInfo, pAllocator, pSetLayout) returns VkResult VK_SUCCESS (0):
device: VkDevice = 0x7f811c6cfb40
pCreateInfo: const VkDescriptorSetLayoutCreateInfo* = 0x7ffe6343a6b0:
sType: VkStructureType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO (32)
pNext: const void* = NULL
flags: VkDescriptorSetLayoutCreateFlags = 0
bindingCount: uint32_t = 6
pBindings: const VkDescriptorSetLayoutBinding* = 0x7ffe6343a728
pBindings[0]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a728:
binding: uint32_t = 0
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER (1)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = NULL
pBindings[1]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a740:
binding: uint32_t = 1
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER (1)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = NULL
pBindings[2]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a758:
binding: uint32_t = 2
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE (3)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[3]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a770:
binding: uint32_t = 3
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[4]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a788:
binding: uint32_t = 4
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[5]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a7a0:
binding: uint32_t = 5
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE (3)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pAllocator: const VkAllocationCallbacks* = NULL
pSetLayout: VkDescriptorSetLayout* = 0xd1
Thread 0, Frame 0:
vkCreateDescriptorSetLayout(device, pCreateInfo, pAllocator, pSetLayout) returns VkResult VK_SUCCESS (0):
device: VkDevice = 0x7f811c6cfb40
pCreateInfo: const VkDescriptorSetLayoutCreateInfo* = 0x7ffe6343a6b0:
sType: VkStructureType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO (32)
pNext: const void* = NULL
flags: VkDescriptorSetLayoutCreateFlags = 0
bindingCount: uint32_t = 4
pBindings: const VkDescriptorSetLayoutBinding* = 0x7ffe6343a728
pBindings[0]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a728:
binding: uint32_t = 0
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[1]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a740:
binding: uint32_t = 1
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[2]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a758:
binding: uint32_t = 2
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER (7)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pBindings[3]: const VkDescriptorSetLayoutBinding = 0x7ffe6343a770:
binding: uint32_t = 3
descriptorType: VkDescriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER (6)
descriptorCount: uint32_t = 1
stageFlags: VkShaderStageFlags = 32 (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)
pImmutableSamplers: const VkSampler* = UNUSED
pAllocator: const VkAllocationCallbacks* = NULL
pSetLayout: VkDescriptorSetLayout* = 0xd2
Thread 0, Frame 0:
vkCreatePipelineLayout(device, pCreateInfo, pAllocator, pPipelineLayout) returns VkResult VK_SUCCESS (0):
device: VkDevice = 0x7f811c6cfb40
pCreateInfo: const VkPipelineLayoutCreateInfo* = 0x7ffe6343aa90:
sType: VkStructureType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO (30)
pNext: const void* = NULL
flags: VkPipelineLayoutCreateFlags = 0
setLayoutCount: uint32_t = 2
pSetLayouts: const VkDescriptorSetLayout* = 0x7ffe6343ac00
pSetLayouts[0]: const VkDescriptorSetLayout = 0xd1
pSetLayouts[1]: const VkDescriptorSetLayout = 0xd2
pushConstantRangeCount: uint32_t = 0
pPushConstantRanges: const VkPushConstantRange* = 0x7ffe6343aaec
pAllocator: const VkAllocationCallbacks* = NULL
pPipelineLayout: VkPipelineLayout* = 0xd3
[1] 32135 segmentation fault (core dumped) cargo run --example simple --release
Looks good to me. But the problem is still there.
SiebenCorgie
on 27 Sep 2018
Can you provide a backtrace using gdb?
Otherwise I'll try to clone your repo and reproduce later.
This might very well be a bug in
- vulkano or
- glsl or
- vulkan driver
can you also post the output of vulkaninfo?
nicokoch
on 28 Sep 2018
👍1
Something fishy I noticed: You do not create the window on the main thread.
On MacOS this panics: RUST_BACKTRACE=1 cargo run --example simple
Finished dev [unoptimized + debuginfo] target(s) in 1.33s
Runningtarget/debug/examples/simple
Starting render builder
Created App Info
Created Instance
thread '<unnamed>' panicked at 'Windows can only be created on the main thread on macOS'
nicokoch
on 28 Sep 2018
👍1
Hi,
Yes, the window and its input run in another thread. I did not know that this is a problem on Mac (works on Linux and Windows so far). The problem is, that it is not so easy to move it to the main thread atm. Do you have a access to a Vulkan capable computer with Windows or Linux?
Anyways, I attached the output of vulkan info.
vulkaninfo.txt
The gdb backtrace looks like this:
(gdb) backtrace full
#0 0x00007ffff2fcddc3 in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
#1 0x00007ffff2fcbd59 in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
#2 0x00007ffff42cc61e in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
#3 0x00007ffff2fb1d58 in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
#4 0x00007ffff2f9e494 in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
#5 0x00007ffff2f9f43f in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
#6 0x00007ffff2f954fd in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
#7 0x00007ffff2f6f4e6 in ?? () from /usr/lib/amdvlk64.so
No symbol table info available.
[1] 6096 segmentation fault (core dumped) gdb target/debug/examples/simple
I never used gdb before (or debuged correctly in general) so I uploaded the core file as well for you. I only runed it once and then generate-core-file. I hope it helps.
SiebenCorgie
on 28 Sep 2018
I got my shader now working. What I did was that I now handle the three variables of the struct TriangleHit explicit instead. So I got a float tvec vec2 uv and int has_hit which are changed if the return value of the line
TriangleHit leafhit = ray_intersect_mesh(ray, node.left, node.right);
has actually hit something.
So I suspect that something gets screwed up when trying to overwrite a custom struct variable with a new one.
I also tried the old code on windows (with AMD and intel cards) again and it fails there as well, so I think its not a driver problem but more in the layer of the shader compiler / vulkano.
SiebenCorgie
on 30 Sep 2018
Okay, then it's probably an issue with shaderc.
Any reason you use amdvlk instead of radv?
nicokoch
on 1 Oct 2018
I hoped to resolve the issue by changing the driver and then was to lazy to go back. Also, there is a realy nice GPU-Profiler you can use with AMDVLK (RadeonProfiler). But I'll probably go back to RADV at some point.
SiebenCorgie
on 1 Oct 2018
Was this page helpful?
0 / 5 - 0 ratings
Related issues
croraf
路
5Comments
norru
路
11Comments
franckv
路
6Comments
Vengarioth
路
7Comments
CleanCut
路
6Comments
Most helpful comment
Have you tried enabling validation layers?