metal validation crash on macos

[aentity]

Using: ff15ebc (current master, wgpu 0.6)

Error:

validateFunctionArguments:3478: failed assertion `Vertex Function(main0): argument _17[0] from buffer(0) with offset(0) and length(68) has space for 68 bytes, but argument has a length(80).'

Backtrace (release, sorry):

thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
    frame #0: 0x00007fff683a92c6 libsystem_kernel.dylib`__pthread_kill + 10
    frame #1: 0x000000010167f80d libsystem_pthread.dylib`pthread_kill + 284
    frame #2: 0x00007fff683136a6 libsystem_c.dylib`abort + 127
    frame #3: 0x00007fff682dc20d libsystem_c.dylib`__assert_rtn + 324
    frame #4: 0x00007fff409886b6 Metal`MTLReportFailure + 567
    frame #5: 0x00007fff5ab1719b MetalTools`validateFunctionArguments(id<MTLDevice>, NSString*, NSString, NSArray*, MTLDebugFunctionArgument*, unsigned long, MTLDebugFunctionArgument, unsigned long, MTLDebugFunctionArgument, unsigned long) + 1019
    frame #6: 0x00007fff5ab16854 MetalTools`-[MTLDebugRenderCommandEncoder validateCommonDrawErrors:instanceCount:baseInstance:maxVertexID:] + 805
    frame #7: 0x00007fff5ab17fe2 MetalTools`-[MTLDebugRenderCommandEncoder validateDrawIndexedPrimitives:indexCount:indexType:indexBuffer:indexBufferOffset:instanceCount:function:] + 187
    frame #8: 0x00007fff5ab183af MetalTools`-[MTLDebugRenderCommandEncoder drawIndexedPrimitives:indexCount:indexType:indexBuffer:indexBufferOffset:] + 133
    frame #9: 0x0000000100053711 _foo`gfx_backend_metal::command::exec_render::h123d51eaa38d7465(encoder=0x0000000102964200, command=<unavailable>, resources=<unavailable>) + 2497 [opt]
    frame #10: 0x000000010005d1f2 _foo`_$LT$gfx_backend_metal..command..CommandBuffer$u20$as$u20$gfx_hal..command..CommandBuffer$LT$gfx_backend_metal..Backend$GT$$GT$::draw_indexed::hcd6421e98acd2ef3(self=<unavailable>, indices=<unavailable>, base_vertex=<unavailable>, instances=<unavailable>) + 210 [opt]
    frame #11: 0x00000001006dee5b _foo`wgpu_core::command::render::_$LT$impl$u20$wgpu_core..hub..Global$LT$G$GT$$GT$::command_encoder_run_render_pass_impl::h4fb173dda8bd6f8d(self=<unavailable>, encoder_id=<unavailable>, base=<unavailable>, color_attachments=<unavailable>, depth_stencil_attachment=<unavailable>) + 43851 [opt]
    frame #12: 0x000000010070ddad _foo`_$LT$wgpu..RenderPass$u20$as$u20$core..ops..drop..Drop$GT$::drop::hf0bc6f2fdaf89448(self=<unavailable>) + 157 [opt]
    frame #13: 0x00000001000d11d8 _foo`iced_wgpu::backend::Backend::flush::h274c2929d22fb18e(self=<unavailable>, device=<unavailable>, scale_factor=<unavailable>, transformation=<unavailable>, layer=<unavailable>, staging_belt=<unavailable>, encoder=<unavailable>, target=<unavailable>, target_width=<unavailable>, target_height=<unavailable>) + 1080 [opt]

Is uniform iced uses aligned ok?

[aentity]

I think it's quad.vert, 4*4*4 + 4 = 68 (repr(C) sizeof)), which explains 68 in above error message:

layout (set = 0, binding = 0) uniform Globals {
    mat4 u_Transform;
    float u_Scale;
};

so I think fix can be to have last float be a vec4 or have vec3 unused, so the expected sizeof matches:

layout (set = 0, binding = 0) uniform Globals {
    mat4 u_Transform;
    float u_Scale;
    vec3 unused;
};

[hecrj]

Does the validation crash happen in a particular example?

[aentity]

Sorry for delay, no it is just when I run iced integration in project inside xcode; if "Metal Validation" (default) is enabled, I see this crash.

My proposed fix did not fix issue; I'm trying now with:

layout (set = 0, binding = 0) uniform Globals {
    mat4 u_Transform;
    vec4 u_Scales;
};

to see if same error occurs. A workaround is to turn off metal validation in xcode.

[aentity]

FYI: PR I posted fixes metal gpu validation error on macos/ios