Divide by zero crash validating single-draw vkCmdDrawMeshTasksIndirectEXT

[baldurk]

Environment:

• OS: Windows 11
• GPU and driver version: AMD 7900 XTX driver 24.10.1
• SDK or header version if building from repo: Built from b10d21a
• Options enabled (synchronization, best practices, etc.): GPU validation & synchronization

Describe the Issue

If calling vkCmdDrawMeshTasksIndirectEXT (possibly others, but this was where it crashed for me) with drawCount == 1 the spec says stride is ignored, but on this line:

Vulkan-ValidationLayers/layers/gpu/cmd_validation/gpuav_draw.cpp

Line 657 in 07e3442

max_held_draw_cmds = static_cast<uint32_t>((draw_buffer_size - draw_buffer_offset) / draw_cmds_byte_stride);

There's an unconditional divide by stride, I believe it's to clamp the draw count if he draws read out of bounds. In the event that stride is set to 0 this causes a crash.

Expected behavior

When calling an indirect draw with count of 1 stride is not required to be non-zero so this should be ignored.

Valid Usage ID
N/A

Additional context
Adding a simple drawCount > 1 to the check worked for me, though I'm assuming there's earlier validation that stride is properly set when requried.

[spencer-lunarg]

Thanks for catching this (we literately just merged this change in this week)

We will get some mesh tests (and other things that usedrawCount) in for this

[arno-lunarg]

Thank you for spotting that! Especially since the error spat at me when testing that was awful, and the debugger would not even break ...