Vulkan driver instancing bug

[Marcell_K_]

System Setup Information:
-----------------------------------------

System Used: -
CPU SKU: i5 6600K
GPU SKU: HD 530
Processor Line: (Y-Series, U-Series, etc.)
System BIOS Version:
CMOS settings:
Graphics Driver Version: 20.19.15.4473
GOP/VBIOS Version:
Operating System: Windows
OS Version: 10
API: Vulkan 

        apiVersion     = 0x400000  (1.0.0)

        driverVersion  = 65538 (0x10002)
Occurs on non-Intel GPUs?: No (tested on NV)

Steps to Reproduce:
-------------------------------
1.  Create RenderPass & CommandBuffer, set up 2 triangles side - by - side
2.  Use a shader program which outputs gl_InstanceIndex as colour for debugging
3.  Issue vkCmdDraw* against command buffer in pass with firstInstance = 0 on 1st triangle
4.  Issue vkCmdDraw* against command buffer in pass with firstInstance = 1 on 2nd triangle
5.  Observe results (gl_InstanceIndex == 0 for both)
6.  Change code where the two draws are switched
7.  Observe results (gl_InstanceIndex == 1 for both)

Expected Results:
-------------------------------

vkCmdDraw* should use firstInstance given as parameter for the first instance index

Actual Results:
-------------------------------

vkCmdDraw* uses firstInstance set previously and ignores firstInstance argument starting from second draw

Additional Information:
-------------------------------

When viewed in RenderDoc, the second draw appears normally (with correct gl_InstanceIndex), so this parameter might get recorded in the command buffer or the renderpass.

[Michael_C_Intel2]

Hi Marcell,

I am talking with the Vulkan driver team about your issue I will update you when we learn anything.

-Michael 

[Michael_C_Intel2]

Hi Marcell,

Can you share the code your using to create this issue. Or sample that will reproduce the issue? This will help our investigation. Also which vkCmdDraw* command is it? 

Thanks,

-Michael 

[Marcell_K_]

Sure, here you go:

https://github.com/martty/Vulkan/commit/f61c8014123eb474b970a46db2d72ea40434e07e

I modified Sascha Willems's triangle example to show the bug.

I tried both vkCmdDraw and vkCmdDrawIndexed and got the same result. It might work with the indirect ones, those I did not try.

Best regards,

Marcell

[Michael_C_Intel2]

Thanks! I have passed this along to the team. They reproduced the issue yesterday and are looking for the root cause now.

[Michael_C_Intel2]

Hi Marcell,

We have found a fix for your issue. I do not have an ETA when it we will have a public driver with the fix yet. It should make it into the first 15.45 driver (current drivers are 15.40). 

-Michael