In my case, I'm working on migrating away from my variable descriptor array pattern, and towards Slang's buffer pointers.

I have a differentiable slang test, which optimizes the rotation of an oriented bounding box by doing gradient descent on the surface area.

That function looks like this:

[Differentiable] float computeOBB(float3 eul, no_diff float3 a, no_diff float3 b, no_diff float3 c, no_diff gprt::Buffer aabbBuffer, no_diff float3 *aabbsPtr)

I'm migrating away from this gprt::Buffer aabbBuffer type, and towards a direct float3* type.

In Vulkan, I can see that I'm getting the following validation error:

VUID-VkShaderModuleCreateInfo-pCode-08737(ERROR / SPEC): msgNum: -1520283006 - Validation Error: [ VUID-VkShaderModuleCreateInfo-pCode-08737 ] | MessageID = 0xa5625282 | vkCreateShaderModule(): pCreateInfo->pCode (spirv-val produced an error):
OpFunctionCall Argument '2133[%2133]'s type does not match Function '313[%313]'s parameter type.
%2140 = OpFunctionCall %float %s_primal_ctx_computeOBB %2139 %2126 %2127 %2128 %2129 %2133 %2137
The Vulkan spec states: If pCode is a pointer to SPIR-V code, pCode must adhere to the validation rules described by the Validation Rules within a Module section of the SPIR-V Environment appendix (https://vulkan.lunarg.com/doc/view/1.3.296.0/windows/1.3-extensions/vkspec.html#VUID-VkShaderModuleCreateInfo-pCode-08737)

It looks like the IR code for the fwd_diff is wrong:
this function

float sumOfSquares(float x, float y, no_diff float* test)

generates following IR for fwd_diff:

func %sx5Ffwdx5FsumOfSquares : Func(DiffPairUserCode(Float, %70), DiffPairUserCode(Float, %70), DiffPairUserCode(Float, %70), Attributed(DiffPairUserCode(Float, %70)))

The last parameter should be something like:

Attributed(Ptr(Float, 0 : UInt64), %no_diff)