David Neto
David Neto
If I change the assignment to local_a.x to 66 from zero, I get this at replace pointer bitcasts: ``` define spir_kernel void @foo(%struct.a addrspace(1)* nocapture %struct_out, i32 %n) local_unnamed_addr #0...
Also, with the 66, if I do replace LLVM intrinsics early, I get this IR before descriptor allocation, and this is reasonable: ``` define spir_kernel void @foo(%struct.a addrspace(1)* %struct_out, i32...
Looking at changes in Clspv tests if I replace LLVM intrinsics early. The code generated is noticeably worse. E.g. test/composite_construct.cl has this happen: The boo function is: ``` typedef struct...
I've recently fixed a somewhat similar problem in replace-pointer-bitcasts to handle arrays vs. scalars. But I've confirmed this issue still persists. In the first example, the unhandled case looks like...
+1 to the request to write out SPIR-V disassembly.
I assume this is only doable from the API. That is, we don't need AmberScript or vkscript syntax for this.
Seeing if this works before I request review
I'm not intending this to be included in a CTS, not in this form anyway. I mainly wanted a place to stash this interesting test case.
> Do we still want to keep this open? I converted it to draft. I'm referencing it from WebGPU CTS as an interesting case.
Image comparison seems like a completely separable function. ?! It would be useful and quite usable as a separate utility. It seems overly fussy to *force* someone to write an...