[PATCH] D98469: [NVPTX] Avoid temp copy of byval kernel parameters.

[Artem Belevich via Phabricator via llvm-commits]

tra added inline comments.


================
Comment at: llvm/test/CodeGen/NVPTX/lower-kernel-ptr-arg.ll:35
 ; CHECK: cvta.to.global.u64 %[[optr_g:.*]], %[[optr]];
 ; CHECK: ld.param.u64 	%[[iptr:rd.*]], [ptr_in_byval_kernel_param_0+8]
 ; CHECK: cvta.to.global.u64 %[[iptr_g:.*]], %[[iptr]];
----------------
With a temporary copy of the argument, the alignment used to be inferred to be 8 based on the struct alignment and that allowed LLVM to generate 64-bit load. W/o a temporary copy, we're only looking at the load's alignment of  4, which probably evolved from the original 32-bit IR.
Making load alignment to be 8 makes the test work predictably regardless of whether we have a temp copy.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D98469/new/

https://reviews.llvm.org/D98469