[PATCH] D52846: [AMDGPU] Add FixupVectorISel pass, currently Supports SREGs in GLOBAL LD/ST

[Ron Lieberman via Phabricator via llvm-commits]

ronlieb added inline comments.


================
Comment at: test/CodeGen/AMDGPU/conv2d-saddr.ll:16
+; Function Attrs: convergent nounwind
+define hidden amdgpu_kernel void @simpleConv2d(i32 addrspace(1)* nocapture %dst_image, i32 addrspace(1)* nocapture readonly %src_image, i32 addrspace(1)* nocapture readonly %conv_kernel) local_unnamed_addr #0 !kernel_arg_addr_space !5 !kernel_arg_access_qual !6 !kernel_arg_type !7 !kernel_arg_base_type !7 !kernel_arg_type_qual !8 {
+entry:
----------------
arsenm wrote:
> ronlieb wrote:
> > arsenm wrote:
> > > This testcase is way too big. Most of the tests should be only a handful of instructions. There should be ones stressing the immediate limits. I also don't see ones for the atomics that the comments say work
> > reduced test case to 47 lines total.
> This is still too big. You only need not much more than load, a store, and a GEP in each function
is this comment about only needing "load store gep per function?  intended for  a different test? This test, conv2d-saddr.ll only has one function.  perhaps you were thinking of global-saddr-atomics.ll ?


https://reviews.llvm.org/D52846