[llvm] [AMDGPU] Allow amdgpu-waves-per-eu to lower target occupancy range (PR #168358)

[Matt Arsenault via llvm-commits]

================
@@ -1,61 +0,0 @@
-; RUN: opt -S -mtriple=amdgcn-unknown-unknown -mcpu=tahiti -passes=amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck %s
-
-; Both of these kernels have the same value for
-; amdgpu-flat-work-group-size, except one explicitly sets it. This is
-; a program visible property which should always take precedence over
-; the amdgpu-waves-per-eu optimization hint.
-;
-; The range is incompatible with the amdgpu-waves-per-eu value, so the
-; flat work group size should take precedence implying a requirement
-; to support 1024 size workgroups (which exceeds the available LDS
-; amount).
-
-; CHECK-NOT: @no_flat_workgroup_size.stack
-; CHECK-NOT: @explicit_default_workgroup_size.stack
-
-; CHECK-LABEL: @no_flat_workgroup_size(
-; CHECK: alloca [5 x i32]
----------------
arsenm wrote:

I don't understand why this test was deleted, flat-workgroup-size still needs to win 

https://github.com/llvm/llvm-project/pull/168358