[llvm] [AMDGPU] Update PromoteAlloca to handle GEPs with variable offset. (PR #122342)

[Sumanth Gundapaneni via llvm-commits]

================
@@ -122,6 +122,34 @@ define amdgpu_vs void @promote_load_from_store_aggr() #0 {
   ret void
 }
 
+%Block4 = type { [2 x i32], i32 }
+ at block4 = external addrspace(1) global %Block4
+%gl_PV = type { <4 x i32>, i32, [1 x i32], [1 x i32] }
+ at pv1 = external addrspace(1) global %gl_PV
+
+; This should should not crash on variable offset that can be
+; optimized out (variable foo4 in the test)
+define amdgpu_vs void @promote_load_from_store_aggr_varoff() local_unnamed_addr {
+; CHECK-LABEL: @promote_load_from_store_aggr_varoff(
+; CHECK-NEXT:    [[FOO3_UNPACK2:%.*]] = load i32, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @block4, i64 8), align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <3 x i32> undef, i32 [[FOO3_UNPACK2]], i32 2
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <3 x i32> [[TMP1]], i32 [[FOO3_UNPACK2]]
+; CHECK-NEXT:    [[FOO12:%.*]] = insertelement <4 x i32> poison, i32 [[TMP2]], i64 3
+; CHECK-NEXT:    store <4 x i32> [[FOO12]], ptr addrspace(1) @pv1, align 16
+; CHECK-NEXT:    ret void
+;
+  %f1 = alloca [3 x i32], align 4, addrspace(5)
+  %G1 = getelementptr inbounds i8, ptr addrspace(5) %f1, i32 8
+  %foo3.unpack2 = load i32, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @block4, i64 8), align 4
+  store i32 %foo3.unpack2, ptr addrspace(5) %G1, align 4
+  %foo4 = load i32, ptr addrspace(5) %G1, align 4
----------------
sgundapa wrote:

Sure

https://github.com/llvm/llvm-project/pull/122342