1*fcefe957SJeffrey Byrnes; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 2*fcefe957SJeffrey Byrnes; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 < %s | FileCheck -check-prefix=GCN %s 3*fcefe957SJeffrey Byrnes 4*fcefe957SJeffrey Byrnesdefine void @scalarize_insert_subvector(ptr addrspace(3) %inptr, ptr addrspace(3) %inptr1, ptr addrspace(3) %outptr) { 5*fcefe957SJeffrey Byrnes; GCN-LABEL: scalarize_insert_subvector: 6*fcefe957SJeffrey Byrnes; GCN: ; %bb.0: 7*fcefe957SJeffrey Byrnes; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) 8*fcefe957SJeffrey Byrnes; GCN-NEXT: ds_read_b64 v[4:5], v0 9*fcefe957SJeffrey Byrnes; GCN-NEXT: s_waitcnt lgkmcnt(0) 10*fcefe957SJeffrey Byrnes; GCN-NEXT: ds_read_b32 v5, v1 offset:4 11*fcefe957SJeffrey Byrnes; GCN-NEXT: s_waitcnt lgkmcnt(0) 12*fcefe957SJeffrey Byrnes; GCN-NEXT: ds_write_b64 v2, v[4:5] 13*fcefe957SJeffrey Byrnes; GCN-NEXT: s_waitcnt lgkmcnt(0) 14*fcefe957SJeffrey Byrnes; GCN-NEXT: s_setpc_b64 s[30:31] 15*fcefe957SJeffrey Byrnes %load0 = load <2 x i32>, ptr addrspace(3) %inptr, align 8 16*fcefe957SJeffrey Byrnes %load1= load <2 x i32>, ptr addrspace(3) %inptr1, align 8 17*fcefe957SJeffrey Byrnes %shuffle0 = shufflevector <2 x i32> %load1, <2 x i32> poison, <1 x i32> <i32 1> 18*fcefe957SJeffrey Byrnes %bitcast0 = bitcast <1 x i32> %shuffle0 to <2 x half> 19*fcefe957SJeffrey Byrnes %bitcast1 = bitcast <2 x i32> %load0 to <4 x half> 20*fcefe957SJeffrey Byrnes %shuffle1 = shufflevector <2 x half> %bitcast0, <2 x half> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 21*fcefe957SJeffrey Byrnes %shuffle2 = shufflevector <4 x half> %bitcast1, <4 x half> %shuffle1, <4 x i32> <i32 0, i32 1, i32 4, i32 5> 22*fcefe957SJeffrey Byrnes store <4 x half> %shuffle2, ptr addrspace(3) %outptr 23*fcefe957SJeffrey Byrnes ret void 24*fcefe957SJeffrey Byrnes} 25