1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck %s -check-prefix=MUBUF 3; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -mattr=+enable-flat-scratch < %s | FileCheck %s -check-prefix=FLATSCR 4 5; Make sure there's no assertion from passing a 0 alignment value 6define void @memcpy_fixed_align(ptr addrspace(5) %dst, ptr addrspace(1) %src) { 7; MUBUF-LABEL: memcpy_fixed_align: 8; MUBUF: ; %bb.0: 9; MUBUF-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) 10; MUBUF-NEXT: global_load_dwordx2 v[11:12], v[1:2], off offset:32 11; MUBUF-NEXT: global_load_dwordx4 v[3:6], v[1:2], off 12; MUBUF-NEXT: global_load_dwordx4 v[7:10], v[1:2], off offset:16 13; MUBUF-NEXT: s_lshr_b32 s4, s32, 6 14; MUBUF-NEXT: s_waitcnt vmcnt(2) 15; MUBUF-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:32 16; MUBUF-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:36 17; MUBUF-NEXT: s_waitcnt vmcnt(3) 18; MUBUF-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:12 19; MUBUF-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:8 20; MUBUF-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:4 21; MUBUF-NEXT: buffer_store_dword v3, off, s[0:3], s32 22; MUBUF-NEXT: s_waitcnt vmcnt(6) 23; MUBUF-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:28 24; MUBUF-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:24 25; MUBUF-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:20 26; MUBUF-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:16 27; MUBUF-NEXT: ;;#ASMSTART 28; MUBUF-NEXT: ; use s4 29; MUBUF-NEXT: ;;#ASMEND 30; MUBUF-NEXT: s_waitcnt vmcnt(0) 31; MUBUF-NEXT: s_setpc_b64 s[30:31] 32; 33; FLATSCR-LABEL: memcpy_fixed_align: 34; FLATSCR: ; %bb.0: 35; FLATSCR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) 36; FLATSCR-NEXT: global_load_dwordx4 v[3:6], v[1:2], off 37; FLATSCR-NEXT: global_load_dwordx4 v[7:10], v[1:2], off offset:16 38; FLATSCR-NEXT: global_load_dwordx2 v[11:12], v[1:2], off offset:32 39; FLATSCR-NEXT: s_mov_b32 s0, s32 40; FLATSCR-NEXT: s_waitcnt vmcnt(2) 41; FLATSCR-NEXT: scratch_store_dwordx4 off, v[3:6], s32 42; FLATSCR-NEXT: s_waitcnt vmcnt(2) 43; FLATSCR-NEXT: scratch_store_dwordx4 off, v[7:10], s32 offset:16 44; FLATSCR-NEXT: s_waitcnt vmcnt(2) 45; FLATSCR-NEXT: scratch_store_dwordx2 off, v[11:12], s32 offset:32 46; FLATSCR-NEXT: ;;#ASMSTART 47; FLATSCR-NEXT: ; use s0 48; FLATSCR-NEXT: ;;#ASMEND 49; FLATSCR-NEXT: s_waitcnt vmcnt(0) 50; FLATSCR-NEXT: s_setpc_b64 s[30:31] 51 %alloca = alloca [40 x i8], addrspace(5) 52 call void @llvm.memcpy.p5.p1.i64(ptr addrspace(5) align 4 dereferenceable(40) %alloca, ptr addrspace(1) align 4 dereferenceable(40) %src, i64 40, i1 false) 53 call void asm sideeffect "; use $0", "s"(ptr addrspace(5) %alloca) #0 54 ret void 55} 56 57declare void @llvm.memcpy.p5.p1.i64(ptr addrspace(5) noalias nocapture writeonly, ptr addrspace(1) noalias nocapture readonly, i64, i1 immarg) #0 58 59attributes #0 = { argmemonly nounwind willreturn } 60