80 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc -mtriple=amdgcn-amd-amdpal < %s | FileCheck -check-prefix=GCN %s3 4define amdgpu_kernel void @global_atomic_fadd_ret_f32_wrong_subtarget(ptr addrspace(1) %ptr) #1 {5; GCN-LABEL: global_atomic_fadd_ret_f32_wrong_subtarget:6; GCN: ; %bb.0:7; GCN-NEXT: s_mov_b64 s[0:1], exec8; GCN-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 09; GCN-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v010; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v011; GCN-NEXT: ; implicit-def: $vgpr112; GCN-NEXT: s_and_saveexec_b64 s[2:3], vcc13; GCN-NEXT: s_cbranch_execz .LBB0_414; GCN-NEXT: ; %bb.1:15; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x016; GCN-NEXT: s_bcnt1_i32_b64 s0, s[0:1]17; GCN-NEXT: v_cvt_f32_ubyte0_e32 v1, s018; GCN-NEXT: s_mov_b64 s[6:7], 019; GCN-NEXT: v_mul_f32_e32 v2, 4.0, v120; GCN-NEXT: s_waitcnt lgkmcnt(0)21; GCN-NEXT: s_load_dword s8, s[4:5], 0x022; GCN-NEXT: v_mov_b32_e32 v3, 023; GCN-NEXT: s_waitcnt lgkmcnt(0)24; GCN-NEXT: v_mov_b32_e32 v1, s825; GCN-NEXT: .LBB0_2: ; %atomicrmw.start26; GCN-NEXT: ; =>This Inner Loop Header: Depth=127; GCN-NEXT: v_mov_b32_e32 v5, v128; GCN-NEXT: v_add_f32_e32 v4, v5, v229; GCN-NEXT: global_atomic_cmpswap v1, v3, v[4:5], s[4:5] glc30; GCN-NEXT: s_waitcnt vmcnt(0)31; GCN-NEXT: buffer_wbinvl132; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], v1, v533; GCN-NEXT: s_or_b64 s[6:7], s[0:1], s[6:7]34; GCN-NEXT: s_andn2_b64 exec, exec, s[6:7]35; GCN-NEXT: s_cbranch_execnz .LBB0_236; GCN-NEXT: ; %bb.3: ; %Flow37; GCN-NEXT: s_or_b64 exec, exec, s[6:7]38; GCN-NEXT: .LBB0_4: ; %Flow239; GCN-NEXT: s_or_b64 exec, exec, s[2:3]40; GCN-NEXT: v_readfirstlane_b32 s0, v141; GCN-NEXT: v_cvt_f32_ubyte0_e32 v0, v042; GCN-NEXT: v_mov_b32_e32 v1, s043; GCN-NEXT: v_mad_f32 v0, v0, 4.0, s044; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc45; GCN-NEXT: global_store_dword v[0:1], v0, off46; GCN-NEXT: s_endpgm47 %result = atomicrmw fadd ptr addrspace(1) %ptr, float 4.0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !048 store float %result, ptr addrspace(1) poison49 ret void50}51 52define amdgpu_kernel void @global_atomic_fadd_noret_f32_wrong_subtarget(ptr addrspace(1) %ptr) #1 {53; GCN-LABEL: global_atomic_fadd_noret_f32_wrong_subtarget:54; GCN: ; %bb.0:55; GCN-NEXT: s_mov_b64 s[0:1], exec56; GCN-NEXT: v_mbcnt_lo_u32_b32 v0, s0, 057; GCN-NEXT: v_mbcnt_hi_u32_b32 v0, s1, v058; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v059; GCN-NEXT: s_and_saveexec_b64 s[2:3], vcc60; GCN-NEXT: s_cbranch_execz .LBB1_261; GCN-NEXT: ; %bb.1:62; GCN-NEXT: s_load_dwordx2 s[2:3], s[4:5], 0x063; GCN-NEXT: s_bcnt1_i32_b64 s0, s[0:1]64; GCN-NEXT: v_cvt_f32_ubyte0_e32 v1, s065; GCN-NEXT: v_mov_b32_e32 v0, 066; GCN-NEXT: v_mul_f32_e32 v1, 4.0, v167; GCN-NEXT: s_waitcnt lgkmcnt(0)68; GCN-NEXT: global_atomic_add_f32 v0, v1, s[2:3]69; GCN-NEXT: s_waitcnt vmcnt(0)70; GCN-NEXT: buffer_wbinvl171; GCN-NEXT: .LBB1_2:72; GCN-NEXT: s_endpgm73 %result = atomicrmw fadd ptr addrspace(1) %ptr, float 4.0 syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !074 ret void75}76 77attributes #1 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "target-cpu"="gfx803" "target-features"="+atomic-fadd-no-rtn-insts" }78 79!0 = !{}80