201 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=amdgcn -mcpu=gfx90a < %s | FileCheck %s3 4define protected amdgpu_kernel void @InferNothing(i32 %a, ptr %b, double %c) {5; CHECK-LABEL: InferNothing:6; CHECK: ; %bb.0: ; %entry7; CHECK-NEXT: s_load_dword s6, s[4:5], 0x248; CHECK-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2c9; CHECK-NEXT: s_waitcnt lgkmcnt(0)10; CHECK-NEXT: s_ashr_i32 s7, s6, 3111; CHECK-NEXT: v_mov_b32_e32 v2, s212; CHECK-NEXT: v_mov_b32_e32 v3, s313; CHECK-NEXT: s_lshl_b64 s[2:3], s[6:7], 314; CHECK-NEXT: s_add_u32 s0, s0, s215; CHECK-NEXT: s_addc_u32 s1, s1, s316; CHECK-NEXT: s_add_u32 s0, s0, -817; CHECK-NEXT: s_addc_u32 s1, s1, -118; CHECK-NEXT: v_pk_mov_b32 v[0:1], s[0:1], s[0:1] op_sel:[0,1]19; CHECK-NEXT: flat_atomic_add_f64 v[0:1], v[2:3]20; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)21; CHECK-NEXT: buffer_wbinvl1_vol22; CHECK-NEXT: s_endpgm23entry:24 %i = add nsw i32 %a, -125 %i.2 = sext i32 %i to i6426 %i.3 = getelementptr inbounds double, ptr %b, i64 %i.227 %i.4 = atomicrmw fadd ptr %i.3, double %c syncscope("agent") seq_cst, align 8, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !028 ret void29}30 31define protected amdgpu_kernel void @InferFadd(i32 %a, ptr addrspace(1) %b, double %c) {32; CHECK-LABEL: InferFadd:33; CHECK: ; %bb.0: ; %entry34; CHECK-NEXT: s_load_dword s6, s[4:5], 0x2435; CHECK-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2c36; CHECK-NEXT: s_waitcnt lgkmcnt(0)37; CHECK-NEXT: s_ashr_i32 s7, s6, 3138; CHECK-NEXT: v_mov_b32_e32 v2, s239; CHECK-NEXT: v_mov_b32_e32 v3, s340; CHECK-NEXT: s_lshl_b64 s[2:3], s[6:7], 341; CHECK-NEXT: s_add_u32 s0, s0, s242; CHECK-NEXT: s_addc_u32 s1, s1, s343; CHECK-NEXT: s_add_u32 s0, s0, -844; CHECK-NEXT: s_addc_u32 s1, s1, -145; CHECK-NEXT: v_pk_mov_b32 v[0:1], s[0:1], s[0:1] op_sel:[0,1]46; CHECK-NEXT: flat_atomic_add_f64 v[0:1], v[2:3]47; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)48; CHECK-NEXT: buffer_wbinvl1_vol49; CHECK-NEXT: s_endpgm50entry:51 %i = add nsw i32 %a, -152 %i.2 = sext i32 %i to i6453 %i.3 = getelementptr inbounds double, ptr addrspace(1) %b, i64 %i.254 %i.4 = addrspacecast ptr addrspace(1) %i.3 to ptr55 %0 = atomicrmw fadd ptr %i.4, double %c syncscope("agent") seq_cst, align 8, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !056 ret void57}58 59define protected amdgpu_kernel void @InferMixed(i32 %a, ptr addrspace(1) %b, double %c, ptr %d) {60; CHECK-LABEL: InferMixed:61; CHECK: ; %bb.0: ; %entry62; CHECK-NEXT: s_load_dword s6, s[4:5], 0x2463; CHECK-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0x3c64; CHECK-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2c65; CHECK-NEXT: s_waitcnt lgkmcnt(0)66; CHECK-NEXT: s_ashr_i32 s7, s6, 3167; CHECK-NEXT: v_pk_mov_b32 v[0:1], s[8:9], s[8:9] op_sel:[0,1]68; CHECK-NEXT: v_pk_mov_b32 v[2:3], s[2:3], s[2:3] op_sel:[0,1]69; CHECK-NEXT: s_lshl_b64 s[2:3], s[6:7], 370; CHECK-NEXT: s_add_u32 s0, s0, s271; CHECK-NEXT: s_addc_u32 s1, s1, s372; CHECK-NEXT: s_add_u32 s0, s0, -873; CHECK-NEXT: s_addc_u32 s1, s1, -174; CHECK-NEXT: s_add_u32 s0, s0, 175; CHECK-NEXT: s_addc_u32 s1, s1, 076; CHECK-NEXT: flat_atomic_add_f64 v[0:1], v[2:3]77; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)78; CHECK-NEXT: buffer_wbinvl1_vol79; CHECK-NEXT: v_pk_mov_b32 v[0:1], s[0:1], s[0:1] op_sel:[0,1]80; CHECK-NEXT: flat_atomic_add_f64 v[0:1], v[2:3]81; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)82; CHECK-NEXT: buffer_wbinvl1_vol83; CHECK-NEXT: s_endpgm84entry:85 %i = add nsw i32 %a, -186 %i.2 = sext i32 %i to i6487 %i.3 = getelementptr inbounds double, ptr addrspace(1) %b, i64 %i.288 br label %bb189 90bb1: ; preds = %entry91 %i.7 = ptrtoint ptr addrspace(1) %i.3 to i6492 %i.8 = add nsw i64 %i.7, 193 %i.9 = inttoptr i64 %i.8 to ptr addrspace(1)94 %0 = atomicrmw fadd ptr %d, double %c syncscope("agent") seq_cst, align 8, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !095 %i.11 = addrspacecast ptr addrspace(1) %i.9 to ptr96 %1 = atomicrmw fadd ptr %i.11, double %c syncscope("agent") seq_cst, align 8, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !097 ret void98}99 100define protected amdgpu_kernel void @InferPHI(i32 %a, ptr addrspace(1) %b, double %c) {101; CHECK-LABEL: InferPHI:102; CHECK: ; %bb.0: ; %entry103; CHECK-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0104; CHECK-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1105; CHECK-NEXT: s_load_dword s6, s[4:5], 0x24106; CHECK-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2c107; CHECK-NEXT: s_mov_b32 s14, -1108; CHECK-NEXT: s_mov_b32 s15, 0xe00000109; CHECK-NEXT: s_add_u32 s12, s12, s11110; CHECK-NEXT: s_addc_u32 s13, s13, 0111; CHECK-NEXT: s_waitcnt lgkmcnt(0)112; CHECK-NEXT: s_ashr_i32 s7, s6, 31113; CHECK-NEXT: s_lshl_b64 s[4:5], s[6:7], 3114; CHECK-NEXT: s_add_u32 s0, s0, s4115; CHECK-NEXT: s_addc_u32 s1, s1, s5116; CHECK-NEXT: s_add_u32 s4, s0, -8117; CHECK-NEXT: s_addc_u32 s5, s1, -1118; CHECK-NEXT: s_cmp_eq_u64 s[4:5], 1119; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0120; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1]121; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 1, v0122; CHECK-NEXT: .LBB3_1: ; %bb0123; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1124; CHECK-NEXT: s_and_b64 vcc, exec, s[0:1]125; CHECK-NEXT: s_cbranch_vccnz .LBB3_1126; CHECK-NEXT: ; %bb.2: ; %bb1127; CHECK-NEXT: s_mov_b64 s[0:1], src_shared_base128; CHECK-NEXT: s_cmp_eq_u32 s5, s1129; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0130; CHECK-NEXT: s_andn2_b64 vcc, exec, s[0:1]131; CHECK-NEXT: s_mov_b64 s[0:1], -1132; CHECK-NEXT: s_cbranch_vccnz .LBB3_5133; CHECK-NEXT: ; %bb.3: ; %Flow6134; CHECK-NEXT: s_andn2_b64 vcc, exec, s[0:1]135; CHECK-NEXT: s_cbranch_vccz .LBB3_10136; CHECK-NEXT: .LBB3_4: ; %atomicrmw.phi137; CHECK-NEXT: s_endpgm138; CHECK-NEXT: .LBB3_5: ; %atomicrmw.check.private139; CHECK-NEXT: s_mov_b64 s[0:1], src_private_base140; CHECK-NEXT: s_cmp_eq_u32 s5, s1141; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0142; CHECK-NEXT: s_andn2_b64 vcc, exec, s[0:1]143; CHECK-NEXT: s_mov_b64 s[0:1], -1144; CHECK-NEXT: s_cbranch_vccz .LBB3_7145; CHECK-NEXT: ; %bb.6: ; %atomicrmw.global146; CHECK-NEXT: v_mov_b32_e32 v0, 0147; CHECK-NEXT: v_pk_mov_b32 v[2:3], s[2:3], s[2:3] op_sel:[0,1]148; CHECK-NEXT: global_atomic_add_f64 v0, v[2:3], s[4:5]149; CHECK-NEXT: s_waitcnt vmcnt(0)150; CHECK-NEXT: buffer_wbinvl1_vol151; CHECK-NEXT: s_mov_b64 s[0:1], 0152; CHECK-NEXT: .LBB3_7: ; %Flow153; CHECK-NEXT: s_andn2_b64 vcc, exec, s[0:1]154; CHECK-NEXT: s_cbranch_vccnz .LBB3_9155; CHECK-NEXT: ; %bb.8: ; %atomicrmw.private156; CHECK-NEXT: s_cmp_lg_u64 s[4:5], 0157; CHECK-NEXT: s_cselect_b32 s0, s4, -1158; CHECK-NEXT: v_mov_b32_e32 v2, s0159; CHECK-NEXT: buffer_load_dword v0, v2, s[12:15], 0 offen160; CHECK-NEXT: buffer_load_dword v1, v2, s[12:15], 0 offen offset:4161; CHECK-NEXT: s_waitcnt vmcnt(0)162; CHECK-NEXT: v_add_f64 v[0:1], v[0:1], s[2:3]163; CHECK-NEXT: buffer_store_dword v0, v2, s[12:15], 0 offen164; CHECK-NEXT: buffer_store_dword v1, v2, s[12:15], 0 offen offset:4165; CHECK-NEXT: .LBB3_9: ; %Flow5166; CHECK-NEXT: s_cbranch_execnz .LBB3_4167; CHECK-NEXT: .LBB3_10: ; %atomicrmw.shared168; CHECK-NEXT: s_cmp_lg_u64 s[4:5], 0169; CHECK-NEXT: s_cselect_b32 s0, s4, -1170; CHECK-NEXT: v_mov_b32_e32 v0, s0171; CHECK-NEXT: v_pk_mov_b32 v[2:3], s[2:3], s[2:3] op_sel:[0,1]172; CHECK-NEXT: ds_add_f64 v0, v[2:3]173; CHECK-NEXT: s_waitcnt lgkmcnt(0)174; CHECK-NEXT: s_endpgm175entry:176 %i = add nsw i32 %a, -1177 %i.2 = sext i32 %i to i64178 %i.3 = getelementptr inbounds double, ptr addrspace(1) %b, i64 %i.2179 %i.4 = ptrtoint ptr addrspace(1) %i.3 to i64180 br label %bb0181 182bb0: ; preds = %bb0, %entry183 %phi = phi ptr addrspace(1) [ %i.3, %entry ], [ %i.9, %bb0 ]184 %i.7 = ptrtoint ptr addrspace(1) %phi to i64185 %i.8 = sub nsw i64 %i.7, 1186 %cmp2 = icmp eq i64 %i.8, 0187 %i.9 = inttoptr i64 %i.7 to ptr addrspace(1)188 br i1 %cmp2, label %bb1, label %bb0189 190bb1: ; preds = %bb0191 %i.10 = addrspacecast ptr addrspace(1) %i.9 to ptr192 %0 = atomicrmw fadd ptr %i.10, double %c syncscope("agent") seq_cst, align 8, !amdgpu.no.fine.grained.memory !0193 ret void194}195 196attributes #0 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }197attributes #1 = { mustprogress nounwind willreturn memory(argmem: readwrite) "target-cpu"="gfx90a" }198 199!0 = !{}200!1 = !{i32 5, i32 6}201