| Matt Arsenault | cb3fa37 | 2017-02-08 06:44:58 +0000 | [diff] [blame^] | 1 | ; RUN: opt -S -mtriple=amdgcn-- -mcpu=bonaire -loop-reduce < %s | FileCheck -check-prefix=OPT %s |
| 2 | |
| 3 | target datalayout = "e-p:32:32-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-p24:64:64-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64" |
| 4 | |
| 5 | ; Make sure the pointer / address space of AtomicRMW is considered |
| 6 | |
| 7 | ; OPT-LABEL: @test_local_atomicrmw_addressing_loop_uniform_index_max_offset_i32( |
| 8 | |
| 9 | ; OPT-NOT: getelementptr |
| 10 | |
| 11 | ; OPT: .lr.ph: |
| 12 | ; OPT: %lsr.iv2 = phi i32 addrspace(3)* [ %scevgep3, %.lr.ph ], [ %arg1, %.lr.ph.preheader ] |
| 13 | ; OPT: %lsr.iv1 = phi i32 addrspace(3)* [ %scevgep, %.lr.ph ], [ %arg0, %.lr.ph.preheader ] |
| 14 | ; OPT: %lsr.iv = phi i32 [ %lsr.iv.next, %.lr.ph ], [ %n, %.lr.ph.preheader ] |
| 15 | ; OPT: %scevgep4 = getelementptr i32, i32 addrspace(3)* %lsr.iv2, i32 16383 |
| 16 | ; OPT: %tmp4 = atomicrmw add i32 addrspace(3)* %scevgep4, i32 undef seq_cst |
| 17 | ; OPT: %tmp7 = atomicrmw add i32 addrspace(3)* %lsr.iv1, i32 undef seq_cst |
| 18 | ; OPT: %0 = atomicrmw add i32 addrspace(3)* %lsr.iv1, i32 %tmp8 seq_cst |
| 19 | ; OPT: br i1 %exitcond |
| 20 | define void @test_local_atomicrmw_addressing_loop_uniform_index_max_offset_i32(i32 addrspace(3)* noalias nocapture %arg0, i32 addrspace(3)* noalias nocapture readonly %arg1, i32 %n) #0 { |
| 21 | bb: |
| 22 | %tmp = icmp sgt i32 %n, 0 |
| 23 | br i1 %tmp, label %.lr.ph.preheader, label %._crit_edge |
| 24 | |
| 25 | .lr.ph.preheader: ; preds = %bb |
| 26 | br label %.lr.ph |
| 27 | |
| 28 | ._crit_edge.loopexit: ; preds = %.lr.ph |
| 29 | br label %._crit_edge |
| 30 | |
| 31 | ._crit_edge: ; preds = %._crit_edge.loopexit, %bb |
| 32 | ret void |
| 33 | |
| 34 | .lr.ph: ; preds = %.lr.ph, %.lr.ph.preheader |
| 35 | %indvars.iv = phi i32 [ %indvars.iv.next, %.lr.ph ], [ 0, %.lr.ph.preheader ] |
| 36 | %tmp1 = add nuw nsw i32 %indvars.iv, 16383 |
| 37 | %tmp3 = getelementptr inbounds i32, i32 addrspace(3)* %arg1, i32 %tmp1 |
| 38 | %tmp4 = atomicrmw add i32 addrspace(3)* %tmp3, i32 undef seq_cst |
| 39 | %tmp6 = getelementptr inbounds i32, i32 addrspace(3)* %arg0, i32 %indvars.iv |
| 40 | %tmp7 = atomicrmw add i32 addrspace(3)* %tmp6, i32 undef seq_cst |
| 41 | %tmp8 = add nsw i32 %tmp7, %tmp4 |
| 42 | atomicrmw add i32 addrspace(3)* %tmp6, i32 %tmp8 seq_cst |
| 43 | %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1 |
| 44 | %exitcond = icmp eq i32 %indvars.iv.next, %n |
| 45 | br i1 %exitcond, label %._crit_edge.loopexit, label %.lr.ph |
| 46 | } |
| 47 | |
| 48 | ; OPT-LABEL: test_local_cmpxchg_addressing_loop_uniform_index_max_offset_i32( |
| 49 | ; OPT-NOT: getelementptr |
| 50 | |
| 51 | ; OPT: .lr.ph: |
| 52 | ; OPT: %lsr.iv2 = phi i32 addrspace(3)* [ %scevgep3, %.lr.ph ], [ %arg1, %.lr.ph.preheader ] |
| 53 | ; OPT: %lsr.iv1 = phi i32 addrspace(3)* [ %scevgep, %.lr.ph ], [ %arg0, %.lr.ph.preheader ] |
| 54 | ; OPT: %lsr.iv = phi i32 [ %lsr.iv.next, %.lr.ph ], [ %n, %.lr.ph.preheader ] |
| 55 | ; OPT: %scevgep4 = getelementptr i32, i32 addrspace(3)* %lsr.iv2, i32 16383 |
| 56 | ; OPT: %tmp4 = cmpxchg i32 addrspace(3)* %scevgep4, i32 undef, i32 undef seq_cst monotonic |
| 57 | define void @test_local_cmpxchg_addressing_loop_uniform_index_max_offset_i32(i32 addrspace(3)* noalias nocapture %arg0, i32 addrspace(3)* noalias nocapture readonly %arg1, i32 %n) #0 { |
| 58 | bb: |
| 59 | %tmp = icmp sgt i32 %n, 0 |
| 60 | br i1 %tmp, label %.lr.ph.preheader, label %._crit_edge |
| 61 | |
| 62 | .lr.ph.preheader: ; preds = %bb |
| 63 | br label %.lr.ph |
| 64 | |
| 65 | ._crit_edge.loopexit: ; preds = %.lr.ph |
| 66 | br label %._crit_edge |
| 67 | |
| 68 | ._crit_edge: ; preds = %._crit_edge.loopexit, %bb |
| 69 | ret void |
| 70 | |
| 71 | .lr.ph: ; preds = %.lr.ph, %.lr.ph.preheader |
| 72 | %indvars.iv = phi i32 [ %indvars.iv.next, %.lr.ph ], [ 0, %.lr.ph.preheader ] |
| 73 | %tmp1 = add nuw nsw i32 %indvars.iv, 16383 |
| 74 | %tmp3 = getelementptr inbounds i32, i32 addrspace(3)* %arg1, i32 %tmp1 |
| 75 | %tmp4 = cmpxchg i32 addrspace(3)* %tmp3, i32 undef, i32 undef seq_cst monotonic |
| 76 | %tmp4.0 = extractvalue { i32, i1 } %tmp4, 0 |
| 77 | %tmp6 = getelementptr inbounds i32, i32 addrspace(3)* %arg0, i32 %indvars.iv |
| 78 | %tmp7 = cmpxchg i32 addrspace(3)* %tmp6, i32 undef, i32 undef seq_cst monotonic |
| 79 | %tmp7.0 = extractvalue { i32, i1 } %tmp7, 0 |
| 80 | %tmp8 = add nsw i32 %tmp7.0, %tmp4.0 |
| 81 | atomicrmw add i32 addrspace(3)* %tmp6, i32 %tmp8 seq_cst |
| 82 | %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1 |
| 83 | %exitcond = icmp eq i32 %indvars.iv.next, %n |
| 84 | br i1 %exitcond, label %._crit_edge.loopexit, label %.lr.ph |
| 85 | } |
| 86 | |
| 87 | attributes #0 = { nounwind } |