1; RUN: llc -march=hexagon -mcpu=hexagonv5 -disable-hsdr < %s | FileCheck %s 2 3; Check that store is post-incremented. 4; CHECK: memh(r{{[0-9]+}}++#2) 5 6target datalayout = "e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:32:32-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a0:0-n16:32" 7target triple = "hexagon" 8 9define void @matrix_add_const(i32 %N, i16* nocapture %A, i16 signext %val) #0 { 10entry: 11 %cmp5 = icmp eq i32 %N, 0 12 br i1 %cmp5, label %for.end, label %polly.cond 13 14for.end.loopexit: ; preds = %polly.stmt.for.body29 15 br label %for.end 16 17for.end: ; preds = %for.end.loopexit, %polly.loop_header24.preheader, %entry 18 ret void 19 20polly.cond: ; preds = %entry 21 %0 = icmp sgt i32 %N, 3 22 br i1 %0, label %polly.then, label %polly.loop_header24.preheader 23 24polly.then: ; preds = %polly.cond 25 %1 = add i32 %N, -1 26 %leftover_lb = and i32 %1, -4 27 %2 = icmp sgt i32 %leftover_lb, 0 28 br i1 %2, label %polly.loop_body.lr.ph, label %polly.loop_header24.preheader 29 30polly.loop_body.lr.ph: ; preds = %polly.then 31 %3 = insertelement <4 x i16> undef, i16 %val, i32 0 32 %4 = insertelement <4 x i16> %3, i16 %val, i32 1 33 %5 = insertelement <4 x i16> %4, i16 %val, i32 2 34 %6 = insertelement <4 x i16> %5, i16 %val, i32 3 35 br label %polly.loop_body 36 37polly.loop_header24.preheader.loopexit: ; preds = %polly.loop_body 38 br label %polly.loop_header24.preheader 39 40polly.loop_header24.preheader: ; preds = %polly.loop_header24.preheader.loopexit, %polly.then, %polly.cond 41 %polly.loopiv27.ph = phi i32 [ 0, %polly.cond ], [ %leftover_lb, %polly.then ], [ %leftover_lb, %polly.loop_header24.preheader.loopexit ] 42 %7 = icmp slt i32 %polly.loopiv27.ph, %N 43 br i1 %7, label %polly.stmt.for.body29.preheader, label %for.end 44 45polly.stmt.for.body29.preheader: ; preds = %polly.loop_header24.preheader 46 br label %polly.stmt.for.body29 47 48polly.loop_body: ; preds = %polly.loop_body.lr.ph, %polly.loop_body 49 %p_arrayidx.phi = phi i16* [ %A, %polly.loop_body.lr.ph ], [ %p_arrayidx.inc, %polly.loop_body ] 50 %polly.loopiv34 = phi i32 [ 0, %polly.loop_body.lr.ph ], [ %polly.next_loopiv, %polly.loop_body ] 51 %polly.next_loopiv = add nsw i32 %polly.loopiv34, 4 52 %vector_ptr = bitcast i16* %p_arrayidx.phi to <4 x i16>* 53 %_p_vec_full = load <4 x i16>, <4 x i16>* %vector_ptr, align 2 54 %addp_vec = add <4 x i16> %_p_vec_full, %6 55 store <4 x i16> %addp_vec, <4 x i16>* %vector_ptr, align 2 56 %8 = icmp slt i32 %polly.next_loopiv, %leftover_lb 57 %p_arrayidx.inc = getelementptr i16, i16* %p_arrayidx.phi, i32 4 58 br i1 %8, label %polly.loop_body, label %polly.loop_header24.preheader.loopexit 59 60polly.stmt.for.body29: ; preds = %polly.stmt.for.body29.preheader, %polly.stmt.for.body29 61 %polly.loopiv2733 = phi i32 [ %polly.next_loopiv28, %polly.stmt.for.body29 ], [ %polly.loopiv27.ph, %polly.stmt.for.body29.preheader ] 62 %polly.next_loopiv28 = add nsw i32 %polly.loopiv2733, 1 63 %p_arrayidx30 = getelementptr i16, i16* %A, i32 %polly.loopiv2733 64 %_p_scalar_ = load i16, i16* %p_arrayidx30, align 2 65 %p_add = add i16 %_p_scalar_, %val 66 store i16 %p_add, i16* %p_arrayidx30, align 2 67 %exitcond = icmp eq i32 %polly.next_loopiv28, %N 68 br i1 %exitcond, label %for.end.loopexit, label %polly.stmt.for.body29 69} 70 71attributes #0 = { nounwind "fp-contract-model"="standard" "no-frame-pointer-elim-non-leaf" "realign-stack" "relocation-model"="static" "ssp-buffers-size"="8" } 72