1; RUN: llc -O3 -mtriple=hexagon -mcpu=hexagonv5 < %s | FileCheck %s 2; 3; Generate hardware loop when unknown trip count loop is vectorized. 4 5; CHECK: loop0(.LBB{{[0-9]*}}_{{[0-9]*}},r{{[0-9]+}}) 6; CHECK: endloop0 7; CHECK: loop0(.LBB{{[0-9]*}}_{{[0-9]*}},r{{[0-9]+}}) 8; CHECK: endloop0 9 10@A = common global [1000 x i32] zeroinitializer, align 8 11@B = common global [1000 x i32] zeroinitializer, align 8 12 13define i32 @dotprod2(i32 %count) #0 { 14entry.split: 15 %cmp6 = icmp sgt i32 %count, 0 16 br i1 %cmp6, label %polly.cond, label %for.end 17 18for.end.loopexit: 19 br label %for.end 20 21for.end: 22 %sum.0.lcssa.reg2mem.0.load37 = phi i32 [ 0, %entry.split ], [ %p_add34, %polly.loop_if13 ], [ %p_add, %for.end.loopexit ] 23 ret i32 %sum.0.lcssa.reg2mem.0.load37 24 25polly.cond: 26 %0 = icmp sgt i32 %count, 1 27 br i1 %0, label %polly.loop_if, label %polly.loop_if13 28 29polly.loop_exit.loopexit: 30 br label %polly.loop_exit 31 32polly.loop_exit: 33 %1 = phi <2 x i32> [ zeroinitializer, %polly.loop_if ], [ %addp_vec, %polly.loop_exit.loopexit ] 34 %2 = extractelement <2 x i32> %1, i32 0 35 %3 = extractelement <2 x i32> %1, i32 1 36 %add_sum = add i32 %2, %3 37 br label %polly.loop_if13 38 39polly.loop_if: 40 %4 = add i32 %count, -1 41 %leftover_lb = and i32 %4, -2 42 %polly.loop_guard = icmp eq i32 %leftover_lb, 0 43 br i1 %polly.loop_guard, label %polly.loop_exit, label %polly.loop_preheader 44 45polly.stmt.for.body: 46 %addp_vec28 = phi <2 x i32> [ zeroinitializer, %polly.loop_preheader ], [ %addp_vec, %polly.stmt.for.body ] 47 %scevgep.phi = phi ptr [ @A, %polly.loop_preheader ], [ %scevgep.inc, %polly.stmt.for.body ] 48 %scevgep9.phi = phi ptr [ @B, %polly.loop_preheader ], [ %scevgep9.inc, %polly.stmt.for.body ] 49 %polly.indvar = phi i32 [ 0, %polly.loop_preheader ], [ %polly.indvar_next, %polly.stmt.for.body ] 50 %_p_vec_full = load <2 x i32>, ptr %scevgep.phi, align 8 51 %_p_vec_full11 = load <2 x i32>, ptr %scevgep9.phi, align 8 52 %mulp_vec = mul <2 x i32> %_p_vec_full11, %_p_vec_full 53 %addp_vec = add <2 x i32> %mulp_vec, %addp_vec28 54 %polly.indvar_next = add nsw i32 %polly.indvar, 2 55 %polly.loop_cond = icmp eq i32 %polly.indvar, %polly.adjust_ub 56 %scevgep.inc = getelementptr i32, ptr %scevgep.phi, i32 2 57 %scevgep9.inc = getelementptr i32, ptr %scevgep9.phi, i32 2 58 br i1 %polly.loop_cond, label %polly.loop_exit.loopexit, label %polly.stmt.for.body 59 60polly.loop_preheader: 61 %polly.adjust_ub = add i32 %leftover_lb, -2 62 br label %polly.stmt.for.body 63 64polly.loop_if13: 65 %p_add34 = phi i32 [ 0, %polly.cond ], [ %add_sum, %polly.loop_exit ] 66 %merge.lb = phi i32 [ 0, %polly.cond ], [ %leftover_lb, %polly.loop_exit ] 67 %polly.loop_guard17 = icmp slt i32 %merge.lb, %count 68 br i1 %polly.loop_guard17, label %polly.loop_preheader15, label %for.end 69 70polly.stmt.for.body22: 71 %p_add30 = phi i32 [ %p_add34, %polly.loop_preheader15 ], [ %p_add, %polly.stmt.for.body22 ] 72 %polly.indvar18 = phi i32 [ %merge.lb, %polly.loop_preheader15 ], [ %polly.indvar_next19, %polly.stmt.for.body22 ] 73 %5 = tail call i32 @llvm.annotation.i32(i32 %polly.indvar18, ptr null, ptr null, i32 0), !polly.loop.smallTripCount !0 74 %scevgep23 = getelementptr [1000 x i32], ptr @A, i32 0, i32 %polly.indvar18 75 %_p_scalar_ = load i32, ptr %scevgep23, align 4 76 %scevgep24 = getelementptr [1000 x i32], ptr @B, i32 0, i32 %polly.indvar18 77 %_p_scalar_25 = load i32, ptr %scevgep24, align 4 78 %p_mul = mul nsw i32 %_p_scalar_25, %_p_scalar_ 79 %p_add = add nsw i32 %p_mul, %p_add30 80 %polly.indvar_next19 = add nsw i32 %polly.indvar18, 1 81 %polly.loop_cond21 = icmp slt i32 %polly.indvar18, %polly.adjust_ub20 82 br i1 %polly.loop_cond21, label %polly.stmt.for.body22, label %for.end.loopexit 83 84polly.loop_preheader15: 85 %polly.adjust_ub20 = add i32 %count, -1 86 br label %polly.stmt.for.body22 87} 88 89declare i32 @llvm.annotation.i32(i32, ptr, ptr, i32) #1 90 91!0 = !{} 92