1 ; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -indvars %s | FileCheck %s 2 3 ; Bug 21148 4 5 ; Induction variables should not be widened for 64-bit integers, 6 ; despite being a legal type. 7 ; 8 ; The cost of basic arithmetic instructions on a 64-bit integer are 9 ; twice as expensive as that on a 32-bit integer, or split into 2 10 ; 32-bit components. 11 12 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" 13 14 ; CHECK-LABEL: @indvar_32_bit( 15 ; CHECK-NOT: sext i32 16 ; CHECK: phi i32 17 define void @indvar_32_bit(i32 %n, i32* nocapture %output) { 18 entry: 19 %cmp5 = icmp sgt i32 %n, 0 20 br i1 %cmp5, label %for.body.preheader, label %for.end 21 22 for.body.preheader: ; preds = %entry 23 br label %for.body 24 25 for.body: ; preds = %for.body.preheader, %for.body 26 %i.06 = phi i32 [ 0, %for.body.preheader ], [ %add, %for.body ] 27 %mul = mul nsw i32 %i.06, %i.06 28 %tmp0 = sext i32 %i.06 to i64 29 %arrayidx = getelementptr inbounds i32, i32* %output, i64 %tmp0 30 store i32 %mul, i32* %arrayidx, align 4 31 %add = add nsw i32 %i.06, 3 32 %cmp = icmp slt i32 %add, %n 33 br i1 %cmp, label %for.body, label %for.end.loopexit 34 35 for.end.loopexit: ; preds = %for.body 36 br label %for.end 37 38 for.end: ; preds = %for.end.loopexit, %entry 39 ret void 40 } 41 42 ; CHECK-LABEL: @no_promote_i32( 43 ; CHECK-NOT: sext i32 44 ; CHECK: br 45 ; CHECK-NOT: shl i64 46 ; CHECK-NOT: ashr i64 47 ; CHECK-NOT: mul nsw i64 48 ; CHECK-NOT: add nsw i64 49 define void @no_promote_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) { 50 entry: 51 br label %for.body 52 53 for.body: 54 %inc = phi i32 [ 0, %entry ], [ %inc.i, %for.body ] 55 %tmp0 = add i32 %a, %inc 56 %shl = shl i32 %inc, 8 57 %shr = ashr exact i32 %shl, 8 58 %mul = mul nsw i32 %shr, %a 59 %add = add nsw i32 %mul, %b 60 %tmp1 = sext i32 %add to i64 61 %arrayidx1 = getelementptr inbounds i32, i32 addrspace(1)* %out, i64 %tmp1 62 store i32 %tmp0, i32 addrspace(1)* %arrayidx1, align 4 63 %inc.i = add nsw i32 %inc, 1 64 %cmp = icmp slt i32 %inc.i, 16 65 br i1 %cmp, label %for.body, label %for.end 66 67 for.end: 68 ret void 69 } 70 71 ; FIXME: This should really be promoted to i64, since it will need to 72 ; be legalized anyway. 73 74 ; CHECK-LABEL: @indvar_48_bit( 75 define void @indvar_48_bit(i48 %n, i48* nocapture %output) { 76 entry: 77 %cmp5 = icmp sgt i48 %n, 0 78 br i1 %cmp5, label %for.body.preheader, label %for.end 79 80 for.body.preheader: ; preds = %entry 81 br label %for.body 82 83 for.body: ; preds = %for.body.preheader, %for.body 84 %i.06 = phi i48 [ 0, %for.body.preheader ], [ %add, %for.body ] 85 %mul = mul nsw i48 %i.06, %i.06 86 %tmp0 = sext i48 %i.06 to i64 87 %arrayidx = getelementptr inbounds i48, i48* %output, i64 %tmp0 88 store i48 %mul, i48* %arrayidx, align 4 89 %add = add nsw i48 %i.06, 3 90 %cmp = icmp slt i48 %add, %n 91 br i1 %cmp, label %for.body, label %for.end.loopexit 92 93 for.end.loopexit: ; preds = %for.body 94 br label %for.end 95 96 for.end: ; preds = %for.end.loopexit, %entry 97 ret void 98 } 99
