1 ; RUN: opt < %s -indvars -S | FileCheck %s 2 3 target triple = "nvptx64-unknown-unknown" 4 5 ; For the nvptx64 architecture, the cost of an arithmetic instruction on a 6 ; 64-bit integer is twice as expensive as that on a 32-bit integer, because the 7 ; hardware needs to simulate a 64-bit integer using two 32-bit integers. 8 ; Therefore, in this particular architecture, we should not widen induction 9 ; variables to 64-bit integers even though i64 is a legal type in the 64-bit 10 ; PTX ISA. 11 12 define void @indvar_32_bit(i32 %n, i32* nocapture %output) { 13 ; CHECK-LABEL: @indvar_32_bit 14 entry: 15 %cmp5 = icmp sgt i32 %n, 0 16 br i1 %cmp5, label %for.body.preheader, label %for.end 17 18 for.body.preheader: ; preds = %entry 19 br label %for.body 20 21 for.body: ; preds = %for.body.preheader, %for.body 22 %i.06 = phi i32 [ 0, %for.body.preheader ], [ %add, %for.body ] 23 ; CHECK: phi i32 24 %mul = mul nsw i32 %i.06, %i.06 25 %0 = sext i32 %i.06 to i64 26 %arrayidx = getelementptr inbounds i32, i32* %output, i64 %0 27 store i32 %mul, i32* %arrayidx, align 4 28 %add = add nsw i32 %i.06, 3 29 %cmp = icmp slt i32 %add, %n 30 br i1 %cmp, label %for.body, label %for.end.loopexit 31 32 for.end.loopexit: ; preds = %for.body 33 br label %for.end 34 35 for.end: ; preds = %for.end.loopexit, %entry 36 ret void 37 } 38
