1 ; RUN: opt -loop-idiom -mtriple=amdgcn-- -S < %s | FileCheck %s 2 3 ; Mostly copied from x86 version. 4 5 ;To recognize this pattern: 6 ;int popcount(unsigned long long a) { 7 ; int c = 0; 8 ; while (a) { 9 ; c++; 10 ; a &= a - 1; 11 ; } 12 ; return c; 13 ;} 14 ; 15 16 ; CHECK-LABEL: @popcount_i64 17 ; CHECK: entry 18 ; CHECK: llvm.ctpop.i64 19 ; CHECK: ret 20 define i32 @popcount_i64(i64 %a) nounwind uwtable readnone ssp { 21 entry: 22 %tobool3 = icmp eq i64 %a, 0 23 br i1 %tobool3, label %while.end, label %while.body 24 25 while.body: ; preds = %entry, %while.body 26 %c.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ] 27 %a.addr.04 = phi i64 [ %and, %while.body ], [ %a, %entry ] 28 %inc = add nsw i32 %c.05, 1 29 %sub = add i64 %a.addr.04, -1 30 %and = and i64 %sub, %a.addr.04 31 %tobool = icmp eq i64 %and, 0 32 br i1 %tobool, label %while.end, label %while.body 33 34 while.end: ; preds = %while.body, %entry 35 %c.0.lcssa = phi i32 [ 0, %entry ], [ %inc, %while.body ] 36 ret i32 %c.0.lcssa 37 } 38 39 ; CHECK-LABEL: @popcount_i32 40 ; CHECK: entry 41 ; CHECK: llvm.ctpop.i32 42 ; CHECK: ret 43 define i32 @popcount_i32(i32 %a) nounwind uwtable readnone ssp { 44 entry: 45 %tobool3 = icmp eq i32 %a, 0 46 br i1 %tobool3, label %while.end, label %while.body 47 48 while.body: ; preds = %entry, %while.body 49 %c.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ] 50 %a.addr.04 = phi i32 [ %and, %while.body ], [ %a, %entry ] 51 %inc = add nsw i32 %c.05, 1 52 %sub = add i32 %a.addr.04, -1 53 %and = and i32 %sub, %a.addr.04 54 %tobool = icmp eq i32 %and, 0 55 br i1 %tobool, label %while.end, label %while.body 56 57 while.end: ; preds = %while.body, %entry 58 %c.0.lcssa = phi i32 [ 0, %entry ], [ %inc, %while.body ] 59 ret i32 %c.0.lcssa 60 } 61 62 ; CHECK-LABEL: @popcount_i128 63 ; CHECK: entry 64 ; CHECK: llvm.ctpop.i128 65 ; CHECK: ret 66 define i32 @popcount_i128(i128 %a) nounwind uwtable readnone ssp { 67 entry: 68 %tobool3 = icmp eq i128 %a, 0 69 br i1 %tobool3, label %while.end, label %while.body 70 71 while.body: ; preds = %entry, %while.body 72 %c.05 = phi i32 [ %inc, %while.body ], [ 0, %entry ] 73 %a.addr.04 = phi i128 [ %and, %while.body ], [ %a, %entry ] 74 %inc = add nsw i32 %c.05, 1 75 %sub = add i128 %a.addr.04, -1 76 %and = and i128 %sub, %a.addr.04 77 %tobool = icmp eq i128 %and, 0 78 br i1 %tobool, label %while.end, label %while.body 79 80 while.end: ; preds = %while.body, %entry 81 %c.0.lcssa = phi i32 [ 0, %entry ], [ %inc, %while.body ] 82 ret i32 %c.0.lcssa 83 } 84 85 ; To recognize this pattern: 86 ;int popcount(unsigned long long a, int mydata1, int mydata2) { 87 ; int c = 0; 88 ; while (a) { 89 ; c++; 90 ; a &= a - 1; 91 ; mydata1 *= c; 92 ; mydata2 *= (int)a; 93 ; } 94 ; return c + mydata1 + mydata2; 95 ;} 96 97 ; CHECK-LABEL: @popcount2 98 ; CHECK: entry 99 ; CHECK: llvm.ctpop.i64 100 ; CHECK: ret 101 define i32 @popcount2(i64 %a, i32 %mydata1, i32 %mydata2) nounwind uwtable readnone ssp { 102 entry: 103 %tobool9 = icmp eq i64 %a, 0 104 br i1 %tobool9, label %while.end, label %while.body 105 106 while.body: ; preds = %entry, %while.body 107 %c.013 = phi i32 [ %inc, %while.body ], [ 0, %entry ] 108 %mydata2.addr.012 = phi i32 [ %mul1, %while.body ], [ %mydata2, %entry ] 109 %mydata1.addr.011 = phi i32 [ %mul, %while.body ], [ %mydata1, %entry ] 110 %a.addr.010 = phi i64 [ %and, %while.body ], [ %a, %entry ] 111 %inc = add nsw i32 %c.013, 1 112 %sub = add i64 %a.addr.010, -1 113 %and = and i64 %sub, %a.addr.010 114 %mul = mul nsw i32 %inc, %mydata1.addr.011 115 %conv = trunc i64 %and to i32 116 %mul1 = mul nsw i32 %conv, %mydata2.addr.012 117 %tobool = icmp eq i64 %and, 0 118 br i1 %tobool, label %while.end, label %while.body 119 120 while.end: ; preds = %while.body, %entry 121 %c.0.lcssa = phi i32 [ 0, %entry ], [ %inc, %while.body ] 122 %mydata2.addr.0.lcssa = phi i32 [ %mydata2, %entry ], [ %mul1, %while.body ] 123 %mydata1.addr.0.lcssa = phi i32 [ %mydata1, %entry ], [ %mul, %while.body ] 124 %add = add i32 %mydata2.addr.0.lcssa, %mydata1.addr.0.lcssa 125 %add2 = add i32 %add, %c.0.lcssa 126 ret i32 %add2 127 } 128
