1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -march=amdgcn -mcpu=gfx900 < %s | FileCheck %s 3 4declare i64 @llvm.amdgcn.ballot.i64(i1) 5 6; Test ballot(0) 7 8define amdgpu_cs i64 @constant_false() { 9; CHECK-LABEL: constant_false: 10; CHECK: ; %bb.0: 11; CHECK-NEXT: s_mov_b32 s0, 0 12; CHECK-NEXT: s_mov_b32 s1, 0 13; CHECK-NEXT: ; return to shader part epilog 14 %ballot = call i64 @llvm.amdgcn.ballot.i64(i1 0) 15 ret i64 %ballot 16} 17 18; Test ballot(1) 19 20define amdgpu_cs i64 @constant_true() { 21; CHECK-LABEL: constant_true: 22; CHECK: ; %bb.0: 23; CHECK-NEXT: s_mov_b32 s0, exec_lo 24; CHECK-NEXT: s_mov_b32 s1, exec_hi 25; CHECK-NEXT: ; return to shader part epilog 26 %ballot = call i64 @llvm.amdgcn.ballot.i64(i1 1) 27 ret i64 %ballot 28} 29 30; Test ballot of a non-comparison operation 31 32define amdgpu_cs i64 @non_compare(i32 %x) { 33; CHECK-LABEL: non_compare: 34; CHECK: ; %bb.0: 35; CHECK-NEXT: v_and_b32_e32 v0, 1, v0 36; CHECK-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v0 37; CHECK-NEXT: ; return to shader part epilog 38 %trunc = trunc i32 %x to i1 39 %ballot = call i64 @llvm.amdgcn.ballot.i64(i1 %trunc) 40 ret i64 %ballot 41} 42 43; Test ballot of comparisons 44 45define amdgpu_cs i64 @compare_ints(i32 %x, i32 %y) { 46; CHECK-LABEL: compare_ints: 47; CHECK: ; %bb.0: 48; CHECK-NEXT: v_cmp_eq_u32_e64 s[0:1], v0, v1 49; CHECK-NEXT: ; return to shader part epilog 50 %cmp = icmp eq i32 %x, %y 51 %ballot = call i64 @llvm.amdgcn.ballot.i64(i1 %cmp) 52 ret i64 %ballot 53} 54 55define amdgpu_cs i64 @compare_int_with_constant(i32 %x) { 56; CHECK-LABEL: compare_int_with_constant: 57; CHECK: ; %bb.0: 58; CHECK-NEXT: s_movk_i32 s0, 0x62 59; CHECK-NEXT: v_cmp_lt_i32_e64 s[0:1], s0, v0 60; CHECK-NEXT: ; return to shader part epilog 61 %cmp = icmp sge i32 %x, 99 62 %ballot = call i64 @llvm.amdgcn.ballot.i64(i1 %cmp) 63 ret i64 %ballot 64} 65 66define amdgpu_cs i64 @compare_floats(float %x, float %y) { 67; CHECK-LABEL: compare_floats: 68; CHECK: ; %bb.0: 69; CHECK-NEXT: v_cmp_gt_f32_e64 s[0:1], v0, v1 70; CHECK-NEXT: ; return to shader part epilog 71 %cmp = fcmp ogt float %x, %y 72 %ballot = call i64 @llvm.amdgcn.ballot.i64(i1 %cmp) 73 ret i64 %ballot 74} 75