| ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --filter-out-after "ir-bb<loop>" --version 6 | |
| ; RUN: opt -passes=loop-vectorize -force-vector-width=4 -enable-epilogue-vectorization \ | |
| ; RUN: -epilogue-vectorization-force-VF=2 -vplan-print-after="printFinalVPlan$" \ | |
| ; RUN: -disable-output %s 2>&1 | FileCheck %s | |
| ; We CSE the reduction start vector in the main loop. | |
| define i32 @cse_reduction_start_vector_multiple_users(ptr %a, i64 %n) { | |
| ; CHECK-LABEL: VPlan for loop in 'cse_reduction_start_vector_multiple_users' | |
| ; CHECK: VPlan 'Final VPlan for VF={4},UF={1}' { | |
| ; CHECK-NEXT: Live-in ir<%n> = original trip-count | |
| ; CHECK-EMPTY: | |
| ; CHECK-NEXT: ir-bb<entry>: | |
| ; CHECK-NEXT: EMIT vp<%min.iters.check> = icmp ult ir<%n>, ir<2> | |
| ; CHECK-NEXT: EMIT branch-on-cond vp<%min.iters.check> | |
| ; CHECK-NEXT: Successor(s): ir-bb<scalar.ph>, vector.main.loop.iter.check | |
| ; CHECK-EMPTY: | |
| ; CHECK-NEXT: vector.main.loop.iter.check: | |
| ; CHECK-NEXT: EMIT vp<%min.iters.check>.1 = icmp ult ir<%n>, ir<4> | |
| ; CHECK-NEXT: EMIT branch-on-cond vp<%min.iters.check>.1 | |
| ; CHECK-NEXT: Successor(s): ir-bb<scalar.ph>, vector.ph | |
| ; CHECK-EMPTY: | |
| ; CHECK-NEXT: vector.ph: | |
| ; CHECK-NEXT: EMIT vp<[[VP4:%[0-9]+]]> = and ir<%n>, ir<3> | |
| ; CHECK-NEXT: EMIT vp<%n.vec> = sub ir<%n>, vp<[[VP4]]> | |
| ; CHECK-NEXT: EMIT vp<[[VP5:%[0-9]+]]> = reduction-start-vector ir<0>, ir<0>, ir<1> | |
| ; CHECK-NEXT: Successor(s): vector.body | |
| ; CHECK-EMPTY: | |
| ; CHECK-NEXT: vector.body: | |
| ; CHECK-NEXT: EMIT-SCALAR vp<%index> = phi [ ir<0>, vector.ph ], [ vp<%index.next>, vector.body ] | |
| ; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%rdx1> = phi (add) vp<[[VP5]]>, ir<%rdx1.next> | |
| ; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%rdx2> = phi (xor) vp<[[VP5]]>, ir<%rdx2.next> | |
| ; CHECK-NEXT: CLONE ir<%gep.a.iv> = getelementptr inbounds ir<%a>, vp<%index> | |
| ; CHECK-NEXT: WIDEN ir<%ld.a> = load ir<%gep.a.iv> | |
| ; CHECK-NEXT: WIDEN ir<%rdx1.next> = add ir<%rdx1>, ir<%ld.a> | |
| ; CHECK-NEXT: WIDEN ir<%rdx2.next> = xor ir<%rdx2>, ir<%ld.a> | |
| ; CHECK-NEXT: EMIT vp<%index.next> = add nuw vp<%index>, ir<4> | |
| ; CHECK-NEXT: EMIT vp<[[VP6:%[0-9]+]]> = icmp eq vp<%index.next>, vp<%n.vec> | |
| ; CHECK-NEXT: EMIT branch-on-cond vp<[[VP6]]> | |
| ; CHECK-NEXT: Successor(s): middle.block, vector.body | |
| ; CHECK-EMPTY: | |
| ; CHECK-NEXT: middle.block: | |
| ; CHECK-NEXT: EMIT-SCALAR vp<[[VP8:%[0-9]+]]> = resume-for-epilogue vp<%n.vec>, ir<0> | |
| ; CHECK-NEXT: EMIT vp<[[VP9:%[0-9]+]]> = compute-reduction-result (add) ir<%rdx1.next> | |
| ; CHECK-NEXT: EMIT vp<[[VP10:%[0-9]+]]> = compute-reduction-result (xor) ir<%rdx2.next> | |
| ; CHECK-NEXT: EMIT vp<%cmp.n> = icmp eq ir<%n>, vp<%n.vec> | |
| ; CHECK-NEXT: EMIT-SCALAR vp<[[VP11:%[0-9]+]]> = resume-for-epilogue vp<%n.vec>, ir<0> | |
| ; CHECK-NEXT: EMIT-SCALAR vp<[[VP12:%[0-9]+]]> = resume-for-epilogue vp<[[VP9]]>, ir<0> | |
| ; CHECK-NEXT: EMIT-SCALAR vp<[[VP13:%[0-9]+]]> = resume-for-epilogue vp<[[VP10]]>, ir<0> | |
| ; CHECK-NEXT: EMIT branch-on-cond vp<%cmp.n> | |
| ; CHECK-NEXT: Successor(s): ir-bb<exit>, ir-bb<scalar.ph> | |
| ; CHECK-EMPTY: | |
| ; CHECK-NEXT: ir-bb<exit>: | |
| ; CHECK-NEXT: IR %rdx1.next.lcssa = phi i32 [ %rdx1.next, %loop ] (extra operand: vp<[[VP9]]> from middle.block) | |
| ; CHECK-NEXT: IR %rdx2.next.lcssa = phi i32 [ %rdx2.next, %loop ] (extra operand: vp<[[VP10]]> from middle.block) | |
| ; CHECK-NEXT: IR %r = sub i32 %rdx1.next.lcssa, %rdx2.next.lcssa | |
| ; CHECK-NEXT: No successors | |
| ; CHECK-EMPTY: | |
| ; CHECK-NEXT: ir-bb<scalar.ph>: | |
| ; CHECK-NEXT: Successor(s): ir-bb<loop> | |
| ; | |
| entry: | |
| br label %loop | |
| loop: | |
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] | |
| %rdx1 = phi i32 [ 0, %entry ], [ %rdx1.next, %loop ] | |
| %rdx2 = phi i32 [ 0, %entry ], [ %rdx2.next, %loop ] | |
| %gep.a.iv = getelementptr inbounds i32, ptr %a, i64 %iv | |
| %ld.a = load i32, ptr %gep.a.iv, align 4 | |
| %rdx1.next = add i32 %rdx1, %ld.a | |
| %rdx2.next = xor i32 %rdx2, %ld.a | |
| %iv.next = add nuw nsw i64 %iv, 1 | |
| %ec = icmp eq i64 %iv.next, %n | |
| br i1 %ec, label %exit, label %loop | |
| exit: | |
| %r = sub i32 %rdx1.next, %rdx2.next | |
| ret i32 %r | |
| } |