[llvm] r270885 - Do not rename registers that do not start an independent live range

[Evgenii Stepanov via llvm-commits]

Hi,

this test is failing on ASan:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap/builds/11794/steps/check-llvm%20asan/logs/stdio

On Thu, May 26, 2016 at 11:22 AM, Krzysztof Parzyszek via llvm-commits
<llvm-commits at lists.llvm.org> wrote:
> Author: kparzysz
> Date: Thu May 26 13:22:53 2016
> New Revision: 270885
>
> URL: http://llvm.org/viewvc/llvm-project?rev=270885&view=rev
> Log:
> Do not rename registers that do not start an independent live range
>
> Added:
>     llvm/trunk/test/CodeGen/MIR/Hexagon/
>     llvm/trunk/test/CodeGen/MIR/Hexagon/anti-dep-partial.mir
> Modified:
>     llvm/trunk/lib/CodeGen/AggressiveAntiDepBreaker.cpp
>
> Modified: llvm/trunk/lib/CodeGen/AggressiveAntiDepBreaker.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AggressiveAntiDepBreaker.cpp?rev=270885&r1=270884&r2=270885&view=diff
> ==============================================================================
> --- llvm/trunk/lib/CodeGen/AggressiveAntiDepBreaker.cpp (original)
> +++ llvm/trunk/lib/CodeGen/AggressiveAntiDepBreaker.cpp Thu May 26 13:22:53 2016
> @@ -787,6 +787,8 @@ unsigned AggressiveAntiDepBreaker::Break
>    DEBUG(dbgs() << '\n');
>  #endif
>
> +  BitVector RegAliases(TRI->getNumRegs());
> +
>    // Attempt to break anti-dependence edges. Walk the instructions
>    // from the bottom up, tracking information about liveness as we go
>    // to help determine which registers are available.
> @@ -898,6 +900,29 @@ unsigned AggressiveAntiDepBreaker::Break
>            }
>
>            if (AntiDepReg == 0) continue;
> +
> +          // If the definition of the anti-dependency register does not start
> +          // a new live range, bail out. This can happen if the anti-dep
> +          // register is a sub-register of another register whose live range
> +          // spans over PathSU. In such case, PathSU defines only a part of
> +          // the larger register.
> +          RegAliases.reset();
> +          for (MCRegAliasIterator AI(AntiDepReg, TRI, true); AI.isValid(); ++AI)
> +            RegAliases.set(*AI);
> +          for (SDep S : PathSU->Succs) {
> +            SDep::Kind K = S.getKind();
> +            if (K != SDep::Data && K != SDep::Output && K != SDep::Anti)
> +              continue;
> +            unsigned R = S.getReg();
> +            if (!RegAliases[R])
> +              continue;
> +            if (R == AntiDepReg || TRI->isSubRegister(AntiDepReg, R))
> +              continue;
> +            AntiDepReg = 0;
> +            break;
> +          }
> +
> +          if (AntiDepReg == 0) continue;
>          }
>
>          assert(AntiDepReg != 0);
>
> Added: llvm/trunk/test/CodeGen/MIR/Hexagon/anti-dep-partial.mir
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MIR/Hexagon/anti-dep-partial.mir?rev=270885&view=auto
> ==============================================================================
> --- llvm/trunk/test/CodeGen/MIR/Hexagon/anti-dep-partial.mir (added)
> +++ llvm/trunk/test/CodeGen/MIR/Hexagon/anti-dep-partial.mir Thu May 26 13:22:53 2016
> @@ -0,0 +1,35 @@
> +# RUN: llc -march=hexagon -post-RA-scheduler -run-pass post-RA-sched %s 2>&1 -o /dev/null | FileCheck %s
> +
> +--- |
> +  declare void @check(i64, i32, i32, i64)
> +  define void @foo() {
> +    ret void
> +  }
> +...
> +
> +---
> +name: foo
> +tracksRegLiveness: true
> +allVRegsAllocated: true
> +body: |
> +  bb.0:
> +    successors:
> +    liveins: %r0, %r1, %d1, %d2, %r16, %r17, %r19, %r22, %r23
> +        %r2 = A2_add %r23, killed %r17
> +        %r6 = M2_mpyi %r16, %r16
> +        %r22 = M2_accii %r22, killed %r2, 2
> +        %r7 = A2_tfrsi 12345678
> +        %r3 = A2_tfr killed %r16
> +        %d2 = A2_tfrp killed %d0
> +        %r2 = L2_loadri_io %r29, 28
> +        %r2 = M2_mpyi killed %r6, killed %r2
> +        %r23 = S2_asr_i_r %r22, 31
> +        S2_storeri_io killed %r29, 0, killed %r7
> +        ; The anti-dependency on r23 between the first A2_add and the
> +        ; S2_asr_i_r was causing d11 to be renamed, while r22 remained
> +        ; unchanged. Check that the renaming of d11 does not happen.
> +        ; CHECK: d11
> +        %d0 = A2_tfrp killed %d11
> +        J2_call @check, implicit-def %d0, implicit-def %d1, implicit-def %d2, implicit %d0, implicit %d1, implicit %d2
> +...
> +
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits