weixin_39856208
2020-12-09 02:43 阅读 0

Clover issues

invertQuda gives me

CALLING QUDA SOLVER ERROR: Parameter clover_coeff undefined (rank 0, host workstation, check_params.h:365 in checkInvertParam()) last kernel called was (name=N4quda10CopyCloverIffLi72ENS_6clover11FloatNOrderIfLi72ELi4EEENS1_11QDPJITOrderIfLi72EEEEE,volume=16x16x16x32,aux=out_stride=65536,in_stride=65536) QMP m0,n1@ error: abort: 1

This comes from an HMC (Chroma) input file that worked before. It's the MdagM clover operator, biCGstab (CG shows the same behavior). I cross-checked pure Wilson and that one works (Bicgstab and CG). Where/when is the quda invert parameter clover_coeff set?

What wonders me is knowing the fact that we checked quda 0.8 against these tests.

该提问来源于开源项目:lattice/quda

  • 点赞
  • 写回答
  • 关注问题
  • 收藏
  • 复制链接分享

5条回答 默认 最新

  • weixin_39642622 weixin_39642622 2020-12-09 02:43

    Hi Frank,

    We came across this in MG work but I thought we fixed it. I think this comes from the fact that we didnt set it at all since we passed the precomputed clover term so we didn't need to. I think at some point Kate tightened up the parameters checking. I can't remember if I set a default value in the Mg branch or whether Kate did it on the MG QUDA branch.

    Best, B

    On May 22, 2016 4:04:36 PM EDT, Frank Winter notifications.com wrote:

    invertQuda gives me

    CALLING QUDA SOLVER ERROR: Parameter clover_coeff undefined (rank 0, host workstation, check_params.h:365 in checkInvertParam()) last kernel called was (name=N4quda10CopyCloverIffLi72ENS_6clover11FloatNOrderIfLi72ELi4EEENS1_11QDPJITOrderIfLi72EEEEE,volume=16x16x16x32,aux=out_stride=65536,in_stride=65536) QMP m0,n1@ error: abort: 1

    This comes from an HMC (Chroma) input file that worked before. It's the MdagM clover operator, biCGstab (CG shows the same behavior). I cross-checked pure Wilson and that one works (Bicgstab and CG). Where/when is the quda invert parameter clover_coeff set?

    What wonders me is knowing the fact that we checked quda 0.8 against these tests.

    You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/lattice/quda/issues/470

    Balint Joó, Scientific Computing Group, Jefferson Lab Email: bjoo.org Tel: +1 757 269 5339 Sent form my mobile phone

    点赞 评论 复制链接分享
  • weixin_39718460 weixin_39718460 2020-12-09 02:43

    This works if you use the quda-multigrid branch of Chroma. We've increased some interface checking robustness so more parameters need to be explicitly set. I guess Balint hasn't merged this branch into master yet.

    点赞 评论 复制链接分享
  • weixin_39856208 weixin_39856208 2020-12-09 02:43

    chroma/feature/quda_multigrid gets me a step further, but

    CG: Convergence at 22 iterations, L2 relative residual: iterated = 6.872258e-06, true = 6.872258e-06 Cuda Space Required Spinor:0.0146484375 GiB Gauge :0 GiB InvClover :0 GiB QUDA_CG_CLOVER_SOLVER: time=1.471436 s Performance=4.80858746421863 GFLOPS Total Time (incl. load gauge)=2.793924 s QUDA_CG_CLOVER_SOLVER: 22 iterations. Rsd = 2922.468 Relative Rsd = 1.00000003903682 QUDA_CG_CLOVER_SOLVER: Total time (with prediction)=2.794021 ERROR: QUDA Solver residuum is outside tolerance: QUDA resid=1.00000003903682 Desired =1e-05 Max Tolerated = 0.0001 SOLVER FAILED: Aborting

    and

    BiCGstab: Convergence at 12 iterations, L2 relative residual: iterated = 9.648832e-07, true = 9.648832e-07 Cuda Space Required Spinor:0.0205078125 GiB Gauge :0 GiB InvClover :0 GiB QUDA_BICGSTAB_CLOVER_SOLVER: time=1.495165 s Performance=2.69356109325727 GFLOPS Total Time (incl. load gauge)=2.240986 s Null Predictor Predict X: Leaving guess unchanged ERROR: Source has zero norm (rank 0, host workstation, interface_quda.cpp:2348 in invertQuda()) last kernel called was (name=N4quda4blas5Norm2Id6float26float4EE,volume=8x16x16x16,aux=vol=32768,stride=32768,precision=4)

    Have we seen this before?

    点赞 评论 复制链接分享
  • weixin_39856208 weixin_39856208 2020-12-09 02:43

    "a step further" here means it obviously passes the parameter check but fails on the first solve.

    点赞 评论 复制链接分享
  • weixin_39856208 weixin_39856208 2020-12-09 02:43

    I got it to work with 0.8 quda release branch and chroma/master. We can leave this issue open for the master branch.

    点赞 评论 复制链接分享

相关推荐