From ba82a8d84089ccd257f38c857082d12e46178c9f Mon Sep 17 00:00:00 2001 From: Junru Shao Date: Thu, 7 Sep 2023 17:02:49 -0700 Subject: [PATCH] Switch to GPU workers --- ci/jenkins/mlc_jenkinsfile.groovy | 100 +++++++++++++++--------------- 1 file changed, 50 insertions(+), 50 deletions(-) diff --git a/ci/jenkins/mlc_jenkinsfile.groovy b/ci/jenkins/mlc_jenkinsfile.groovy index 2c5b690345..6c767cb72a 100644 --- a/ci/jenkins/mlc_jenkinsfile.groovy +++ b/ci/jenkins/mlc_jenkinsfile.groovy @@ -247,48 +247,48 @@ fi } stage('Unittest') { - parallel 'Relax Core IR': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/relax-core-ir')) { + parallel 'Relax Core IR': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/relax-core-ir')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-ir", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-ir", label: "Relax IR", ) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-op", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-op", label: "Relax Operators", ) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-pass", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-pass", label: "Relax Pass", ) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-nn-module", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-nn-module", label: "Relax nn.Module", ) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-dist", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-dist", label: "Relax DistIR", ) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-pattern-lang", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-pattern-lang", label: "Relax Pattern Lang", ) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-training", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-training", label: "Relax Training", ) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-tvmscript", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-tvmscript", label: "Relax TVMScript", ) }}}, - 'Relax Runtime': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/relax-vm')) { + 'Relax Runtime': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/relax-vm')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-vm", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-vm", label: "Relax VM", ) }}}, @@ -302,60 +302,60 @@ stage('Unittest') { ) }}}, - 'Relax Integration': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/relax-meta-schedule')) { + 'Relax Integration': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/relax-meta-schedule')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 8 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-meta-schedule", + script: "${docker_run} --cpus 8 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-meta-schedule", label: "Relax MetaSchedule", ) sh ( - script: "${docker_run} --cpus 8 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-relay", + script: "${docker_run} --cpus 8 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-relay", label: "Relax Relay", ) }}}, - 'Relax DNNL': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/relax-dnnl')) { + 'Relax DNNL': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/relax-dnnl')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-dnnl", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-dnnl", label: "Relax DNNL", ) }}}, - 'Relax ONNX': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/relax-onnx')) { + 'Relax ONNX': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/relax-onnx')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-onnx", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-onnx", label: "Relax ONNX", ) }}}, - 'Relax StableHLO': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/relax-stablehlo')) { + 'Relax StableHLO': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/relax-stablehlo')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-stablehlo", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-stablehlo", label: "Relax StableHLO", ) }}}, - 'Relax Torch': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/relax-torch')) { + 'Relax Torch': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/relax-torch')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh relax-torch", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh relax-torch", label: "Relax Torch", ) }}}, - 'Arith': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/arith')) { + 'Arith': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/arith')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh arith", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh arith", label: "Arith", ) }}}, @@ -369,11 +369,11 @@ stage('Unittest') { // ) // }}}, - 'Dlight': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/dlight')) { + 'Dlight': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/dlight')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh dlight", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh dlight", label: "Dlight", ) }}}, @@ -405,48 +405,48 @@ stage('Unittest') { // ) // }}}, - 'Codegen': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/codegen')) { + 'Codegen': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/codegen')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh codegen", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh codegen", label: "Codegen", ) }}}, - 'Uncategorized': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/uncategorized')) { + 'Uncategorized': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/uncategorized')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh unittest", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh unittest", label: "Uncategorized", ) }}}, - 'TIR TVMScript': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/tir-tvmscript')) { + 'TIR TVMScript': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/tir-tvmscript')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh tir-tvmscript", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh tir-tvmscript", label: "TIR TVMScript", ) }}}, - 'TIR Pass': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/tir-pass')) { + 'TIR Pass': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/tir-pass')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh tir-pass", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh tir-pass", label: "TIR Pass", ) }}}, - 'TIR Schedule': { node('JUNRU-CPU-SMALL') { ws(per_exec_ws('tvm/unittest/tir-schedule')) { + 'TIR Schedule': { node('JUNRU-GPU') { ws(per_exec_ws('tvm/unittest/tir-schedule')) { init_git() - unpack_lib("cpu", tvm_lib) + unpack_lib("cuda", tvm_lib) sh ( - script: "${docker_run} --cpus 1 ${ci_cpu} ./tests/scripts/unity/task_python_relax.sh tir-schedule", + script: "${docker_run} --cpus 1 ${ci_cuda} ./tests/scripts/unity/task_python_relax.sh tir-schedule", label: "TIR Schedule", ) }}}