This is an automated email from the ASF dual-hosted git repository.
apeforest pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
The following commit(s) were added to refs/heads/master by this push:
new 3ab0509 Fixcmakegpubuild (#17552)
3ab0509 is described below
commit 3ab0509f9b1dbd642bc281620af11b5b94c3f089
Author: Leonard Lausen <[email protected]>
AuthorDate: Sun Feb 9 21:19:28 2020 +0000
Fixcmakegpubuild (#17552)
* Fix missing MXNET_ENABLE_CUDA_RTC check
* Add tests
* Update config/config.cmake
---
ci/docker/runtime_functions.sh | 24 ++++++++++++++++++++++++
ci/jenkins/Jenkins_steps.groovy | 13 +++++++++++++
ci/jenkins/Jenkinsfile_unix_gpu | 1 +
config/config.cmake | 2 +-
src/imperative/cached_op.h | 10 ++++++++--
5 files changed, 47 insertions(+), 3 deletions(-)
diff --git a/ci/docker/runtime_functions.sh b/ci/docker/runtime_functions.sh
index e0be19e..673ea9a 100755
--- a/ci/docker/runtime_functions.sh
+++ b/ci/docker/runtime_functions.sh
@@ -889,6 +889,30 @@ build_ubuntu_gpu_cmake() {
ninja -v
}
+build_ubuntu_gpu_cmake_no_rtc() {
+ set -ex
+ cd /work/build
+ build_ccache_wrappers
+ cmake \
+ -DUSE_SIGNAL_HANDLER=ON \
+ -DUSE_CUDA=ON \
+ -DUSE_CUDNN=ON \
+ -DUSE_TVM_OP=ON \
+ -DPython3_EXECUTABLE=/usr/bin/python3 \
+ -DUSE_MKL_IF_AVAILABLE=OFF \
+ -DUSE_MKLML_MKL=OFF \
+ -DUSE_MKLDNN=ON \
+ -DUSE_DIST_KVSTORE=ON \
+ -DCMAKE_BUILD_TYPE=Release \
+ -DMXNET_CUDA_ARCH="$CI_CMAKE_CUDA_ARCH" \
+ -DBUILD_CYTHON_MODULES=1 \
+ -DENABLE_CUDA_RTC=OFF \
+ -G Ninja \
+ /work/mxnet
+
+ ninja -v
+}
+
build_ubuntu_gpu_cmake_no_tvm_op() {
set -ex
cd /work/build
diff --git a/ci/jenkins/Jenkins_steps.groovy b/ci/jenkins/Jenkins_steps.groovy
index 7a37826..2ce8860 100644
--- a/ci/jenkins/Jenkins_steps.groovy
+++ b/ci/jenkins/Jenkins_steps.groovy
@@ -311,6 +311,19 @@ def compile_unix_cmake_gpu_no_tvm_op() {
}]
}
+def compile_unix_cmake_gpu_no_rtc() {
+ return ['GPU: CMake CUDA RTC OFF': {
+ node(NODE_LINUX_CPU) {
+ ws('workspace/build-cmake-gpu-no-rtc') {
+ timeout(time: max_time, unit: 'MINUTES') {
+ utils.init_git()
+ utils.docker_run('ubuntu_gpu_cu101',
'build_ubuntu_gpu_cmake_no_rtc', false)
+ }
+ }
+ }
+ }]
+}
+
def compile_unix_tensorrt_gpu() {
return ['TensorRT': {
node(NODE_LINUX_CPU) {
diff --git a/ci/jenkins/Jenkinsfile_unix_gpu b/ci/jenkins/Jenkinsfile_unix_gpu
index f3ed638..f8c28d5 100644
--- a/ci/jenkins/Jenkinsfile_unix_gpu
+++ b/ci/jenkins/Jenkinsfile_unix_gpu
@@ -43,6 +43,7 @@ core_logic: {
custom_steps.compile_unix_int64_gpu(),
custom_steps.compile_unix_full_gpu_no_tvm_op(),
custom_steps.compile_unix_cmake_gpu_no_tvm_op(),
+ custom_steps.compile_unix_cmake_gpu_no_rtc(),
custom_steps.compile_unix_full_gpu_mkldnn_cpp_test()
])
diff --git a/config/config.cmake b/config/config.cmake
index cfd6f26..101e43f 100644
--- a/config/config.cmake
+++ b/config/config.cmake
@@ -125,5 +125,5 @@ set(USE_INT64_TENSOR_SIZE OFF CACHE BOOL "Use int64_t to
represent the total num
# Other GPU features
set(USE_NCCL "Use NVidia NCCL with CUDA" OFF)
set(NCCL_ROOT "" CACHE BOOL "NCCL install path. Supports autodetection.")
-set(ENABLE_CUDA_RTC OFF CACHE BOOL "Build with CUDA runtime compilation
support")
+set(ENABLE_CUDA_RTC ON CACHE BOOL "Build with CUDA runtime compilation
support")
set(USE_NVTX ON CACHE BOOL "Build with NVTX support")
diff --git a/src/imperative/cached_op.h b/src/imperative/cached_op.h
index 239e3e8..c56d8cf 100644
--- a/src/imperative/cached_op.h
+++ b/src/imperative/cached_op.h
@@ -230,7 +230,7 @@ void SetRefCounts(nnvm::Graph* fwd_graph, const
nnvm::Graph& full_graph) {
void OptimizeGraph(nnvm::Graph * full_graph, nnvm::Graph * fwd_graph,
nnvm::Graph * grad_graph,
const Context& context, size_t num_forward_outputs, const
bool inlining) {
-#if MXNET_USE_CUDA && !defined(_WIN32)
+#if MXNET_USE_CUDA && MXNET_ENABLE_CUDA_RTC && !defined(_WIN32)
if (context.dev_mask() == kGPU &&
!inlining &&
dmlc::GetEnv("MXNET_USE_FUSION", true)) {
@@ -265,7 +265,13 @@ void OptimizeGraph(nnvm::Graph * full_graph, nnvm::Graph *
fwd_graph, nnvm::Grap
<< "Graph contains duplicate names for some of its inputs - fusion is
NOT enabled!";
}
}
-#endif // MXNET_USE_CUDA
+#else
+ // Only warn user if MXNET_USE_FUSION env var is explicitly set
+ if (context.dev_mask() == kGPU && !inlining &&
+ dmlc::GetEnv("MXNET_USE_FUSION", false)) {
+ exec::WarnFusionNotSupported();
+ }
+#endif // MXNET_USE_CUDA && MXNET_ENABLE_CUDA_RTC && !defined(_WIN32)
*fwd_graph = nnvm::Graph();
fwd_graph->outputs =
std::vector<nnvm::NodeEntry>(full_graph->outputs.begin(),
