Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
88 changes: 0 additions & 88 deletions configuration.yaml
Original file line number Diff line number Diff line change
@@ -1,93 +1,5 @@
---
variables:
xrt_variables:
XRT_MESH_SERVICE_ADDRESS:
description:
- The mesh service address used to create the XRT mesh client.
type: string
default_value: ""
XRT_GRPC_COMPRESSION:
description:
- Configures compression for rpc options on the XRT client.
type: string
default_value: ""
XRT_TPU_CONFIG:
description:
- Addresses for the TPU services to be used by the XRT client. ";"
separated list of addresses, e.g. localservice;0;localhost:51011.
type: string
XRT_WORKERS:
description:
- Addresses for the XRT workers to be used by the XRT client, for
example localservice:0;grpc://localhost:51011
type: string
XRT_DEVICE_MAP:
description:
- Maps devices to metadata about the job, replica, task that the device
is responsible for. e.g.
CPU:0;/job:localservice/replica:0/task:0/device:XLA_CPU:0
type: string
XRT_GRPC_MULTISTREAM:
description:
- Used to disable session conneciton sharing for XRT.
type: bool
default_value: true
XRT_START_LOCAL_SERVER:
description:
- Whether or not XRT should start the local service. If true, and if
the devices are CPU or GPU, XRT will try to start the local server.
type: bool
default_value: false
XRT_SHARD_LOCAL_ORDINAL:
description:
- Ordinal to be appended to the paths used by this thread of the xla
client.
type: int
XRT_GRPC_COMPRESSION_LEVEL:
description:
- Configures compression level for rpc options on the XRT client.
type: int
default_value: 3
XRT_MESH_MAX_MSGSIZE:
description:
- Max size of the mesh used by the XRT mesh service, product of
dimensions e.g. 1024 * 1024 * 1024
type: int
default_value: 1073741824 # (1024^3)
XRT_MESH_CONNECT_WAIT:
description:
- Number of seconds to wait for a connection to the XRT mesh service,
particularly the client mesh master
type: int
default_value: 300
XRT_LOCAL_WORKER:
description:
- Local service address for XRT local worker, e.g. localhost:8000.
type: string
default_value: ""
XRT_SHARD_WORLD_SIZE:
description:
- Total number of XRT shards to consider in this client instance. Does
not have a default because there's special behavior when the flag is
not set.
type: int
XRT_MULTI_PROCESSING_DEVICE:
description:
- Service address of the XRT device to be used as a multi processing
device.
type: string
default_value: ""
XRT_HOST_ORDINAL:
description:
- Sets the host ordinal for the XRT computation client. Used to
identify the rank of current device. Does not have a default because
there's special behavior when the flag is not set.
type: int
XRT_SHARD_ORDINAL:
description:
- Sets the shard ordinal for the XRT computation client.
type: int
default_value: -1
pjrt_variables:
PJRT_DEVICE:
description:
Expand Down
6 changes: 0 additions & 6 deletions setup.py
Original file line number Diff line number Diff line change
Expand Up @@ -40,9 +40,6 @@
# TPUVM_MODE=0
# whether to build for TPU
#
# DISABLE_XRT=0
# whether to exclude XRT from the build
#
# SILO_NAME=""
# name of the remote build cache silo
#
Expand Down Expand Up @@ -253,9 +250,6 @@ def bazel_build(self, ext):
if _check_env_flag('TPUVM_MODE'):
bazel_argv.append('--config=tpu')

if _check_env_flag('DISABLE_XRT'):
bazel_argv.append('--config=disable_xrt')

# Remote cache authentication.
if _check_env_flag('BAZEL_REMOTE_CACHE'):
bazel_argv.append('--config=remote_cache')
Expand Down
56 changes: 0 additions & 56 deletions test/allreduce_torchrun.py

This file was deleted.

10 changes: 0 additions & 10 deletions test/cpp/BUILD
Original file line number Diff line number Diff line change
Expand Up @@ -70,16 +70,6 @@ ptxla_cc_test(
],
)

ptxla_cc_test(
name = "test_op_by_op_executor",
srcs = ["test_op_by_op_executor.cpp"],
deps = [
":cpp_test_util",
"//torch_xla/csrc:tensor",
"@com_google_googletest//:gtest_main",
],
)

ptxla_cc_test(
name = "test_replication",
srcs = ["test_replication.cpp"],
Expand Down
85 changes: 0 additions & 85 deletions test/cpp/test_op_by_op_executor.cpp

This file was deleted.

12 changes: 4 additions & 8 deletions test/pjrt/test_runtime.py
Original file line number Diff line number Diff line change
Expand Up @@ -55,19 +55,15 @@ def test_xla_device_error(self):
}, False), ('pjrt_cpu', {
'PJRT_DEVICE': 'CPU',
'PJRT_SELECT_DEFAULT_DEVICE': '0'
}, True), ('xrt_tpu', {
'XRT_TPU_CONFIG': 'localservice;0;localhost:51011'
}, False), ('pjrt_tpu_precedence', {
}, True), ('pjrt_tpu_precedence', {
'PJRT_DEVICE': 'TPU',
'XRT_TPU_CONFIG': 'localservice;0;localhost:51011',
}, True), ('xrt_gpu', {
}, True), ('gpu_num_devives', {
'GPU_NUM_DEVICES': '4'
}, False), ('pjrt_gpu', {
}, True), ('pjrt_gpu', {
'PJRT_DEVICE': 'GPU',
'GPU_NUM_DEVICES': '4'
}, True), ('xla_dist_worker', {
'XRT_LOCAL_WORKER': 'c_localservice:2'
}, False))
}, True))
def test_pjrt_default_device(self, env_vars, expect_using_pjrt):
with mock.patch.dict(os.environ, env_vars, clear=True):
# Print a warningif we had to select a default runtime
Expand Down
40 changes: 0 additions & 40 deletions test/run_tests.sh
Original file line number Diff line number Diff line change
Expand Up @@ -108,42 +108,6 @@ function run_xla_backend_mp {
MASTER_ADDR=localhost MASTER_PORT=6000 run_test "$@"
}

function run_xrt {
if [ -x "$(command -v nvidia-smi)" ] && [ "$XLA_CUDA" != "0" ]; then
GPU_NUM_DEVICES=2 run_coverage "$@"
else
XRT_DEVICE_MAP="CPU:0;/job:localservice/replica:0/task:0/device:XLA_CPU:0" XRT_WORKERS="localservice:0;grpc://localhost:$(shuf -i 40701-40999 -n 1)" run_coverage "$@"
fi
}

function run_opbyop {
echo "Running in OpByOp mode: $@"
XLA_GET_TENSORS_OPBYOP=1 XLA_SYNC_TENSORS_OPBYOP=1 run_xrt "$@"
}

function run_async_scalar {
echo "Running in Async Scalar Upload mode: $@"
XLA_TRANSFER_SCALAR_ASYNC=1 run_xrt "$@"
}

function run_torchrun {
echo "Running tests spawned by torchrun"
if [ -x "$(command -v nvidia-smi)" ]; then
run_xrt "$@"
else
echo "the tests need atleast two XLA workers to validate"
fi
}

function run_xrt_tests {
# For features not supported in PJRT
echo "Running XRT tests"
run_xrt "$CDIR/test_operations.py" "$@" --verbosity=$VERBOSITY
run_opbyop "$CDIR/test_operations.py" "$@" --verbosity=$VERBOSITY
run_async_scalar "$CDIR/test_operations.py" "$@" --verbosity=$VERBOSITY
run_torchrun "$CDIR/test_allreduce_torchrun.py"
}

function run_torch_op_tests {
run_dynamic "$CDIR/../../test/test_view_ops.py" "$@" -v TestViewOpsXLA
run_test_without_functionalization "$CDIR/../../test/test_view_ops.py" "$@" -v TestViewOpsXLA
Expand Down Expand Up @@ -171,7 +135,6 @@ function run_xla_op_tests {
run_test_without_functionalization "$CDIR/test_operations.py" "$@" --verbosity=$VERBOSITY
run_test "$CDIR/test_async_closures.py"
run_test "$CDIR/test_autocast.py"
run_test "$CDIR/test_xla_dist.py"
run_test "$CDIR/test_profiler.py"
run_test "$CDIR/test_ops.py"
run_test "$CDIR/test_metrics.py"
Expand Down Expand Up @@ -235,9 +198,6 @@ function run_tests {
if [[ "$XLA_SKIP_MP_OP_TESTS" != "1" ]]; then
run_mp_op_tests
fi
if [[ "$XLA_SKIP_XRT_TESTS" != "1" ]]; then
run_xrt_tests
fi
}

if [ "$LOGFILE" != "" ]; then
Expand Down
26 changes: 0 additions & 26 deletions test/test_allreduce_torchrun.py

This file was deleted.

Loading