2016-08-08 11:10:22 +08:00
|
|
|
@LIT_SITE_CFG_IN_HEADER@
|
|
|
|
|
|
|
|
|
|
# Tool-specific config options.
|
|
|
|
|
config.name_suffix = "@XRAY_TEST_CONFIG_SUFFIX@"
|
|
|
|
|
config.xray_lit_source_dir = "@XRAY_LIT_SOURCE_DIR@"
|
|
|
|
|
config.target_cflags = "@XRAY_TEST_TARGET_CFLAGS@"
|
|
|
|
|
config.target_arch = "@XRAY_TEST_TARGET_ARCH@"
|
[XRay] [compiler-rt] - Fix standalone and non-deterministic test issue
Summary:
The thread order test fails sometimes my machine independently of standalone
build.
From testing both standalone and in-tree build, I see I configured it wrong.
The other hypothesis for an issue is that cold starts can interfere with whether
record unwriting happens. Once this happens more than once, we can naively
FileCheck on the wrong test output, which compounds the issue.
While "rm blah.* || true" will print to stderr if the glob can't expand, this is
mostly harmless and makes sure earlier failing tests don't sabotage us.
Example failure:
---
header:
version: 1
type: 1
constant-tsc: true
nonstop-tsc: true
cycle-frequency: 3800000000
records:
- { type: 0, func-id: 1, function: 'f1()', cpu: 9, thread: 21377, kind: function-enter, tsc: 2413745203147228 }
- { type: 0, func-id: 1, function: 'f1()', cpu: 9, thread: 21377, kind: function-exit, tsc: 2413745203304238 }
...
The CMAKE related change fixes the expectation that COMPILER_RT_STANDALONE_BUILD will be explicitly FALSE instead
of empty string when it is not "TRUE".
Reviewers: dberris
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D32259
llvm-svn: 300822
2017-04-20 13:59:26 +08:00
|
|
|
config.built_with_llvm = ("@COMPILER_RT_STANDALONE_BUILD@" != "TRUE")
|
2017-04-20 01:20:47 +08:00
|
|
|
|
[XRay] [compiler-rt] - Fix standalone and non-deterministic test issue
Summary:
The thread order test fails sometimes my machine independently of standalone
build.
From testing both standalone and in-tree build, I see I configured it wrong.
The other hypothesis for an issue is that cold starts can interfere with whether
record unwriting happens. Once this happens more than once, we can naively
FileCheck on the wrong test output, which compounds the issue.
While "rm blah.* || true" will print to stderr if the glob can't expand, this is
mostly harmless and makes sure earlier failing tests don't sabotage us.
Example failure:
---
header:
version: 1
type: 1
constant-tsc: true
nonstop-tsc: true
cycle-frequency: 3800000000
records:
- { type: 0, func-id: 1, function: 'f1()', cpu: 9, thread: 21377, kind: function-enter, tsc: 2413745203147228 }
- { type: 0, func-id: 1, function: 'f1()', cpu: 9, thread: 21377, kind: function-exit, tsc: 2413745203304238 }
...
The CMAKE related change fixes the expectation that COMPILER_RT_STANDALONE_BUILD will be explicitly FALSE instead
of empty string when it is not "TRUE".
Reviewers: dberris
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D32259
llvm-svn: 300822
2017-04-20 13:59:26 +08:00
|
|
|
# TODO: Look into whether we can run a capability test on the standalone build to
|
|
|
|
|
# see whether it can run 'llvm-xray convert' instead of turning off tests for a
|
|
|
|
|
# standalone build.
|
2017-04-20 01:20:47 +08:00
|
|
|
if config.built_with_llvm:
|
|
|
|
|
config.available_features.add('built-in-llvm-tree')
|
2016-08-08 11:10:22 +08:00
|
|
|
|
|
|
|
|
# Load common config for all compiler-rt lit tests
|
|
|
|
|
lit_config.load_config(config, "@COMPILER_RT_BINARY_DIR@/test/lit.common.configured")
|
|
|
|
|
|
|
|
|
|
# Load tool-specific config that would do the real work.
|
2017-08-31 08:50:12 +08:00
|
|
|
lit_config.load_config(config, "@CMAKE_CURRENT_SOURCE_DIR@/lit.cfg")
|