Improve CI DX#4352
Merged
Merged
Conversation
github-actions
Bot
added
component: tests
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
narendasan
force-pushed
the
narendasan/push-xunsslrxsyuv
branch
from
bab6b00 to
c98b447
Compare
github-actions
Bot
added
component: conversion
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-25 22:25:33.200414+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-25 22:25:55.991197+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-25 22:25:33.222901+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-25 22:26:00.920103+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-25 22:31:05.627569+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-25 22:31:27.810024+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-25 22:31:05.647957+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-25 22:31:32.328158+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:18:12.420127+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:18:38.112307+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:18:12.440866+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:18:42.935385+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:35:10.973121+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:35:34.656487+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:35:10.994104+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:35:39.628278+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:43:56.293360+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:44:21.277666+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:43:56.306533+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:44:26.387007+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:56:03.821999+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-26 18:56:30.939609+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:56:03.842721+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-26 18:56:35.876986+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
narendasan
force-pushed
the
narendasan/push-xunsslrxsyuv
branch
from
c775700 to
6754fac
Compare
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-29 20:27:03.234833+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-29 20:27:28.200091+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/utils/junit_summary.py 2026-06-29 20:27:03.258834+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/utils/junit_summary.py 2026-06-29 20:27:32.916111+00:00
@@ -43,12 +43,20 @@
return sys.stdout.isatty()
_USE_COLOR = _color_enabled()
_CODES = {
- "reset": "0", "bold": "1", "dim": "2", "red": "31", "green": "32",
- "yellow": "33", "blue": "34", "magenta": "35", "cyan": "36", "grey": "90",
+ "reset": "0",
+ "bold": "1",
+ "dim": "2",
+ "red": "31",
+ "green": "32",
+ "yellow": "33",
+ "blue": "34",
+ "magenta": "35",
+ "cyan": "36",
+ "grey": "90",
}
def c(text: str, *styles: str) -> str:
if not _USE_COLOR or not styles:
@@ -166,12 +174,19 @@
try:
root = ET.parse(xml).getroot()
except ET.ParseError as e:
suite.errors += 1
report.failures.append(
- Failure("PARSE", os.path.basename(xml), f"could not parse XML: {e}",
- "", "", suite.label, xml)
+ Failure(
+ "PARSE",
+ os.path.basename(xml),
+ f"could not parse XML: {e}",
+ "",
+ "",
+ suite.label,
+ xml,
+ )
)
report.suites.append(suite)
continue
for ts in root.iter("testsuite"):
@@ -181,18 +196,34 @@
test = f"{classname}::{name}" if classname else name
file_attr = tc.get("file", "")
fa, er, sk = tc.find("failure"), tc.find("error"), tc.find("skipped")
if fa is not None:
suite.failed += 1
- report.failures.append(Failure(
- "FAIL", test, _first_line(fa.get("message") or fa.text, "failed"),
- (fa.text or "").strip(), file_attr, suite.label, xml))
+ report.failures.append(
+ Failure(
+ "FAIL",
+ test,
+ _first_line(fa.get("message") or fa.text, "failed"),
+ (fa.text or "").strip(),
+ file_attr,
+ suite.label,
+ xml,
+ )
+ )
elif er is not None:
suite.errors += 1
- report.failures.append(Failure(
- "ERROR", test, _first_line(er.get("message") or er.text, "errored"),
- (er.text or "").strip(), file_attr, suite.label, xml))
+ report.failures.append(
+ Failure(
+ "ERROR",
+ test,
+ _first_line(er.get("message") or er.text, "errored"),
+ (er.text or "").strip(),
+ file_attr,
+ suite.label,
+ xml,
+ )
+ )
elif sk is not None:
suite.skipped += 1
else:
suite.passed += 1
report.suites.append(suite)
@@ -228,29 +259,40 @@
print()
return 1
print()
print(rule)
- label_w = min(max(max((_visible_len(s.label) for s in report.suites), default=20), 12), width - 40)
+ label_w = min(
+ max(max((_visible_len(s.label) for s in report.suites), default=20), 12),
+ width - 40,
+ )
for s in sorted(report.suites, key=lambda s: (s.bad == 0, s.label)):
if s.bad:
icon, lstyle = c("✗", "red", "bold"), "red"
elif s.passed == 0 and s.skipped:
icon, lstyle = c("⊘", "yellow"), "yellow"
else:
icon, lstyle = c("✓", "green"), "green"
- cells = " ".join([
- _count(s.passed, "pass", "green"), _count(s.failed, "fail", "red"),
- _count(s.errors, "err", "magenta"), _count(s.skipped, "skip", "yellow"),
- ])
+ cells = " ".join(
+ [
+ _count(s.passed, "pass", "green"),
+ _count(s.failed, "fail", "red"),
+ _count(s.errors, "err", "magenta"),
+ _count(s.skipped, "skip", "yellow"),
+ ]
+ )
print(f" {icon} {_pad(c(s.label, lstyle), label_w)} {cells}")
print(rule)
- totals = " ".join([
- _count(report.passed, "pass", "green"), _count(report.failed, "fail", "red"),
- _count(report.errors, "err", "magenta"), _count(report.skipped, "skip", "yellow"),
- ])
+ totals = " ".join(
+ [
+ _count(report.passed, "pass", "green"),
+ _count(report.failed, "fail", "red"),
+ _count(report.errors, "err", "magenta"),
+ _count(report.skipped, "skip", "yellow"),
+ ]
+ )
meta = c(f"({report.total} tests · {len(report.suites)} suites)", "grey")
print(f" {_pad(c('TOTAL', 'bold'), label_w + 3)}{totals} {meta}")
if report.failures:
print()
@@ -260,24 +302,32 @@
for i, f in enumerate(report.failures, 1):
tag = c(f"{f.kind:<5}", "magenta" if f.kind == "ERROR" else "red", "bold")
print(f" {c(f'{i:>{num_w}}.', 'grey')} {tag} {c(f.suite, 'grey')}")
print(f" {' ' * num_w} {c('•', 'grey')} {c(f.test, 'bold')}")
if f.message:
- print(f" {' ' * num_w} {c(_truncate(f.message, width - num_w - 6), 'yellow')}")
+ print(
+ f" {' ' * num_w} {c(_truncate(f.message, width - num_w - 6), 'yellow')}"
+ )
print(f" {' ' * num_w} {c('↻ ' + _repro(f), 'grey')}")
print()
# Footer: point at the raw results + the agent report.
print(c(" JUnit XMLs:", "grey"), c(report.results_dir, "grey"))
- print(c(" For a paste-to-Claude report:", "grey"),
- c("just test-summary --agent", "cyan"))
+ print(
+ c(" For a paste-to-Claude report:", "grey"),
+ c("just test-summary --agent", "cyan"),
+ )
if report.bad:
- print(c(f" ✗ {report.failed} failed · {report.errors} errored ", "bold", "red")
- + c(f" ({report.total} tests · {len(report.suites)} suites)", "grey"))
+ print(
+ c(f" ✗ {report.failed} failed · {report.errors} errored ", "bold", "red")
+ + c(f" ({report.total} tests · {len(report.suites)} suites)", "grey")
+ )
else:
- print(c(f" ✓ all {report.passed} passed ", "bold", "green")
- + c(f" ({report.total} tests · {len(report.suites)} suites)", "grey"))
+ print(
+ c(f" ✓ all {report.passed} passed ", "bold", "green")
+ + c(f" ({report.total} tests · {len(report.suites)} suites)", "grey")
+ )
print()
return 1 if report.bad else 0
# ── agent (Markdown) rendering ────────────────────────────────────────────────
@@ -287,30 +337,36 @@
out = print # plain, no color, capturable
out("# Torch-TensorRT test report")
out("")
out(f"- results dir: `{report.results_dir}`")
- out(f"- totals: {report.passed} passed, {report.failed} failed, "
+ out(
+ f"- totals: {report.passed} passed, {report.failed} failed, "
f"{report.errors} errored, {report.skipped} skipped "
- f"({report.total} tests across {len(report.suites)} suites)")
+ f"({report.total} tests across {len(report.suites)} suites)"
+ )
if not report.suites:
out("")
out("No JUnit XML files found — the run did not write results.")
return 1
out("- to read a full traceback, open the `junit` path listed for the failure")
- out("- repro commands use `-n0` (serial); drop `-k` to run the whole file, "
- "lower `-n`/free GPU memory if a test OOMs")
+ out(
+ "- repro commands use `-n0` (serial); drop `-k` to run the whole file, "
+ "lower `-n`/free GPU memory if a test OOMs"
+ )
out("")
# Suite table (compact, Markdown).
out("## Suites")
out("")
out("| suite | passed | failed | errors | skipped |")
out("|---|--:|--:|--:|--:|")
for s in sorted(report.suites, key=lambda s: (s.bad == 0, s.label)):
mark = "" if s.bad == 0 else " ⚠"
- out(f"| `{s.label}`{mark} | {s.passed} | {s.failed} | {s.errors} | {s.skipped} |")
+ out(
+ f"| `{s.label}`{mark} | {s.passed} | {s.failed} | {s.errors} | {s.skipped} |"
+ )
out("")
if not report.failures:
out("## Result: all green ✅")
return 0
@@ -332,29 +388,37 @@
out("- detail:")
out("```")
for ln in clipped:
out(ln)
if len(lines) > _AGENT_DETAIL_MAX_LINES:
- out(f"... [{len(lines) - _AGENT_DETAIL_MAX_LINES} more lines; "
- f"read the full traceback in the junit XML above]")
+ out(
+ f"... [{len(lines) - _AGENT_DETAIL_MAX_LINES} more lines; "
+ f"read the full traceback in the junit XML above]"
+ )
out("```")
out("")
return 1
def _results_dir(value) -> str:
if value:
return value
return os.environ.get("RUNNER_TEST_RESULTS_DIR") or os.path.join(
- os.environ.get("TMPDIR", "/tmp"), "trt_test_results")
+ os.environ.get("TMPDIR", "/tmp"), "trt_test_results"
+ )
def main(argv: list) -> int:
p = argparse.ArgumentParser(description="Summarize JUnit XML test results.")
p.add_argument("results_dir", nargs="?", help="dir of JUnit *.xml files")
- p.add_argument("--agent", "--claude", action="store_true", dest="agent",
- help="emit a plain Markdown report for handing to an AI agent")
+ p.add_argument(
+ "--agent",
+ "--claude",
+ action="store_true",
+ dest="agent",
+ help="emit a plain Markdown report for handing to an AI agent",
+ )
args = p.parse_args(argv[1:])
report = parse(_results_dir(args.results_dir))
return render_agent(report) if args.agent else render_pretty(report)
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-29 20:27:03.257256+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-29 20:27:32.991406+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
narendasan
force-pushed
the
narendasan/push-xunsslrxsyuv
branch
from
6754fac to
ac54fef
Compare
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp b/tmp/changes.txt
index 83bf2b7..83b3fe7 100644
--- a/home/runner/work/TensorRT/TensorRT/core/conversion/converters/impl/conv_deconv.cpp
+++ b/tmp/changes.txt
@@ -131,8 +131,8 @@ bool add_conv_deconv(ConversionCtx* ctx, const torch::jit::Node* n, args& args)
// the constant-weights path below for why TRT requires this.
if (in->getType() != kernel->getType()) {
LOG_DEBUG(
- "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type ("
- << kernel->getType() << "); casting input to match.");
+ "Conv/deconv input type (" << in->getType() << ") differs from kernel tensor type (" << kernel->getType()
+ << "); casting input to match.");
in = castITensor(ctx, in, kernel->getType());
}
auto kernel_dims = kernel->getDimensions();
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
--- /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-29 20:55:34.966498+00:00
+++ /home/runner/work/TensorRT/TensorRT/py/torch_tensorrt/runtime/_runtime_config.py 2026-06-29 20:55:55.983257+00:00
@@ -296,13 +296,11 @@
# already materialized the underlying IRuntimeCache on ``_live``.
wrapped = RuntimeCache(path=rc, autosave_on_del=True)
try:
wrapped.load()
except Exception as e:
- logger.warning(
- f"Failed to warm-load runtime cache from {rc!r}: {e}"
- )
+ logger.warning(f"Failed to warm-load runtime cache from {rc!r}: {e}")
cache = wrapped.ensure_cache(self._live)
self._live.set_runtime_cache(cache)
else:
raise TypeError(
f"runtime_cache must be None, str, or RuntimeCache by the "
--- /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-29 20:55:34.988590+00:00
+++ /home/runner/work/TensorRT/TensorRT/tests/py/ts/api/test_classes.py 2026-06-29 20:56:00.654964+00:00
@@ -380,12 +380,10 @@
io_key, "Neither 'I/O Tensors' nor 'Bindings' key is present"
)
self.assertTrue(
len(trt_json["Layers"]) == num_layers
), "Not enough layers found"
- self.assertTrue(
- len(trt_json[io_key]) == 2, "Not enough I/O tensors found"
- )
+ self.assertTrue(len(trt_json[io_key]) == 2, "Not enough I/O tensors found")
if __name__ == "__main__":
unittest.main()
narendasan
force-pushed
the
narendasan/push-xunsslrxsyuv
branch
from
7de8fcc to
acbf0f1
Compare
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
PRs previously ran the full 8-variant matrix (py3.10–3.13 × cu130/cu132)
through all L0/L1/L2 tiers. Two changes cut feedback time significantly:
1. Implement --limit-pr-builds in filter-matrix.py: PR builds now use a
single representative config (Python 3.12 × cu130) instead of 8. The
full matrix still runs on main / nightly / release. The flag was already
wired in filter-matrix.py but the logic was never implemented.
2. L2 (slow model-compilation tests) is now opt-in for PRs. Add the
`ci: run-l2` label to enable it; `Force All Tests[L0+L1+L2]` continues
to work as before. `ci: skip-l2` becomes a no-op on PRs (kept for
backward compat, still suppresses L2 on branch-push runs).
Also adds `permissions: {}` to the ci-rollup jobs in both workflows to
satisfy CodeQL's GITHUB_TOKEN scope requirement.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Post /rerun on any PR to re-run failed jobs, or /rerun all to re-run every job from scratch. Requires write access to the repo. Uses the GitHub Actions re-run API so the new run stays linked to the PR's head SHA and updates the existing check statuses. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The standard and RTX Linux x86_64 pipelines were ~80% duplicated — every
test command, if-condition, and the rollup script existed in two files, so
changes (like the recent ci: run-l2 / --limit-pr-builds work) had to be
applied twice and drifted over time.
Consolidate into one reusable workflow, _linux-x86_64-core.yml, parameterized
by use-rtx. The two entry workflows shrink to a ~10-line `core:` caller plus a
ci-rollup renderer:
build-test-linux-x86_64.yml → core(use-rtx: false)
build-test-linux-x86_64_rtx.yml → core(use-rtx: true)
RTX differences are expressed two ways, both verified to reproduce the old
behavior byte-for-byte:
* Standard-only jobs (L0/L1/L2 torchscript, dynamo-distributed, executorch
static build) are gated `if: ${{ !inputs.use-rtx && ... }}`.
* Jobs whose pytest scope differs (L0 converter --dist flag; L0/L1 core
partitioning + hlo subsets; L2 core executorch; L2 plugin kernels) branch
inside the script on the $USE_TRT_RTX env var that linux-test.yml exports.
The per-job pass/fail map is bubbled out of the core via a `results` output so
each entry workflow renders ONE rollup status check whose name is unchanged
(`CI / Linux x86_64`, `CI / Linux x86_64 (RTX)`) — branch protection keeps
working. The rollup also fails defensively if the core call concludes
failure/cancelled with an empty results payload, so a reusable-workflow output
edge case can't mask a real failure.
Net: 1263 → 998 lines, single source of truth for all build/test logic.
NOTE: branch-protection required checks are unchanged (CI / Linux x86_64 and
CI / Linux x86_64 (RTX) are still top-level jobs). The individual per-tier
job check names now nest under the `core` caller (e.g. `core / L0 dynamo
converter tests`); if any of those were individually required in branch
protection they'd need updating, but the rollup is the intended required check.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
nox's per-suite test paths had drifted from what CI actually runs, and the
"reproduce locally with: uv run pytest ..." hint failed because plain uv run
tries to rebuild torch-tensorrt from source.
Add uv-native just recipes (just already drives docs here) that mirror the
pytest selectors in _linux-x86_64-core.yml so contributors can run exactly
what a CI tier runs before pushing:
just lint / lint-changed pre-commit (all files / changed-vs-main)
just test <args> pytest in the uv env
just l0 / l1 full tier reproduction
just l0-converter ... individual sub-suites
Key local-DX fixes baked into the recipes:
* uv run --no-sync — use the already-built .venv instead of rebuilding.
* per-user $TMPDIR (/tmp/torch_tensorrt_$USER) — the engine/timing cache
defaults to a shared /tmp path that PermissionErrors for everyone after
the first user creates it; a per-user dir sidesteps that.
* {{jobs}} knob — single-GPU machines can dial back -n auto to avoid OOM
from too many concurrent TRT engine builds.
Also register the load-bearing `critical`/`unit` pytest markers (CI tiers
split on -m critical / -m "not critical" but they were never declared), fix
the ci_helpers.sh repro hint to use --no-sync, and replace the stale nox
section in CONTRIBUTING (it still listed Python 3.7-3.10) with the just guide.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously the test selectors existed in two places — inline pytest commands
in _linux-x86_64-core.yml and hand-mirrored recipes in the justfile — with only
a "keep these in sync" comment guarding against drift (exactly how nox rotted).
Make tests/py/ci_helpers.sh the single source of truth: one trt_tier_* function
per CI tier, baking in the paths/markers/RTX branching. Both consumers now call
the same functions:
* CI: each linux-test.yml script block is now `source ci_helpers.sh;
trt_tier_<name>` (the 14 inline blocks collapse to 3 lines each).
* Local: the just recipes call `trt_tier_<name>` too.
Only environment policy differs, injected via env vars the library reads:
PYTHON launcher — CI uses container python; just sets
"uv run --no-sync python" (run against the built .venv).
TRT_JOBS xdist workers for the parallel suites (CI 8; just passes its
jobs= knob).
RUNNER_TEST_RESULTS_DIR junit dir (CI-set; locally defaults under $TMPDIR).
TRT_PYTEST_RERUNS flaky-test reruns — default on (CI), the just recipes set
it to 0 (the pytest-rerunfailures plugin may be absent locally
and you want to see flakes, not retry them).
Verified equivalent: a harness resolves every trt_tier_* function for
USE_TRT_RTX in {false,true}, preserving arg boundaries, and diffs the argv
against the original inline commands from git — all 28 tier/variant
combinations match byte-for-byte. Also smoke-ran a tier end-to-end via
`just l0-converter` (engine builds, junit written under the per-user dir).
just now uses bash as its recipe shell (the library is bash). CONTRIBUTING
updated to note local and CI can no longer drift.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
narendasan
force-pushed
the
narendasan/push-xunsslrxsyuv
branch
from
acbf0f1 to
9982af0
Compare
Collaborator
Author
|
/rerun all |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change.
Fixes # (issue)
Type of change
Please delete options that are not relevant and/or add your own.
Checklist: