b := s.heapBits()
		clear(b)
	}
}

// heapBits returns the heap ptr/scalar bits stored at the end of the span for
// small object spans and heap arena spans.
//
// Note that the uintptr of each element means something different for small object
// spans and for heap arena spans. Small object spans are easy: they're never interpreted
// as anything but uintptr, so they're immune to differences in endianness. However, the

build:windows --cxxopt=/std:c++17
build:windows --host_cxxopt=/std:c++17

# On windows, we still link everything into a single DLL.
build:windows --config=monolithic

# On linux, we dynamically link small amount of kernels
build:linux --config=dynamic_kernels

# Make sure to include as little of windows.h as possible
build:windows --copt=-DWIN32_LEAN_AND_MEAN
build:windows --host_copt=-DWIN32_LEAN_AND_MEAN

    ],
    hdrs = [
        "utils/size_utils.h",
    ],
    deps = [
        "@llvm-project//mlir:IR",
    ],
)

tf_cc_test(
    name = "size_utils_test",
    size = "small",
    srcs = ["utils/size_utils_test.cc"],
    deps = [
        ":size_utils",
        "//tensorflow/core:test",
        "//tensorflow/core:test_main",
        "@llvm-project//mlir:IR",
    ],
)

   * {@link TreeSet} or the {@link Map#keySet} of an {@code IdentityHashMap}.
   *
   * <p><b>Note:</b> The returned view performs slightly better when {@code set1} is the smaller of
   * the two sets. If you have reason to believe one of your sets will generally be smaller than the
   * other, pass it first. Unfortunately, since this method sets the generic type of the returned

// its rate and RSS goal. For this, it requires the current heapGoal,
// and the heapGoal for the previous GC cycle.
//
// The RSS goal is based on the current heap goal with a small overhead
// to accommodate non-determinism in the allocator.
//
// The pacing is based on scavengePageRate, which applies to both regular and
// huge pages. See that constant for more information.
//

        "//tensorflow/core:test",
        "//tensorflow/core:test_main",
        "//tensorflow/core:testlib",
    ],
)

tf_cc_test(
    name = "deadness_analysis_test",
    size = "small",
    srcs = [
        "deadness_analysis_internal.h",
        "deadness_analysis_test.cc",
    ],
    deps = [
        ":common",
        ":compilation_passes",
        "//tensorflow/cc:cc_ops",

//     larger TLS records better amortize framing and encryption overheads.
//
// A simple heuristic that works well in practice is to use small records for
// the first 1MB of data, then use larger records for subsequent data, and
// reset back to smaller records after the connection becomes idle. See "High
// Performance Web Networking", Chapter 4, or:

			buf = make([]byte, 100)
			for i := range buf {
				buf[i] = 0xff
			}
			checkResult(t, x.FillBytes(buf), x)

			// Too small.
			if byteLen > 0 {
				buf = make([]byte, byteLen-1)
				if !panics(func() { x.FillBytes(buf) }) {
					t.Errorf("expected panic for small buffer and value %x", x)
				}
			}
		})
	}
}

func TestNewIntMinInt64(t *testing.T) {

}

// DeleteObjects deletes objects/versions in bulk, this function will still automatically split objects list
// into smaller bulks if some object names are found to be duplicated in the delete list, splitting
// into smaller bulks will avoid holding twice the write lock of the duplicated object names.

		t.Fatalf("open %q failed: %v", dir, err2)
	}
	defer file1.Close()
	small := smallReaddirnames(file1, len(all)+100, t) // +100 in case we screw up
	if len(small) < len(all) {
		t.Fatalf("len(small) is %d, less than %d", len(small), len(all))
	}
	for i, n := range all {
		if small[i] != n {
			t.Errorf("small read %q mismatch: %v", small[i], n)
		}
	}
}

func TestReaddirNValues(t *testing.T) {