' binary GPU code, or as "sm_xy" to only include the binary '
        'code.\nPlease note that each additional compute capability '
        'significantly increases your build time and binary size, and that '
        'TensorFlow only supports compute capabilities >= 3.5 [Default is: '
        '%s]: ' % default_cuda_compute_capabilities)

	return -1
}

// asciiSet is a 32-byte value, where each bit represents the presence of a
// given ASCII character in the set. The 128-bits of the lower 16 bytes,
// starting with the least-significant bit of the lowest word to the
// most-significant bit of the highest word, map to the full range of all
// 128 ASCII characters. The 128-bits of the upper 16 bytes will be zeroed,

  // A note on memory visibility.
  // Many of the utilities in this class (transform, withFallback, withTimeout, asList, combine)
  // have two requirements that significantly complicate their design.
  // 1. Cancellation should propagate from the returned future to the input future(s).
  // 2. The returned futures shouldn't unnecessarily 'pin' their inputs after completion.
  //

 ([#64792](https://github.com/kubernetes/kubernetes/pull/64792), [@luxas](https://github.com/luxas))

identifiers not by parsing C code but by feeding carefully constructed
programs into the system C compiler and interpreting the generated
error messages, debug information, and object files. In practice,
parsing these is significantly less work and more robust than parsing
C source.

Cgo first invokes gcc -E -dM on the preamble, in order to find out
about simple #defines for constants and the like. These are recorded
for later use.

- Number of errors reported to the metric `storage_operation_duration_seconds_count` for emptyDir decreased significantly because previously one error was reported for each projected volume created. ([#117022](https://github.com/kubernetes/kubernetes/pull/117022), [@mpatlasov](https://github.com/mpatlasov)) [SIG Storage]

- Improved exponential backoff in Reflector, significantly reducing the load on Kubernetes apiserver in case of throttling of requests. ([#118132](https://github.com/kubernetes/kubernetes/pull/118132), [@marseel](https://github.com/marseel)) [SIG API Machinery and Scalability]

### Bug or Regression

- An inefficient lock in EndpointSlice controller metrics cache has been reworked. Network programming latency may be significantly reduced in certain scenarios, especially in clusters with a large number of Services. ([#107169](https://github.com/kubernetes/kubernetes/pull/107169), [@robscott](https://github.com/robscott)) [SIG Apps and Network]

    Potential impacts:
    If an SLO/alert has been defined based on `volume_operation_total_seconds`, it might get violated because of the more accurate metric might be significantly larger than previously reported. The metric is defined to be a histogram and the new semantic could change the distribution. ([#78061](https://github.com/kubernetes/kubernetes/pull/78061), [@yuxiangqian](https://github.com/yuxiangqian))

- Number of errors reported to the metric `storage_operation_duration_seconds_count` for emptyDir decreased significantly because previously one error was reported for each projected volume created. ([#117022](https://github.com/kubernetes/kubernetes/pull/117022), [@mpatlasov](https://github.com/mpatlasov)) [SIG Storage]