get_calibrated_timestamps should only return one value for maxDeviation

[kanashimia]

It should return VkResult<(Vec<u64>, u64)> instead of VkResult<(Vec<u64>, Vec<u64>)> as per spec

Bug here:

/// <https://www.khronos.org/registry/vulkan/specs/1.3-extensions/man/html/vkGetCalibratedTimestampsEXT.html>
///
/// Returns a tuple containing `(timestamps, max_deviation)`
#[inline]
pub unsafe fn get_calibrated_timestamps(
    &self,
    device: vk::Device,
    info: &[vk::CalibratedTimestampInfoEXT],
) -> VkResult<(Vec<u64>, Vec<u64>)> {
    let mut timestamps = vec![0u64; info.len()];
    let mut max_deviation = vec![0u64; info.len()];
    (self.fp.get_calibrated_timestamps_ext)(
        device,
        info.len() as u32,
        info.as_ptr(),
        timestamps.as_mut_ptr(),
        max_deviation.as_mut_ptr(),
    )
    .result_with_success((timestamps, max_deviation))
}

Spec xml:

<command successcodes="VK_SUCCESS" errorcodes="VK_ERROR_OUT_OF_HOST_MEMORY,VK_ERROR_OUT_OF_DEVICE_MEMORY">
    <proto><type>VkResult</type> <name>vkGetCalibratedTimestampsEXT</name></proto>
    <param><type>VkDevice</type> <name>device</name></param>
    <param><type>uint32_t</type> <name>timestampCount</name></param>
    <param len="timestampCount">const <type>VkCalibratedTimestampInfoEXT</type>* <name>pTimestampInfos</name></param>
    <param len="timestampCount"><type>uint64_t</type>* <name>pTimestamps</name></param>
    <param><type>uint64_t</type>* <name>pMaxDeviation</name></param>
</command>

Spec: pMaxDeviation is a pointer to a 64-bit unsigned integer value in which the strictly positive maximum deviation, in nanoseconds, of the calibrated timestamp values is returned.
https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/vkGetCalibratedTimestampsEXT.html

[Ralith]

Thanks; I confirm your analysis. This is not a soundness issue since we initialize the data regardless, but it is misleading. Want to make a PR with a fix?

[MarijnS95]

Whoops, I agree. I probably misunderstood the last sentence of the quoted docs to say that there was one deviation per timestamp when implementing this, but the vk.xml spec makes it clear this is not the case.