Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-17491

micro optimize page_id_t

Details

    Description

      Remove page_id_t::m_fold. This was added as an optimization but my benchmark shows it actually slows things down. http://quick-bench.com/p9tQCIFqk2MJgBrK_vp80YD6r8Y

      Updated page_id_t will be consisted of a two uint32_t. It can be transferred in one register in 64 bit platforms. So, pass it by value everywhere.

      Attachments

        Issue Links

          relates to

          Bug - A problem which impairs or prevents the functions of the product. MDEV-23719 Make lock_sys use page_id_t

          • Major - Major loss of function.
          • Closed

          Activity

            Descending order - Click to sort in ascending order
            marko Marko Mäkelä added a comment -

            In 10.5, page_id_t was optimized further to be a wrapper of a single 64-bit integer, so that we can simplify comparisons.

            marko Marko Mäkelä added a comment - In 10.5, page_id_t was optimized further to be a wrapper of a single 64-bit integer , so that we can simplify comparisons.
            nikitamalyavin Nikita Malyavin added a comment - - edited

            kevg
            >aarch64: I don't know assembly. It looks like a struct was passed on a stack https://godbolt.org/z/q1ucnG

            Because it was 32-bit arm

            https://godbolt.org/z/GT9999Y5q

            nikitamalyavin Nikita Malyavin added a comment - - edited kevg >aarch64: I don't know assembly. It looks like a struct was passed on a stack https://godbolt.org/z/q1ucnG Because it was 32-bit arm https://godbolt.org/z/GT9999Y5q
            marko Marko Mäkelä added a comment -

            Thank you! I merged this along with some further cleanup.

            marko Marko Mäkelä added a comment - Thank you! I merged this along with some further cleanup.
            kevg Eugene Kosov (Inactive) added a comment -

            Cleanup done, PR updated.

            kevg Eugene Kosov (Inactive) added a comment - Cleanup done, PR updated.
            kevg Eugene Kosov (Inactive) added a comment -

            amd64: really passed in one `rdi` https://godbolt.org/z/YBAX0e

            i686: passed in a stack. And I don't know why https://godbolt.org/z/AZ3L9H

            aarch64: I don't know assembly. It looks like a struct was passed on a stack https://godbolt.org/z/q1ucnG

            kevg Eugene Kosov (Inactive) added a comment - amd64: really passed in one `rdi` https://godbolt.org/z/YBAX0e i686: passed in a stack. And I don't know why https://godbolt.org/z/AZ3L9H aarch64: I don't know assembly. It looks like a struct was passed on a stack https://godbolt.org/z/q1ucnG
            marko Marko Mäkelä added a comment -

            Looks OK, but I’d request some related additional cleanup.
            Also, I’d appreciate an example of the generated code for passing `page_id_t` by value on i386, amd64, aarch64. (I would like to confirm that it is only one 64-bit register or two 32-bit registers.)

            marko Marko Mäkelä added a comment - Looks OK, but I’d request some related additional cleanup. Also, I’d appreciate an example of the generated code for passing `page_id_t` by value on i386, amd64, aarch64. (I would like to confirm that it is only one 64-bit register or two 32-bit registers.)

            People

              kevg Eugene Kosov (Inactive)
              kevg Eugene Kosov (Inactive)
              Votes:
              0 Vote for this issue
              Watchers:
              3 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved:

                Git Integration

                  Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.