• I Was Wrong About Itanium

    From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Sat Nov 7 13:55:16 2026
    From Newsgroup: comp.arch

    When I looked at the Itanium instruction set for the first time, I had
    thought that it had the severe weakness of being built around a single microarchitectural implementation.

    However, I have now learned that one of the later models of the
    Itanium, the series codenamed "Poulson", was 12-issue. So the Itanium architecture was able to survive advances in chip technology.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From anton@anton@mips.complang.tuwien.ac.at (Anton Ertl) to comp.arch on Sat Nov 7 16:21:53 2026
    From Newsgroup: comp.arch

    quadibloc@invalid.com (John Savard) writes:
    When I looked at the Itanium instruction set for the first time, I had >thought that it had the severe weakness of being built around a single >microarchitectural implementation.

    It already had two at the very start: Merced and McKinley.

    However, I have now learned that one of the later models of the
    Itanium, the series codenamed "Poulson", was 12-issue. So the Itanium >architecture was able to survive advances in chip technology.

    Given IA-64's fate, "survive" is an overstatement. A major advance
    which killed IA-64 (and EPIC in general) for branchy code already
    happened in the first half of the 1990s, before the first IA-64
    implementation was originally planned to come out: Dynamic branch
    prediction left static branch prediction in the dust, and therefore
    also compiler-based speculative execution as designed into IA-64.

    The architecture certainly is designed for arbitrary width, but given
    the restrictions encoded in a bundle, implementing a multiple of a
    bundle (three instructions) is probably a good idea.

    As for Poulson, either the design team hit difficulties that delayed
    its release, or Intel realized pretty soon after Poulsons announcement
    in 2005 that IA-64 would flop, and reduced the investment into further
    IA-64 products as far as their contractual obligations allowed in
    order to cut their losses.

    Very few Itanium 9500 (Poulson) systems are on sale on ebay (currently
    I find one configured one for USD 1700, which is more than last time I
    looked), so apparently few have been sold, and most of the few
    customers still need them, or something. Iterestingly, there are
    several Itanium 9700 CPUs on offer, but no complete systems.

    - anton
    --
    'Anyone trying for "industrial quality" ISA should avoid undefined behavior.'
    Mitch Alsup, <c17fcd89-f024-40e7-a594-88a85ac10d20o@googlegroups.com>
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From Chris M. Thomasson@chris.m.thomasson.1@gmail.com to comp.arch on Sat Nov 7 12:37:33 2026
    From Newsgroup: comp.arch

    On 11/7/2026 5:55 AM, John Savard wrote:
    When I looked at the Itanium instruction set for the first time, I had thought that it had the severe weakness of being built around a single microarchitectural implementation.

    However, I have now learned that one of the later models of the
    Itanium, the series codenamed "Poulson", was 12-issue. So the Itanium architecture was able to survive advances in chip technology.[...]

    I never really understood why Itanium had this odd ball instruction,
    Volume 3: Instruction Set Reference...

    LOCK CMP8XCHG16 iirc...

    Why not just CMPXCHG16?
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From Chris M. Thomasson@chris.m.thomasson.1@gmail.com to comp.arch on Sat Nov 7 13:06:35 2026
    From Newsgroup: comp.arch

    On 11/7/2026 12:37 PM, Chris M. Thomasson wrote:
    On 11/7/2026 5:55 AM, John Savard wrote:
    When I looked at the Itanium instruction set for the first time, I had
    thought that it had the severe weakness of being built around a single
    microarchitectural implementation.

    However, I have now learned that one of the later models of the
    Itanium, the series codenamed "Poulson", was 12-issue. So the Itanium
    architecture was able to survive advances in chip technology.[...]

    I never really understood why Itanium had this odd ball instruction,
    Volume 3: Instruction Set Reference...

    LOCK CMP8XCHG16 iirc...

    I am guessing that the CMP8 part works with the version monotonic
    counter. So, the data structure for say a classic lock-free LIFO, needs
    to be:

    struct anchor
    {
    uint64_t m_aba;
    node* m_head;
    };

    this would be useless:

    struct anchor
    {
    node* m_head;
    uint64_t m_aba;
    };

    struct anchor padded to l2 cacheline and aligned of course.


    There is a way to use CMPXCHG for the push and CMPXCHG8/16 for the pop
    over on x86/x64. But it depends on what you load first for the pop. We
    need to load the aba counter first!

    So for Itanium, the aba needs to be first in the structure!


    Why not just CMPXCHG16?

    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From Lawrence =?iso-8859-13?q?D=FFOliveiro?=@ldo@nz.invalid to comp.arch on Sat Nov 7 22:22:15 2026
    From Newsgroup: comp.arch

    On Sat, 07 Nov 2026 16:21:53 GMT, Anton Ertl wrote:

    Very few Itanium 9500 (Poulson) systems are on sale on ebay
    (currently I find one configured one for USD 1700, which is more
    than last time I looked), so apparently few have been sold, and most
    of the few customers still need them, or something.

    Have you seen many Alpha systems offered?

    I just point out that Alpha is still supported in the current Linux
    kernel, while Itanium was dropped a few years ago. This in spite of
    the fact that Alpha went out of production long before Itanium.

    Illustrating the difference between the enthusiasm of their respective
    user communities, I guess ...
    --- Synchronet 3.22a-Linux NewsLink 1.2
  • From quadibloc@quadibloc@invalid.com (John Savard) to comp.arch on Sun Nov 8 18:15:12 2026
    From Newsgroup: comp.arch

    On Sat, 07 Nov 2026 16:21:53 GMT, anton@mips.complang.tuwien.ac.at
    (Anton Ertl) wrote:

    Given IA-64's fate, "survive" is an overstatement.

    I said "survived", not "thrived"; IA-64 still _existed_ at the time of
    Poulson. Some specimens of it were actually sold.

    John Savard
    --- Synchronet 3.22a-Linux NewsLink 1.2