Require BML_USE_POSIX_MEMALIGN to use posix_memalign() #754
Conversation
mewall
requested review from
nicolasbock,
cnegre,
suemni,
jeanlucf22,
tokshgithub and
jmohdyusof
as code owners
|
Hi @mewall Do you see this kind of slow-down on all machines you run the code on or just on a particular architecture / OS combination? If this is a general issue we might simply want to remove this code path all together. |
|
I don't know but it would be good to look into that. I'll try that next week. For now it's possible avoid the code path using compiler flag -UHAVE_POSIX_MEMALIGN.
…________________________________
From: Nicolas Bock ***@***.***>
Sent: Friday, June 28, 2024 8:15 PM
To: lanl/bml ***@***.***>
Cc: Wall, Michael E ***@***.***>; Mention ***@***.***>
Subject: [EXTERNAL] Re: [lanl/bml] Require BML_USE_POSIX_MEMALIGN to use posix_memalign() (PR #754)
Hi @mewall<https://urldefense.com/v3/__https://github.com/mewall__;!!Bt8fGhp8LhKGRg!D50ERlVT-COdnX_ko_oqxrlQihZZmfZkx1QkqwxQ8Nb8IAnW5bHqSmH6iVwVsodS5SeAQZNrQLnT62Cqhs74cpOM$>
Do you see this kind of slow-down on all machines you run the code on or just on a particular architecture / OS combination? If this is a general issue we might simply want to remove this code path all together.
—
Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https://github.com/lanl/bml/pull/754*issuecomment-2197856904__;Iw!!Bt8fGhp8LhKGRg!D50ERlVT-COdnX_ko_oqxrlQihZZmfZkx1QkqwxQ8Nb8IAnW5bHqSmH6iVwVsodS5SeAQZNrQLnT62CqhoqfKfmc$>, or unsubscribe<https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AA67VEILYRO7QDSXJ3AROWLZJYKEZAVCNFSM6AAAAABKCMAAAWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCOJXHA2TMOJQGQ__;!!Bt8fGhp8LhKGRg!D50ERlVT-COdnX_ko_oqxrlQihZZmfZkx1QkqwxQ8Nb8IAnW5bHqSmH6iVwVsodS5SeAQZNrQLnT62CqhnOv2bbN$>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
|
It would be good to enable the old behavior by default and optionally allow to disable it. This would make sure that the old behavior does not unexpectedly change. |
|
I'll do this and revise the PR.
…________________________________
From: Nicolas Bock ***@***.***>
Sent: Saturday, June 29, 2024 11:37 PM
To: lanl/bml ***@***.***>
Cc: Wall, Michael E ***@***.***>; Mention ***@***.***>
Subject: [EXTERNAL] Re: [lanl/bml] Require BML_USE_POSIX_MEMALIGN to use posix_memalign() (PR #754)
It would be good to enable the old behavior by default and optionally allow to disable it. This would make sure that the old behavior does not unexpectedly change.
—
Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https://github.com/lanl/bml/pull/754*issuecomment-2198441341__;Iw!!Bt8fGhp8LhKGRg!HhDawgltKfRAbIQpR7xKeTkJ-q9vsTmWVMbkkVW6LVXskIQmO1gElnRz4F4ukV5SrNIrx6Wj-5OPyH9r6hRK2I9v$>, or unsubscribe<https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AA67VELFHQFKLC6NUXJKPH3ZJ6KP3AVCNFSM6AAAAABKCMAAAWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCOJYGQ2DCMZUGE__;!!Bt8fGhp8LhKGRg!HhDawgltKfRAbIQpR7xKeTkJ-q9vsTmWVMbkkVW6LVXskIQmO1gElnRz4F4ukV5SrNIrx6Wj-5OPyH9r6kH7E-eA$>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
Using the HAVE_POSIX_MEMALIGN code path results in ~1.7x slowdown of an MD code which frequently allocates arrays. Require a new macro BML_USE_POSIX_MEMALIGN to be defined, to use the code path. Define by default
|
@nicolasbock The behavior is now the same as before by default. Please merge if OK.
|
Using the HAVE_POSIX_MEMALIGN code path results in ~1.7x slowdown of an MD code which frequently allocates arrays.
Require a new macro BML_USE_POSIX_MEMALIGN to be defined, to use the code path. Define the macro by default.