Reenable high accuracy timers #285
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
for MPI_Wtime and MPI_Wtick.
MPI_Wtime and MPI_Wtick. Merge branch 'topic/accurate_timers'
|
retest this please |
|
Test PASSed. |
|
Test PASSed. |
| } | ||
|
|
||
| #else | ||
|
|
There was a problem hiding this comment.
The code (in the context) that was in this #else block is buggy. rdtsc must be used with a serializing cpuid instruction to ensure that you are timing what you think you are timing. There's an even better pattern that should be used in general, though I'm not sure we can apply it to MPI_Wtime() because we don't know if we are being called at the beginning of a timing region or at the end. For more information see http://www.intel.com/content/www/us/en/intelligent-systems/embedded-systems-training/ia-32-ia-64-benchmark-code-execution-paper.html
|
George, I've finished my review, see my inline comments on the PR (not sure if they show up in email or not). |
together with RDTSC. It is still not perfect, but hopefully much better than before.
|
Test FAILed. Build Log Test FAILed. |
| __asm__ __volatile__ ("cpuid\n\t" | ||
| "rdtsc\n\t" | ||
| : "=a" (a), "=d" (d) | ||
| :: "%rax", "%rbx", "%rcx", "%rdx"); |
There was a problem hiding this comment.
I think you want to drop the % characters in the clobber list (two places in this file, once in ia32/timer.h). Untested, but that's my recollection of the syntax.
There was a problem hiding this comment.
The syntax seems to be correct. The problem is that I over-specified the
clobber list (rax and rdx) were already implied by the "=a" and "=d" one
line above.
George.
On Mon, Nov 24, 2014 at 3:29 PM, Dave Goodell [email protected]
wrote:
In opal/include/opal/sys/amd64/timer.h:
static inline opal_timer_t
opal_sys_timer_get_cycles(void)
{- opal_timer_t ret;
- asm volatile("rdtsc" : "=A"(ret));
- return ret;
-}unsigned a, d;**asm** **volatile** ("cpuid\n\t""rdtsc\n\t": "=a" (a), "=d" (d):: "%rax", "%rbx", "%rcx", "%rdx");I think you want to drop the % characters in the clobber list (two places
in this file, once in ia32/timer.h). Untested, but that's my recollection
of the syntax.—
Reply to this email directly or view it on GitHub
https://github.com/open-mpi/ompi/pull/285/files#r20823713.
There was a problem hiding this comment.
I addressed all your concerns/suggestions. Thanks, they really make the
code cleaner and hopefully more robust. This is now ready to be merged.
On Mon, Nov 24, 2014 at 3:35 PM, George Bosilca [email protected] wrote:
The syntax seems to be correct. The problem is that I over-specified the
clobber list (rax and rdx) were already implied by the "=a" and "=d" one
line above.George.
On Mon, Nov 24, 2014 at 3:29 PM, Dave Goodell [email protected]
wrote:In opal/include/opal/sys/amd64/timer.h:
static inline opal_timer_t
opal_sys_timer_get_cycles(void)
{- opal_timer_t ret;
- asm volatile("rdtsc" : "=A"(ret));
- return ret;
-}unsigned a, d;**asm** **volatile** ("cpuid\n\t""rdtsc\n\t": "=a" (a), "=d" (d):: "%rax", "%rbx", "%rcx", "%rdx");I think you want to drop the % characters in the clobber list (two
places in this file, once in ia32/timer.h). Untested, but that's my
recollection of the syntax.—
Reply to this email directly or view it on GitHub
https://github.com/open-mpi/ompi/pull/285/files#r20823713.
|
Test FAILed. Build Log Test FAILed. |
|
Test FAILed. Build Log Test FAILed. |
so they must be saved before.
|
retest this please. |
|
Test PASSed. |
|
Test PASSed. |
|
The current version looks good to me too. Thanks for fixing this, George. I'm assuming you'll push it. |
jsquyres
pushed a commit
to jsquyres/ompi
that referenced
this pull request
Sep 19, 2016
If the user specifies a --map-by <foo> option, then default to bind-t…
dong0321
pushed a commit
to dong0321/ompi
that referenced
this pull request
Feb 17, 2020
Fix hwloc version check ordering
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
As discussed on the user mailing list the current implementation of MPI_Wtime defaults to gettimeofday in most cases. This patch proposes a better approach by:
@goodell please review. Based on our discussion at SC I made sure that when available we are using the RDTSC-based timers.