Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-21 Thread Stephane Eranian
On Sat, Jul 19, 2014 at 2:49 AM, Andi Kleen wrote: > On Tue, Jul 15, 2014 at 12:49:43AM +0200, Stephane Eranian wrote: >> On Tue, Jul 15, 2014 at 12:39 AM, Andi Kleen wrote: >> >> I have a problem with this patch. >> >> >> >> It makes: perf mem -t store rec record OP_NA for the store. >> >> It wa

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-18 Thread Andi Kleen
On Tue, Jul 15, 2014 at 12:49:43AM +0200, Stephane Eranian wrote: > On Tue, Jul 15, 2014 at 12:39 AM, Andi Kleen wrote: > >> I have a problem with this patch. > >> > >> It makes: perf mem -t store rec record OP_NA for the store. > >> It was recording OP_STORE before. > >> > >> I think we need to k

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-17 Thread Stephane Eranian
On Tue, Jul 15, 2014 at 8:19 PM, Stephane Eranian wrote: > On Tue, Jul 15, 2014 at 10:59 AM, Peter Zijlstra wrote: >> On Mon, Jul 14, 2014 at 09:05:42PM -0700, Andi Kleen wrote: >> >>> Yes, there are only three store events: >>> >>> MEM_UOPS_RETIRED.STLB_MISS_STORES >>> MEM_UOPS_RETIRED.SPLIT_STO

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-15 Thread Stephane Eranian
On Tue, Jul 15, 2014 at 10:59 AM, Peter Zijlstra wrote: > On Mon, Jul 14, 2014 at 09:05:42PM -0700, Andi Kleen wrote: > >> Yes, there are only three store events: >> >> MEM_UOPS_RETIRED.STLB_MISS_STORES >> MEM_UOPS_RETIRED.SPLIT_STORES >> MEM_UOPS_RETIRED.ALL_STORES >> >> These can be added. >> >>

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-15 Thread Peter Zijlstra
On Mon, Jul 14, 2014 at 09:05:42PM -0700, Andi Kleen wrote: > Yes, there are only three store events: > > MEM_UOPS_RETIRED.STLB_MISS_STORES > MEM_UOPS_RETIRED.SPLIT_STORES > MEM_UOPS_RETIRED.ALL_STORES > > These can be added. > > But most others are loads, so if you wanted loads too (besides me

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-14 Thread Andi Kleen
On Tue, Jul 15, 2014 at 12:50:27AM +0200, Stephane Eranian wrote: > On Tue, Jul 15, 2014 at 12:49 AM, Stephane Eranian wrote: > > On Tue, Jul 15, 2014 at 12:39 AM, Andi Kleen wrote: > >>> I have a problem with this patch. > >>> > >>> It makes: perf mem -t store rec record OP_NA for the store. > >

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-14 Thread Stephane Eranian
On Tue, Jul 15, 2014 at 12:49 AM, Stephane Eranian wrote: > On Tue, Jul 15, 2014 at 12:39 AM, Andi Kleen wrote: >>> I have a problem with this patch. >>> >>> It makes: perf mem -t store rec record OP_NA for the store. >>> It was recording OP_STORE before. >>> >>> I think we need to keep LD/ST inf

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-14 Thread Stephane Eranian
On Tue, Jul 15, 2014 at 12:39 AM, Andi Kleen wrote: >> I have a problem with this patch. >> >> It makes: perf mem -t store rec record OP_NA for the store. >> It was recording OP_STORE before. >> >> I think we need to keep LD/ST info. This is useful for analysis >> especially if we collect loads/st

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-14 Thread Andi Kleen
> I have a problem with this patch. > > It makes: perf mem -t store rec record OP_NA for the store. > It was recording OP_STORE before. > > I think we need to keep LD/ST info. This is useful for analysis > especially if we collect loads/stores simultaneously. > > Was working before for the mem-l

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-14 Thread Stephane Eranian
Andi, On Mon, Jul 14, 2014 at 9:24 PM, Andi Kleen wrote: >> You could do better if you tagged the event during setup as load vs. store. >> And then you could simply propagate the flag to the data source struct. > > This would require listing all PEBS events in the table again. > The whole point o

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-14 Thread Andi Kleen
> You could do better if you tagged the event during setup as load vs. store. > And then you could simply propagate the flag to the data source struct. This would require listing all PEBS events in the table again. The whole point of the other patch was to get rid of that. Besides it wouldn't wor

Re: [PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-14 Thread Stephane Eranian
On Sat, Jul 12, 2014 at 2:01 AM, Andi Kleen wrote: > From: Andi Kleen > > Haswell supports reporting the data address for a range > of PEBS events, including: > > UOPS_RETIRED.ALL > MEM_UOPS_RETIRED.STLB_MISS_LOADS > MEM_UOPS_RETIRED.STLB_MISS_STORES > MEM_UOPS_RETIRED.LOCK_LOADS > MEM_UOPS_RETIR

[PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-11 Thread Andi Kleen
From: Andi Kleen Haswell supports reporting the data address for a range of PEBS events, including: UOPS_RETIRED.ALL MEM_UOPS_RETIRED.STLB_MISS_LOADS MEM_UOPS_RETIRED.STLB_MISS_STORES MEM_UOPS_RETIRED.LOCK_LOADS MEM_UOPS_RETIRED.SPLIT_LOADS MEM_UOPS_RETIRED.SPLIT_STORES MEM_UOPS_RETIRED.ALL_LOAD

[PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-07-07 Thread Andi Kleen
From: Andi Kleen Haswell supports reporting the data address for a range of PEBS events, including: UOPS_RETIRED.ALL MEM_UOPS_RETIRED.STLB_MISS_LOADS MEM_UOPS_RETIRED.STLB_MISS_STORES MEM_UOPS_RETIRED.LOCK_LOADS MEM_UOPS_RETIRED.SPLIT_LOADS MEM_UOPS_RETIRED.SPLIT_STORES MEM_UOPS_RETIRED.ALL_LOAD

[PATCH 2/2] perf, x86: Don't mark DataLA addresses as store

2014-03-13 Thread Andi Kleen
From: Andi Kleen Haswell supports reporting the data address for a range of events, including UOPS_RETIRED.ALL and some load events. Currently these addresses were always marked as stores, which is wrong, as they could be loads too. Change it to NA instead. Signed-off-by: Andi Kleen --- arch/x