[PATCH] perf test: Compile named_threads workload with -O0

[PATCH] perf test: Compile named_threads workload with -O0

[James Clark]

The work loop relies on the compiler not optimizing it away, although
named_threads_work is not static for that reason, the compiler could
still do it.

Fix it by compiling without optimization. Also add -fno-inline for
consistency and in case anyone wants to look at callstacks.

Signed-off-by: James Clark <james.clark@linaro.org>
---
One trailing Sashiko fix from the "perf cs-etm: Queue context packets
for frontend" patch series that has been applied already.
---
 tools/perf/tests/workloads/Build | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tools/perf/tests/workloads/Build b/tools/perf/tests/workloads/Build
index 75b377934a0e..7bb4b9829ba2 100644
--- a/tools/perf/tests/workloads/Build
+++ b/tools/perf/tests/workloads/Build
@@ -26,3 +26,4 @@ CFLAGS_datasym.o          = -g -O0 -fno-inline -U_FORTIFY_SOURCE
 CFLAGS_traploop.o         = -g -O0 -fno-inline -U_FORTIFY_SOURCE
 CFLAGS_inlineloop.o       = -g -O2
 CFLAGS_deterministic.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
+CFLAGS_named_threads.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE

---
base-commit: 7336514f41e75d44782fee7e0990d4195a3d3161
change-id: 20260611-james-cs-unformatted-per-thread-fix-followup-babf28a4e640

Best regards,
-- 
James Clark <james.clark@linaro.org>

Re: [PATCH] perf test: Compile named_threads workload with -O0

[Leo Yan]

On Thu, Jun 11, 2026 at 12:13:46PM +0100, James Clark wrote:
> The work loop relies on the compiler not optimizing it away, although
> named_threads_work is not static for that reason, the compiler could
> still do it.
> 
> Fix it by compiling without optimization. Also add -fno-inline for
> consistency and in case anyone wants to look at callstacks.
> 
> Signed-off-by: James Clark <james.clark@linaro.org>

Reviewed-by: Leo Yan <leo.yan@arm.com>

Re: [PATCH] perf test: Compile named_threads workload with -O0

[Arnaldo Carvalho de Melo]

On Thu, Jun 11, 2026 at 12:13:46PM +0100, James Clark wrote:
> The work loop relies on the compiler not optimizing it away, although
> named_threads_work is not static for that reason, the compiler could
> still do it.
> 
> Fix it by compiling without optimization. Also add -fno-inline for
> consistency and in case anyone wants to look at callstacks.
> 
> Signed-off-by: James Clark <james.clark@linaro.org>
> ---
> One trailing Sashiko fix from the "perf cs-etm: Queue context packets
> for frontend" patch series that has been applied already.

I'll add a Fixes tag pointing to that cset, also a reported-by sashiko
and a closes, for completeness.

Thanks, applying.

- Arnaldo

> ---
>  tools/perf/tests/workloads/Build | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/tools/perf/tests/workloads/Build b/tools/perf/tests/workloads/Build
> index 75b377934a0e..7bb4b9829ba2 100644
> --- a/tools/perf/tests/workloads/Build
> +++ b/tools/perf/tests/workloads/Build
> @@ -26,3 +26,4 @@ CFLAGS_datasym.o          = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>  CFLAGS_traploop.o         = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>  CFLAGS_inlineloop.o       = -g -O2
>  CFLAGS_deterministic.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
> +CFLAGS_named_threads.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
> 
> ---
> base-commit: 7336514f41e75d44782fee7e0990d4195a3d3161
> change-id: 20260611-james-cs-unformatted-per-thread-fix-followup-babf28a4e640
> 
> Best regards,
> -- 
> James Clark <james.clark@linaro.org>

Re: [PATCH] perf test: Compile named_threads workload with -O0

[James Clark]

On 11/06/2026 4:08 pm, Arnaldo Carvalho de Melo wrote:
> On Thu, Jun 11, 2026 at 12:13:46PM +0100, James Clark wrote:
>> The work loop relies on the compiler not optimizing it away, although
>> named_threads_work is not static for that reason, the compiler could
>> still do it.
>>
>> Fix it by compiling without optimization. Also add -fno-inline for
>> consistency and in case anyone wants to look at callstacks.
>>
>> Signed-off-by: James Clark <james.clark@linaro.org>
>> ---
>> One trailing Sashiko fix from the "perf cs-etm: Queue context packets
>> for frontend" patch series that has been applied already.
> 
> I'll add a Fixes tag pointing to that cset, also a reported-by sashiko
> and a closes, for completeness.
> 
> Thanks, applying.
> 
> - Arnaldo
> 

Thanks, makes sense yeah. I probably could have been a bit more thorough.

>> ---
>>   tools/perf/tests/workloads/Build | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/tools/perf/tests/workloads/Build b/tools/perf/tests/workloads/Build
>> index 75b377934a0e..7bb4b9829ba2 100644
>> --- a/tools/perf/tests/workloads/Build
>> +++ b/tools/perf/tests/workloads/Build
>> @@ -26,3 +26,4 @@ CFLAGS_datasym.o          = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>>   CFLAGS_traploop.o         = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>>   CFLAGS_inlineloop.o       = -g -O2
>>   CFLAGS_deterministic.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>> +CFLAGS_named_threads.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>>
>> ---
>> base-commit: 7336514f41e75d44782fee7e0990d4195a3d3161
>> change-id: 20260611-james-cs-unformatted-per-thread-fix-followup-babf28a4e640
>>
>> Best regards,
>> -- 
>> James Clark <james.clark@linaro.org>

Re: [PATCH] perf test: Compile named_threads workload with -O0

[David Laight]

On Thu, 11 Jun 2026 12:13:46 +0100
James Clark <james.clark@linaro.org> wrote:

> The work loop relies on the compiler not optimizing it away, although
> named_threads_work is not static for that reason, the compiler could
> still do it.

Isn't in enough to put barrier() in the loop?

	David

> 
> Fix it by compiling without optimization. Also add -fno-inline for
> consistency and in case anyone wants to look at callstacks.
> 
> Signed-off-by: James Clark <james.clark@linaro.org>
> ---
> One trailing Sashiko fix from the "perf cs-etm: Queue context packets
> for frontend" patch series that has been applied already.
> ---
>  tools/perf/tests/workloads/Build | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/tools/perf/tests/workloads/Build b/tools/perf/tests/workloads/Build
> index 75b377934a0e..7bb4b9829ba2 100644
> --- a/tools/perf/tests/workloads/Build
> +++ b/tools/perf/tests/workloads/Build
> @@ -26,3 +26,4 @@ CFLAGS_datasym.o          = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>  CFLAGS_traploop.o         = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>  CFLAGS_inlineloop.o       = -g -O2
>  CFLAGS_deterministic.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
> +CFLAGS_named_threads.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
> 
> ---
> base-commit: 7336514f41e75d44782fee7e0990d4195a3d3161
> change-id: 20260611-james-cs-unformatted-per-thread-fix-followup-babf28a4e640
> 
> Best regards,

Re: [PATCH] perf test: Compile named_threads workload with -O0

[James Clark]

On 11/06/2026 7:54 pm, David Laight wrote:
> On Thu, 11 Jun 2026 12:13:46 +0100
> James Clark <james.clark@linaro.org> wrote:
> 
>> The work loop relies on the compiler not optimizing it away, although
>> named_threads_work is not static for that reason, the compiler could
>> still do it.
> 
> Isn't in enough to put barrier() in the loop?
> 
> 	David
> 

Probably, but I don't think it would make a difference for this specific 
test workload, and I don't think that style has ever been used for any 
of the Perf tests.

I think Sashiko's comment leading to this was overly pedantic, neither 
GCC or Clang optimize this away even with -O2. So in this specific case 
it was only fixing a hypothetical which is why I didn't add a fixes tag. 
Although I admit there is some chance it could happen with other changes.

However, I think -O0 is the appropriate fix for a workload because it 
makes the stacks and order of execution predictable, which is what the 
tests look at. And that also has the effect of guaranteeing this isn't 
optimized away, so we probably don't need both.

Thanks
James

>>
>> Fix it by compiling without optimization. Also add -fno-inline for
>> consistency and in case anyone wants to look at callstacks.
>>
>> Signed-off-by: James Clark <james.clark@linaro.org>
>> ---
>> One trailing Sashiko fix from the "perf cs-etm: Queue context packets
>> for frontend" patch series that has been applied already.
>> ---
>>   tools/perf/tests/workloads/Build | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/tools/perf/tests/workloads/Build b/tools/perf/tests/workloads/Build
>> index 75b377934a0e..7bb4b9829ba2 100644
>> --- a/tools/perf/tests/workloads/Build
>> +++ b/tools/perf/tests/workloads/Build
>> @@ -26,3 +26,4 @@ CFLAGS_datasym.o          = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>>   CFLAGS_traploop.o         = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>>   CFLAGS_inlineloop.o       = -g -O2
>>   CFLAGS_deterministic.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>> +CFLAGS_named_threads.o    = -g -O0 -fno-inline -U_FORTIFY_SOURCE
>>
>> ---
>> base-commit: 7336514f41e75d44782fee7e0990d4195a3d3161
>> change-id: 20260611-james-cs-unformatted-per-thread-fix-followup-babf28a4e640
>>
>> Best regards,
>