[25/30] libcamera: software_isp: gpu_pipeline_shader_pass: Add shader DEBUG time logging
diff mbox series

Message ID 20260618122245.946138-26-bryan.odonoghue@linaro.org
State RFC
Headers show
Series
  • RFC/RFT: gpuisp: Multipass with speed optimisations on top
Related show

Commit Message

Bryan O'Donoghue June 18, 2026, 12:22 p.m. UTC 
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
---
 .../software_isp/gpu_pipeline_shader_pass.cpp          | 10 ++++++++++
 src/libcamera/software_isp/gpu_pipeline_shader_pass.h  |  4 ++++
 2 files changed, 14 insertions(+)

Comments

Milan Zamazal June 25, 2026, 6:49 a.m. UTC | #1 
Bryan O'Donoghue <bryan.odonoghue@linaro.org> writes:

> Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
> ---
>  .../software_isp/gpu_pipeline_shader_pass.cpp          | 10 ++++++++++
>  src/libcamera/software_isp/gpu_pipeline_shader_pass.h  |  4 ++++
>  2 files changed, 14 insertions(+)
>
> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
> index d0d13eef9..57d60f50e 100644
> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
> @@ -57,6 +57,8 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>  	glViewport(0, 0, width, height);
>  	glClear(GL_COLOR_BUFFER_BIT);
>  
> +	eglBenchMark_.begin(egl_);
> +
>  	glDrawArrays(GL_TRIANGLE_FAN, 0, DEBAYER_OPENGL_COORDS);
>  	err = glGetError();
>  	if (err != GL_NO_ERROR) {
> @@ -64,9 +66,17 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>  		return -ENODEV;
>  	}
>  
> +	eglBenchMark_.end(egl_);
> +
>  	return 0;
>  }
>  
> +void GpuIspShaderPass::printShaderBenchMark(void)
> +{
> +	LOG(GpuShaderPass, Debug) <<
> +	    "ShaderPass = " << this->name() << " took " << eglBenchMark_.getTimeElapsedSync(egl_) / 1000 << " us";
> +}
> +
>  int GpuIspShaderPass::start()
>  {
>  	return 0;
> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
> index a329845ee..0cff531e9 100644
> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
> @@ -74,8 +74,12 @@ public:
>  	 */
>  	GLint glFormat_;
>  
> +	void initShaderBenchMark(void) { eglBenchMark_.init(); }
> +	void printShaderBenchMark(void);

1,$s/void//

Otherwise it looks good to me as far as I understand it.

> +
>  protected:
>  	eGL& egl_;
> +	eGLBenchMark eglBenchMark_;
>  
>  	/* Shader calculates this getter provides ability to interrogate if needed */
>  	unsigned int bytesPerPixel_;
Milan Zamazal July 6, 2026, 4:06 p.m. UTC | #2 
Bryan O'Donoghue <bryan.odonoghue@linaro.org> writes:

> Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
> ---
>  .../software_isp/gpu_pipeline_shader_pass.cpp          | 10 ++++++++++
>  src/libcamera/software_isp/gpu_pipeline_shader_pass.h  |  4 ++++
>  2 files changed, 14 insertions(+)
>
> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
> index d0d13eef9..57d60f50e 100644
> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
> @@ -57,6 +57,8 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>  	glViewport(0, 0, width, height);
>  	glClear(GL_COLOR_BUFFER_BIT);
>  
> +	eglBenchMark_.begin(egl_);
> +
>  	glDrawArrays(GL_TRIANGLE_FAN, 0, DEBAYER_OPENGL_COORDS);
>  	err = glGetError();
>  	if (err != GL_NO_ERROR) {
> @@ -64,9 +66,17 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>  		return -ENODEV;
>  	}
>  
> +	eglBenchMark_.end(egl_);

On RPi 4:

- Debian 12: It doesn't work as the facility is unsupported.

- Debian 13: It works, but if I comment out the eglBenchMark_ calls, the
  per-frame time reported by software ISP is about 15% faster (and a bit
  faster than on master).

> +
>  	return 0;
>  }
>  
> +void GpuIspShaderPass::printShaderBenchMark(void)
> +{
> +	LOG(GpuShaderPass, Debug) <<
> +	    "ShaderPass = " << this->name() << " took " << eglBenchMark_.getTimeElapsedSync(egl_) / 1000 << " us";

When running on RPi 4 Debian 13 with DEBUG logging, I get the following
segmentation fault:

0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from /lib/aarch64-linux-gnu/libstdc++.so.6
(gdb) where
#0  0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from /lib/aarch64-linux-gnu/libstdc++.so.6
#1  0x0000007ff7821d10 in std::basic_ostream<char, std::char_traits<char> >& std::basic_ostream<char, std::char_traits<char> >::_M_insert<unsigned long>(unsigned long) ()
   from /lib/aarch64-linux-gnu/libstdc++.so.6
#2  0x0000007ff7f485c4 in std::basic_ostream<char, std::char_traits<char> >::operator<< (
    this=0x7feefdd270, __n=<optimized out>) at /usr/include/c++/14/ostream:179
#3  libcamera::GpuIspShaderPass::printShaderBenchMark (this=this@entry=0x7fe8027d08)
    at ../src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp:78

When I extract

  eglBenchMark_.getTimeElapsedSync(egl_) / 1000

to a local `unsigned int' variable just before the LOG call and use the
variable there instead of the expression, it works fine.  Don't ask me
why.

> +}
> +
>  int GpuIspShaderPass::start()
>  {
>  	return 0;
> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
> index a329845ee..0cff531e9 100644
> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
> @@ -74,8 +74,12 @@ public:
>  	 */
>  	GLint glFormat_;
>  
> +	void initShaderBenchMark(void) { eglBenchMark_.init(); }
> +	void printShaderBenchMark(void);
> +
>  protected:
>  	eGL& egl_;
> +	eGLBenchMark eglBenchMark_;
>  
>  	/* Shader calculates this getter provides ability to interrogate if needed */
>  	unsigned int bytesPerPixel_;
Bryan O'Donoghue July 6, 2026, 7:43 p.m. UTC | #3 
On 06/07/2026 17:06, Milan Zamazal wrote:
> Bryan O'Donoghue <bryan.odonoghue@linaro.org> writes:
> 
>> Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
>> ---
>>   .../software_isp/gpu_pipeline_shader_pass.cpp          | 10 ++++++++++
>>   src/libcamera/software_isp/gpu_pipeline_shader_pass.h  |  4 ++++
>>   2 files changed, 14 insertions(+)
>>
>> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>> index d0d13eef9..57d60f50e 100644
>> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>> @@ -57,6 +57,8 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>>   	glViewport(0, 0, width, height);
>>   	glClear(GL_COLOR_BUFFER_BIT);
>>
>> +	eglBenchMark_.begin(egl_);
>> +
>>   	glDrawArrays(GL_TRIANGLE_FAN, 0, DEBAYER_OPENGL_COORDS);
>>   	err = glGetError();
>>   	if (err != GL_NO_ERROR) {
>> @@ -64,9 +66,17 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>>   		return -ENODEV;
>>   	}
>>
>> +	eglBenchMark_.end(egl_);
> 
> On RPi 4:
> 
> - Debian 12: It doesn't work as the facility is unsupported.
> 
> - Debian 13: It works, but if I comment out the eglBenchMark_ calls, the
>    per-frame time reported by software ISP is about 15% faster (and a bit
>    faster than on master).

Out of curiosity why not RaspiOS ?

BTW which version of Mesa is that ?

Actually this is pretty useful to know, I have a bunch of RPI 3/4 so, I 
should be able to "play along at home" and verify the tests we put in 
place for the different levels of functionality work.

>> +
>>   	return 0;
>>   }
>>
>> +void GpuIspShaderPass::printShaderBenchMark(void)
>> +{
>> +	LOG(GpuShaderPass, Debug) <<
>> +	    "ShaderPass = " << this->name() << " took " << eglBenchMark_.getTimeElapsedSync(egl_) / 1000 << " us";
> 
> When running on RPi 4 Debian 13 with DEBUG logging, I get the following
> segmentation fault:
> 
> 0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from /lib/aarch64-linux-gnu/libstdc++.so.6
> (gdb) where
> #0  0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from /lib/aarch64-linux-gnu/libstdc++.so.6
> #1  0x0000007ff7821d10 in std::basic_ostream<char, std::char_traits<char> >& std::basic_ostream<char, std::char_traits<char> >::_M_insert<unsigned long>(unsigned long) ()
>     from /lib/aarch64-linux-gnu/libstdc++.so.6
> #2  0x0000007ff7f485c4 in std::basic_ostream<char, std::char_traits<char> >::operator<< (
>      this=0x7feefdd270, __n=<optimized out>) at /usr/include/c++/14/ostream:179
> #3  libcamera::GpuIspShaderPass::printShaderBenchMark (this=this@entry=0x7fe8027d08)
>      at ../src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp:78
> 
> When I extract
> 
>    eglBenchMark_.getTimeElapsedSync(egl_) / 1000
> 
> to a local `unsigned int' variable just before the LOG call and use the
> variable there instead of the expression, it works fine.  Don't ask me
> why.

what happens with unsigned long long ?

> 
>> +}
>> +
>>   int GpuIspShaderPass::start()
>>   {
>>   	return 0;
>> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>> index a329845ee..0cff531e9 100644
>> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>> @@ -74,8 +74,12 @@ public:
>>   	 */
>>   	GLint glFormat_;
>>
>> +	void initShaderBenchMark(void) { eglBenchMark_.init(); }
>> +	void printShaderBenchMark(void);
>> +
>>   protected:
>>   	eGL& egl_;
>> +	eGLBenchMark eglBenchMark_;
>>
>>   	/* Shader calculates this getter provides ability to interrogate if needed */
>>   	unsigned int bytesPerPixel_;
>
Milan Zamazal July 7, 2026, 9:10 p.m. UTC | #4 
Bryan O'Donoghue <bod.linux@nxsw.ie> writes:

> On 06/07/2026 17:06, Milan Zamazal wrote:
>> Bryan O'Donoghue <bryan.odonoghue@linaro.org> writes:
>> 
>
>>> Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
>>> ---
>>>   .../software_isp/gpu_pipeline_shader_pass.cpp          | 10 ++++++++++
>>>   src/libcamera/software_isp/gpu_pipeline_shader_pass.h  |  4 ++++
>>>   2 files changed, 14 insertions(+)
>>>
>>> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>>> index d0d13eef9..57d60f50e 100644
>>> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>>> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>>> @@ -57,6 +57,8 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>>>   	glViewport(0, 0, width, height);
>>>   	glClear(GL_COLOR_BUFFER_BIT);
>>>
>>> +	eglBenchMark_.begin(egl_);
>>> +
>>>   	glDrawArrays(GL_TRIANGLE_FAN, 0, DEBAYER_OPENGL_COORDS);
>>>   	err = glGetError();
>>>   	if (err != GL_NO_ERROR) {
>>> @@ -64,9 +66,17 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>>>   		return -ENODEV;
>>>   	}
>>>
>>> +	eglBenchMark_.end(egl_);
>> On RPi 4:
>> - Debian 12: It doesn't work as the facility is unsupported.
>> - Debian 13: It works, but if I comment out the eglBenchMark_ calls, the
>>    per-frame time reported by software ISP is about 15% faster (and a bit
>>    faster than on master).
>
> Out of curiosity why not RaspiOS ?

I had had a frustrating experience with it (unstable graphics etc.), so
I reinstalled to Debian, some years ago.

> BTW which version of Mesa is that ?

EGL: GL_VERSION: OpenGL ES 3.1 Mesa 25.0.7-2

> Actually this is pretty useful to know, I have a bunch of RPI 3/4 so, I should be able to "play along at
> home" and verify the tests we put in place for the different levels of functionality work.
>
>>> +
>>>   	return 0;
>>>   }
>>>
>>> +void GpuIspShaderPass::printShaderBenchMark(void)
>>> +{
>>> +	LOG(GpuShaderPass, Debug) <<
>>> +	    "ShaderPass = " << this->name() << " took " << eglBenchMark_.getTimeElapsedSync(egl_) / 1000 << " us";
>> When running on RPi 4 Debian 13 with DEBUG logging, I get the following
>> segmentation fault:
>> 0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char>
>> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from
>> /lib/aarch64-linux-gnu/libstdc++.so.6
>> (gdb) where
>> #0  0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from /lib/aarch64-linux-gnu/libstdc++.so.6
>> #1  0x0000007ff7821d10 in std::basic_ostream<char, std::char_traits<char> >& std::basic_ostream<char, std::char_traits<char> >::_M_insert<unsigned long>(unsigned long) ()
>>     from /lib/aarch64-linux-gnu/libstdc++.so.6
>> #2  0x0000007ff7f485c4 in std::basic_ostream<char, std::char_traits<char> >::operator<< (
>>      this=0x7feefdd270, __n=<optimized out>) at /usr/include/c++/14/ostream:179
>> #3  libcamera::GpuIspShaderPass::printShaderBenchMark (this=this@entry=0x7fe8027d08)
>>      at ../src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp:78
>> When I extract
>>    eglBenchMark_.getTimeElapsedSync(egl_) / 1000
>> to a local `unsigned int' variable just before the LOG call and use the
>> variable there instead of the expression, it works fine.  Don't ask me
>> why.
>
> what happens with unsigned long long ?

No problem, GLuint64 is fine too.  The only thing that seems to matter
is using the local variable.

gcc (Debian 14.2.0-19) 14.2.0

>> 
>>> +}
>>> +
>>>   int GpuIspShaderPass::start()
>>>   {
>>>   	return 0;
>>> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>>> index a329845ee..0cff531e9 100644
>>> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>>> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>>> @@ -74,8 +74,12 @@ public:
>>>   	 */
>>>   	GLint glFormat_;
>>>
>>> +	void initShaderBenchMark(void) { eglBenchMark_.init(); }
>>> +	void printShaderBenchMark(void);
>>> +
>>>   protected:
>>>   	eGL& egl_;
>>> +	eGLBenchMark eglBenchMark_;
>>>
>>>   	/* Shader calculates this getter provides ability to interrogate if needed */
>>>   	unsigned int bytesPerPixel_;
>>
Milan Zamazal July 8, 2026, 8:09 a.m. UTC | #5 
Milan Zamazal <mzamazal@redhat.com> writes:

> Bryan O'Donoghue <bod.linux@nxsw.ie> writes:
>
>> On 06/07/2026 17:06, Milan Zamazal wrote:
>>> Bryan O'Donoghue <bryan.odonoghue@linaro.org> writes:
>>> 
>>
>>>> Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
>>>> ---
>>>>   .../software_isp/gpu_pipeline_shader_pass.cpp          | 10 ++++++++++
>>>>   src/libcamera/software_isp/gpu_pipeline_shader_pass.h  |  4 ++++
>>>>   2 files changed, 14 insertions(+)
>>>>
>>>> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>>>> index d0d13eef9..57d60f50e 100644
>>>> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>>>> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
>>>> @@ -57,6 +57,8 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>>>>   	glViewport(0, 0, width, height);
>>>>   	glClear(GL_COLOR_BUFFER_BIT);
>>>>
>>>> +	eglBenchMark_.begin(egl_);
>>>> +
>>>>   	glDrawArrays(GL_TRIANGLE_FAN, 0, DEBAYER_OPENGL_COORDS);
>>>>   	err = glGetError();
>>>>   	if (err != GL_NO_ERROR) {
>>>> @@ -64,9 +66,17 @@ int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
>>>>   		return -ENODEV;
>>>>   	}
>>>>
>>>> +	eglBenchMark_.end(egl_);
>>> On RPi 4:
>>> - Debian 12: It doesn't work as the facility is unsupported.
>>> - Debian 13: It works, but if I comment out the eglBenchMark_ calls, the
>>>    per-frame time reported by software ISP is about 15% faster (and a bit
>>>    faster than on master).
>>
>> Out of curiosity why not RaspiOS ?
>
> I had had a frustrating experience with it (unstable graphics etc.), so
> I reinstalled to Debian, some years ago.
>
>> BTW which version of Mesa is that ?
>
> EGL: GL_VERSION: OpenGL ES 3.1 Mesa 25.0.7-2
>
>> Actually this is pretty useful to know, I have a bunch of RPI 3/4 so, I should be able to "play along at
>> home" and verify the tests we put in place for the different levels of functionality work.
>>
>>>> +
>>>>   	return 0;
>>>>   }
>>>>
>>>> +void GpuIspShaderPass::printShaderBenchMark(void)
>>>> +{
>>>> +	LOG(GpuShaderPass, Debug) <<
>>>> +	    "ShaderPass = " << this->name() << " took " << eglBenchMark_.getTimeElapsedSync(egl_) / 1000 << " us";
>>> When running on RPi 4 Debian 13 with DEBUG logging, I get the following
>>> segmentation fault:
>>> 0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char>
>>> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from
>>> /lib/aarch64-linux-gnu/libstdc++.so.6
>>> (gdb) where
>>> #0  0x0000007ff78207e0 in std::basic_ostream<char, std::char_traits<char> >::sentry::sentry(std::basic_ostream<char, std::char_traits<char> >&) () from /lib/aarch64-linux-gnu/libstdc++.so.6
>>> #1  0x0000007ff7821d10 in std::basic_ostream<char, std::char_traits<char> >& std::basic_ostream<char, std::char_traits<char> >::_M_insert<unsigned long>(unsigned long) ()
>>>     from /lib/aarch64-linux-gnu/libstdc++.so.6
>>> #2  0x0000007ff7f485c4 in std::basic_ostream<char, std::char_traits<char> >::operator<< (
>>>      this=0x7feefdd270, __n=<optimized out>) at /usr/include/c++/14/ostream:179
>>> #3  libcamera::GpuIspShaderPass::printShaderBenchMark (this=this@entry=0x7fe8027d08)
>>>      at ../src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp:78
>>> When I extract
>>>    eglBenchMark_.getTimeElapsedSync(egl_) / 1000
>>> to a local `unsigned int' variable just before the LOG call and use the
>>> variable there instead of the expression, it works fine.  Don't ask me
>>> why.
>>
>> what happens with unsigned long long ?
>
> No problem, GLuint64 is fine too.  The only thing that seems to matter
> is using the local variable.

AI discovered the problem -- a 32-bit variable is used in a place where
a 64-bit memory location is expected.  This patch fixes the problem:

diff --git a/include/libcamera/internal/egl.h b/include/libcamera/internal/egl.h
index c9b3a4e81..8d98faf75 100644
--- a/include/libcamera/internal/egl.h
+++ b/include/libcamera/internal/egl.h
@@ -152,7 +152,7 @@ private:
        PFNGLBEGINQUERYPROC glBeginQuery;
        PFNGLENDQUERYPROC glEndQuery;
        PFNGLGETQUERYOBJECTUIVPROC glGetQueryObjectuiv;
-       PFNGLGETQUERYOBJECTUIVPROC glGetQueryObjectui64v;
+       PFNGLGETQUERYOBJECTUI64VEXTPROC glGetQueryObjectui64v;
        PFNGLGENQUERIESPROC glGenQueries;
 };

@@ -192,7 +192,7 @@ private:

        GLuint64 getTimeElapsed(eGL &egl, bool sync)
        {
-               GLuint time_elapsed_ns = 0;
+               GLuint64 time_elapsed_ns = 0;
                GLuint available = 0;

                if (!sync)
diff --git a/src/libcamera/egl.cpp b/src/libcamera/egl.cpp
index fc1516634..237ce59bc 100644
--- a/src/libcamera/egl.cpp
+++ b/src/libcamera/egl.cpp
@@ -435,7 +435,7 @@ int eGL::initEGLContext()
                goto fail;
        }

-       glGetQueryObjectui64v = (PFNGLGETQUERYOBJECTUIVPROC)eglGetProcAddress("glGetQueryObjectui64v");
+       glGetQueryObjectui64v = (PFNGLGETQUERYOBJECTUI64VEXTPROC)eglGetProcAddress("glGetQueryObjectui64v");
        if (!glGetQueryObjectuiv) {
                LOG(eGL, Error) << "glGetQueryObjectui64v not found";
                goto fail;


> gcc (Debian 14.2.0-19) 14.2.0
>
>>> 
>>>> +}
>>>> +
>>>>   int GpuIspShaderPass::start()
>>>>   {
>>>>   	return 0;
>>>> diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>>>> index a329845ee..0cff531e9 100644
>>>> --- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>>>> +++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
>>>> @@ -74,8 +74,12 @@ public:
>>>>   	 */
>>>>   	GLint glFormat_;
>>>>
>>>> +	void initShaderBenchMark(void) { eglBenchMark_.init(); }
>>>> +	void printShaderBenchMark(void);
>>>> +
>>>>   protected:
>>>>   	eGL& egl_;
>>>> +	eGLBenchMark eglBenchMark_;
>>>>
>>>>   	/* Shader calculates this getter provides ability to interrogate if needed */
>>>>   	unsigned int bytesPerPixel_;
>>>

Patch
diff mbox series 

diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
index d0d13eef9..57d60f50e 100644
--- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
+++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.cpp
@@ -57,6 +57,8 @@  int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
 	glViewport(0, 0, width, height);
 	glClear(GL_COLOR_BUFFER_BIT);
 
+	eglBenchMark_.begin(egl_);
+
 	glDrawArrays(GL_TRIANGLE_FAN, 0, DEBAYER_OPENGL_COORDS);
 	err = glGetError();
 	if (err != GL_NO_ERROR) {
@@ -64,9 +66,17 @@  int GpuIspShaderPass::process(eGLImage &eglImageIn, eGLImage &eglImageOut, uint3
 		return -ENODEV;
 	}
 
+	eglBenchMark_.end(egl_);
+
 	return 0;
 }
 
+void GpuIspShaderPass::printShaderBenchMark(void)
+{
+	LOG(GpuShaderPass, Debug) <<
+	    "ShaderPass = " << this->name() << " took " << eglBenchMark_.getTimeElapsedSync(egl_) / 1000 << " us";
+}
+
 int GpuIspShaderPass::start()
 {
 	return 0;
diff --git a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
index a329845ee..0cff531e9 100644
--- a/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
+++ b/src/libcamera/software_isp/gpu_pipeline_shader_pass.h
@@ -74,8 +74,12 @@  public:
 	 */
 	GLint glFormat_;
 
+	void initShaderBenchMark(void) { eglBenchMark_.init(); }
+	void printShaderBenchMark(void);
+
 protected:
 	eGL& egl_;
+	eGLBenchMark eglBenchMark_;
 
 	/* Shader calculates this getter provides ability to interrogate if needed */
 	unsigned int bytesPerPixel_;