Temporary shader files can overlap each other

[TIMONz1535]

I run a lot of RGA instances to generate all my shaders, however, if the log files have a timestamp that somehow guarantees uniqueness for each instance with the same pid

rga_cli7040-20230216-194032.log
rga_cli7040-20230216-194043.log

then the temporary DXIL files are created only using pid. So statistics may be collected for the wrong shader BUT I'm not sure.

dxil_rga-temp-out7040_vert.obj
dxil_rga-temp-out7040_pixel.obj

// used by rga_cli7040-20230216-194032.log with isa, livereg, stats for "object.0063"
// used by rga_cli7040-20230216-194043.log with isa, livereg, stats for "object.0024"

rga_cli7040-20230216-194032.log rga_cli7040-20230216-194043.log

Perhaps temporary DXIL files should also have a timestamp. At the moment I can compile DXIL shaders myself with unique filenames and specify the correct binary shaders in RGA.

p.s. It seems like this can't happen. But I get varying data in the statistics table. I will answer later

[TIMONz1535]

okay I realized that described problem only occurs if I use a hidden parameter --retain so I don't need binary files. the statistics that are obtained at the output are always correct - because until one instance with some pid ends, another with the same pid will not start

I found an error in my code because due to the RGA crash I restarted it with incorrect shader "id". but I don't understand why because the crash happens randomly:

• once the application started, it was silent for two seconds and crash. sometimes it created an empty log file.
• once it collected the shader and statistics successfully. but for some reason it crashed at the exit

it seems to be related to the system issues, like #103 because on another computer it happens much less often and it is observed only after multiple launches

well, I just restart it and everything is fine. I don't think it's possible to catch it. so I'm closing this ticket