Sam Price commented: 
https://gitlab.rtems.org/rtems/tools/rtems-source-builder/-/work_items/169#note_153528


>From a brief glance Xilinx GCC is carrying 4 patches that are not in the 
>mailine gcc for the 32 bit version.

- [8 Stage pipeline model](https://github.com/thesamprice/gcc/commit/89a6cc55a
I am not sure if we are compiled for this.

- [Disable -fivopts](github.com/thesamprice/gcc/commit/05227fb9d)
Could be disabled by `-fno-ivopts` in the bsp opts.
Would need performance results before / after.


Minor ones
- [bitfield opt](github.com/thesamprice/gcc/commit/c3220f3f8)
- [zero_extend opt](github.com/thesamprice/gcc/commit/ac1e8cde5)

These look like optimization changes to me.
I am guessing going from gcc4 to 15.2 will probably be more of a speedup than 
these optimizations.

I have a modified version of qemu that I can track basic cycle counts.  
- Xilinx 4, 13 versions.
- Mainline 15, and 17

![image](/uploads/47b906c95d45da773b1ea69be9217016/image.png){width=900 
height=554}

Arith stood out to me, I ran the optimizations without the -fno-ivopts flag.

Diffing the assembly between the two, 15/17 is not making use of the delay 
slot. 

I will post -no-ivopts performance tomorrow.
Compiling withh -no-ivopts did put the delay slot into the assembly.  I expect 
better

-- 
View it on GitLab: 
https://gitlab.rtems.org/rtems/tools/rtems-source-builder/-/work_items/169#note_153528
You're receiving this email because of your account on gitlab.rtems.org. 
Unsubscribe from this thread: 
https://gitlab.rtems.org/-/sent_notifications/4-5nwbxmui8856en00zj6e7oytr-e/unsubscribe
 | Manage all notifications: https://gitlab.rtems.org/-/profile/notifications | 
Help: https://gitlab.rtems.org/help


_______________________________________________
bugs mailing list
[email protected]
http://lists.rtems.org/mailman/listinfo/bugs

Reply via email to