Messages in this thread | | | From | David Laight <> | Subject | RE: [PATCH v11 2/3] x86,/proc/pid/status: Add AVX-512 usage elapsed time | Date | Wed, 20 Feb 2019 15:35:47 +0000 |
| |
From: Sent: 16 February 2019 12:56 > To: Li, Aubrey ... > The above experiment just confirms what I said: The numbers are inaccurate > and potentially misleading to a large extent when the AVX using task is not > scheduled out for a longer time.
Not only that, they won't detect programs that use AVX-512 but never context switch with live AVX-512 registers.
Since the registers are all caller-saved they can be zeroed after being saved (or used) in order to speed up any context switches.
I believe that all system calls are now done through libc 'stub' functions (I haven't seen inlined system calls for ages). So by the time the kernel is entered (by a system call) none of the AVX (or even SSE2?) registers can contain live values. This should mean it is save for the system call entry code to zero all the AVX registers - so none will be saved by a context switch.
Of course, is the kernel is entered by a trap or hardware interrupt the registers are live and do need saving.
David
- Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
| |