Open peterhchen opened 1 year ago
No, but why would you want this? Power = energy/time, you need to estimate time in order to estimate power, and time = performance.
Thanks Since Aladdin only support X86, we need RISCV and ARM . I used gem5 for RISCV and ARM for performance I need power and area from Aladdin (or CACTI/MCPAT or area from Yosys)
Peter
On Thu, Sep 1, 2022 at 3:40 PM Sam Xi @.***> wrote:
No, but why would you want this? Power = energy/time, you need to estimate time in order to estimate power, and time = performance.
— Reply to this email directly, view it on GitHub https://github.com/harvard-acc/gem5-aladdin/issues/47#issuecomment-1234858891, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADBSBD3RAR6G75DPIZYJ75LV4EWGVANCNFSM6AAAAAAQBOG3SQ . You are receiving this because you authored the thread.Message ID: @.***>
-- Sincerely Yours, Peter H. Chen
I'm still not sure I understand. gem5-aladdin doesn't support other architectures; if you want to simulate RISCV or ARM, you'll need to disable all of Aladdin's accelerator modeling capabilities, so then what are you going to use to feed into Aladdin's power models?
I was thinking to run twice. One for gem5 for RISCV, ARM and power and area from Aladdin. But it does not make sense. Fir X86, gem5-Aladdin can have all information I need for hypermaper MOO (multi-objectives optimization)
Customer need RISCV and ARM, I cannot use grm5-Aladdin anymore. I have use Gem5. Gem5 does not have area and power (may be only ARM only). I will need to use McPAT or cacti to find power and area. Or yosys for more accurate area.
Peter
On Thu, Sep 1, 2022 at 10:52 PM Sam Xi @.***> wrote:
I'm still not sure I understand. gem5-aladdin doesn't support other architectures; if you want to simulate RISCV or ARM, you'll need to disable all of Aladdin's accelerator modeling capabilities, so then what are you going to use to feed into Aladdin's power models?
— Reply to this email directly, view it on GitHub https://github.com/harvard-acc/gem5-aladdin/issues/47#issuecomment-1235094438, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADBSBD4DX4LJO5ZUQDCHJ2TV4GIZHANCNFSM6AAAAAAQBOG3SQ . You are receiving this because you authored the thread.Message ID: @.***>
-- Sincerely Yours, Peter H. Chen
You could run the simulation twice and just pick the parts from each that you care about (i.e. just the Aladdin part from the X86 simulation), but it would be a pretty crude approximation.
Is there a way to control