Suggestion Use Flag MMX ,SSE for compile

[Zilog]

I Tom,
suggestion you of use mmx and sse flag in then compile option.
The emulator are more fast....at least in my pc.

Thanks a lot for your work.
By.

[leilei]

Probably placebo at this point. -march=amdfam10/-mtune=amdfam10 which has SSE and more doesn't make a faster PCem than -march=i686 for me.

[SarahWalker]

I'd want to see some numbers before I add this. My view up until now has been that there was extremely little performance to be gained from playing with the compiler flags like this.

[Zilog]

I'd want to see some numbers before I add this. My view up until now has been that there was extremely little performance to be gained from playing with the compiler flags like this.

I have utilized this parameters:
-march=i686 -fomit-frame-pointer -DDYNAREC -mmmx -msse

Problay that with the cpu more fast the different is small, but on cpu not more fast the difference are more.
My Cpu is Intel Core 7 m620 (Cpu of mobile computer, clock is more slow rispect a desktop cpu).

[startmenu]

I just use "-fprofile-use" to get a PGO build.

[leilei]

I just use "-fprofile-use" to get a PGO build.

I dont' see much performance gain on generating profiles either. march=amdfam10 doesn't help either, and made me a much slower build than a plain march=i686 build

[SarahWalker]

I tend to see about 10-15% benefit from generating profiles, which is why all the release builds are built with -fprofile-use.

[leilei]

Code: Select all

CFLAGS = -O3 -march=amdfam10 -fomit-frame-pointer -DDYNAREC -fprofile-arcs -ftest-coverage -fprofile-generate

Code: Select all

CFLAGS = -O3 -march=amdfam10 -fomit-frame-pointer -DDYNAREC -fprofile-use -flto 

and just running an entire TRIP run at PMMX233 infinite before going back to use the profiles

[Zilog]

Code: Select all

CFLAGS = -O3 -march=amdfam10 -fomit-frame-pointer -DDYNAREC -fprofile-arcs -ftest-coverage -fprofile-generate

Code: Select all

CFLAGS = -O3 -march=amdfam10 -fomit-frame-pointer -DDYNAREC -fprofile-use -flto 

and just running an entire TRIP run at PMMX233 infinite before going back to use the profiles

Ok for optimization, but if use -march=amdfam10 can create problem with intel platform?

[SarahWalker]

For the release builds I use

Code: Select all

CFLAGS = -O3 -march=i686 -fomit-frame-pointer -fprofile-generate
CFLAGS = -O3 -march=i686 -fomit-frame-pointer -fprofile-use