How are you dealing with binaries?

[Youlean]

If you make plugin with a name "Plugin.vst" do you make it "Plugin64.vst" for the 64bit, or leave it the same? If you rename it what should I change, bundle id?

[Tale]

If you are using default WDL-OL (or even my WDL/IPlug) settings, then the OSX binaries will contain both 32 and 64-bit i.e. there are no separate 32/64-bit bundles.

[Youlean]

Originally Posted by Tale

If you are using default WDL-OL (or even my WDL/IPlug) settings, then the OSX binaries will contain both 32 and 64-bit i.e. there are no separate 32/64-bit bundles.

Really? But why there is an option to select target architecture (32bit/64bit)?

[Tale]

Originally Posted by Youlean

Really? But why there is an option to select target architecture (32bit/64bit)?

Because you can build only for a single architecture, e.g. if you no longer want to support 32-bit. But by default Architectures is set to Universal (32/64-bit Intel), which builds a universal binary.

[jack461]

Originally Posted by Youlean

Really? But why there is an option to select target architecture (32bit/64bit)?

This puzzles me too. While I can generate the 32/64 bit plug-in, in a 64-bit scheme, Xcode don't let me run REAPER, but only REAPER64:

the specified architecture 'x86_64' is not compatible with 'i386-apple-' in '/Applications/REAPER.app/Contents/MacOS/REAPER'

So I have one VST-32 scheme, with My Mac (32-bit) and REAPER, and one VST-64 scheme, with My Mac (64-bit) and REAPER64. This is just for fast testing from Xcode, because the generated VST is the same in both cases.

Of course, if I start REAPER or REAPER64 independently of Xcode, the VST works correctly in both applications, and displays in its window "32b" or "64b", based on the size of (void*).

char txt[24];
sprintf(txt, "V%d.%d.%d%s%s%s %s", (PLUG_VER>>16)&0xff, (PLUG_VER>>8)&0xff, PLUG_VER&0xff,
            (bldflgs ? "-" : ""), (bldflgs&1 ? "D" : ""), (bldflgs&2 ? "M" : ""),
            (sizeof(void*)==4?"32b":"64b"));
pGraphics->AttachControl(new ITextControl(this, tmpRectTT, &textPropsTT, txt));

By the way, It seems (very informally) that the 64bits version runs 10 to 20% faster than the 32bits versions (with Xcode 6.2, on an iMac 3.5 GHz Core I7), according to REAPER Performance Meter. Is there some good reason for this ?

[Tale]

Originally Posted by jack461

By the way, It seems (very informally) that the 64bits version runs 10 to 20% faster than the 32bits versions (with Xcode 6.2, on an iMac 3.5 GHz Core I7), according to REAPER Performance Meter. Is there some good reason for this ?

x86-64 automatically implies SSE2, and proper alignment of doubles. I would kinda expect the same performance from i386, provided you enable SSE2, and properly align your doubles.

[jack461]

Originally Posted by Tale

x86-64 automatically implies SSE2, and proper alignment of doubles. I would kinda expect the same performance from i386, provided you enable SSE2, and properly align your doubles.

Tale, thanks for this idea. I'll try to find where to put this option.

[Tale]

BTW, if you do a lot of 64-bit int caclulations, then x64 will also be faster than x86.

[jack461]

Originally Posted by Tale

BTW, if you do a lot of 64-bit int caclulations, then x64 will also be faster than x86.

Tale, I'm not sure I understand your remark.
I'm using in Xcode "Architectures" with the value "Universal (32/64-bit Intel)", and "Valid Architectures" with the value "i386 x86_64".
Do you say that I should replace this by "x64", or something like this ?

[Tale]

Originally Posted by jack461

Do you say that I should replace this by "x64", or something like this ?

No, I was just being lazy... On Windows 32-bit is usually called "x86" and 64-bit "x64" vs i386 and x86-64 on OSX. My remark about ints (and about SSE2) is valid on both Windows and OSX. Note that AFAIK double alignment is an issue only on OSX (I guess because the Microsoft compiler automatically properly aligns doubles).

[jack461]

Originally Posted by Tale

No, I was just being lazy... On Windows 32-bit is usually called "x86" and 64-bit "x64" vs i386 and x86-64 on OSX. My remark about ints (and about SSE2) is valid on both Windows and OSX. Note that AFAIK double alignment is an issue only on OSX (I guess because the Microsoft compiler automatically properly aligns doubles).

OK, thanks ! (As you can guess, I know little about Windows :-)

[Youlean]

Also on 64bit architecture you can move 64bit of data in one cycle...