win32/Makefile: update compilation flags for Visual C++

[xenu]

Makefile contains the following comment from 2002:

# -O1 yields smaller code, which turns out to be faster than -O2

Visual C++ documentation says that -O2 "creates fast code", while -O1 "creates small code". It's unlikely that, after 20 years, -O2 still generates worse code. If it does, it should be reported to Microsoft.

Also added -Gw (optimize global data).

[bulk88]

I disagree with -O2, extra code growth every where. Our Perl_croak()/Perl_die() and "panic: " type fatal error branches, will be INLINED EVERYWHERE if you do that.

-Gw I def agree, it cut 70 bodies_by_type copies to 50 when I turned it on. Instead of -O2, add -Oi https://learn.microsoft.com/en-us/cpp/build/reference/oi-generate-intrinsic-functions?view=msvc-170

Inlined memcmp(ptr, "exit", 4) is push, push, push call, clean c stack, right now, -Oi makes that 1 CPU op, I shoudve added -Oi years ago to core.

[bulk88]

I disagree with -O2, extra code growth every where. Our Perl_croak()/Perl_die() and "panic: " type fatal error branches, will be INLINED EVERYWHERE if you do that.

-Gw I def agree, it cut 70 bodies_by_type copies to 50 when I turned it on. Instead of -O2, add -Oi https://learn.microsoft.com/en-us/cpp/build/reference/oi-generate-intrinsic-functions?view=msvc-170

Inlined memcmp(ptr, "exit", 4) is push, push, push call, clean c stack, right now, -Oi makes that 1 CPU op, I shoudve added -Oi years ago to core.

__forceinline macro support exists in P5 Core headers already, and that overrides all inline "cost analysis" by MSVC even on -O1, If there is a very very good justification, in very limited situations, just use the proper perl macro that already exists. Plus if you change -O1 to -O2, you are affecting all of CPAN.

edit: -O2 has a terrible side effect, that EVERY CONSTANT in machine code is extended to 4 bytes.

char ch = 0x1"

under -O1 that is 3 bytes of machine code, under MSVC -O2, that becomes 7 bytes every time.

[xenu]

I don't have the energy to argue with you.

I did this to partially address your complaints about the lack of inlining, as this change actually helps with e.g. newSV(iv|nv).

If you don't want it, that's fine by me. I don't care about Visual C++. Almost none of our users use it. Both Strawberry and ActivePerl are built using MinGW.