Re: [PATCH 2/2] BaseTools/tools_def AARCH64: avoid SIMD register in XIP code

[Laszlo Ersek <lersek@redhat.com>]

On 07/13/17 19:31, Leif Lindholm wrote:
> How are you ending up with all of those -O3?

Sorry for butting in, but personally I consider -O3 "suicidal".

The only -O3's I can currently see in BaseTools are:

> Conf/tools_def.template:DEBUG_CLANG38_IA32_DLINK2_FLAGS     = DEF(GCC5_IA32_DLINK2_FLAGS) -O3
> Conf/tools_def.template:RELEASE_CLANG38_IA32_DLINK2_FLAGS   = DEF(GCC5_IA32_DLINK2_FLAGS) -O3
> Conf/tools_def.template:DEBUG_CLANG38_X64_DLINK2_FLAGS     = DEF(GCC5_X64_DLINK2_FLAGS) -O3
> Conf/tools_def.template:RELEASE_CLANG38_X64_DLINK2_FLAGS   = DEF(GCC5_X64_DLINK2_FLAGS) -O3

Since I don't use CLANG, I don't mind these, but for gcc, I would.

In the past, Liming once suggested -O3 for the build tools themselves
(via -Ofast) -- meaning full well, of course! --, and I protested:

  http://mid.mail-archive.com/ed237dbc-25f0-d0a9-3252-9a503de67ee4@redhat.com

And so we have commit 202726b3ceb3 ("BaseTools Makefile: Enable O2
option for GCC tool chain", 2016-09-29) today.

... I realize my email can appear as FUD, but I think there's a reason
most distros build their packages with -O2. Even upstream Linux uses -O2
generally (the only mention of -O3 I find right now in "Makefile" is
about problems it causes with gcc-4.8).

Thanks
Laszlo