i386 pcpu: fix clobbers, suppress warnings, and clean up

[rlibby]

Summary

• Add missing cc clobber to __PCPU_ADD (which is currently unused).
• Allow the compiler the opportunity to marginally improve code generation from __PCPU_PTR by letting it figure out how to do the add (also removing the addition fixes a missing cc clobber).
• Quiet gcc -Warray-bounds by using constant operands instead of bogus memory references.
• Remove the struct s s temporaries, just cast through the type.

[rlibby]

Test patch used to examine codegen:

0001-XXX-pcpu-demo-functions.patch1 KBDownload

(I also #if 0'd the __SEG_GS section of amd64's pcpu.h in order to check the changes.)

Disassembly before the patch for amd64 clang, amd64 gcc, i386 clang, and i386 gcc:

pcpu_demo.before-all.disas13 KBDownload

Disassembly after both D45826 and D45827 for amd64 clang, amd64 gcc, i386 clang, and i386 gcc:

pcpu_demo.after-all.disas13 KBDownload

diff -u1000 of the above before/after disassemblies:

pcpu_demo.before-after.diff15 KBDownload

[rlibby]

By the way, I thought about also making these strict-aliasing safe with a union. It's straightforward and also gets rid of the casts. However, since we explicitly build with -fno-strict-aliasing, there's not much motivation. But in case you'd prefer that, I can tweak these patches.

[kib]

In D45826#1044991, @rlibby wrote:

By the way, I thought about also making these strict-aliasing safe with a union. It's straightforward and also gets rid of the casts. However, since we explicitly build with -fno-strict-aliasing, there's not much motivation. But in case you'd prefer that, I can tweak these patches.

Why would a special treat for aliasing needed? There is the "memory" clobber.

[rlibby]

In D45826#1045245, @kib wrote:

In D45826#1044991, @rlibby wrote:

By the way, I thought about also making these strict-aliasing safe with a union. It's straightforward and also gets rid of the casts. However, since we explicitly build with -fno-strict-aliasing, there's not much motivation. But in case you'd prefer that, I can tweak these patches.

Why would a special treat for aliasing needed? There is the "memory" clobber.

Yes, I think you're right that the memory clobbers avoid any actual strict aliasing safety issue.

Gcc would still issue a diagnostic (despite the clobbers) if we didn't use -fno-strict-aliasing, but we do, so, it would be an exercise in suppressing a diagnostic that's not being issued. And I agree that sounds pointless.