LLVM 6: Win32 inline assembly regression

[kinke]

core.cpuid fails to build on Win32 with LLVM 6.0.0 and enabled optimizations, with symptoms of issue #2257.

I tried replacing that DMD-style inline asm with LLVM asm:

            import ldc.llvmasm;
            __asm(`mov %eax, 2
                   cpuid
                   mov $0, %eax
                   mov $1, %ebx
                   mov $2, %ecx
                   mov $3, %edx`, "=*m,=*m,=*m,=*m,~{eax},~{ebx},~{ecx},~{edx}",
                   &a[0], &a[1], &a[2], &a[3]);

That makes it work with enabled optimizations, but debug druntime fails to compile with inline assembly requires more registers than available. That error vanishes when dropping all clobbers...

What's strange is that it can apparently still be compiled fine for 32-bit Linux and OSX, the CircleCI multilib jobs with preliminary LLVM 6.0 are green.

[ZILtoid1991]

I'm getting errors like this one:

 <inline asm>:6:2: error: assembler label 'L7CPUblit9composing11blitter8bit_eightpixel' can not be   undefined
         jl L7CPUblit9composing11blitter8bit_eightpixel

This is the same on both Win32 and Win64, have not tried to build for Linux or OSX.

The project in question: https://github.com/ZILtoid1991/CPUblit

[kinke]

@ZILtoid1991: That's a separate issue as that 5k lines asm monster errors out with the same errors for LDC 1.8 and LLVM 5.0.1. Please try to come up with a nice small testcase and file a new issue.

[dnadlinger]

@ZILtoid1991: See #2667 for that issue.

[kinke]

Worked around in #2770.