ghc-HEAD (only in -O -fllvm aka 'optllvm') prunes yield in -fno-omit-yields mode

Found on ./validate --slow today:

Unexpected failures:
   concurrent/should_run  T367_letnoescape [bad exit code] (optllvm)

# run manually:
$ LANG=C make fulltest stage=2 TEST=T367_letnoescape WAY=optllvm

=====> T367_letnoescape(optllvm) 2419 of 4083 [0, 0, 0] 
cd ./concurrent/should_run && '/home/st/fun/ghc/inplace/bin/ghc-stage2' -fforce-recomp -dcore-lint -dcmm-lint -dno-debug-output -no-user-package-db -rtsopts -fno-ghci-history -o T367_letnoescape T367_letnoescape.hs -O -fllvm -fno-omit-yields  >T367_letnoescape.comp.stderr 2>&1
cd ./concurrent/should_run && ./T367_letnoescape    </dev/null >T367_letnoescape.run.stdout 2>T367_letnoescape.run.stderr
Timeout happened...killing process...
Wrong exit code (expected 0 , actual 99 )
Stdout:

Stderr:

*** unexpected failure for T367_letnoescape(optllvm)

I've loaded test in gdb and manually interrupted it:

^C
Program received signal SIGINT, Interrupt.
0x0000000000407110 in r5Wp_info ()
(gdb) disassemble 
Dump of assembler code for function r5Wp_info:
   0x00000000004070f8 <+0>:     mov    0x358(%r13),%rax
   0x00000000004070ff <+7>:     dec    %r14
   0x0000000000407102 <+10>:    data32 data32 data32 data32 nopw %cs:0x0(%rax,%rax,1)

=> 0x0000000000407110 <+24>:    inc    %r14
   0x0000000000407113 <+27>:    test   %rax,%rax
   0x0000000000407116 <+30>:    jne    0x407110 <r5Wp_info+24>

   0x0000000000407118 <+32>:    mov    -0x8(%r13),%rax
   0x000000000040711c <+36>:    mov    $0x8e0690,%ebx
   0x0000000000407121 <+41>:    jmpq   *%rax
End of assembler dump

Here we see 3 instructions-long dead loop.

llvm-3.4.2, amd64-linux.

Looks like yield code was not emitted or was moved away.