[Bug target/68803] gcc.vect/powerpc/20050603-3.c failures since r230167

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68803

--- Comment #6 from Martin Sebor <msebor at gcc dot gnu.org> ---
I would expect them to if they ran it in 64-bit mode since the test in 5.3
looks for the "inm" pattern which matches the rlwinm instruction in gcc's
output.  But in 5.3 and before the test was constrained to ilp32, and so it in
most configurations didn't run on powerpc64le (I don't have a 32-bit GCC
configured to see what it might do).  It's only after r230167 that the failures
started showing up on powerpc6le because the change enabled the test for lp64.

I also didn't run the test suite with 5.3.0 and only inspected the assembly for
the instructions the updated test looks for.

This is what it looks like:

$ /build/gcc-5.3.0/gcc/xgcc -B /build/gcc-5.3.0/gcc -O2 -S -Wall -o/dev/stdout
/src/gcc/trunk/gcc/testsuite/gcc.target/powerpc/20050603-3.c
        .file   "20050603-3.c"
        .machine power8
        .abiversion 2
        .section        ".toc","aw"
        .section        ".text"
        .section        ".toc","aw"
.LC0:
        .quad   b
        .section        ".text"
        .align 2
        .p2align 4,,15
        .globl rotins
        .type   rotins, @function
rotins:
0:      addis 2,12,.TOC.-0b@ha
        addi 2,2,.TOC.-0b@l
        .localentry     rotins,.-rotins
        addis 8,2,.LC0@toc@ha           # gpr load fusion, type long
        ld 8,.LC0@toc@l(8)
        rlwinm 3,3,0,8,11
        lwz 10,0(8)
        rlwinm 9,10,8,4,31
        rotlwi 9,9,24
        or 9,9,3
        stw 9,0(8)
        blr
        .long 0
        .byte 0,0,0,0,0,0,0,0
        .size   rotins,.-rotins
        .comm   b,8,8
        .ident  "GCC: (GNU) 5.3.0"
        .section        .note.GNU-stack,"",@progbits