Optimisation 01: The predicated instructions

[shahab-vahedi]

Having the following code:

long long func(long long num, unsigned n)
{
  return (num << n);
}

results in an assembly like:

$ arc-elf32-gcc -O3 func.c -S -o -
        .file   "func.c"
        .cpu HS
        .arc_attribute Tag_ARC_PCS_config, 2
        .arc_attribute Tag_ARC_ABI_rf16, 0
        .arc_attribute Tag_ARC_ABI_pic, 0
        .arc_attribute Tag_ARC_ABI_tls, 0
        .arc_attribute Tag_ARC_ABI_sda, 2
        .arc_attribute Tag_ARC_ABI_exceptions, 0
        .arc_attribute Tag_ARC_CPU_variation, 2
        .section        .text
        .align 4
        .global func
        .type   func, @function
func:
        not_s   r12,r2
        lsr_s   r3,r0
        asl_s   r1,r1,r2
        asl_s   r0,r0,r2
        lsr_s   r3,r3,r12
        btst_s  r2,5
        or_s    r1,r1,r3
        mov_s   r2,r0   ;4
        mov.eq  r2,r1
        mov_s   r1,r2   ;4
        j_s.d   [blink]
        sub.ne  r0,r0,r0
        .size   func, .-func
        .ident  "GCC: (ARCompact/ARCv2 ISA elf32 toolchain) 10.2.0"
        .section        .note.GNU-stack,"",@progbits

But there are rooms for improvement:

  btst_s  r2,5      \
  or_s    r1,r1,r3  |            btst_s  r2,5
  mov_s   r2,r0     |========>   or_s    r1,r1,r3
  mov.eq  r2,r1     |            mov.ne  r1,r0
  mov_s   r1,r2     /