[Bug target/96768] New: -mpure-code produces switch tables for thumb-1

[Bug target/96768] New: -mpure-code produces switch tables for thumb-1

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96768

            Bug ID: 96768
           Summary: -mpure-code produces switch tables for thumb-1
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: clyon at gcc dot gnu.org
  Target Milestone: ---

As discussed in PR94538, -mpure-code produces switch tables for thumb-1.

int f2 (int x, int y)
{
  switch (x)
  {
    case 0: return y + 0;
    case 1: return y + 1;
    case 2: return y + 2;
    case 3: return y + 3;
    case 4: return y + 4;
    case 5: return y + 5;
  }
  return y;
}

Compiled with -O2 -mpure-code,
-mcpu=cortex-m0:
f2:
        cmp     r0, #5
        bhi     .L9
        movs    r2, #:upper8_15:#.LC0
        lsls    r2, #8
        adds    r2, #:upper0_7:#.LC0
        lsls    r2, #8
        adds    r2, #:lower8_15:#.LC0
        lsls    r2, #8
        adds    r2, #:lower0_7:#.LC0
        ldr     r2, [r2]
        lsls    r0, r0, #2
        ldr     r3, [r2, r0]
        mov     pc, r3
        .section        .rodata
        .align  2
.L4:
        .word   .L9
        .word   .L8
        .word   .L7
        .word   .L6
        .word   .L5
        .word   .L3
        .section .text,"0x20000006",%progbits
.L3:
        adds    r0, r1, #5
.L1:
        @ sp needed
        bx      lr
.L8:
        adds    r0, r1, #1
        b       .L1
.L7:
        adds    r0, r1, #2
        b       .L1
.L6:
        adds    r0, r1, #3
        b       .L1
.L5:
        adds    r0, r1, #4
        b       .L1
.L9:
        movs    r0, r1
        b       .L1


For cortex-m23:
f2:
        cmp     r0, #5
        bhi     .L9
        movw    r2, #:lower16:.LC0
        movt    r2, #:upper16:.LC0
        ldr     r2, [r2]
        lsls    r0, r0, #2
        ldr     r3, [r2, r0]
        mov     pc, r3


For reference, for cortex-m3:
f2:
        cmp     r0, #3
        beq     .L2
        ble     .L11
        cmp     r0, #4
        beq     .L7
        cmp     r0, #5
        bne     .L9
        adds    r0, r1, #5
        bx      lr
.L11:
        cmp     r0, #1
        beq     .L4
        cmp     r0, #2
        bne     .L9
        adds    r0, r1, #2
        bx      lr
.L2:
        adds    r0, r1, #3
        bx      lr
.L7:
        adds    r0, r1, #4
        bx      lr
.L4:
        adds    r0, r1, #1
        bx      lr
.L9:
        mov     r0, r1
        bx      lr

[Bug target/96768] -mpure-code produces switch tables for thumb-1

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96768

--- Comment #1 from Christophe Lyon <clyon at gcc dot gnu.org> ---
This is related to comments 10,11,14,15 and 16 in the original PR94538.

In comment 14, Wilco suggested: "The best option is to do the same as
Cortex-M3: just switch off branch tables altogether and fall back to
compare+branch. That completely avoids loading data from flash and is always
smaller than emitting 32-bit tables."

[Bug target/96768] -mpure-code produces switch tables for thumb-1

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96768

--- Comment #2 from Christophe Lyon <clyon at gcc dot gnu.org> ---
Send patch proposal:
https://gcc.gnu.org/pipermail/gcc-patches/2020-August/552798.html

[Bug target/96768] -mpure-code produces switch tables for thumb-1

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96768

Christophe Lyon <clyon at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-08-27
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED

[Bug target/96768] -mpure-code produces switch tables for thumb-1

[rearnsha at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96768

--- Comment #3 from Richard Earnshaw <rearnsha at gcc dot gnu.org> ---
Note that the switch table is in the .rodata section, so that's not a problem.

[Bug target/96768] -mpure-code produces switch tables for thumb-1

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96768

--- Comment #4 from Christophe Lyon <clyon at gcc dot gnu.org> ---
That's what I replied in the original PR94538, but Wilco said the best option
was to turn off switch tables:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94538#c14

See also another comment from him:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94538#c16 related to immediate
loads from rodata/bss.

[Bug target/96768] -mpure-code produces switch tables for thumb-1

[wilco at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96768

Wilco <wilco at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |wilco at gcc dot gnu.org

--- Comment #5 from Wilco <wilco at gcc dot gnu.org> ---
(In reply to Christophe Lyon from comment #4)
> That's what I replied in the original PR94538, but Wilco said the best
> option was to turn off switch tables:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94538#c14
> 
> See also another comment from him:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94538#c16 related to immediate
> loads from rodata/bss.

Yes, switching off switch tables is obviously the best option - just look at
the m0 vs m3 code above. It's the fastest and smallest, and is also "pure code"
since there are no extra literals generated anywhere.

In AArch32 switch tables are very inefficient. An easy improvement would be to
copy the AArch64 case_values_threshold to disable tables if there are fewer
than 16 cases.