GNU as recursive/loop macro expected output

Question

In this assembly file below, the macro jump_table should automagically create ... a jump table to consecutively numbered labels like jump_0, jump_1, ... jump_<n>.

It seems there is no loop feature in as .macro directive and the official documentation seems to suggest what appears to be a recursive macro.

Anyway, this is the file:

.section .text
.align 4

.macro jump_table name init last
  j \name\()_\init
  .if \last > \init
  jump_table \name \init+1 \last
  .endif
.endm

jump_table jump 0 3

jump_0:
  nop
jump_1:
  nop
  nop
jump_2:
  nop
  nop
  nop
jump_3:
  nop
  nop
  nop
  nop

The resulting binary, once disassembled with objdump, yields this:

0000000000000000 <jump_0-0x1c>:
   0:   00000013            nop
   4:   00000013            nop
   8:   00000013            nop
   c:   0100006f            j   1c <jump_0>
  10:   00c0006f            j   1c <jump_0>
  14:   00a0006f            j   1e <jump_0+0x2>
  18:   0060006f            j   1e <jump_0+0x2>

000000000000001c <jump_0>:
  1c:   00000013            nop

0000000000000020 <jump_1>:
  20:   00000013            nop
  24:   00000013            nop

0000000000000028 <jump_2>:
  28:   00000013            nop
  2c:   00000013            nop
  30:   00000013            nop

0000000000000034 <jump_3>:
  34:   00000013            nop
  38:   00000013            nop
  3c:   00000013            nop
  40:   00000013            nop
  44:   00000013            nop
  48:   00000013            nop
  4c:   00000013            nop

The jump table looks weird to me: offsets are decreasing while I would expect them to increase. Moreover, labelling also looks weird.

If I understood correctly, I should expect something like:

  j   1c <jump_0>
  j   20 <jump_1>
  j   28 <jump_2>
  j   34 <jump_3>

I bet I am missing something and some explanation would help me greatly.

I am using riscv64-elf-as and riscv64-elf-objdump v2.44 on MacOS M4.

I think \init will not evaluate the actual number, but just the text with +1 appended, thus, you get j jump_0+1, j jump_0+1+1, etc (which the assembler then rounds to the next valid address, i.e., jump_0, jump_0, jump_0+2, jump_0+2, ... — chtz
– chtz, Commented May 14 at 9:11
Interesting... Do I need to use those "(...)" as per documentation? I was getting errors from as. It's not clear to me how those assembler directives evaluate stuff. — EnzoR
– EnzoR, Commented May 14 at 10:10
I don't know if that works directly. I know that the C/C++ preprocessor also will not simplify arithmetic expressions, probably that is is the same for as. — chtz
– chtz, Commented May 14 at 10:51

chtz · Accepted Answer · 2025-05-14 11:06:38Z

3

\init+1 will not be evaluated before it gets passed recursively, therefore your macro creates something like this:

j  jump_0
j  jump_0+1
j  jump_0+1+1
j  jump_0+1+1+1

Which the assembler appears to round to the next valid address, i.e., in your case:

j   1c <jump_0>
j   1c <jump_0>
j   1e <jump_0+0x2>
j   1e <jump_0+0x2>

If you enable .altmacro-mode, you can have expressions evaluated before they get passed as parameters, using %(expr):

.altmacro

.macro jump_table name init last
  j \name\()_\init
  .if \last > \init
  jump_table \name, %(\init+1), \last
  .endif
.endm

Godbolt-Demo: https://godbolt.org/z/hGjEG6TE1

answered May 14 at 11:06

chtz

19.4k5 gold badges30 silver badges64 bronze badges

Sign up to request clarification or add additional context in comments.

8 Comments

EnzoR May 14 at 11:47

It doesn't work for me (as v2.44). Please see here: pastebin.com/L25DrKTz . I get 3 entries, should be 2. I get only jump_0. Hex assembler seems not to match symbolic assembly.

chtz May 14 at 12:08

jump_table jump 0 2 should be 3 entries, shouldn't it? But I don't know if you need something else to make .altmacro work properly, or if MacOS has a weird version of as.

EnzoR May 14 at 12:13

Nope. It should be 2 as the test is "\last > \init" so 0 and 1 alone. And it should be 3 in your example. Right? Then I have no idea on how as works in general and on MacOS in particular. I only can use v2.44 there.

Peter Cordes May 14 at 13:20

@EnzoR: Works for me with clang -target riscv32 (on x86-64 Linux), producing the same results as chtz's Godbolt link. llvm-objdump -drwC shows the relocation entries in the .o, otherwise you just see placeholders. (IDK if that's helpful; I saw macOS and thought of clang, but then I realized you're using actual GNU binutils with the version number you specified as a cross-assembler since you're targeting RV32, not MachO64 AArch64 or x86-64.)

chtz May 14 at 14:55

@EnzoR Why do you expect 2 entries when you create an entry outside the .if block already? Even in your original question you were expecting 4 entries for jump_table jump 0 3.

|

Collectives™ on Stack Overflow

GNU as recursive/loop macro expected output

1 Answer 1

8 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

8 Comments

Your Answer

Sign up or log in

Post as a guest

Related