Skip to content

Commit

Permalink
Relax GOT indirection into PC-relative addressing
Browse files Browse the repository at this point in the history
Some psABIs define the linker optimizations to relax a GOT load into a
PC-relative address materialization. AArch64 [1] allows the linker to
rewrite ADRP+ADD into ADR. x86-64 does the same thing with the
`R_X86_64_GOTPCRELX` and `R_X86_64_REX_GOTPCRELX` relocations.

In our case, we have lots of AUIPC+LD instruction pairs to load an
address from the GOT in our RISC-V programs because `la` assembly pseudo
instruction is expanded to that instruction pair. If the PC-relative
address loaded by the instruction pair is a link-time constant, we can
rewrite the instructions with AUIPC+ADDI to directly materialize the
value into a register, which eliminates one memory load.

[1] https://github.com/ARM-software/abi-aa/blob/844a79fd4c77252a11342709e3b27b2c9f590cf1/aaelf64/aaelf64.rst#relocation-optimization
  • Loading branch information
rui314 committed Oct 24, 2023
1 parent df8b80b commit e89046a
Showing 1 changed file with 74 additions and 0 deletions.
74 changes: 74 additions & 0 deletions riscv-elf.adoc
Original file line number Diff line number Diff line change
Expand Up @@ -1573,6 +1573,80 @@ optimize code size and performance of the symbol accessing.

NOTE: Tag_RISCV_x3_reg_usage is treated as 0 if it is not present.

==== GOT load relaxation

Target Relocation:: R_RISCV_GOT_HI20, R_RISCV_PCREL_LO12_I

Description:: This relaxation can relax a GOT indirection into load
immediate or PC-relative addressing. This relaxation is intended to
optimize the `lga` assembly pseudo-instruction (and thus `la` for
PIC objects), which loads a symbol's address from a GOT entry with
an `auipc` + `l[w|d]` instruction pair.

Condition::
- Both `R_RISCV_GOT_HI20` and `R_RISCV_PCREL_LO12_I` are marked with
`R_RISCV_RELAX`.

- The symbol pointed to by `R_RISCV_PCREL_LO12_I` is at the location to
which `R_RISCV_GOT_HI20` refers.

- If the symbol is absolute, its address is within `0x0` ~ `0x7ff` or
`0xfffffffffffff800` ~ `0xffffffffffffffff` for RV64 and
`0xfffff800` ~ `0xffffffff` for RV32.
Note that an undefined weak symbol satisfies this condition because
such a symbol is handled as if it were an absolute symbol at address 0.

- If the symbol is relative, it's bound at link time to be within the
object. It should not be of the GNU ifunc type. Additionally, the offset
between the location to which `R_RISCV_GOT_HI20` refers and the target
symbol should be within a range of +-2GiB.

Relaxation::
- The `auipc` instruction associated with `R_RISCV_GOT_HI20` can be
removed if the symbol is absolute.

- The instruction or instructions associated with `R_RISCV_PCREL_LO12_I`
can be rewritten to either `c.li` or `addi` to materialize the symbol's
address directly in a register.

- If this relaxation eliminates all references to the symbol's GOT slot,
the linker may opt not to create a GOT slot for that symbol.

Example::
+
--
Relaxation candidate:
[,asm]
----
label:
auipc tX, 0 # R_RISCV_GOT_HI20 (symbol), R_RISCV_RELAX
l[w|d] tY, 0(tX) # R_RISCV_PCREL_LO12_I (label), R_RISCV_RELAX
----

Relaxation result (absolute symbol whose address can be represented as
a 6-bit signed integer and if the RVC instruction is permitted):

[,asm]
----
c.li tY, <symbol-value>
----

Relaxation result (absolute symbol whose address cannot be represented
as a 12-bit signed integer or if the RVC instruction is not permitted):

[,asm]
----
addi tY, zero, <symbol-value>
----

Relaxation result (relative symbol):
[,asm]
----
auipc tX, <hi>
addi tY, tX, <lo>
----
--

==== Zero-page Relaxation

Target Relocation:: R_RISCV_HI20, R_RISCV_LO12_I, R_RISCV_LO12_S
Expand Down

0 comments on commit e89046a

Please sign in to comment.