Mirror
/
cpython
mirror of https://github.com/python/cpython
4
1
Fork 0
Code Issues Releases Wiki Activity

GH-119726: Emit AArch64 trampolines out-of-line (GH-121280)

This commit is contained in:
Diego Russo 2024-07-03 14:22:21 +01:00 committed by GitHub
parent 26d24eeb90
commit 84512c0e7f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Misc/NEWS.d/next/Core and Builtins/2024-06-28-18-34-49.gh-issue-119726.Fjv_Ab.rst Normal file
View File

@ -0,0 +1,2 @@
Optimize code layout for calls to C functions from the JIT on AArch64.
Patch by Diego Russo.

14
Tools/jit/_stencils.py
View File

@ -184,7 +184,7 @@ class Stencil:
self.disassembly.append(f"{offset:x}: {' '.join(['00'] * padding)}")
self.body.extend([0] * padding)
def emit_aarch64_trampoline(self, hole: Hole, alignment: int) -> None:
def emit_aarch64_trampoline(self, hole: Hole, alignment: int) -> Hole:
"""Even with the large code model, AArch64 Linux insists on 28-bit jumps."""
assert hole.symbol is not None
reuse_trampoline = hole.symbol in self.trampolines
@ -194,14 +194,10 @@ class Stencil:
else:
self.pad(alignment)
base = len(self.body)
where = slice(hole.offset, hole.offset + 4)
instruction = int.from_bytes(self.body[where], sys.byteorder)
instruction &= 0xFC000000
instruction |= ((base - hole.offset) >> 2) & 0x03FFFFFF
self.body[where] = instruction.to_bytes(4, sys.byteorder)
new_hole = hole.replace(addend=base, symbol=None, value=HoleValue.DATA)
if reuse_trampoline:
return
return new_hole
self.disassembly += [
f"{base + 4 * 0:x}: 58000048 ldr x8, 8",
@ -219,6 +215,7 @@ class Stencil:
self.body.extend(code)
self.holes.append(hole.replace(offset=base + 8, kind="R_AARCH64_ABS64"))
self.trampolines[hole.symbol] = base
return new_hole
def remove_jump(self, *, alignment: int = 1) -> None:
"""Remove a zero-length continuation jump, if it exists."""
@ -294,8 +291,9 @@ class StencilGroup:
in {"R_AARCH64_CALL26", "R_AARCH64_JUMP26", "ARM64_RELOC_BRANCH26"}
and hole.value is HoleValue.ZERO
):
self.code.emit_aarch64_trampoline(hole, alignment)
new_hole = self.data.emit_aarch64_trampoline(hole, alignment)
self.code.holes.remove(hole)
self.code.holes.append(new_hole)
self.code.remove_jump(alignment=alignment)
self.code.pad(alignment)
self.data.pad(8)