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

GH-119726: Use LDR for AArch64 trampolines (GH-121001)

This commit is contained in:
Diego Russo 2024-07-01 23:52:33 +01:00 committed by GitHub
parent 1dc9a4f6b2
commit 966260841b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 11 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Misc/NEWS.d/next/Core and Builtins/2024-06-25-16-26-44.gh-issue-119726.WqvHxB.rst Normal file
View File

@ -0,0 +1,2 @@
Improve the speed and memory use of C function calls from JIT code on AArch64.
Patch by Diego Russo

31
Tools/jit/_stencils.py
View File

@ -204,33 +204,20 @@ class Stencil:
return return
self.disassembly += [ self.disassembly += [
f"{base + 4 * 0:x}: d2800008 mov x8, #0x0", f"{base + 4 * 0:x}: 58000048 ldr x8, 8",
f"{base + 4 * 0:016x}: R_AARCH64_MOVW_UABS_G0_NC {hole.symbol}", f"{base + 4 * 1:x}: d61f0100 br x8",
f"{base + 4 * 1:x}: f2a00008 movk x8, #0x0, lsl #16", f"{base + 4 * 2:x}: 00000000",
f"{base + 4 * 1:016x}: R_AARCH64_MOVW_UABS_G1_NC {hole.symbol}", f"{base + 4 * 2:016x}: R_AARCH64_ABS64 {hole.symbol}",
f"{base + 4 * 2:x}: f2c00008 movk x8, #0x0, lsl #32", f"{base + 4 * 3:x}: 00000000",
f"{base + 4 * 2:016x}: R_AARCH64_MOVW_UABS_G2_NC {hole.symbol}",
f"{base + 4 * 3:x}: f2e00008 movk x8, #0x0, lsl #48",
f"{base + 4 * 3:016x}: R_AARCH64_MOVW_UABS_G3 {hole.symbol}",
f"{base + 4 * 4:x}: d61f0100 br x8",
] ]
for code in [ for code in [
0xD2800008.to_bytes(4, sys.byteorder), 0x58000048.to_bytes(4, sys.byteorder),
0xF2A00008.to_bytes(4, sys.byteorder),
0xF2C00008.to_bytes(4, sys.byteorder),
0xF2E00008.to_bytes(4, sys.byteorder),
0xD61F0100.to_bytes(4, sys.byteorder), 0xD61F0100.to_bytes(4, sys.byteorder),
0x00000000.to_bytes(4, sys.byteorder),
0x00000000.to_bytes(4, sys.byteorder),
]: ]:
self.body.extend(code) self.body.extend(code)
for i, kind in enumerate( self.holes.append(hole.replace(offset=base + 8, kind="R_AARCH64_ABS64"))
[
"R_AARCH64_MOVW_UABS_G0_NC",
"R_AARCH64_MOVW_UABS_G1_NC",
"R_AARCH64_MOVW_UABS_G2_NC",
"R_AARCH64_MOVW_UABS_G3",
]
):
self.holes.append(hole.replace(offset=base + 4 * i, kind=kind))
self.trampolines[hole.symbol] = base self.trampolines[hole.symbol] = base
def remove_jump(self, *, alignment: int = 1) -> None: def remove_jump(self, *, alignment: int = 1) -> None: