Repository navigation
The code generated by the JIT for loading 16 bit and 32 bit constants is inefficient. #140683
Description
This affects both x86-64 and AArch64.
Most guard micro-ops are checking against 16 or 32 bit version numbers, so this is an important use case.
The code we generate is too large and, on AArch64, performs extra memory accesses.
| Constant | x86-64 now | x86-64 ideal | AArch64 now | AArch64 ideal |
|---|---|---|---|---|
| 16 bit | 1 10 byte instruction | 1 5 byte instruction | 2 instructions + 1 memory load | 1 instruction |
| 32 bit | 1 10 byte instruction | 1 5 byte instruction | 2 instructions + 1 memory load | 2 instructions |
Activity
brandtbucher commentedon Oct 27, 2025
Is the plan to handle this in the textual assembly transformation step? Or by relaxing the relocations at runtime?
markshannon commentedon Oct 28, 2025
I think this needs to handled all the way through the tools chain, including the code generator to avoid Clang generating the load instructions. Or it might be possible to fix that with different compiler flags. I'll need to experiment. Have you given this any thought?
GH-140683: JIT: Improve machine code for loading smaller constants on…
pythonGH-140683: JIT: Improve machine code for loading smaller consta…
sergey-miryanov commentedon May 4, 2026
Completed via #142511