Michal Schulz has published a new update development in which it addresses one of the most frequently asked questions among users of Emu68Whether it would be possible to run ARM code directly through the system. He explains that he always answered no, mainly to avoid further dividing the community. Amiga and generate executables exclusive to Emu68.
However, he acknowledges that motherboards with ARM CPUs already exist and that, if this possibility were to become available at some point, it would ideally be done in an open manner. Schulz even mentions potential variants such as AArch32 , AArch64 , Thumb2 mode , and little-endian or big-endian configurations , but ultimately concludes that the ARM landscape is not as stable as it might seem.
Instead, they've decided to move in a different direction: adding PowerPC support to Emu68 . They clarify that the name remains the same—the M68k isn't being dropped, but rather PPC is simply being added. The idea, they explain, arose years ago and has gradually solidified over time, since both architectures are RISC and share certain principles.
To achieve this, Schulz took the main components of Emu68—the translator, the caches, and the main JIT loop—and began experimenting. In a short time, he managed to execute nearly 90% of the PowerPC's complete instructions, condensing the entire translator into about 7.000 lines of C++ code.
During initial testing it achieved remarkable results: around 3.600 MIPS , with an average performance of 1,8 PowerPC instructions per ARM cycle , writing to cached memory with 16-bit operations.
Schulz explains that these values are due to the efficiency of the GCC compiler in generating PowerPC code, optimized to take advantage of the registers available in a load/store architecture . Since Emu68 uses a simple JIT compiler, it avoids heavy optimizations, but in this case they are not necessary because the PowerPC code itself is already well-organized.
The new PowerPC translator operates completely independently of the M68k core , running on a separate CPU core. This ensures that the 68k's performance remains unaffected. If the user doesn't need PowerPC, they can simply ignore it: the additional core remains inactive.
Schulz justifies this decision by explaining that running both architectures on the same kernel would lead to performance losses and context complications. He further details how he maps the registers: the 16 most frequently used integers and 14 most frequently used floating-point numbers are kept in AArch64 registers, while the rest are loaded dynamically under an LRU policy. Temporary integer registers are stored in SIMD spaces, allowing for fast crossovers with a penalty of only 2-3 cycles.
It also points out that context switching, a long-standing problem in real-world PPC accelerators due to a lack of cache consistency, is greatly simplified under Emu68 , as the AArch64 cores maintain automatic consistency. In practice, a single memory barrier is sufficient to synchronize the data.
Currently, the entire PowerPC component is virtually complete. The emulator supports interrupts between both sides (M68k and PPC) and has its own ROM that starts with the system, awaiting its companion, powerpc.library , which is still under development. This library aims for compatibility with WarpOS in both API and ABI, in collaboration with Hedeon , the manufacturer of PowerPC PCI hardware such as the Sonnet Crescendo 7200 cards.
The ROM includes a dedicated scheduler for PPC tasks, allowing them to run in parallel with M68k processes. Although floating-point emulation is in an early stage, tests have already been conducted with an adapted version of the Buddhabrot fractal generator , which offers significantly faster execution times than on M68k, as well as an approximate 2x improvement in Dhrystone 2.1 tests.
Schulz concludes by mentioning that this text was intended to be part of the third installment of “What's New in 1.1,” but the topic ended up growing more than anticipated. He adds that he is exhausted after several colds and needs a break before moving forward. He anticipates that he will offer more information in the coming days. Emu68 1.1, the support PPC, the meeting Amiga40 and a significant technical change: the progressive migration of the project to C++.


Schulz is incredible!!!! He never ceases to amaze us.