Unicore
Unicore is the name of a computer instruction set architecture designed by the Microprocessor Research and Development Center (MPRC) of Peking University in the PRC. The computer built on this architecture is called the Unity-863.[1] The CPU is integrated into a fully functional SoC to make a PC-like system.[2]
- For the grid computing middleware, see UNICORE.
| Designer | Microprocessor Research and Development Center |
|---|---|
| Bits | 32-bit |
| Introduced | 1999 |
| Design | RISC |
| Encoding | Fixed |
| Branching | Condition code |
| Endianness | Little |
| Page size | 4 KiB |
| Registers | |
| General purpose | 31 |
| Floating point | 32 |
The processor is very similar to the ARM architecture, but uses a different instruction set.[3]
It is supported by the Linux kernel as of version 2.6.39.[4] Support will be removed in Linux kernel version 5.9 as nobody seems to maintain it and the code is falling behind the rest of the kernel code and compiler requirements.[5]
Instruction set
The instructions are almost identical to the standard ARM formats, except that conditional execution has been removed, and the bits reassigned to expand all the register specifiers to 5 bits.[6][7] Likewise, the immediate format is 9 bits rotated by a 5-bit amount (rather than 8 bit rotated by 4), the load/store offset sizes are 14 bits for byte/word and 10 bits for signed byte or half-word. Conditional moves are provided by encoding the condition in the (unused by ARM) second source register field Rn for MOV and MVN instructions.
| Description | ||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 0 | 0 | opcode | S | Rn | Rd | shift | 0 | Sh | 0 | Rm | ALU operation, Rd = Rn op Rm shift #shift | ||||||||||||||||||||
| 0 | 0 | 0 | opcode | S | Rn | Rd | Rs | 0 | Sh | 1 | Rm | ALU operation, Rd = Rn op Rm shift Rs | ||||||||||||||||||||
| 0 | 0 | 1 | opcode | S | Rn | Rd | shift | imm9 | ALU operation, Rd = Rn op #imm9 ROTL #shift | |||||||||||||||||||||||
| 0 | 1 | 0 | P | U | B | W | L | Rn | Rd | shift | 0 | Sh | 0 | Rm | Load/store Rd to address Rn ± Rm shift #shift | |||||||||||||||||
| 0 | 1 | 1 | P | U | B | W | L | Rn | Rd | offset14 | Load/store Rd to address Rn ± offset14 | |||||||||||||||||||||
| 1 | 0 | 0 | P | U | S | W | L | Rn | Bitmap high | 0 | 0 | H | Bitmap low | Load/store multiple registers | ||||||||||||||||||
| 1 | 0 | 1 | cond | L | offset24 | Branch (and link) if condition true | ||||||||||||||||||||||||||
| 1 | 1 | 0 | Coprocessor (FPU) instructions | |||||||||||||||||||||||||||||
| 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | Trap number | Software interrupt | |||||||||||||||||||||||
| 0 | 0 | 0 | 0 | 0 | 0 | A | S | Rn | Rd | Rs | 1 | 0 | 0 | 1 | Rm | Multiply, Rd = Rm * Rs (+ Rn) | ||||||||||||||||
| 0 | 0 | 0 | 1 | 0 | 0 | 0 | L | 11111 | 11111 | 00000 | 1 | 0 | 0 | 1 | Rm | Branch and exchange (BX, BLX) | ||||||||||||||||
| 0 | 1 | 0 | P | U | 0 | W | L | Rn | Rd | 00000 | 1 | S | H | 1 | Rm | Load/store Rd to address Rn ± Rm (16-bit) | ||||||||||||||||
| 0 | 1 | 0 | P | U | 1 | W | L | Rn | Rd | imm_hi | 1 | S | H | 1 | imm_lo | Load/store Rd to address Rn ± #imm10 (16-bit) | ||||||||||||||||
The meaning of various flag bits (such as S=1 enables setting the condition codes) is identical to the ARM instruction set. The load/store multiple instruction can only access half of the register set, depending on the H bit. If H=0, the 16 bits indicate R0–R15; if H=1, R16–R31.
References
- "Introduction to MPRC". Microprocessor Research and Develop Center, Peking University.
- Xu Cheng; Xiaoyin Wang; Junlin Lu; Jiangfang Yi; Dong Tong; Xuetao Guan; Feng Liu; Xianhua Liu; Chun Yang; Yi Feng (March 2010), "Research Progress of UniCore CPUs and PKUnity SoCs" (PDF), Journal of Computer Science and Technology (JCST), 25 (2): 200–213, retrieved 2012-07-11
- Bergmann, Arnd (2012-07-09). "Re: [PATCH 00/36] AArch64 Linux kernel port". linux-kernel (Mailing list). Retrieved 2012-07-11.
Another interesting example is unicore32, which actually shares more code with arch/arm than the proposed arch/aarch64 does. I think the unicore32 code base would benefit from being merged back into arch/arm as a third instruction set, but the additional maintenance cost for everyone working on ARM makes that unrealistic.
- "Merge window closed - 2.6.39-rc1 out". Linus Torvalds.
- "remove unicore32 support". Mike Rapoport.
- Hsu-Hung Chiang; Huang-Jia Cheng; Yuan-Shin Hwan (2012-02-25), "Doubling the Number of Registers on ARM Processors" (PDF), 16th Workshop on Interaction between Compilers and Computer Architectures (INTERACT), pp. 1–8, doi:10.1109/INTERACT.2012.6339620, ISBN 1-4673-2613-5
- Unicore processor simulator source code. Instruction formats are in decode.c, disassembly in interpret.c, and emulation in instEx.c.
- QEMU Unicore32 emulator source code