Home

Overview

This processor is designed for small size, high clock speed, and simplicity. There are 8 general purpose registers, each of which can contain a 16 bit value. There are also 4 condition code flags (zero, negative, carry, and overflow), which are set by arithmetic operations and can be used for conditional branches.

The pipeline does perform bypassing of intermediate arithmetic results, but it does not have any form of interlocking. As such, it has two "branch delay slots," which is a fancy way of saying it will execute the next two instructions after a branch because the instructions are already in the pipeline. NOPs can be inserted to avoid side effects, or the code can be structured to take advantage of this. Likewise, loads have 2 cycles of latency. If you attempt to access a load destination register within two instructions, it will not contain the correct value.

Instruction Formats

	15	14	13	12	11	10	9	8	7	6	5	4	3	2	1	0
Arithmetic	0	0	0	operation				opb			opa			dest
Load	0	0	1	offset							ptr			dest
Store	0	1	0	offset h				src			ptr			offset l
Addi	0	1	1	immediate							opa			dest
Lui	1	0	0	immediate										dest
Conditional branch	1	0	1	cond				offset
Unconditional branch	1	1	0	link	offset
Jump to reg	1	1	1	link	unused			target			unused

Instructions

Name	Params	Instruction Format	Flags Affected	operation/cond/	Description
and	dest, srca, srcb	Arithmetic	NZ	0	Bitwise logical and
or	dest, srca, srcb	Arithmetic	NZ	1	Bitwise logical or
shl	dest, srca	Arithmetic	CNZ	2	Logical shift left one position
shr	dest, srca	Arithmetic	CNZ	3	Logical shift right one position
add	dest, srca, srcb	Arithmetic	CNZO	4	Add without carry
sub	dest, srca, srcb	Arithmetic	CNZO	5	Subtract without carry
xor	dest, srca, srcb	Arithmetic	NZ	6	Bitwise logical exclusive or
not	dest, srca	Arithmetic	NZ	7	Bitwise logical not
adc	dest, srca, srcb	Arithmetic	CNZO	8	Add with carry in
sbc	dest, srca, srcb	Arithmetic	CNZO	9	Subtract with carry[borrow] in
rol	dest, srca	Arithmetic	NZ	10	Rotate left (carry bit loaded into LSB)
ror	dest, srca	Arithmetic	NZ	11	Rotate right (carry bit loaded into MSB)
load	dest, [offset](ptr)	Load			Load word
store	src, [offset](ptr)	Store			Store word
addi	dest, srca, immediate	Addi	CZNO1		Add signed immediate value -31 to 31
lui	dest, immediate	Lui			Load upper immediate. Value is loaded into top 10 bits of dest. Low 6 bits are cleared.
jump	label	Unconditional branch		link=0	Jump to label
call	label	Unconditional branch		link=1	Call to label (return address saved in r7)
jumpr	reg	Jump to reg		link=0	Jump to address in register
callr	reg	Jump to reg		link=1	Call to address in register (return address saved in r7)
bcc	label	Conditional branch		6	Branch if carry flag clear
bcs	label	Conditional branch		2	Branch if carry flag set
bzc	label	Conditional branch		4	Branch if zero flag clear
bzs	label	Conditional branch		0	Branch if zero flag set
bnc	label	Conditional branch		5	Branch if negative flag clear
bns	label	Conditional branch		1	Branch if negative flag set
boc	label	Conditional branch		7	Branch if overflow flag clear
bos	label	Conditional branch		3	Branch if overflow flag set

1. Note that C and O bits are currently not implemented. I guess that's technically a bug.

Pseudo Ops

ldi	immediate	Load 16-bit immediate into register (creates LDI/ADDI pair)
nop		No-operation (and r0, r0, r0)
lea	label	Load effective address of label into register

Memory map

+------------------+  0000
|     Boot ROM     |
+------------------+  0010
|  Local memory    |
+------------------+  4000
|  Global memory   |
+------------------+  (global memory size)
|                  |
/                  /
|                  |
+------------------+  FC00
| Device Registers |
+------------------+  FFFF

Note that Internally, the processor is a harvard architecture. Instructions and data are fetched on independent busses. For the most part, however, the address spaces access the same memory. One exception is that the instruction bus cannot access global memory or device registers.

When each core comes out of reset, it begins executing code at address 0 in its local address space. A small chunk of ROM at that location contains copies the program from global memory into local memory at address 16, then branches to it.

(Currently, the ROM is emulated using $readmemh into each local memory)

Hardware Mutexes

Writing a one to a hardware mutex location will attempt to acquire it and writing a zero will release it if it is held by the owning core. A core may read the location to determine if it has acquired the mutex: it will return one if so, zero if not.