Does AMOXOR.H zero-extend rd?
No. Zabha AMO.H sign-extends the old value into rd; the U in AMOMINU.H/AMOMAXU.H only means unsigned comparison.
AMOXOR.H
RISC-V AMOXOR.H Instruction Details
Instruction ManualAMO-typeAMOXOR.H atomically reads the old halfword, sign-extends it into rd, and bitwise-XORs the old halfword with rs2[15:0] and writes the result.
Instruction Syntax
amoxor.h rd, rs2, (rs1)
Syntax is rd, rs2, (rs1). The three fields select x0–x31 or ABI names, with no offset immediate or rs3.
32-bit encoding: funct5=00100, funct3=001, opcode=0101111; aq is bit26 and rl is bit25.
RV32/RV64 Zabha (requires Zaamo): rd receives sign-extended old 16-bit memory; memory receives old memory XOR the low 16 rs2 bits.
The region must support this AMO width and operation. Use two-byte alignment, or a PMA with all accessed bytes in one misaligned atomicity granule; otherwise address-misaligned or access-fault is raised.
Sources use pre-execution values; rd=x0 discards only register writeback and memory still undergoes read-modify-write. aq/rl orders only the accessed memory/I/O domain; suffixes are .aq, .rl and .aqrl.
ZabhaAtomic
Instruction Behavior
AMOXOR.H is a Zabha (and therefore Zaamo) halfword AMO. It atomically reads the old 16-bit value at rs1, sign-extends it into rd, then writes old_halfword XOR rs2[15:0]; aq/rl annotate memory ordering.
AMOXOR.H Atomic Decode And Execute Animation
Shows Zabha (requires Zaamo) halfword atomic field decode, aq/rl ordering bits, a memory read, and RMW state update.
rd
rs2
rs1
amoxor.h,,()
Execution Context
31..27
26
25
24..20
19..15
14..12
11..7
6..0
00100
funct5
0
aq
0
rl
01011
rs2
01010
rs1
001
funct3
00101
rd
0101111
opcode
Atomic Data Path
instruction
0x20B512AF
opcode
0101111 -> AMO
funct3
001 -> halfword width
funct5
00100 -> AMOXOR.H
aq / rl
0/0 -> unordered
rd / rs2 / rs1
t0(x5) / a1(x11)[15:0]=0x0003 / a0(x10)=0x00001000
memory read
mem[0x00001000] -> old halfword 0x0005
RMW
Not computed
rd / memory
t0(x5) = unspecified initial value; mem[0x00001000] = 0x0005
Current Step
Concept Step: receive the 32-bit atomic instruction encoding
Zabha depends on Zaamo; Zabha-extension halfword atomic instructions use a 32-bit encoding; the animation starts by splitting the fields.
encoding: 0x20B512AF
syntax : amoxor.h t0(x5), a1(x11), (a0(x10))
Pending result : Not computed
Current architectural state
t0(x5) = unspecified initial value; mem[0x00001000] = 0x0005
Assumes Zabha and Zaamo at the selected XLEN, successful fetch, a region supporting this width and operation, and a successful access. A misaligned halfword requires the stated PMA condition. Sources use pre-execution values; aq/rl orders only the accessed domain. Steps are a teaching sequence.
Quick Understanding & Search Notes
AMOXOR.H is a Zabha halfword AMO: it atomically reads the old 16-bit memory value, sign-extends that old value into rd, and bitwise-XORs the old halfword with rs2[15:0] and writes the result.
The encoding uses AMO major opcode 0101111, funct3=001 for halfword width, and funct5 to select the specific AMO operation.
All AMO.H forms write the sign-extended old halfword to rd; the unsigned suffix only changes min/max comparison interpretation.
Zabha depends on Zaamo; rs2 contributes only its low 16 bits. Normally the address is naturally aligned to 2 bytes; with a qualifying misaligned atomicity granule PMA, it does not raise an exception solely for address alignment. aq/rl only affect memory-ordering semantics.
Common Usage Scenarios
Atomic & Sync
RV32/RV64 Zabha+Zaamo, a1=0x1000, old memory=5 and a2=3; assuming a supporting region and successful access, a0 receives old value 5 and the 16-bit memory operand becomes 6.
Pre-Use Checklist
Syntax Check
- Syntax is rd, rs2, (rs1). The three fields select x0–x31 or ABI names, with no offset immediate or rs3.
- 32-bit encoding: funct5=00100, funct3=001, opcode=0101111; aq is bit26 and rl is bit25.
Semantic Check
- RV32/RV64 Zabha (requires Zaamo): rd receives sign-extended old 16-bit memory; memory receives old memory XOR the low 16 rs2 bits.
- The region must support this AMO width and operation. Use two-byte alignment, or a PMA with all accessed bytes in one misaligned atomicity granule; otherwise address-misaligned or access-fault is raised.
- Sources use pre-execution values; rd=x0 discards only register writeback and memory still undergoes read-modify-write. aq/rl orders only the accessed memory/I/O domain; suffixes are .aq, .rl and .aqrl.
Pitfalls / Common Confusions
rd receives the old memory halfword sign-extended to XLEN, not the newly written memory value.
Only the low 16 bits of rs2 participate in the halfword computation; upper rs2 bits do not affect this AMO.H data result.
AMOXOR.H is a 16-bit bitwise operation; it is not a 32-bit word AMO over neighboring bytes or halfwords.
Normally the target address is naturally aligned to 2 bytes; when a misaligned atomicity granule PMA is present and all accessed bytes lie in one granule, it does not raise an exception solely for address alignment. aq/rl does not change the data computation.
FAQ
How much memory does AMOXOR.H modify?
Architecturally it modifies only the target 16-bit halfword and computes the new memory value at halfword width; neighboring bytes are not part of that data result.