Does AMOMAXU.H return the post-operation value?
No. Like A-extension AMOs, rd receives the old value loaded from memory.
AMOMAXU.H
RISC-V AMOMAXU.H Instruction Details
Instruction ManualR-typeAtomic halfword unsigned max: rd = *rs1 (zero-ext), *rs1 = umax(*rs1, rs2[15:0])
Instruction Syntax
amomaxu.h rd, rs2, (rs1)
Operand Breakdown
Destination rd: register receiving the operation result.
Source rs1: register holding the first operand.
Source rs2: register holding the second operand.
ZabhaAtomic
Instruction Behavior
AMOMAXU.H (Zabha extension) atomically loads a halfword from the address in rs1 into rd (zero-extended), compares with the low 16 bits of rs2, stores unsigned maximum back. Supports aq/rl bits.
Quick Understanding & Search Notes
AMOMAXU.H is a Zabha 16-bit unsigned atomic maximum instruction. It atomically reads the old memory value, performs the read-modify-write at 16-bit width, and writes the sign-extended old value to rd.
Only the 16-bit subword is operated on; upper bits of rs2 are ignored.
rd receives the old memory value sign-extended to XLEN, not the newly updated value.
The address must be naturally aligned to the operand size; aq/rl bits can provide release-consistency ordering semantics.
Common Usage Scenarios
Atomic & Sync
Understand this scenario with real code like «amomaxu.h a0, a2, (a1)».
Network & Byte Order
Understand this scenario with real code like «amomaxu.h a0, a2, (a1)».
Pre-Use Checklist
Syntax Check
- Confirm the current instruction format is R-type.
- Confirm the operand order matches the example.
Semantic Check
- Ensure the destination register usage is compatible with the calling convention.
- Confirm this is not the lower-level form of a pseudo-instruction expansion.
Pitfalls / Common Confusions
rd returns old memory value (sign-extended to XLEN), not the operation result
Unsigned maximum comparison; negative values become very large when interpreted as unsigned
Atomicity depends on natural alignment and implementation support
FAQ
Can AMOMAXU.H affect neighboring bytes?
Architecturally it modifies only the halfword operand; Zabha provides native halfword-granule atomic read-modify-write operations.