x86sra.w
LBT intrinsic · group: x86 Shift
Synopsis
Instruction: x86sra.w rj, rk
CPU Flags: LBT
Description
x86-style shift right arithmetic: shift 32-bit value in rj right arithmetically by the amount specified in rk. If shift count is non-zero, update CF, PF, ZF, SF (OF is set to 0 if count is 1). If shift count is zero, flags are unchanged. Only EFLAGS (LBT4) is updated; the GPR is not modified.
Operation
uint32_t v = (uint32_t)a;
unsigned c = (unsigned)(b & 0x1f);
if (c != 0) {
uint8_t carry_out = c > 32 ? ((int32_t)v < 0) : ((v >> (c - 1)) & 1);
uint32_t r = c >= 32 ? (uint32_t)((int32_t)v >> (32 - 1))
: (uint32_t)((int32_t)v >> c);
EFLAGS.CF = carry_out;
EFLAGS.PF = parity_even((uint8_t)r);
EFLAGS.ZF = r == 0;
EFLAGS.SF = (int32_t)r < 0;
if (c == 1) {
EFLAGS.OF = 0;
}
}
Tested on real machine.
Latency and Throughput
| CPU | µarch | Latency | Throughput (IPC) |
|---|---|---|---|
| 3C5000 | LA464 | 2 | 0.5(1/2) |
| 3C6000 | LA664 | 2 | 0.5(1/2) |