__lsx_vshuf_b

LSX intrinsic · group: Shuffling

Synopsis

__m128i __lsx_vshuf_b (__m128i a, __m128i b, __m128i c)
#include <lsxintrin.h>
Instruction: vshuf.b vr, vr, vr, vr
CPU Flags: LSX

Description

For each 128-bit lane, shuffle bytes from a and b using the low 5 bits of the indices from the corresponding lane of c.

Caveat: the indices are placed in c, while in other vshuf intrinsics, they are placed in a.

Caution: the result is undefined when the highest 3 bits of the index byte are non-zero.

Examples

__m128i __lsx_vshuf_b(__m128i{0x1122334455667788, 0x99aabbccddeeff00}, __m128i{0xabcdef1314156678, 0x1234123443214321}, __m128i{0x0011021304050607, 0x0811120213031404})
= 0x7877155513efcdab 0x2177661555144413

Operation

for (int i = 0; i < 16; i++) {
  if (c.byte[i] >= 64 && (UARCH_LA264 || UARCH_LA464)) {
    // Caveat: observed in LA264 and LA464
    dst.byte[i] = 0;
  } else if ((c.byte[i] % 32) < 16) {
    dst.byte[i] = b.byte[c.byte[i] % 16];
  } else {
    dst.byte[i] = a.byte[c.byte[i] % 16];
  }
}

Tested on real machine.

Latency and Throughput

CPU	µarch	Latency	Throughput (IPC)
3C5000	LA464	1	2
3A6000	LA664	1	2
3C6000	LA664	1	2
2K1000LA	LA264	1	1
2K3000	LA364E	1	1

---

All intrinsics in alphabetical order