__lsx_vmaddwod_h_bu
LSX intrinsic · group: Integer Computation
Synopsis
__m128i __lsx_vmaddwod_h_bu (__m128i a, __m128i b, __m128i c)
#include <lsxintrin.h>
Instruction: vmaddwod.h.bu vr, vr, vr
CPU Flags: LSX
Description
Multiply the odd-positioned unsigned 8-bit elements in b and the corresponding unsigned 8-bit elements in c, extending both to 16-bit, and add the product to the 16-bit elements in a.
Operation
for (int i = 0; i < 8; i++) {
dst.half[i] =
(u16)(u8)b.byte[2 * i + 1] * (u16)(u8)c.byte[2 * i + 1] + (u16)a.half[i];
}
// Expands to:
if (0) {
dst.half[0] =
(((u16)((u8)b.byte[1])) * ((u16)((u8)c.byte[1]))) + ((u16)a.half[0]);
dst.half[1] =
(((u16)((u8)b.byte[3])) * ((u16)((u8)c.byte[3]))) + ((u16)a.half[1]);
dst.half[2] =
(((u16)((u8)b.byte[5])) * ((u16)((u8)c.byte[5]))) + ((u16)a.half[2]);
dst.half[3] =
(((u16)((u8)b.byte[7])) * ((u16)((u8)c.byte[7]))) + ((u16)a.half[3]);
dst.half[4] =
(((u16)((u8)b.byte[9])) * ((u16)((u8)c.byte[9]))) + ((u16)a.half[4]);
dst.half[5] =
(((u16)((u8)b.byte[11])) * ((u16)((u8)c.byte[11]))) + ((u16)a.half[5]);
dst.half[6] =
(((u16)((u8)b.byte[13])) * ((u16)((u8)c.byte[13]))) + ((u16)a.half[6]);
dst.half[7] =
(((u16)((u8)b.byte[15])) * ((u16)((u8)c.byte[15]))) + ((u16)a.half[7]);
}
Tested on real machine.
Latency and Throughput
| CPU | µarch | Latency | Throughput (IPC) |
|---|---|---|---|
| 3C5000 | LA464 | 4 | 2 |
| 3A6000 | LA664 | 4 | 2 |
| 3C6000 | LA664 | 4 | 2 |
| 2K1000LA | LA264 | 4 | 1 |
| 2K3000 | LA364E | 4 | 1 |