__lasx_xvftintrne_w_d
LASX intrinsic · group: Floating Point Conversion
Synopsis
__m256i __lasx_xvftintrne_w_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvftintrne.w.d xr, xr, xr
CPU Flags: LASX
Description
For each 128-bit lane, convert double-precision floating point elements in the corresponding lane of a into the upper half and elements in b into the lower half, as signed 32-bit integers, rounding towards nearest even.
Operation
for (int i = 0; i < 4; i++) {
dst.word[i] = (i < 2)
? (s32)b.fp64[i]
: (s32)a.fp64[i - 2]; // rounding mode is not expressed in C
}
for (int i = 4; i < 8; i++) {
dst.word[i] = (i < 6)
? (s32)b.fp64[i - 2]
: (s32)a.fp64[i - 4]; // rounding mode is not expressed in C
}
// Expands to:
if (0) {
dst.word[0] = (s32)b.fp64[0];
dst.word[1] = (s32)b.fp64[1];
dst.word[2] = (s32)a.fp64[0];
dst.word[3] = (s32)a.fp64[1];
dst.word[4] = (s32)b.fp64[2];
dst.word[5] = (s32)b.fp64[3];
dst.word[6] = (s32)a.fp64[2];
dst.word[7] = (s32)a.fp64[3];
}
Tested on real machine.
Latency and Throughput
| CPU | µarch | Latency | Throughput (IPC) |
|---|---|---|---|
| 3C5000 | LA464 | 5 | 1 |
| 3A6000 | LA664 | 5 | 2 |
| 3C6000 | LA664 | 5 | 2 |