Floating Point Computation

__m256 __lasx_xvfadd_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfadd_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfadd.s xr, xr, xr
CPU Flags: LASX

Description

Add single precision floating point elements in a to elements in b.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = a.fp32[i] + b.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	3	4
3C5000	5	2

__m256d __lasx_xvfadd_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfadd_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfadd.d xr, xr, xr
CPU Flags: LASX

Description

Add double precision floating point elements in a to elements in b.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = a.fp64[i] + b.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	3	4
3C5000	5	2

__m256 __lasx_xvfdiv_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfdiv_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfdiv.s xr, xr, xr
CPU Flags: LASX

Description

Divide single precision floating point elements in a by elements in b.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = a.fp32[i] / b.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	11	0.18(1/5.5)
3C5000	11, 19.5	0.1(1/10.5)

__m256d __lasx_xvfdiv_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfdiv_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfdiv.d xr, xr, xr
CPU Flags: LASX

Description

Divide double precision floating point elements in a by elements in b.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = a.fp64[i] / b.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	8, 21.5	0.25(1/4)
3C5000	8, 17	0.08(1/12.5)

__m256 __lasx_xvfmax_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfmax_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfmax.s xr, xr, xr
CPU Flags: LASX

Description

Compute maximum of single precision floating point elements in a and b.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = +(a.fp32[i], b.fp32[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256d __lasx_xvfmax_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfmax_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfmax.d xr, xr, xr
CPU Flags: LASX

Description

Compute maximum of double precision floating point elements in a and b.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = +(a.fp64[i], b.fp64[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256 __lasx_xvfmaxa_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfmaxa_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfmaxa.s xr, xr, xr
CPU Flags: LASX

Description

Compute maximum of single precision floating point elements in a and b by magnitude.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = (abs(a.fp32[i]) > abs(b.fp32[i])) ? a.fp32[i] : b.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256d __lasx_xvfmaxa_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfmaxa_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfmaxa.d xr, xr, xr
CPU Flags: LASX

Description

Compute maximum of double precision floating point elements in a and b by magnitude.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = (abs(a.fp64[i]) > abs(b.fp64[i])) ? a.fp64[i] : b.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256 __lasx_xvfmin_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfmin_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfmax.s xr, xr, xr
CPU Flags: LASX

Description

Compute minimum of single precision floating point elements in a and b.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = +(a.fp32[i], b.fp32[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256d __lasx_xvfmin_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfmin_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfmax.d xr, xr, xr
CPU Flags: LASX

Description

Compute minimum of double precision floating point elements in a and b.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = +(a.fp64[i], b.fp64[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256 __lasx_xvfmina_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfmina_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfmina.s xr, xr, xr
CPU Flags: LASX

Description

Compute minimum of single precision floating point elements in a and b by magnitude.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = (abs(a.fp32[i]) < abs(b.fp32[i])) ? a.fp32[i] : b.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256d __lasx_xvfmina_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfmina_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfmina.d xr, xr, xr
CPU Flags: LASX

Description

Compute minimum of double precision floating point elements in a and b by magnitude.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = (abs(a.fp64[i]) < abs(b.fp64[i])) ? a.fp64[i] : b.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	2	4
3C5000	2	2

__m256 __lasx_xvfmul_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfmul_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfmul.s xr, xr, xr
CPU Flags: LASX

Description

Multiply single precision floating point elements in a and elements in b.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = a.fp32[i] * b.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	5	2
3C5000	5	2

__m256d __lasx_xvfmul_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfmul_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfmul.d xr, xr, xr
CPU Flags: LASX

Description

Multiply double precision floating point elements in a and elements in b.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = a.fp64[i] * b.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	5	2
3C5000	5	2

__m256 __lasx_xvfsub_s (__m256 a, __m256 b)

Synopsis

__m256 __lasx_xvfsub_s (__m256 a, __m256 b)
#include <lasxintrin.h>
Instruction: xvfsub.s xr, xr, xr
CPU Flags: LASX

Description

Subtract single precision floating point elements in a by elements in b.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = a.fp32[i] - b.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	3	4
3C5000	5	2

__m256d __lasx_xvfsub_d (__m256d a, __m256d b)

Synopsis

__m256d __lasx_xvfsub_d (__m256d a, __m256d b)
#include <lasxintrin.h>
Instruction: xvfsub.d xr, xr, xr
CPU Flags: LASX

Description

Subtract double precision floating point elements in a by elements in b.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = a.fp64[i] - b.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	3	4
3C5000	5	2

__m256 __lasx_xvflogb_s (__m256 a)

Synopsis

__m256 __lasx_xvflogb_s (__m256 a)
#include <lasxintrin.h>
Instruction: xvflogb.s xr, xr
CPU Flags: LASX

Description

Compute 2-based logarithm of single precision floating point elements in a.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = log2(a.fp32[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	4	4
3C5000	4	2

__m256d __lasx_xvflogb_d (__m256d a)

Synopsis

__m256d __lasx_xvflogb_d (__m256d a)
#include <lasxintrin.h>
Instruction: xvflogb.d xr, xr
CPU Flags: LASX

Description

Compute 2-based logarithm of double precision floating point elements in a.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = log2(a.fp64[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	4	4
3C5000	4	2

__m256 __lasx_xvfsqrt_s (__m256 a)

Synopsis

__m256 __lasx_xvfsqrt_s (__m256 a)
#include <lasxintrin.h>
Instruction: xvfsqrt.s xr, xr
CPU Flags: LASX

Description

Compute square root of single precision floating point elements in a.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = sqrt(a.fp32[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	15	0.08(1/12)
3C5000	15	0.07(1/13.5)

__m256d __lasx_xvfsqrt_d (__m256d a)

Synopsis

__m256d __lasx_xvfsqrt_d (__m256d a)
#include <lasxintrin.h>
Instruction: xvfsqrt.d xr, xr
CPU Flags: LASX

Description

Compute square root of double precision floating point elements in a.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = sqrt(a.fp64[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	36	0.06(1/17.5)
3C5000	36	0.05(1/18.5)

__m256 __lasx_xvfrsqrt_s (__m256 a)

Synopsis

__m256 __lasx_xvfrsqrt_s (__m256 a)
#include <lasxintrin.h>
Instruction: xvfrsqrt.s xr, xr
CPU Flags: LASX

Description

Compute reciprocal of square root of single precision floating point elements in a.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = 1.0 / sqrt(a.fp32[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	25	0.05(1/19)
3C5000	25	0.03(1/32)

__m256d __lasx_xvfrsqrt_d (__m256d a)

Synopsis

__m256d __lasx_xvfrsqrt_d (__m256d a)
#include <lasxintrin.h>
Instruction: xvfrsqrt.d xr, xr
CPU Flags: LASX

Description

Compute reciprocal of square root of double precision floating point elements in a.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = 1.0 / sqrt(a.fp64[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	15	0.04(1/26.5)
3C5000	15	0.04(1/27.5)

__m256 __lasx_xvfrecip_s (__m256 a)

Synopsis

__m256 __lasx_xvfrecip_s (__m256 a)
#include <lasxintrin.h>
Instruction: xvfrecip.s xr, xr
CPU Flags: LASX

Description

Compute reciprocal of single precision floating point elements in a.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = 1 / a.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	27	0.18(1/5.5)
3C5000	27	0.14(1/7)

__m256d __lasx_xvfrecip_d (__m256d a)

Synopsis

__m256d __lasx_xvfrecip_d (__m256d a)
#include <lasxintrin.h>
Instruction: xvfrecip.d xr, xr
CPU Flags: LASX

Description

Compute reciprocal of double precision floating point elements in a.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = 1 / a.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	23	0.25(1/4)
3C5000	23	0.08(1/12)

__m256 __lasx_xvfrsqrte_s (__m256 a)

Synopsis

__m256 __lasx_xvfrsqrte_s (__m256 a)
#include <lasxintrin.h>
Instruction: xvfrsqrte.s xr, xr
CPU Flags: LASX

Description

Compute estimated reciprocal of square root of single precision floating point elements in a.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = 1.0 / sqrt(a.fp32[i]); // estimated
}

__m256d __lasx_xvfrsqrte_d (__m256d a)

Synopsis

__m256d __lasx_xvfrsqrte_d (__m256d a)
#include <lasxintrin.h>
Instruction: xvfrsqrte.d xr, xr
CPU Flags: LASX

Description

Compute estimated reciprocal of square root of double precision floating point elements in a.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = 1.0 / sqrt(a.fp64[i]); // estimated
}

__m256 __lasx_xvfrecipe_s (__m256 a)

Synopsis

__m256 __lasx_xvfrecipe_s (__m256 a)
#include <lasxintrin.h>
Instruction: xvfrecipe.s xr, xr
CPU Flags: LASX

Description

Compute estimated reciprocal of single precision floating point elements in a.

Operation

for (int i = 0; i < 8; i++) {
  dst.fp32[i] = 1 / a.fp32[i]; // estimated
}

__m256d __lasx_xvfrecipe_d (__m256d a)

Synopsis

__m256d __lasx_xvfrecipe_d (__m256d a)
#include <lasxintrin.h>
Instruction: xvfrecipe.d xr, xr
CPU Flags: LASX

Description

Compute estimated reciprocal of double precision floating point elements in a.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp64[i] = 1 / a.fp64[i]; // estimated
}