Floating Point Computation

__m128 __lsx_vfadd_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfadd_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfadd.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = a.fp32[i] + b.fp32[i];
}

Tested on real machine.

Latency and Throughput

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

__m128d __lsx_vfadd_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfadd_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfadd.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = a.fp64[i] + b.fp64[i];
}

Tested on real machine.

Latency and Throughput

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

__m128 __lsx_vfdiv_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfdiv_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfdiv.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; 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	0.13(1/7.5)

__m128d __lsx_vfdiv_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfdiv_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfdiv.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; 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, 16.5	0.08(1/12.5)

__m128 __lsx_vfmax_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfmax_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfmax.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = +(a.fp32[i], b.fp32[i]);
}

Latency and Throughput

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

__m128d __lsx_vfmax_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfmax_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfmax.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = +(a.fp64[i], b.fp64[i]);
}

Latency and Throughput

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

__m128 __lsx_vfmaxa_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfmaxa_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfmaxa.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; 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

__m128d __lsx_vfmaxa_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfmaxa_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfmaxa.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; 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

__m128 __lsx_vfmin_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfmin_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfmax.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = +(a.fp32[i], b.fp32[i]);
}

Latency and Throughput

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

__m128d __lsx_vfmin_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfmin_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfmax.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = +(a.fp64[i], b.fp64[i]);
}

Latency and Throughput

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

__m128 __lsx_vfmina_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfmina_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfmina.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; 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

__m128d __lsx_vfmina_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfmina_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfmina.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; 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

__m128 __lsx_vfmul_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfmul_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfmul.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = a.fp32[i] * b.fp32[i];
}

Latency and Throughput

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

__m128d __lsx_vfmul_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfmul_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfmul.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = a.fp64[i] * b.fp64[i];
}

Latency and Throughput

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

__m128 __lsx_vfsub_s (__m128 a, __m128 b)

Synopsis

__m128 __lsx_vfsub_s (__m128 a, __m128 b)
#include <lsxintrin.h>
Instruction: vfsub.s vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = a.fp32[i] - b.fp32[i];
}

Tested on real machine.

Latency and Throughput

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

__m128d __lsx_vfsub_d (__m128d a, __m128d b)

Synopsis

__m128d __lsx_vfsub_d (__m128d a, __m128d b)
#include <lsxintrin.h>
Instruction: vfsub.d vr, vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = a.fp64[i] - b.fp64[i];
}

Tested on real machine.

Latency and Throughput

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

__m128 __lsx_vflogb_s (__m128 a)

Synopsis

__m128 __lsx_vflogb_s (__m128 a)
#include <lsxintrin.h>
Instruction: vflogb.s vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = log2(a.fp32[i]);
}

Latency and Throughput

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

__m128d __lsx_vflogb_d (__m128d a)

Synopsis

__m128d __lsx_vflogb_d (__m128d a)
#include <lsxintrin.h>
Instruction: vflogb.d vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = log2(a.fp64[i]);
}

Latency and Throughput

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

__m128 __lsx_vfsqrt_s (__m128 a)

Synopsis

__m128 __lsx_vfsqrt_s (__m128 a)
#include <lsxintrin.h>
Instruction: vfsqrt.s vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = sqrt(a.fp32[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	11	0.08(1/12)
3C5000	27	0.17(1/6)

__m128d __lsx_vfsqrt_d (__m128d a)

Synopsis

__m128d __lsx_vfsqrt_d (__m128d a)
#include <lsxintrin.h>
Instruction: vfsqrt.d vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; 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)

__m128 __lsx_vfrsqrt_s (__m128 a)

Synopsis

__m128 __lsx_vfrsqrt_s (__m128 a)
#include <lsxintrin.h>
Instruction: vfrsqrt.s vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = 1.0 / sqrt(a.fp32[i]);
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	17	0.05(1/19)
3C5000	21	0.11(1/9)

__m128d __lsx_vfrsqrt_d (__m128d a)

Synopsis

__m128d __lsx_vfrsqrt_d (__m128d a)
#include <lsxintrin.h>
Instruction: vfrsqrt.d vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; 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)

__m128 __lsx_vfrecip_s (__m128 a)

Synopsis

__m128 __lsx_vfrecip_s (__m128 a)
#include <lsxintrin.h>
Instruction: vfrecip.s vr, vr
CPU Flags: LSX

Description

Compute reciprocal of single precision floating point elements in a.

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = 1 / a.fp32[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	11	0.18(1/5.5)
3C5000	27	0.14(1/7)

__m128d __lsx_vfrecip_d (__m128d a)

Synopsis

__m128d __lsx_vfrecip_d (__m128d a)
#include <lsxintrin.h>
Instruction: vfrecip.d vr, vr
CPU Flags: LSX

Description

Compute reciprocal of double precision floating point elements in a.

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = 1 / a.fp64[i];
}

Latency and Throughput

CPU	Latency	Throughput (CPI)
3A6000	8	0.25(1/4)
3C5000	23	0.08(1/12)

__m128 __lsx_vfrsqrte_s (__m128 a)

Synopsis

__m128 __lsx_vfrsqrte_s (__m128 a)
#include <lsxintrin.h>
Instruction: vfrsqrte.s vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = 1.0 / sqrt(a.fp32[i]); // estimated
}

__m128d __lsx_vfrsqrte_d (__m128d a)

Synopsis

__m128d __lsx_vfrsqrte_d (__m128d a)
#include <lsxintrin.h>
Instruction: vfrsqrte.d vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = 1.0 / sqrt(a.fp64[i]); // estimated
}

__m128 __lsx_vfrecipe_s (__m128 a)

Synopsis

__m128 __lsx_vfrecipe_s (__m128 a)
#include <lsxintrin.h>
Instruction: vfrecipe.s vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 4; i++) {
  dst.fp32[i] = 1 / a.fp32[i]; // estimated
}

__m128d __lsx_vfrecipe_d (__m128d a)

Synopsis

__m128d __lsx_vfrecipe_d (__m128d a)
#include <lsxintrin.h>
Instruction: vfrecipe.d vr, vr
CPU Flags: LSX

Description

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

Operation

for (int i = 0; i < 2; i++) {
  dst.fp64[i] = 1 / a.fp64[i]; // estimated
}