Skip to content

Commit

Permalink
Merge pull request #43 from howjmay/sqrt
Browse files Browse the repository at this point in the history 
feat: Add _mm_sqrt*
  • Loading branch information
@howjmay
howjmay authored Jan 12, 2024
2 parents 69da5f6 + 5ca462f commit 8113ca7
Show file tree
Hide file tree
Showing 2 changed files with 75 additions and 58 deletions.
26 changes: 22 additions & 4 deletions sse2rvv.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2412,13 +2412,31 @@ FORCE_INLINE __m128i _mm_slli_epi64(__m128i a, int imm8) {

// FORCE_INLINE __m128i _mm_slli_si128 (__m128i a, int imm8) {}

// FORCE_INLINE __m128d _mm_sqrt_pd (__m128d a) {}
FORCE_INLINE __m128d _mm_sqrt_pd(__m128d a) {
vfloat64m1_t _a = vreinterpretq_m128d_f64(a);
return vreinterpretq_f64_m128d(
__riscv_vfrec7_v_f64m1(__riscv_vfrsqrt7_v_f64m1(_a, 2), 2));
}

// FORCE_INLINE __m128 _mm_sqrt_ps (__m128 a) {}
FORCE_INLINE __m128 _mm_sqrt_ps(__m128 a) {
vfloat32m1_t _a = vreinterpretq_m128_f32(a);
return vreinterpretq_f32_m128(
__riscv_vfrec7_v_f32m1(__riscv_vfrsqrt7_v_f32m1(_a, 4), 4));
}

// FORCE_INLINE __m128d _mm_sqrt_sd (__m128d a, __m128d b) {}
FORCE_INLINE __m128d _mm_sqrt_sd(__m128d a, __m128d b) {
vfloat64m1_t _a = vreinterpretq_m128d_f64(a);
vfloat64m1_t _b = vreinterpretq_m128d_f64(b);
vfloat64m1_t b_rnd =
__riscv_vfrec7_v_f64m1(__riscv_vfrsqrt7_v_f64m1(_b, 2), 2);
return vreinterpretq_f64_m128d(__riscv_vslideup_vx_f64m1(_a, b_rnd, 0, 1));
}

// FORCE_INLINE __m128 _mm_sqrt_ss (__m128 a) {}
FORCE_INLINE __m128 _mm_sqrt_ss(__m128 a) {
vfloat32m1_t _a = vreinterpretq_m128_f32(a);
vfloat32m1_t rnd = __riscv_vfrec7_v_f32m1(__riscv_vfrsqrt7_v_f32m1(_a, 4), 4);
return vreinterpretq_f32_m128(__riscv_vslideup_vx_f32m1(_a, rnd, 0, 1));
}

// FORCE_INLINE __m128i _mm_sra_epi16 (__m128i a, __m128i count) {}

Expand Down
107 changes: 53 additions & 54 deletions tests/impl.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3215,39 +3215,39 @@ result_t test_mm_shuffle_ps(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
}

result_t test_mm_sqrt_ps(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
// #ifdef ENABLE_TEST_ALL
// const float *_a = (const float *)impl.test_cases_float_pointer1;
//
// float f0 = sqrt(_a[0]);
// float f1 = sqrt(_a[1]);
// float f2 = sqrt(_a[2]);
// float f3 = sqrt(_a[3]);
//
// __m128 a = load_m128(_a);
// __m128 c = _mm_sqrt_ps(a);
//
// return validate_float_error(c, f0, f1, f2, f3, 0.000001f);
// #else
#ifdef ENABLE_TEST_ALL
const float *_a = (const float *)impl.test_cases_float_pointer1;

float f0 = sqrt(_a[0]);
float f1 = sqrt(_a[1]);
float f2 = sqrt(_a[2]);
float f3 = sqrt(_a[3]);

__m128 a = load_m128(_a);
__m128 c = _mm_sqrt_ps(a);

return validate_float_error(c, f0, f1, f2, f3, 0.1f);
#else
return TEST_UNIMPL;
// #endif // ENABLE_TEST_ALL
#endif // ENABLE_TEST_ALL
}

result_t test_mm_sqrt_ss(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
// #ifdef ENABLE_TEST_ALL
// const float *_a = (const float *)impl.test_cases_float_pointer1;
//
// float f0 = sqrt(_a[0]);
// float f1 = _a[1];
// float f2 = _a[2];
// float f3 = _a[3];
//
// __m128 a = load_m128(_a);
// __m128 c = _mm_sqrt_ss(a);
//
// return validate_float_error(c, f0, f1, f2, f3, 0.000001f);
// #else
#ifdef ENABLE_TEST_ALL
const float *_a = (const float *)impl.test_cases_float_pointer1;

float f0 = sqrt(_a[0]);
float f1 = _a[1];
float f2 = _a[2];
float f3 = _a[3];

__m128 a = load_m128(_a);
__m128 c = _mm_sqrt_ss(a);

return validate_float_error(c, f0, f1, f2, f3, 0.1f);
#else
return TEST_UNIMPL;
// #endif // ENABLE_TEST_ALL
#endif // ENABLE_TEST_ALL
}

result_t test_mm_store_ps(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
Expand Down Expand Up @@ -6989,37 +6989,36 @@ result_t test_mm_slli_si128(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
}

result_t test_mm_sqrt_pd(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
// #ifdef ENABLE_TEST_ALL
// const double *_a = (const double *)impl.test_cases_float_pointer1;
//
// double f0 = sqrt(_a[0]);
// double f1 = sqrt(_a[1]);
//
// __m128d a = load_m128d(_a);
// __m128d c = _mm_sqrt_pd(a);
//
// return validate_double_error(c, f0, f1, 1.0e-15);
// #else
#ifdef ENABLE_TEST_ALL
const double *_a = (const double *)impl.test_cases_float_pointer1;
double _c[2];
_c[0] = sqrt(_a[0]);
_c[1] = sqrt(_a[1]);

__m128d a = load_m128d(_a);
__m128d c = _mm_sqrt_pd(a);
return validate_double_error(c, _c[0], _c[1], 0.1f);
#else
return TEST_UNIMPL;
// #endif // ENABLE_TEST_ALL
#endif // ENABLE_TEST_ALL
}

result_t test_mm_sqrt_sd(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
// #ifdef ENABLE_TEST_ALL
// const double *_a = (const double *)impl.test_cases_float_pointer1;
// const double *_b = (const double *)impl.test_cases_float_pointer2;
//
// double f0 = sqrt(_b[0]);
// double f1 = _a[1];
//
// __m128d a = load_m128d(_a);
// __m128d b = load_m128d(_b);
// __m128d c = _mm_sqrt_sd(a, b);
//
// return validate_double_error(c, f0, f1, 1.0e-15);
// #else
#ifdef ENABLE_TEST_ALL
const double *_a = (const double *)impl.test_cases_float_pointer1;
const double *_b = (const double *)impl.test_cases_float_pointer2;

double f0 = sqrt(_b[0]);
double f1 = _a[1];

__m128d a = load_m128d(_a);
__m128d b = load_m128d(_b);
__m128d c = _mm_sqrt_sd(a, b);

return validate_double_error(c, f0, f1, 0.1f);
#else
return TEST_UNIMPL;
// #endif // ENABLE_TEST_ALL
#endif // ENABLE_TEST_ALL
}

result_t test_mm_sra_epi16(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
Expand Down

0 comments on commit 8113ca7

Please sign in to comment.