Skip to content

Commit

Permalink
Merge pull request #19 from howjmay/andnot
Browse files Browse the repository at this point in the history 
feat: Add _mm_andnot_*
  • Loading branch information
@howjmay
howjmay authored Dec 24, 2023
2 parents 3fbe37a + 1008158 commit 92933b4
Show file tree
Hide file tree
Showing 2 changed files with 80 additions and 65 deletions.
23 changes: 19 additions & 4 deletions sse2rvv.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -476,7 +476,7 @@ typedef struct {
FORCE_INLINE __m128 _mm_add_ps(__m128 a, __m128 b) {
vfloat32m1_t _a = vreinterpretq_m128_f32(a);
vfloat32m1_t _b = vreinterpretq_m128_f32(b);
return vreinterpretq_f32_m128(__riscv_vfadd_vv_f32m1(a, b, 4));
return vreinterpretq_f32_m128(__riscv_vfadd_vv_f32m1(_a, _b, 4));
}

// Add the lower single-precision (32-bit) floating-point element in a and b,
Expand All @@ -497,7 +497,12 @@ FORCE_INLINE __m128 _mm_and_ps(__m128 a, __m128 b) {
// Compute the bitwise NOT of packed single-precision (32-bit) floating-point
// elements in a and then AND with b, and store the results in dst.
// https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_andnot_ps
// FORCE_INLINE __m128 _mm_andnot_ps(__m128 a, __m128 b) {}
FORCE_INLINE __m128 _mm_andnot_ps(__m128 a, __m128 b) {
vint32m1_t _a = vreinterpretq_m128_i32(a);
vint32m1_t _b = vreinterpretq_m128_i32(b);
return vreinterpretq_i32_m128(
__riscv_vand_vv_i32m1(__riscv_vnot_v_i32m1(_a, 4), _b, 4));
}

// Average packed unsigned 16-bit integers in a and b, and store the results in
// dst.
Expand Down Expand Up @@ -1480,12 +1485,22 @@ FORCE_INLINE __m128i _mm_and_si128(__m128i a, __m128i b) {
// Compute the bitwise NOT of packed double-precision (64-bit) floating-point
// elements in a and then AND with b, and store the results in dst.
// https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_andnot_pd
// FORCE_INLINE __m128d _mm_andnot_pd(__m128d a, __m128d b) {}
FORCE_INLINE __m128d _mm_andnot_pd(__m128d a, __m128d b) {
vint64m1_t _a = vreinterpretq_m128d_i64(a);
vint64m1_t _b = vreinterpretq_m128d_i64(b);
return vreinterpretq_i64_m128d(
__riscv_vand_vv_i64m1(__riscv_vnot_v_i64m1(_a, 2), _b, 2));
}

// Compute the bitwise NOT of 128 bits (representing integer data) in a and then
// AND with b, and store the result in dst.
// https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_andnot_si128
// FORCE_INLINE __m128i _mm_andnot_si128(__m128i a, __m128i b) {}
FORCE_INLINE __m128i _mm_andnot_si128(__m128i a, __m128i b) {
vint32m1_t _a = vreinterpretq_m128i_i32(a);
vint32m1_t _b = vreinterpretq_m128i_i32(b);
return vreinterpretq_i32_m128i(
__riscv_vand_vv_i32m1(__riscv_vnot_v_i32m1(_a, 4), _b, 4));
}

// Average packed unsigned 16-bit integers in a and b, and store the results in
// dst.
Expand Down
122 changes: 61 additions & 61 deletions tests/impl.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -810,31 +810,31 @@ result_t test_mm_and_ps(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
// r2 := ~a2 & b2
// r3 := ~a3 & b3
result_t test_mm_andnot_ps(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
// #ifdef ENABLE_TEST_ALL
// const float *_a = impl.test_cases_float_pointer1;
// const float *_b = impl.test_cases_float_pointer2;
//
// __m128 a = load_m128(_a);
// __m128 b = load_m128(_b);
// __m128 c = _mm_andnot_ps(a, b);
#ifdef ENABLE_TEST_ALL
const float *_a = impl.test_cases_float_pointer1;
const float *_b = impl.test_cases_float_pointer2;

__m128 a = load_m128(_a);
__m128 b = load_m128(_b);
__m128 c = _mm_andnot_ps(a, b);
// now for the assertion...
// const uint32_t *ia = (const uint32_t *)&a;
// const uint32_t *ib = (const uint32_t *)&b;
// uint32_t r[4];
// r[0] = ~ia[0] & ib[0];
// r[1] = ~ia[1] & ib[1];
// r[2] = ~ia[2] & ib[2];
// r[3] = ~ia[3] & ib[3];
// __m128i ret = do_mm_set_epi32(r[3], r[2], r[1], r[0]);
// result_t res = TEST_FAIL;
// res = VALIDATE_INT32_M128(*(const __m128i *)&c, r);
// if (res) {
// res = VALIDATE_INT32_M128(ret, r);
// }
// return res;
// #else
const uint32_t *ia = (const uint32_t *)&a;
const uint32_t *ib = (const uint32_t *)&b;
uint32_t r[4];
r[0] = ~ia[0] & ib[0];
r[1] = ~ia[1] & ib[1];
r[2] = ~ia[2] & ib[2];
r[3] = ~ia[3] & ib[3];
__m128i ret = do_mm_set_epi32(r[3], r[2], r[1], r[0]);
result_t res = TEST_FAIL;
res = VALIDATE_INT32_M128(*(const __m128i *)&c, r);
if (res) {
res = VALIDATE_INT32_M128(ret, r);
}
return res;
#else
return TEST_UNIMPL;
// #endif // ENABLE_TEST_ALL
#endif // ENABLE_TEST_ALL
}

result_t test_mm_avg_pu16(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
Expand Down Expand Up @@ -4016,53 +4016,53 @@ result_t test_mm_and_si128(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
}

result_t test_mm_andnot_pd(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;
//
// __m128d a = load_m128d(_a);
// __m128d b = load_m128d(_b);
// __m128d c = _mm_andnot_pd(a, b);
//
#ifdef ENABLE_TEST_ALL
const double *_a = (const double *)impl.test_cases_float_pointer1;
const double *_b = (const double *)impl.test_cases_float_pointer2;

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

// Take AND operation a complement of 'a' and 'b'. Bitwise operations are
// not allowed on float/double datatype, so 'a' and 'b' are calculated in
// uint64_t datatype.
// const uint64_t *ia = (const uint64_t *)&a;
// const uint64_t *ib = (const uint64_t *)&b;
// uint64_t r0 = ~ia[0] & ib[0];
// uint64_t r1 = ~ia[1] & ib[1];
// return validate_uint64(*(const __m128i *)&c, r0, r1);
// #else
const uint64_t *ia = (const uint64_t *)&a;
const uint64_t *ib = (const uint64_t *)&b;
uint64_t r0 = ~ia[0] & ib[0];
uint64_t r1 = ~ia[1] & ib[1];
return validate_uint64(*(const __m128i *)&c, r0, r1);
#else
return TEST_UNIMPL;
// #endif // ENABLE_TEST_ALL
#endif // ENABLE_TEST_ALL
}

result_t test_mm_andnot_si128(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
// #ifdef ENABLE_TEST_ALL
// const int32_t *_a = impl.test_cases_int_pointer1;
// const int32_t *_b = impl.test_cases_int_pointer2;
// __m128i a = load_m128i(_a);
// __m128i b = load_m128i(_b);
// __m128 fc = _mm_andnot_ps(*(const __m128 *)&a, *(const __m128 *)&b);
// __m128i c = *(const __m128i *)&fc;
#ifdef ENABLE_TEST_ALL
const int32_t *_a = impl.test_cases_int_pointer1;
const int32_t *_b = impl.test_cases_int_pointer2;
__m128i a = load_m128i(_a);
__m128i b = load_m128i(_b);
__m128 fc = _mm_andnot_ps(*(const __m128 *)&a, *(const __m128 *)&b);
__m128i c = *(const __m128i *)&fc;
// now for the assertion...
// const uint32_t *ia = (const uint32_t *)&a;
// const uint32_t *ib = (const uint32_t *)&b;
// uint32_t r[4];
// r[0] = ~ia[0] & ib[0];
// r[1] = ~ia[1] & ib[1];
// r[2] = ~ia[2] & ib[2];
// r[3] = ~ia[3] & ib[3];
// __m128i ret = do_mm_set_epi32(r[3], r[2], r[1], r[0]);
// result_t res = TEST_SUCCESS;
// res = VALIDATE_INT32_M128(c, r);
// if (res) {
// res = VALIDATE_INT32_M128(ret, r);
// }
// return res;
// #else
const uint32_t *ia = (const uint32_t *)&a;
const uint32_t *ib = (const uint32_t *)&b;
uint32_t r[4];
r[0] = ~ia[0] & ib[0];
r[1] = ~ia[1] & ib[1];
r[2] = ~ia[2] & ib[2];
r[3] = ~ia[3] & ib[3];
__m128i ret = do_mm_set_epi32(r[3], r[2], r[1], r[0]);
result_t res = TEST_SUCCESS;
res = VALIDATE_INT32_M128(c, r);
if (res) {
res = VALIDATE_INT32_M128(ret, r);
}
return res;
#else
return TEST_UNIMPL;
// #endif // ENABLE_TEST_ALL
#endif // ENABLE_TEST_ALL
}

result_t test_mm_avg_epu16(const SSE2RVV_TEST_IMPL &impl, uint32_t iter) {
Expand Down

0 comments on commit 92933b4

Please sign in to comment.