Skip to content

Commit

Permalink
MAINT: Remove unnecessary function optimize attribute
Browse files Browse the repository at this point in the history
  • Loading branch information
@r-devulap
r-devulap committed Feb 1, 2022
1 parent 60e7c19 commit f312f00
Showing 1 changed file with 20 additions and 20 deletions.
40 changes: 20 additions & 20 deletions numpy/core/src/npysort/x86-qsort.dispatch.c.src
View file
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@
#define VROTL(x, k) /* rotate each uint64_t value in vector */ \
_mm256_or_si256(_mm256_slli_epi64((x),(k)),_mm256_srli_epi64((x),64-(k)))

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
__m256i vnext(__m256i* s0, __m256i* s1) {
*s1 = _mm256_xor_si256(*s0, *s1); /* modify vectors s1 and s0 */
*s0 = _mm256_xor_si256(_mm256_xor_si256(VROTL(*s0, 24), *s1),
Expand All @@ -73,7 +73,7 @@ __m256i vnext(__m256i* s0, __m256i* s1) {
}

/* transform random numbers to the range between 0 and bound - 1 */
static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
__m256i rnd_epu32(__m256i rnd_vec, __m256i bound) {
__m256i even = _mm256_srli_epi64(_mm256_mul_epu32(rnd_vec, bound), 32);
__m256i odd = _mm256_mul_epu32(_mm256_srli_epi64(rnd_vec, 32), bound);
Expand Down Expand Up @@ -108,7 +108,7 @@ __m256i rnd_epu32(__m256i rnd_vec, __m256i bound) {
a = _mm256_min_@vsuf1@(a, b); \
b = _mm256_max_@vsuf1@(temp, b);} \

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
@zmm_t@ cmp_merge_@vsuf1@(@zmm_t@ in1, @zmm_t@ in2, __mmask16 mask)
{
@zmm_t@ min = _mm512_min_@vsuf1@(in2, in1);
Expand All @@ -120,7 +120,7 @@ static NPY_INLINE NPY_GCC_OPT_3
* Assumes zmm is random and performs a full sorting network defined in
* https://en.wikipedia.org/wiki/Bitonic_sorter#/media/File:BitonicSort.svg
*/
static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
@zmm_t@ sort_zmm_@vsuf1@(@zmm_t@ zmm)
{
zmm = cmp_merge_@vsuf1@(zmm, SHUFFLE_@vsuf2@(zmm, SHUFFLE_MASK(2,3,0,1)), 0xAAAA);
Expand All @@ -137,7 +137,7 @@ static NPY_INLINE NPY_GCC_OPT_3
}

// Assumes zmm is bitonic and performs a recursive half cleaner
static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
@zmm_t@ bitonic_merge_zmm_@vsuf1@(@zmm_t@ zmm)
{
// 1) half_cleaner[16]: compare 1-9, 2-10, 3-11 etc ..
Expand All @@ -152,7 +152,7 @@ static NPY_INLINE NPY_GCC_OPT_3
}

// Assumes zmm1 and zmm2 are sorted and performs a recursive half cleaner
static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void bitonic_merge_two_zmm_@vsuf1@(@zmm_t@* zmm1, @zmm_t@* zmm2)
{
// 1) First step of a merging network: coex of zmm1 and zmm2 reversed
Expand All @@ -165,7 +165,7 @@ void bitonic_merge_two_zmm_@vsuf1@(@zmm_t@* zmm1, @zmm_t@* zmm2)
}

// Assumes [zmm0, zmm1] and [zmm2, zmm3] are sorted and performs a recursive half cleaner
static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void bitonic_merge_four_zmm_@vsuf1@(@zmm_t@* zmm)
{
@zmm_t@ zmm2r = _mm512_permutexvar_@vsuf2@(_mm512_set_epi32(NETWORK5), zmm[2]);
Expand All @@ -184,7 +184,7 @@ void bitonic_merge_four_zmm_@vsuf1@(@zmm_t@* zmm)
zmm[3] = bitonic_merge_zmm_@vsuf1@(zmm3);
}

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void bitonic_merge_eight_zmm_@vsuf1@(@zmm_t@* zmm)
{
@zmm_t@ zmm4r = _mm512_permutexvar_@vsuf2@(_mm512_set_epi32(NETWORK5), zmm[4]);
Expand Down Expand Up @@ -217,15 +217,15 @@ void bitonic_merge_eight_zmm_@vsuf1@(@zmm_t@* zmm)
zmm[7] = bitonic_merge_zmm_@vsuf1@(zmm_t8);
}

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void sort_16_@vsuf1@(@type_t@* arr, npy_int N)
{
__mmask16 load_mask = (0x0001 << N) - 0x0001;
@zmm_t@ zmm = _mm512_mask_loadu_@vsuf2@(ZMM_MAX_@TYPE@, load_mask, arr);
_mm512_mask_storeu_@vsuf2@(arr, load_mask, sort_zmm_@vsuf1@(zmm));
}

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void sort_32_@vsuf1@(@type_t@* arr, npy_int N)
{
if (N <= 16) {
Expand All @@ -242,7 +242,7 @@ void sort_32_@vsuf1@(@type_t@* arr, npy_int N)
_mm512_mask_storeu_@vsuf2@(arr + 16, load_mask, zmm2);
}

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void sort_64_@vsuf1@(@type_t@* arr, npy_int N)
{
if (N <= 32) {
Expand Down Expand Up @@ -275,7 +275,7 @@ void sort_64_@vsuf1@(@type_t@* arr, npy_int N)
_mm512_mask_storeu_@vsuf2@(arr + 48, load_mask2, zmm[3]);
}

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void sort_128_@vsuf1@(@type_t@* arr, npy_int N)
{
if (N <= 64) {
Expand Down Expand Up @@ -337,15 +337,15 @@ void sort_128_@vsuf1@(@type_t@* arr, npy_int N)
}


static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void swap_@TYPE@(@type_t@ *arr, npy_intp ii, npy_intp jj) {
@type_t@ temp = arr[ii];
arr[ii] = arr[jj];
arr[jj] = temp;
}

// Median of 3 stratergy
//static NPY_INLINE NPY_GCC_OPT_3
//static NPY_INLINE
//npy_intp get_pivot_index(@type_t@ *arr, const npy_intp left, const npy_intp right) {
// return (rand() % (right + 1 - left)) + left;
// //npy_intp middle = ((right-left)/2) + left;
Expand All @@ -362,7 +362,7 @@ void swap_@TYPE@(@type_t@ *arr, npy_intp ii, npy_intp jj) {
* Picking the pivot: Median of 72 array elements chosen at random.
*/

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
@type_t@ get_pivot_@vsuf1@(@type_t@ *arr, const npy_intp left, const npy_intp right) {
/* seeds for vectorized random number generator */
__m256i s0 = _mm256_setr_epi64x(8265987198341093849, 3762817312854612374,
Expand Down Expand Up @@ -414,7 +414,7 @@ static NPY_INLINE NPY_GCC_OPT_3
* Parition one ZMM register based on the pivot and returns the index of the
* last element that is less than equal to the pivot.
*/
static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
npy_int partition_vec_@vsuf1@(@type_t@* arr, npy_intp left, npy_intp right,
const @zmm_t@ curr_vec, const @zmm_t@ pivot_vec,
@zmm_t@* smallest_vec, @zmm_t@* biggest_vec)
Expand All @@ -433,7 +433,7 @@ npy_int partition_vec_@vsuf1@(@type_t@* arr, npy_intp left, npy_intp right,
* Parition an array based on the pivot and returns the index of the
* last element that is less than equal to the pivot.
*/
static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
npy_intp partition_avx512_@vsuf1@(@type_t@* arr, npy_intp left, npy_intp right,
@type_t@ pivot, @type_t@* smallest, @type_t@* biggest)
{
Expand Down Expand Up @@ -503,7 +503,7 @@ npy_intp partition_avx512_@vsuf1@(@type_t@* arr, npy_intp left, npy_intp right,
return l_store;
}

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void qsort_@type@(@type_t@* arr, npy_intp left, npy_intp right, npy_int max_iters)
{
/*
Expand Down Expand Up @@ -532,7 +532,7 @@ void qsort_@type@(@type_t@* arr, npy_intp left, npy_intp right, npy_int max_iter
}
/**end repeat**/

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
npy_intp replace_nan_with_inf(npy_float* arr, npy_intp arrsize)
{
npy_intp nan_count = 0;
Expand All @@ -551,7 +551,7 @@ npy_intp replace_nan_with_inf(npy_float* arr, npy_intp arrsize)
return nan_count;
}

static NPY_INLINE NPY_GCC_OPT_3
static NPY_INLINE
void replace_inf_with_nan(npy_float* arr, npy_intp arrsize, npy_intp nan_count)
{
for (npy_intp ii = arrsize-1; nan_count > 0; --ii) {
Expand Down

0 comments on commit f312f00

Please sign in to comment.