Skip to content

Commit 2de1aac

Browse files
terfendailterfendail
committed
Updated alignment declarations to CV_DECL_ALIGNED macro
1 parent 59373a1 commit 2de1aac

File tree

1 file changed

+35
-138
lines changed

1 file changed

+35
-138
lines changed

‎modules/imgproc/src/imgwarp.cpp

Lines changed: 35 additions & 138 deletions
Original file line numberDiff line numberDiff line change
@@ -435,13 +435,7 @@ class resizeNNInvokerAVX4 :
435435
int y, x, pix_size = (int)src.elemSize();
436436
int width = dsize.width;
437437
int avxWidth = width - (width & 0x7);
438-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
439-
const __declspec(align(64)) __m256i mask = _mm256_set1_epi32(-1);
440-
#elif defined __GNUC__ && __GNUC__ >= 4
441-
const __m256i mask __attribute__((aligned(64))) = _mm256_set1_epi32(-1);
442-
#else
443-
const __m256i mask = _mm256_set1_epi32(-1);
444-
#endif
438+
const __m256i CV_DECL_ALIGNED(64) mask = _mm256_set1_epi32(-1);
445439
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
446440
{
447441
for(y = range.start; y < range.end; y++)
@@ -453,19 +447,9 @@ class resizeNNInvokerAVX4 :
453447
#pragma unroll(4)
454448
for(x = 0; x < avxWidth; x += 8)
455449
{
456-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
457-
__declspec(align(64)) const __m256i *addr = (__m256i*)(x_ofs + x);
458-
__declspec(align(64)) __m256i indices = _mm256_lddqu_si256(addr);
459-
__declspec(align(64)) __m256i pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
460-
#elif defined __GNUC__ && __GNUC__ >= 4
461-
const __m256i *addr __attribute__((aligned(64))) = (__m256i*)(x_ofs + x);
462-
__m256i indices __attribute__((aligned(64))) = _mm256_lddqu_si256(addr);
463-
__m256i pixels __attribute__((aligned(64))) = _mm256_i32gather_epi32((const int*)S, indices, 1);
464-
#else
465-
const __m256i *addr = (__m256i*)(x_ofs + x);
466-
__m256i indices = _mm256_lddqu_si256(addr);
467-
__m256i pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
468-
#endif
450+
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
451+
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
452+
__m256i CV_DECL_ALIGNED(64) pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
469453
_mm256_maskstore_epi32((int*)D, mask, pixels);
470454
D += 32;
471455
}
@@ -486,19 +470,9 @@ class resizeNNInvokerAVX4 :
486470
#pragma unroll(4)
487471
for(x = 0; x < avxWidth; x += 8)
488472
{
489-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
490-
__declspec(align(64)) const __m256i *addr = (__m256i*)(x_ofs + x);
491-
__declspec(align(64)) __m256i indices = _mm256_lddqu_si256(addr);
492-
__declspec(align(64)) __m256i pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
493-
#elif defined __GNUC__ && __GNUC__ >= 4
494-
const __m256i *addr __attribute__((aligned(64))) = (__m256i*)(x_ofs + x);
495-
__m256i indices __attribute__((aligned(64))) = _mm256_lddqu_si256(addr);
496-
__m256i pixels __attribute__((aligned(64))) = _mm256_i32gather_epi32((const int*)S, indices, 1);
497-
#else
498-
const __m256i *addr = (__m256i*)(x_ofs + x);
499-
__m256i indices = _mm256_lddqu_si256(addr);
500-
__m256i pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
501-
#endif
473+
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
474+
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
475+
__m256i CV_DECL_ALIGNED(64) pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
502476
_mm256_storeu_si256((__m256i*)D, pixels);
503477
D += 32;
504478
}
@@ -538,29 +512,12 @@ class resizeNNInvokerAVX2 :
538512
int width = dsize.width;
539513
//int avxWidth = (width - 1) - ((width - 1) & 0x7);
540514
int avxWidth = width - (width & 0xf);
541-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
542-
const __declspec(align(64)) __m256i mask = _mm256_set1_epi32(-1);
543-
const __declspec(align(64)) __m256i shuffle_mask = _mm256_set_epi8(15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0,
544-
15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0);
545-
const __declspec(align(64)) __m256i permute_mask = _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0);
546-
const __declspec(align(64)) __m256i shift_shuffle_mask = _mm256_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2,
547-
13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);
548-
#elif defined __GNUC__ && __GNUC__ >= 4
549-
const __m256i mask __attribute__((aligned(64))) = _mm256_set1_epi32(-1);
550-
const __m256i shuffle_mask __attribute__((aligned(64))) = _mm256_set_epi8(15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0,
551-
15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0);
552-
const __m256i permute_mask __attribute__((aligned(64))) = _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0);
553-
const __m256i shift_shuffle_mask __attribute__((aligned(64))) = _mm256_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2,
554-
13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);
555-
556-
#else
557-
const __m256i mask = _mm256_set1_epi32(-1);
558-
const __m256i shuffle_mask = _mm256_set_epi8(15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0,
559-
15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0);
560-
const __m256i permute_mask = _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0);
561-
const __m256i shift_shuffle_mask = _mm256_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2,
562-
13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);
563-
#endif
515+
const __m256i CV_DECL_ALIGNED(64) mask = _mm256_set1_epi32(-1);
516+
const __m256i CV_DECL_ALIGNED(64) shuffle_mask = _mm256_set_epi8(15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0,
517+
15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0);
518+
const __m256i CV_DECL_ALIGNED(64) permute_mask = _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0);
519+
const __m256i CV_DECL_ALIGNED(64) shift_shuffle_mask = _mm256_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2,
520+
13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);
564521
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
565522
{
566523
for(y = range.start; y < range.end; y++)
@@ -573,40 +530,16 @@ class resizeNNInvokerAVX2 :
573530
#pragma unroll(4)
574531
for(x = 0; x < avxWidth; x += 16)
575532
{
576-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
577-
__declspec(align(64)) const __m256i *addr = (__m256i*)(x_ofs + x);
578-
__declspec(align(64)) __m256i indices = _mm256_lddqu_si256(addr);
579-
__declspec(align(64)) __m256i pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
580-
__declspec(align(64)) const __m256i *addr2 = (__m256i*)(x_ofs + x + 8);
581-
__declspec(align(64)) __m256i indices2 = _mm256_lddqu_si256(addr2);
582-
__declspec(align(64)) __m256i pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
583-
__declspec(align(64)) __m256i unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
584-
585-
__declspec(align(64)) __m256i bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
586-
__declspec(align(64)) __m256i ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
587-
#elif defined __GNUC__ && __GNUC__ >= 4
588-
const __m256i *addr __attribute__((aligned(64))) = (__m256i*)(x_ofs + x);
589-
__m256i indices __attribute__((aligned(64))) = _mm256_lddqu_si256(addr);
590-
__m256i pixels1 __attribute__((aligned(64))) = _mm256_i32gather_epi32((const int*)S, indices, 1);
591-
const __m256i *addr2 __attribute__((aligned(64))) = (__m256i*)(x_ofs + x + 8);
592-
__m256i indices2 __attribute__((aligned(64))) = _mm256_lddqu_si256(addr2);
593-
__m256i pixels2 __attribute__((aligned(64))) = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
594-
__m256i unpacked __attribute__((aligned(64))) = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
595-
596-
__m256i bytes_shuffled __attribute__((aligned(64))) = _mm256_shuffle_epi8(unpacked, shuffle_mask);
597-
__m256i ints_permuted __attribute__((aligned(64))) = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
598-
#else
599-
const __m256i *addr = (__m256i*)(x_ofs + x);
600-
__m256i indices = _mm256_lddqu_si256(addr);
601-
__m256i pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
602-
const __m256i *addr2 = (__m256i*)(x_ofs + x + 8);
603-
__m256i indices2 = _mm256_lddqu_si256(addr2);
604-
__m256i pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
605-
__m256i unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
606-
607-
__m256i bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
608-
__m256i ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
609-
#endif
533+
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
534+
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
535+
__m256i CV_DECL_ALIGNED(64) pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
536+
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
537+
__m256i CV_DECL_ALIGNED(64) indices2 = _mm256_lddqu_si256(addr2);
538+
__m256i CV_DECL_ALIGNED(64) pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
539+
__m256i CV_DECL_ALIGNED(64) unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
540+
541+
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
542+
__m256i CV_DECL_ALIGNED(64) ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
610543
_mm256_maskstore_epi32((int*)D, mask, ints_permuted);
611544
D += 32;
612545
}
@@ -629,40 +562,16 @@ class resizeNNInvokerAVX2 :
629562
#pragma unroll(4)
630563
for(x = 0; x < avxWidth; x += 16)
631564
{
632-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
633-
__declspec(align(64)) const __m256i *addr = (__m256i*)(x_ofs + x);
634-
__declspec(align(64)) __m256i indices = _mm256_lddqu_si256(addr);
635-
__declspec(align(64)) __m256i pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
636-
__declspec(align(64)) const __m256i *addr2 = (__m256i*)(x_ofs + x + 8);
637-
__declspec(align(64)) __m256i indices2 = _mm256_lddqu_si256(addr2);
638-
__declspec(align(64)) __m256i pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
639-
__declspec(align(64)) __m256i unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
640-
641-
__declspec(align(64)) __m256i bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
642-
__declspec(align(64)) __m256i ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
643-
#elif defined __GNUC__ && __GNUC__ >= 4
644-
const __m256i *addr __attribute__((aligned(64))) = (__m256i*)(x_ofs + x);
645-
__m256i indices __attribute__((aligned(64))) = _mm256_lddqu_si256(addr);
646-
__m256i pixels1 __attribute__((aligned(64))) = _mm256_i32gather_epi32((const int*)S, indices, 1);
647-
const __m256i *addr2 __attribute__((aligned(64))) = (__m256i*)(x_ofs + x + 8);
648-
__m256i indices2 __attribute__((aligned(64))) = _mm256_lddqu_si256(addr2);
649-
__m256i pixels2 __attribute__((aligned(64))) = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
650-
__m256i unpacked __attribute__((aligned(64))) = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
651-
652-
__m256i bytes_shuffled __attribute__((aligned(64))) = _mm256_shuffle_epi8(unpacked, shuffle_mask);
653-
__m256i ints_permuted __attribute__((aligned(64))) = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
654-
#else
655-
const __m256i *addr = (__m256i*)(x_ofs + x);
656-
__m256i indices = _mm256_lddqu_si256(addr);
657-
__m256i pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
658-
const __m256i *addr2 = (__m256i*)(x_ofs + x + 8);
659-
__m256i indices2 = _mm256_lddqu_si256(addr2);
660-
__m256i pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
661-
__m256i unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
662-
663-
__m256i bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
664-
__m256i ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
665-
#endif
565+
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
566+
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
567+
__m256i CV_DECL_ALIGNED(64) pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
568+
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
569+
__m256i CV_DECL_ALIGNED(64) indices2 = _mm256_lddqu_si256(addr2);
570+
__m256i CV_DECL_ALIGNED(64) pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
571+
__m256i CV_DECL_ALIGNED(64) unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
572+
573+
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
574+
__m256i CV_DECL_ALIGNED(64) ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
666575
_mm256_storeu_si256((__m256i*)D, ints_permuted);
667576
D += 32;
668577
}
@@ -709,13 +618,7 @@ class resizeNNInvokerSSE2 :
709618
uchar* Dstart = D;
710619
int sy = std::min(cvFloor(y*ify), ssize.height-1);
711620
const uchar* S = src.data + sy*src.step;
712-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
713-
__declspec(align(64)) __m128i pixels = _mm_set1_epi16(0);
714-
#elif defined __GNUC__ && __GNUC__ >= 4
715-
__m128i pixels __attribute__((aligned(64))) = _mm_set1_epi16(0);
716-
#else
717-
__m128i pixels = _mm_set1_epi16(0);
718-
#endif
621+
__m128i CV_DECL_ALIGNED(64) pixels = _mm_set1_epi16(0);
719622
for(x = 0; x < sseWidth; x += 8)
720623
{
721624
ushort imm = *(ushort*)(S + x_ofs[x + 0]);
@@ -776,13 +679,7 @@ class resizeNNInvokerSSE4 :
776679
uchar* Dstart = D;
777680
int sy = std::min(cvFloor(y*ify), ssize.height-1);
778681
const uchar* S = src.data + sy*src.step;
779-
#if (defined WIN32 || defined _WIN32 || defined __CYGWIN__)
780-
__declspec(align(64)) __m128i pixels = _mm_set1_epi16(0);
781-
#elif defined __GNUC__ && __GNUC__ >= 4
782-
__m128i pixels __attribute__((aligned(64))) = _mm_set1_epi16(0);
783-
#else
784-
__m128i pixels = _mm_set1_epi16(0);
785-
#endif
682+
__m128i CV_DECL_ALIGNED(64) pixels = _mm_set1_epi16(0);
786683
for(x = 0; x < sseWidth; x += 4)
787684
{
788685
int imm = *(int*)(S + x_ofs[x + 0]);

0 commit comments

Comments
 (0)