mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-25 19:50:38 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

eliminated some unnecessary instructions

Browse Source
This commit is contained in:
Vadim Pisarevsky 2014-07-29 18:10:18 +04:00
parent 11e9e375a3
commit 101769d26c
1 changed files with 15 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
modules/imgproc/src/demosaicing.cpp
View File

@ -359,10 +359,9 @@ public:
uint16x8_t r1 = vld1q_u16((const ushort*)(bayer + bayer_step));
uint16x8_t r2 = vld1q_u16((const ushort*)(bayer + bayer_step*2));
uint16x8_t b1 = vaddq_u16(vandq_u16(r0, masklo), vandq_u16(r2, masklo));
uint16x8_t nextb1 = vextq_u16(b1, b1, 1);
uint16x8_t b0 = vaddq_u16(b1, nextb1);
b1 = vshlq_n_u16(nextb1, 1);
uint16x8_t b1_ = vaddq_u16(vandq_u16(r0, masklo), vandq_u16(r2, masklo));
uint16x8_t b1 = vextq_u16(b1_, b1_, 1);
uint16x8_t b0 = vaddq_u16(b1_, b1);
// b0 = b0 b2 b4 ...
// b1 = b1 b3 b5 ...
@ -374,24 +373,24 @@ public:
// g1 = b1 b3 b5 ...
r0 = vshrq_n_u16(r1, 8);
r1 = vshlq_n_u16(vaddq_u16(r0, vextq_u16(r0, r0, 1)), 1);
r1 = vaddq_u16(r0, vextq_u16(r0, r0, 1));
r0 = vshlq_n_u16(r0, 2);
// r0 = r0 r2 r4 ...
// r1 = r1 r3 r5 ...
b0 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(b0), (short)(rcoeff*2)));
b1 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(b1), (short)(rcoeff*2)));
b1 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(b1), (short)(rcoeff*4)));
g0 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(g0), (short)(gcoeff*2)));
g1 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(g1), (short)(gcoeff*2)));
r0 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(r0), (short)(bcoeff*2)));
r1 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(r1), (short)(bcoeff*2)));
r1 = vreinterpretq_u16_s16(vqdmulhq_n_s16(vreinterpretq_s16_u16(r1), (short)(bcoeff*4)));
g0 = vshrq_n_u16(vaddq_u16(vaddq_u16(g0, b0), r0), 2);
g1 = vshrq_n_u16(vaddq_u16(vaddq_u16(g1, b1), r1), 2);
g0 = vaddq_u16(vaddq_u16(g0, b0), r0);
g1 = vaddq_u16(vaddq_u16(g1, b1), r1);
uint8x8x2_t p = vzip_u8(vqmovn_u16(g0), vqmovn_u16(g1));
uint8x8x2_t p = vzip_u8(vrshrn_n_u16(g0, 2), vrshrn_n_u16(g1, 2));
vst1_u8(dst, p.val[0]);
vst1_u8(dst + 8, p.val[1]);
}
@ -419,26 +418,22 @@ public:
uint16x8_t b1 = vaddq_u16(vandq_u16(r0, masklo), vandq_u16(r2, masklo));
uint16x8_t nextb1 = vextq_u16(b1, b1, 1);
uint16x8_t b0 = vaddq_u16(b1, nextb1);
b1 = vrshrq_n_u16(nextb1, 1);
b0 = vrshrq_n_u16(b0, 2);
// b0 b1 b2 ...
uint8x8x2_t bb = vzip_u8(vmovn_u16(b0), vmovn_u16(b1));
uint8x8x2_t bb = vzip_u8(vrshrn_n_u16(b0, 2), vrshrn_n_u16(nextb1, 1));
pix.val[1-blue] = vcombine_u8(bb.val[0], bb.val[1]);
uint16x8_t g0 = vaddq_u16(vshrq_n_u16(r0, 8), vshrq_n_u16(r2, 8));
uint16x8_t g1 = vandq_u16(r1, masklo);
g0 = vaddq_u16(g0, vaddq_u16(g1, vextq_u16(g1, g1, 1)));
g1 = vextq_u16(g1, g1, 1);
g0 = vrshrq_n_u16(g0, 2);
// g0 g1 g2 ...
uint8x8x2_t gg = vzip_u8(vmovn_u16(g0), vmovn_u16(g1));
uint8x8x2_t gg = vzip_u8(vrshrn_n_u16(g0, 2), vmovn_u16(g1));
pix.val[1] = vcombine_u8(gg.val[0], gg.val[1]);
r0 = vshrq_n_u16(r1, 8);
r1 = vaddq_u16(r0, vextq_u16(r0, r0, 1));
r1 = vrshrq_n_u16(r1, 1);
// r0 r1 r2 ...
uint8x8x2_t rr = vzip_u8(vmovn_u16(r0), vmovn_u16(r1));
uint8x8x2_t rr = vzip_u8(vmovn_u16(r0), vrshrn_n_u16(r1, 1));
pix.val[1+blue] = vcombine_u8(rr.val[0], rr.val[1]);
vst3q_u8(dst-1, pix);
@ -468,26 +463,22 @@ public:
uint16x8_t b1 = vaddq_u16(vandq_u16(r0, masklo), vandq_u16(r2, masklo));
uint16x8_t nextb1 = vextq_u16(b1, b1, 1);
uint16x8_t b0 = vaddq_u16(b1, nextb1);
b1 = vrshrq_n_u16(nextb1, 1);
b0 = vrshrq_n_u16(b0, 2);
// b0 b1 b2 ...
uint8x8x2_t bb = vzip_u8(vmovn_u16(b0), vmovn_u16(b1));
uint8x8x2_t bb = vzip_u8(vrshrn_n_u16(b0, 2), vrshrn_n_u16(nextb1, 1));
pix.val[1-blue] = vcombine_u8(bb.val[0], bb.val[1]);
uint16x8_t g0 = vaddq_u16(vshrq_n_u16(r0, 8), vshrq_n_u16(r2, 8));
uint16x8_t g1 = vandq_u16(r1, masklo);
g0 = vaddq_u16(g0, vaddq_u16(g1, vextq_u16(g1, g1, 1)));
g1 = vextq_u16(g1, g1, 1);
g0 = vrshrq_n_u16(g0, 2);
// g0 g1 g2 ...
uint8x8x2_t gg = vzip_u8(vmovn_u16(g0), vmovn_u16(g1));
uint8x8x2_t gg = vzip_u8(vrshrn_n_u16(g0, 2), vmovn_u16(g1));
pix.val[1] = vcombine_u8(gg.val[0], gg.val[1]);
r0 = vshrq_n_u16(r1, 8);
r1 = vaddq_u16(r0, vextq_u16(r0, r0, 1));
r1 = vrshrq_n_u16(r1, 1);
// r0 r1 r2 ...
uint8x8x2_t rr = vzip_u8(vmovn_u16(r0), vmovn_u16(r1));
uint8x8x2_t rr = vzip_u8(vmovn_u16(r0), vrshrn_n_u16(r1, 1));
pix.val[1+blue] = vcombine_u8(rr.val[0], rr.val[1]);
vst4q_u8(dst-1, pix);