Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
Greenplum
Opencv
提交
17f0c93b
O
Opencv
项目概览
Greenplum
/
Opencv
11 个月 前同步成功
通知
7
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
DevOps
流水线
流水线任务
计划
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
O
Opencv
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
DevOps
DevOps
流水线
流水线任务
计划
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
流水线任务
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
17f0c93b
编写于
9月 21, 2016
作者:
A
Alexander Alekhin
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #7300 from K-Shinotsuka:issue17
上级
5728795c
4540260d
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
32 addition
and
40 deletion
+32
-40
modules/imgproc/src/color.cpp
modules/imgproc/src/color.cpp
+32
-40
未找到文件。
modules/imgproc/src/color.cpp
浏览文件 @
17f0c93b
...
...
@@ -6072,11 +6072,6 @@ struct Luv2RGB_b
v_scale
=
vdupq_n_f32
(
255.
f
);
v_alpha
=
vdup_n_u8
(
ColorChannel
<
uchar
>::
max
());
#elif CV_SSE2
v_scale_inv
=
_mm_set1_ps
(
100.
f
/
255.
f
);
v_coeff1
=
_mm_set1_ps
(
1.388235294117647
f
);
v_coeff2
=
_mm_set1_ps
(
1.027450980392157
f
);
v_134
=
_mm_set1_ps
(
134.
f
);
v_140
=
_mm_set1_ps
(
140.
f
);
v_scale
=
_mm_set1_ps
(
255.
f
);
v_zero
=
_mm_setzero_si128
();
haveSIMD
=
checkHardwareSupport
(
CV_CPU_SSE2
);
...
...
@@ -6086,6 +6081,7 @@ struct Luv2RGB_b
#if CV_SSE2
// 16s x 8
void
process
(
__m128i
v_l
,
__m128i
v_u
,
__m128i
v_v
,
__m128
v_coeffs
,
__m128
v_res
,
float
*
buf
)
const
{
__m128
v_l0
=
_mm_cvtepi32_ps
(
_mm_unpacklo_epi16
(
v_l
,
v_zero
));
...
...
@@ -6096,15 +6092,26 @@ struct Luv2RGB_b
__m128
v_u1
=
_mm_cvtepi32_ps
(
_mm_unpackhi_epi16
(
v_u
,
v_zero
));
__m128
v_v1
=
_mm_cvtepi32_ps
(
_mm_unpackhi_epi16
(
v_v
,
v_zero
));
v_l0
=
_mm_mul_ps
(
v_l0
,
v_scale_inv
);
v_l1
=
_mm_mul_ps
(
v_l1
,
v_scale_inv
);
v_l0
=
_mm_mul_ps
(
v_l0
,
v_coeffs
);
v_u1
=
_mm_mul_ps
(
v_u1
,
v_coeffs
);
v_l0
=
_mm_sub_ps
(
v_l0
,
v_res
);
v_u1
=
_mm_sub_ps
(
v_u1
,
v_res
);
v_coeffs
=
_mm_castsi128_ps
(
_mm_shuffle_epi32
(
_mm_castps_si128
(
v_coeffs
),
0x49
));
v_res
=
_mm_castsi128_ps
(
_mm_shuffle_epi32
(
_mm_castps_si128
(
v_res
),
0x49
));
v_l1
=
_mm_mul_ps
(
v_l1
,
v_coeffs
);
v_v0
=
_mm_mul_ps
(
v_v0
,
v_coeffs
);
v_l1
=
_mm_sub_ps
(
v_l1
,
v_res
);
v_v0
=
_mm_sub_ps
(
v_v0
,
v_res
);
v_u0
=
_mm_sub_ps
(
_mm_mul_ps
(
v_u0
,
v_coeff1
),
v_134
);
v_u1
=
_mm_sub_ps
(
_mm_mul_ps
(
v_u1
,
v_coeff1
),
v_134
);
v_v0
=
_mm_sub_ps
(
_mm_mul_ps
(
v_v0
,
v_coeff2
),
v_140
);
v_v1
=
_mm_sub_ps
(
_mm_mul_ps
(
v_v1
,
v_coeff2
),
v_140
);
v_coeffs
=
_mm_castsi128_ps
(
_mm_shuffle_epi32
(
_mm_castps_si128
(
v_coeffs
),
0x49
));
v_res
=
_mm_castsi128_ps
(
_mm_shuffle_epi32
(
_mm_castps_si128
(
v_res
),
0x49
));
_mm_interleave_ps
(
v_l0
,
v_l1
,
v_u0
,
v_u1
,
v_v0
,
v_v1
);
v_u0
=
_mm_mul_ps
(
v_u0
,
v_coeffs
);
v_v1
=
_mm_mul_ps
(
v_v1
,
v_coeffs
);
v_u0
=
_mm_sub_ps
(
v_u0
,
v_res
);
v_v1
=
_mm_sub_ps
(
v_v1
,
v_res
);
_mm_store_ps
(
buf
,
v_l0
);
_mm_store_ps
(
buf
+
4
,
v_l1
);
...
...
@@ -6121,6 +6128,11 @@ struct Luv2RGB_b
uchar
alpha
=
ColorChannel
<
uchar
>::
max
();
float
CV_DECL_ALIGNED
(
16
)
buf
[
3
*
BLOCK_SIZE
];
#if CV_SSE2
__m128
v_coeffs
=
_mm_set_ps
(
100.
f
/
255.
f
,
1.027450980392157
f
,
1.388235294117647
f
,
100.
f
/
255.
f
);
__m128
v_res
=
_mm_set_ps
(
0.
f
,
140.
f
,
134.
f
,
0.
f
);
#endif
for
(
i
=
0
;
i
<
n
;
i
+=
BLOCK_SIZE
,
src
+=
BLOCK_SIZE
*
3
)
{
int
dn
=
std
::
min
(
n
-
i
,
(
int
)
BLOCK_SIZE
);
...
...
@@ -6148,36 +6160,16 @@ struct Luv2RGB_b
#elif CV_SSE2
if
(
haveSIMD
)
{
for
(
;
j
<=
(
dn
-
32
)
*
3
;
j
+=
96
)
for
(
;
j
<=
(
dn
-
8
)
*
3
;
j
+=
24
)
{
__m128i
v_r0
=
_mm_loadu_si128
((
__m128i
const
*
)(
src
+
j
));
__m128i
v_r1
=
_mm_loadu_si128
((
__m128i
const
*
)(
src
+
j
+
16
));
__m128i
v_g0
=
_mm_loadu_si128
((
__m128i
const
*
)(
src
+
j
+
32
));
__m128i
v_g1
=
_mm_loadu_si128
((
__m128i
const
*
)(
src
+
j
+
48
));
__m128i
v_b0
=
_mm_loadu_si128
((
__m128i
const
*
)(
src
+
j
+
64
));
__m128i
v_b1
=
_mm_loadu_si128
((
__m128i
const
*
)(
src
+
j
+
80
));
_mm_deinterleave_epi8
(
v_r0
,
v_r1
,
v_g0
,
v_g1
,
v_b0
,
v_b1
);
__m128i
v_src0
=
_mm_loadu_si128
((
__m128i
const
*
)(
src
+
j
));
__m128i
v_src1
=
_mm_loadl_epi64
((
__m128i
const
*
)(
src
+
j
+
16
));
process
(
_mm_unpacklo_epi8
(
v_r0
,
v_zero
),
_mm_unpacklo_epi8
(
v_g0
,
v_zero
),
_mm_unpacklo_epi8
(
v_b0
,
v_zero
),
process
(
_mm_unpacklo_epi8
(
v_src0
,
v_zero
),
_mm_unpackhi_epi8
(
v_src0
,
v_zero
),
_mm_unpacklo_epi8
(
v_src1
,
v_zero
),
v_coeffs
,
v_res
,
buf
+
j
);
process
(
_mm_unpackhi_epi8
(
v_r0
,
v_zero
),
_mm_unpackhi_epi8
(
v_g0
,
v_zero
),
_mm_unpackhi_epi8
(
v_b0
,
v_zero
),
buf
+
j
+
24
);
process
(
_mm_unpacklo_epi8
(
v_r1
,
v_zero
),
_mm_unpacklo_epi8
(
v_g1
,
v_zero
),
_mm_unpacklo_epi8
(
v_b1
,
v_zero
),
buf
+
j
+
48
);
process
(
_mm_unpackhi_epi8
(
v_r1
,
v_zero
),
_mm_unpackhi_epi8
(
v_g1
,
v_zero
),
_mm_unpackhi_epi8
(
v_b1
,
v_zero
),
buf
+
j
+
72
);
}
}
#endif
...
...
@@ -6261,7 +6253,7 @@ struct Luv2RGB_b
float32x4_t
v_scale
,
v_scale_inv
,
v_coeff1
,
v_coeff2
,
v_134
,
v_140
;
uint8x8_t
v_alpha
;
#elif CV_SSE2
__m128
v_scale
,
v_scale_inv
,
v_coeff1
,
v_coeff2
,
v_134
,
v_140
;
__m128
v_scale
;
__m128i
v_zero
;
bool
haveSIMD
;
#endif
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录