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
3f990acc
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
3f990acc
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 #7306 from K-Shinotsuka:issue21
上级
9b12bd91
0d85d2cc
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
25 addition
and
45 deletion
+25
-45
modules/imgproc/src/color.cpp
modules/imgproc/src/color.cpp
+25
-45
未找到文件。
modules/imgproc/src/color.cpp
浏览文件 @
3f990acc
...
...
@@ -4753,7 +4753,6 @@ struct RGB2HLS_b
v_alpha
=
vdup_n_u8
(
ColorChannel
<
uchar
>::
max
());
#elif CV_SSE2
v_scale_inv
=
_mm_set1_ps
(
1.
f
/
255.
f
);
v_scale
=
_mm_set1_ps
(
255.
f
);
v_zero
=
_mm_setzero_si128
();
haveSIMD
=
checkHardwareSupport
(
CV_CPU_SSE2
);
#endif
...
...
@@ -4761,25 +4760,25 @@ struct RGB2HLS_b
#if CV_SSE2
void
process
(
const
float
*
buf
,
__m128
i
&
v_h
,
__m128i
&
v_l
,
__m128i
&
v_s
)
const
__m128
&
v_coeffs
,
uchar
*
dst
)
const
{
__m128
v_h0f
=
_mm_load_ps
(
buf
);
__m128
v_h1f
=
_mm_load_ps
(
buf
+
4
);
__m128
v_l0f
=
_mm_load_ps
(
buf
+
8
);
__m128
v_l1f
=
_mm_load_ps
(
buf
+
12
);
__m128
v_s0f
=
_mm_load_ps
(
buf
+
16
);
__m128
v_s1f
=
_mm_load_ps
(
buf
+
20
);
__m128
v_l0f
=
_mm_load_ps
(
buf
);
__m128
v_l1f
=
_mm_load_ps
(
buf
+
4
);
__m128
v_u0f
=
_mm_load_ps
(
buf
+
8
);
__m128
v_u1f
=
_mm_load_ps
(
buf
+
12
);
_mm_deinterleave_ps
(
v_h0f
,
v_h1f
,
v_l0f
,
v_l1f
,
v_s0f
,
v_s1f
);
v_l0f
=
_mm_mul_ps
(
v_l0f
,
v_coeffs
);
v_u1f
=
_mm_mul_ps
(
v_u1f
,
v_coeffs
);
v_coeffs
=
_mm_castsi128_ps
(
_mm_shuffle_epi32
(
_mm_castps_si128
(
v_coeffs
),
0x92
));
v_u0f
=
_mm_mul_ps
(
v_u0f
,
v_coeffs
);
v_coeffs
=
_mm_castsi128_ps
(
_mm_shuffle_epi32
(
_mm_castps_si128
(
v_coeffs
),
0x92
));
v_l1f
=
_mm_mul_ps
(
v_l1f
,
v_coeffs
);
v_l0f
=
_mm_mul_ps
(
v_l0f
,
v_scale
);
v_l1f
=
_mm_mul_ps
(
v_l1f
,
v_scale
);
v_s0f
=
_mm_mul_ps
(
v_s0f
,
v_scale
);
v_s1f
=
_mm_mul_ps
(
v_s1f
,
v_scale
);
__m128i
v_l
=
_mm_packs_epi32
(
_mm_cvtps_epi32
(
v_l0f
),
_mm_cvtps_epi32
(
v_l1f
));
__m128i
v_u
=
_mm_packs_epi32
(
_mm_cvtps_epi32
(
v_u0f
),
_mm_cvtps_epi32
(
v_u1f
));
__m128i
v_l0
=
_mm_packus_epi16
(
v_l
,
v_u
);
v_h
=
_mm_packs_epi32
(
_mm_cvtps_epi32
(
v_h0f
),
_mm_cvtps_epi32
(
v_h1f
));
v_l
=
_mm_packs_epi32
(
_mm_cvtps_epi32
(
v_l0f
),
_mm_cvtps_epi32
(
v_l1f
));
v_s
=
_mm_packs_epi32
(
_mm_cvtps_epi32
(
v_s0f
),
_mm_cvtps_epi32
(
v_s1f
));
_mm_storeu_si128
((
__m128i
*
)(
dst
),
v_l0
);
}
#endif
...
...
@@ -4787,6 +4786,9 @@ struct RGB2HLS_b
{
int
i
,
j
,
scn
=
srccn
;
float
CV_DECL_ALIGNED
(
16
)
buf
[
3
*
BLOCK_SIZE
];
#if CV_SSE2
__m128
v_coeffs
=
_mm_set_ps
(
1.
f
,
255.
f
,
255.
f
,
1.
f
);
#endif
for
(
i
=
0
;
i
<
n
;
i
+=
BLOCK_SIZE
,
dst
+=
BLOCK_SIZE
*
3
)
{
...
...
@@ -4871,38 +4873,16 @@ struct RGB2HLS_b
#elif CV_SSE2
if
(
haveSIMD
)
{
for
(
;
j
<=
(
dn
-
32
)
*
3
;
j
+=
96
)
for
(
;
j
<=
(
dn
-
16
)
*
3
;
j
+=
48
)
{
__m128i
v_h_0
,
v_l_0
,
v_s_0
;
process
(
buf
+
j
,
v_h_0
,
v_l_0
,
v_s_0
);
__m128i
v_h_1
,
v_l_1
,
v_s_1
;
process
(
buf
+
j
+
24
,
v_h_1
,
v_l_1
,
v_s_1
);
__m128i
v_h0
=
_mm_packus_epi16
(
v_h_0
,
v_h_1
);
__m128i
v_l0
=
_mm_packus_epi16
(
v_l_0
,
v_l_1
);
__m128i
v_s0
=
_mm_packus_epi16
(
v_s_0
,
v_s_1
);
process
(
buf
+
j
+
48
,
v_h_0
,
v_l_0
,
v_s_0
);
v_coeffs
,
dst
+
j
);
process
(
buf
+
j
+
72
,
v_h_1
,
v_l_1
,
v_s_1
);
__m128i
v_h1
=
_mm_packus_epi16
(
v_h_0
,
v_h_1
);
__m128i
v_l1
=
_mm_packus_epi16
(
v_l_0
,
v_l_1
);
__m128i
v_s1
=
_mm_packus_epi16
(
v_s_0
,
v_s_1
);
_mm_interleave_epi8
(
v_h0
,
v_h1
,
v_l0
,
v_l1
,
v_s0
,
v_s1
);
process
(
buf
+
j
+
16
,
v_coeffs
,
dst
+
j
+
16
);
_mm_storeu_si128
((
__m128i
*
)(
dst
+
j
),
v_h0
);
_mm_storeu_si128
((
__m128i
*
)(
dst
+
j
+
16
),
v_h1
);
_mm_storeu_si128
((
__m128i
*
)(
dst
+
j
+
32
),
v_l0
);
_mm_storeu_si128
((
__m128i
*
)(
dst
+
j
+
48
),
v_l1
);
_mm_storeu_si128
((
__m128i
*
)(
dst
+
j
+
64
),
v_s0
);
_mm_storeu_si128
((
__m128i
*
)(
dst
+
j
+
80
),
v_s1
);
process
(
buf
+
j
+
32
,
v_coeffs
,
dst
+
j
+
32
);
}
}
#endif
...
...
@@ -4921,7 +4901,7 @@ struct RGB2HLS_b
float32x4_t
v_scale
,
v_scale_inv
;
uint8x8_t
v_alpha
;
#elif CV_SSE2
__m128
v_scale
,
v_scale
_inv
;
__m128
v_scale_inv
;
__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