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
Xiaomi
Mace
提交
11a55ae0
Mace
项目概览
Xiaomi
/
Mace
通知
106
Star
40
Fork
27
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
Mace
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
11a55ae0
编写于
5月 22, 2018
作者:
L
liuqi
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Update memory layout doc.
上级
e5ada494
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
10 addition
and
10 deletion
+10
-10
docs/development/memory_layout.rst
docs/development/memory_layout.rst
+10
-10
未找到文件。
docs/development/memory_layout.rst
浏览文件 @
11a55ae0
...
@@ -21,10 +21,10 @@ The CPU tensor buffer is organized in the following order:
...
@@ -21,10 +21,10 @@ The CPU tensor buffer is organized in the following order:
* - 1-D Argument, length = W
* - 1-D Argument, length = W
- W
- W
OpenCL
runtime memory layout
GPU
runtime memory layout
-----------------------------
-----------------------------
OpenCL runtime uses 2D image with CL_RGBA channel order as the tensor storage.
GPU runtime implementation base on OpenCL, which uses 2D image with CL_RGBA
This requires OpenCL 1.2 and above.
channel order as the tensor storage.
This requires OpenCL 1.2 and above.
The way of mapping the Tensor data to OpenCL 2D image (RGBA) is critical for
The way of mapping the Tensor data to OpenCL 2D image (RGBA) is critical for
kernel performance.
kernel performance.
...
@@ -53,7 +53,7 @@ The Input/Output Tensor is stored in NHWC format:
...
@@ -53,7 +53,7 @@ The Input/Output Tensor is stored in NHWC format:
- Default Input/Output format
- Default Input/Output format
* - Height-Major Input/Output
* - Height-Major Input/Output
- NHWC
- NHWC
- [W * C, N * (H+3)/4
- [W * C, N * (H+3)/4
]
- Winograd Convolution format
- Winograd Convolution format
* - Width-Major Input/Output
* - Width-Major Input/Output
- NHWC
- NHWC
...
@@ -94,11 +94,11 @@ Filter Tensor
...
@@ -94,11 +94,11 @@ Filter Tensor
- Image size [width, height]
- Image size [width, height]
- Explanation
- Explanation
* - Convolution Filter
* - Convolution Filter
-
HWOI
-
OIHW
- [
RoundUp<4>(I), H * W * (O+3)/4
]
- [
I, (O+3)/4 * W * H
]
- Convolution filter format,There is no difference compared to [H*
w
*I, (O+3)/4]
- Convolution filter format,There is no difference compared to [H*
W
*I, (O+3)/4]
* - Depthwise Convlution Filter
* - Depthwise Convlution Filter
-
HWIM
-
MIHW
- [H * W * M, (I+3)/4]
- [H * W * M, (I+3)/4]
- Depthwise-Convolution filter format
- Depthwise-Convolution filter format
...
@@ -114,10 +114,10 @@ coordination relation between **Image** and **Buffer**.
...
@@ -114,10 +114,10 @@ coordination relation between **Image** and **Buffer**.
- Pixel coordinate relationship
- Pixel coordinate relationship
- Explanation
- Explanation
* - Convolution Filter
* - Convolution Filter
- P[m, n] = {E[
h, w, o, i] | (h=T/W, w=T%W, o=[n/HW*4+k], i=m
)}
- P[m, n] = {E[
o, i, h, w] | (o=[n/HW*4+k], i=m, h=T/W, w=T%W
)}
- HW= H * W, T=n%HW, k=[0, 4)
- HW= H * W, T=n%HW, k=[0, 4)
* - Depthwise Convlution Filter
* - Depthwise Convlution Filter
- P[m, n] = {E[
h, w, i, 0] | (h=m/W, w=m%W, i=[n*4+k]
)}
- P[m, n] = {E[
0, i, h, w] | (i=[n*4+k], h=m/W, w=m%W
)}
- only support multiplier == 1, k=[0, 4)
- only support multiplier == 1, k=[0, 4)
1-D Argument Tensor
1-D Argument Tensor
...
...
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