HLS DATAFLOW & HLS stream ERROR: [XFORM 203-711]

最新推荐文章于 2022-11-06 11:25:40 发布

jerwey

最新推荐文章于 2022-11-06 11:25:40 发布

阅读量2.4k

点赞数 1

分类专栏： HLS

本文链接：https://blog.csdn.net/jerwey/article/details/103896802

版权

HLS 专栏收录该内容

26 篇文章 6 订阅

订阅专栏

http://www.elecfans.com/d/921179.html

ERROR: [XFORM 203-711]

Description
The specified variable is both written to and read from. This violates the requirements for using the DATAFLOW directive, which supports only single-producer and single-consumer variable. The variable must only be read from or written to.
Explanation
The DATAFLOW directive allows functions and loops to consume data as soon as data is available from the producer, allowing task level pipelining where tasks can execute in parallel (each task does not have to complete all operations before the next task can begin).
All variables in a DATAFLOW region must operate in a single-producer single-consumer fashion: data must flow from one task to the next. If the coding style violates this policy, the DATAFLOW directive cannot implement task level pipelining.
Solution
Errors in use of the DATAFLOW directive require coding changes to ensure the code follows a single-producer single-consumer model. To more easily understand the code structure, use the dataflow viewer within the Analysis Perspective in the Vivado HLS GUI. The dataflow viewer is available when the DATAFLOW directive has been specified. Care should be taken to review the code inside of tasks (loops or functions) and notice some cases imply both reading and writing to the same variable. In the following example,
C_matrix[i][j] += A_matrix[i][k]*B_matrix[k][j];
It is easier to see the violation when the code is re-written as (C_matrix is both read from and written to)
C_matrix[i][j] = C_matrix[i][j] + (A_matrix[i][k]*B_matrix[k][j]);
This case is also present when slices are written to a larger variable: HLS uses a read-modify-write operation when writing to parts of a larger word.

ERROR: [XFORM 203-711]
描述
pragma指定的变量既被写又被读。这违反了使用DATAFLOW指令的要求，该指令只支持单个生产者和单个消费者变量。只能从变量中读取或写入变量。
解释
DATAFLOW指令允许functions 和loops在生产者提供数据时立即使用数据，允许任务级流水线，任务可以并行执行(每个任务不必在下一个任务开始之前完成所有操作)。
DATAFLOW 区域中的所有变量必须以单生产者单消费者的方式操作:数据必须从一个任务流向下一个任务。如果编码样式违反此策略，则DATAFLOW指令无法实现任务级流水线。
解决方案
使用DATAFLOW指令时出现的错误要求修改代码，要确保代码遵循单生产者单消费者模型。要更容易地理解代码结构，请在Vivado HLS GUI的分析透视图 Analysis Perspective中使用dataflow查看器。当指定了dataflow指令时，dataflow查看器是可用的。应该仔细检查任务(loops or functions)内部的代码，并注意同时读写相同的变量的情况。在下面的例子中，

C_matrix[i][j] += A_matrix[i][k]*B_matrix[k][j];

当代码被重写为(C_matrix既可以被读，也可以被写)时，就更容易发现冲突

C_matrix[i][j] = C_matrix[i][j] + (A_matrix[i][k]*B_matrix[k][j]);

这种情况也出现在将slices 写入较大的变量时:HLS在写入较大word的部分时，使用read-modify-write操作。

Description
Vivado HLS issues this message when dataflow is requested for code that does not conform to the canonical dataflow form.
Explanation
Use Help > User Guide to open the Vivado HLS User Guide (UG902) and review the “Dataflow Optimization Limitations” section for an explanation of the supported canonical dataflow form. Solution It is recommended that all code use the canonical dataflow form. This message can be disabled via by setting config_dataflow -strict_mode to “off”.
描述
在不符合规范dataflow form的代码要求dataflow时,Vivado HLS发出此消息。
解释
Vivado HLS用户指南(UG902)，并查看“Dataflow Optimization Limitations”一节，了解支持的规范数据流表单。
解决方案
建议所有代码都使用规范的数据流形式。可以通过将config_dataflow -strict_mode设置为“off”来禁用此消息。