HLS generates an unknown error

[hguerard]

The for-loop in the following code generates an unknow error. Any idea ?

producer0.cpp

#include "producer0.h"
#include <stdint.h>
#define OFFSET_DDR 0x20000000
using sc_core::sc_module;

producer0::producer0(sc_core::sc_module_name name)
:sc_module(name) {
	SC_CTHREAD(thread, clock.pos());
	reset_signal_is(reset, true);
}

void producer0::thread() {
	unsigned long data[16] = { 0 };

	while(1) {
		data[0] = 1;
		for(int i=1;i<16;i++)
			data[i] += 1;
		axi_master_0->burst_write(OFFSET_DDR, 16, (int*)data);
	}
}

producer0.h

#ifndef PRODUCER0_H_
#define PRODUCER0_H_

#include "systemc"

using namespace sc_core;
using namespace std;
#include "AXI4_if.h"

class producer0: public sc_core::sc_module {
	public:
		SC_HAS_PROCESS(producer0);

		sc_core::sc_in< bool > clock;
		sc_core::sc_in< bool > reset;
		AXI4M_bus_port< int > axi_master_0;
		
		producer0(sc_core::sc_module_name name);

		void thread();
};

#endif

The error is the following:

INFO: [SCHED 204-11] Starting scheduling ...
INFO: [SCHED 204-61] Pipelining loop 'memcpy.producer0.axi_master_0.m_if.Val.data.gep'.
INFO: [SCHED 204-61] Pipelining result : Target II = 1, Final II = 1, Depth = 3.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
WARNING: [SCHED 204-69] Cannot apply resource constraint between nodes '' and '' due to unknow reason.
INFO: [SCHED 204-11] Finished scheduling.
INFO: [HLS 200-111]  Elapsed time: 28.859 seconds; current allocated memory: 95.070 MB.
INFO: [HLS 200-434] Only 1 loops out of a total 3 loops have been pipelined in this design.
INFO: [BIND 205-100] Starting micro-architecture generation ...
INFO: [BIND 205-101] Performing variable lifetime analysis.
INFO: [BIND 205-101] Exploring resource sharing.
ERROR: [BIND 205-201] The memory core 'RAM' has insufficient ports, for array 'data' in function 'producer0::thread'

[shameera]

Hi @hguerard ,

Can you please share the directives which you are using for the above code ?

[hguerard]

Hi @shameera,

There is no directive.

[archangel-lightworks]

threads? HLS? could you explain how you plan to use threads in hardware?

[hguerard]

This is only the method name. It is not a std::thread nor a posix thread. On the other hand, this is a SystemC module with a SC_CTHREAD which is supported by Vivado HLS. Read page 354 of UG902.

[shameera]

Hi @hguerard ,

Can you please share the producer0.h file ?

[hguerard]

@shameera, I have update the initial post to include the producer0.h file

[archangel-lightworks]

@hguerard 

This:

INFO: [SCHED 204-61] Pipelining result : Target II = 1, Final II = 1, Depth = 3

ERROR: [BIND 205-201] The memory core 'RAM' has insufficient ports, for array 'data' in function 'producer0::thread'

 makes me think you are asking the synthesis to do more things in parallel than possible with a dual(?) port RAM buffer.

[hguerard]

I did not instruct HLS to pipeline the for-loop. It is doing this by his own means. The config_compile command does not specify to automatically pipeline for-loop.

To feed the discussion, this simplified for-loop (which does not use dual-port RAM) outputs the same log.

		data[0] = 0;
		for(int i=1;i<16;i++)
			data[i] = 1;

My end goal is to somehow read a seed and initialize the first element of the array with it as such:

		seed = 0; /* somehow read the seed */
		data[0] = seed;
		for(int i=1;i<16;i++)
			data[i] = data[i-1] + 1;

But no matter I iterates over the array, I always stumble over the same warning and error.

To overcome the dual-port problem, I tried to partition the array with #pragma HLS array_partition variable=data complete but the HLS really chokes on it (complains about a json file wrongly generated):

ERROR: [HLS 200-445] Unexpected error generating RTL model: generate_json error: error "not well-formed (invalid token)" at line 46 character 44
"ster_0.m_if.Val.data[ <--Error-- 0]>

[archangel-lightworks]

You wrote no directives, you didn't constrain HLS to II=1, but there it is doing it.

And you have SystemC stuff in HLS that is "supported" (I've heard that so many times and then...)

The very first thing I would do is to write that thing in HLS and forget about fancy threads. I bet is not even 100 lines. You may now that piece of advice 'Keep It Simple'

[hguerard]

My guess is the for-loop being pipelined is somewhere under Xilinx's implementation of the AXI4M_bus_port class.

I pretty sure if I write pure C HLS code, I can get this thing working. However, in our design flow, we use a CAD tool that outputs SystemC module for Vivado HLS.

[shameera]

Hi @hguerard ,

I am getting the below error when trying to synthesize the shared code

INFO: [HLS 200-489] Unrolling loop with tag "__ssdm_reset__"in function 'producer0::thread' completely with a factor of 15.
ERROR: [XFORM 203-801] Interface parameter 'producer0.axi_master_0.m_if.Val' must have the same bitwidth with the local buffer.
ERROR: [HLS 200-70] Pre-synthesis failed.

I tried it in HLS 2019.2 , can you please share the HLS version you are using ?

[hguerard]

I am using Vivado HLS 2018.3. Note that I have modified the original post to include a zip file that contains all the files.

Below is the TCL script used to create and do the synthesis of the project:

open_project hls_project
add_files producer0.cpp
open_solution "solution1"
set_part {xc7z020clg484-1}
create_clock -period 10.0
set_clock_uncertainty 3.0
config_rtl -reset_level high
csynth_design
exit

[hguerard]

@shameera did you have the chance to try it with HLS 2018.3 ?

[shameera]

Hi @hguerard ,

I tried in HLS 2018.3 too, but still I see the same error which I had mentioned in my last post.

ERROR: [XFORM 203-801] Interface parameter 'producer0.axi_master_0.m_if.Val' must have the same bitwidth with the local buffer.

Here I have attached the HLS files for your reference. Use the command Vivado_hls -f ./build.tcl for building the project.

[hguerard]

@shameera I have execute your tcl script but I have a totally different output. The main different is I am running on Windows 10 64-bit and you appear to be running under CentOS.

I have attached the generated Vivado HLS + log under Windows 10 with Vivado HLS 2018.3. What would you suggest from this point ?

[hguerard]

Hi @shameera,

I run the project under a VM Ubuntu 64 bit and I have the same issue as you noted. To make it work, you have to change the line unsigned long data[16] = { 0 }; for int data[16] = { 0 };

Anyhow, I have added the modified source file with this post.

With this modification, you will be able to reproduce the initial error.

[hguerard]

@shameera were you able to test this zip file ?

[shameera · Accepted solution]

HI @hguerard ,

Yes, I was able to look into it. But somehow I missed to share my observations.

It seems the RAM error was occurring due to the while loop {while (1)}, changing the while loop to use for loop {for ( ; ; )} resolves the error and synthesizes the design without any errors.

[hguerard]

Wow, good finding! Thanks a lot! By any change, do you know if the while(1) error will be corrected in an upcoming version of HLS ?