| Rev |
Log message |
Author |
Age |
Path |
| 71 |
Modelsim whines about missing timescales:
-minsoc_bench.v, minsoc_memory_model.v and dbg_comm_vpi.v updated to include timescale.v
modelsim simulation working under Windows. Modelsim PE Student Edition 10.0a
Simulation<->adv_jtag_bridge<->gdb communication not working perfectly. I believe VPI module has to be recompiled, used pre-compiled one for now. |
rfajardo |
4926d 15h |
/minsoc/trunk/bench/verilog/ |
| 69 |
backend update:
-minsoc_bench_defines.v
-gcc-opt.mk
Both files should now be under minsoc/backend to proper system functionality.
backend subdirectories have been given those files. Configure script updated accordingly.
They are searched there from system scripts and Makefiles.
-sim/bin/minsoc_verilog_files.txt has the files for Icarus Verilog, minsoc_bench_defines.v is now referenced from backend directory.
-sw/support/Makefile.inc now references to gcc-opt.mk inside backend.
backend/spartan3e_starter_kit_eth:
-It is the system configuration for Spartan 3E Starter Kit with Ethernet. |
rfajardo |
4931d 07h |
/minsoc/trunk/bench/verilog/ |
| 64 |
firmware makefiles:
-every firmware makefile has now complete dependency. This also includes dependency on files under minsoc/backend (target specific files). That means, that if some target specific header changes, the support library dependent on it will be compiled. That will always happen, even if you compile the uart firmware. In other words, if you want to use uart firmware, you can always simply compile uart by issuing make all. If anything has changed, backend files, support library, drivers, it will update everything for you.
-TODO: dependency can be automatic created by using make together with gcc. Use it instead of declaring all dependencies manually.
Makefile system for synthesis:
-the dependency for every implementation step has been checked and is working fine.
-Makefile plus support files have been moved to minsoc/syn/src
-make usage is still under syn through files generated by backend bashscripts
Backend:
-files under backend are target specific files used for the system to work
-firmware compilation
-system simulation
-system implementation
-This directory is populated by visiting one of its subdirectories and typing ./configure
backend/spartan3a_dsp_kit:
-working on FPGA
backend/spartan3e_starter_kit:
-has to be tested
backend/ml509:
-missing files have to be copied and adapted, configure script has to be copied and adapted
backend/std:
-include files necessary for firmware compilation and system configuration
-it does not include files for synthesis and is not synthesizable |
rfajardo |
4933d 14h |
/minsoc/trunk/bench/verilog/ |
| 60 |
Selection of memory model or implementation memory is now made on minsoc_bench_defines.v. It is done by a definition instead of including different files for simulation.
minsoc_bench_defines.v definition of reset level was not correct. It based the level decision on defineds POSITIVE_RESET or NEGATIVE_RESET, which couldn't be defined by then, since minsoc_defines.v is not included in minsoc_bench_defines.v. The decision has been moved to minsoc_bench.v and made a localparam instead of a definition. |
rfajardo |
4938d 02h |
/minsoc/trunk/bench/verilog/ |
| 59 |
undefinition of NEGATIVE_RESET on minsoc_bench_defines.v cannot affect other inclusions of minsoc_defines.v. Instead, the testbench now works with the right reset level to avoid the implementation ordering problem. |
rfajardo |
4938d 03h |
/minsoc/trunk/bench/verilog/ |
| 58 |
Standard definitions depended on implementation order. Now, this should be solved.
minsoc_bench_defines.v: when setting reset to be positive (`define POSITIVE_RESET), NEGATIVE_RESET is undefined. This override the implementation order, so that independent of it, POSITIVE_RESET will be used.
minsoc_defines.v: when setting GENERIC_FPGA, FPGA_TAP and FPGA_CLOCK_DIVISION are undefined. This way, even if FPGA_TAP would come prior to GENERIC_TAP on the correspondent implementation, GENERIC_TAP would still be selected.
IMPORTANT: GENERIC_MEMORY must still be implemented first on minsoc_onchip_ram.v, because FPGA's memory is automatically selected from other definitions and cannot be undefined a priori. Since some other memory types can be selected, there is no trivial solution. However, this shouldn't be a big problem, since the beginning of this file will probably not be modified. |
rfajardo |
4938d 03h |
/minsoc/trunk/bench/verilog/ |
| 28 |
1) Period calculations through 1/freq on testbench use now a numerator definition in order to extract nano seconds of the divisions. Previously the number 1e9 was being repeatedly typed as numerator, now FREQ_NUM_FOR_NS is used.
2) There is a possibility of enabling the GENERIC_CLOCK_DIVISION for the testbench, so that you can test the outcome of different system clock inputs and internal clock adjustments. To do so, NO_CLOCK_DIVISION definition of minsoc_bench_defines.v has to be commented out.
-This also requested the initialization of the internal registers clk_int and clock_divisor of the minsoc_clock_manager.v, this is made by the testbench in case the NO_CLOCK_DIVISION definition is NOT defined.
3) Howto part of implementing Ethernet for the Spartan3E Starter Kit has been completely described. (Chapter 7, subitem 3) |
rfajardo |
5296d 10h |
/minsoc/trunk/bench/verilog/ |
| 27 |
Simulation library fpga_memory_primitives.v had an issue with its lpm_ram_dq module, which did not output its data.
The data was being output to doq instead of q, the declared output. doq was also not defined anywhere else.
Icarus Verilog did not detect this, because Verilog-2001 allows internal wires to be used without being defined. To detect this errors, one can define "`default_nettype none". After doing this, Icarus Verilog detected that error and nothing else.
doq changed to q, error corrected. |
rfajardo |
5311d 11h |
/minsoc/trunk/bench/verilog/ |
| 17 |
Ethernet testbench speed penalty solved. Now Ethernet of testbench and minsoc can be enabled by only uncommenting `define ETHERNET on minsoc_defines.v.
send_mac, get_mac and uart_send tasks have been included/improved. Also a testbench, which works for both included firmwares is added. (eth and uart)
If ETHERNET is defined for the SoC, both firmwares will complete successfully. If not, the eth firmware will stall when trying to access the Ethernet module. |
rfajardo |
5465d 11h |
/minsoc/trunk/bench/verilog/ |
| 11 |
External interrupt processing was being run multiple times because:
-external level interrupts have to be cleared
-internal interrupt status register has to be cleared
Since internal interrupt status register was being cleared before external level interrupts clearance, these internal interrupt status was being overwritten inbetween.
Solution:
-move status register reset to end of interrupt handler instead of beginning.
Testbench signal uart_srx initialized now. |
rfajardo |
5490d 10h |
/minsoc/trunk/bench/verilog/ |
| 10 |
Added a file containing models for each FPGA memory instances used in or1200. The file is in bench/verilog/sim_lib/fpga_memory_primitives.v.
With it, people who change the or1200_defines.v inside of the project structure will still be able to simulate, using house-made models, not from manufacturers.
minsoc_bench.v had to be extended by the task, init_fpga_memory, to initialize the dual or two port memories instances of or1200. This has to be done based on the type of memory used, so many different instantiations based on definitions. Somehow or1200 expects all memory values to be 0 upon start, so this is necessary. |
rfajardo |
5504d 10h |
/minsoc/trunk/bench/verilog/ |
| 9 |
Tiny change to testbench gain:
-uart_srx is now reg for future testbench serial input to SoC. |
rfajardo |
5506d 09h |
/minsoc/trunk/bench/verilog/ |
| 8 |
Cosmetic changes to minsoc_bench.v:
-reset and clock initialization are included into the main initial block, it had an own block before |
rfajardo |
5506d 09h |
/minsoc/trunk/bench/verilog/ |
| 7 |
Some changes:
-wb_cabs removed from minsoc_top.v and minsoc_tc_top.v
-added reset polarity control to minsoc_defines.v through:
-POSITIVE_RESET
-NEGATIVE_RESET
-minsoc_onchip_ram_top.v does not use
minsoc_onchip_ram.v oe signals (output enable) anymore,
which are implemented as tristate buffers. Now
minsoc_onchip_ram_top.v has a generated MUX, which
has an arbitrary number of inputs and 1 output.
Input are the internal output of the onchip_rams,
output the wb_dat_o. |
rfajardo |
5511d 09h |
/minsoc/trunk/bench/verilog/ |
| 4 |
minsoc_bench.v had a big memory declaration to load the firmware, which was not necessary.
jp-io-vpi.vpi has to be compiled for different systems so removed from the project. Documentation now explains how to do it.
Documentation now includes:
-vpi module compilation
-gdb patch for adv_jtag_bridge
-ideas for future work on minsoc (To Do v.2) |
rfajardo |
5521d 15h |
/minsoc/trunk/bench/verilog/ |
| 2 |
First commit of project. Beta status:
-testbench: working
-firmware: working
-RTL: Working for: Xilinx Spartan-3A DSP Development Kit |
rfajardo |
5525d 13h |
/minsoc/trunk/bench/verilog/ |
