1080p HD Video on custom FPGA/VDAC/2MBx18 SyncRAM board

[ElEctric_EyE]

I modified Jack's square root algorithm on page 86 to work in Verilog. Takes 50 cycles and actually works on 2 16-bit numbers up to a value of 46,340 (for the formula SQRT(X^2+Y^2)), except I had to perform the last shift right 2x, not 1x like in his algorithm on the last line.

Jack's Code:

Code: Select all

unsigned short sqrt(unsigned long a){
unsigned long rem = 0;
unsigned long root = 0;
for(int i=0; i<16; i++){
root <<= 1;
rem = ((rem << 2) + (a >> 30));
a <<= 2;
root ++;
if(root <= rem){
rem -= root;
root++;
}
else
root--;
}
return (unsigned short)(root >> 1);
}

My copy of his work in Verilog. The variable 'a' is 32-bit.:

Code: Select all

.......//SQRT state machine

			SQRTINIT:
				state <= CALC3;
				
			CALC3:
				state <= CALC4;
			
			CALC4:
				state <= CALC5;
				
			CALC5:
				if (i<16)
					state <= CALC3;
					else state <= CALC6;
					
			CALC6:
				state <= WAIT;
............................................
//square root generator

		SQRTINIT:
			begin
				LGREADY <= 0;
				i <= 0;
				rem <= 0;
				Root <= 0;
				a <= Xs*Xs + Ys*Ys;
			end
			
		CALC3:
			begin
				Root <= Root << 1;
				rem <= ((rem << 2) + (a >> 30));
			end
			
		CALC4:
			begin
				a <= a << 2;
				Root <= Root + 1;
			end
			
		CALC5:
			begin
			i <= i + 1;
			if (Root <= rem) begin
				rem <= rem - Root;
				Root <= Root + 1;
			end
				else Root <= Root - 1;
			end			
				
		CALC6:
			Root <= Root >> 2;

At the conclusion of the chapter on page 87 he says:

Quote:

If there’s any one lesson to be learned from this chapter, it is this:
Never trust a person who merely hands you an equation.

So maybe he did this on purpose?

Looks like the ISE 14.7 has offloaded some work to the DSP48A1 tiles:

[MichaelM]

EEyE:

That's cool. I was reading and thinking about the integer square root function that Dr. Crenshaw published this weekend. Really impressed that it can be implemented succintly in Verilog as your code shows.

[ElEctric_EyE]

MichaelM wrote:

EEyE:

That's cool. I was reading and thinking about the integer square root function that Dr. Crenshaw published this weekend. Really impressed that it can be implemented succintly in Verilog as your code shows.

I hope to use it to control amplitude of the sin or cos table Xilinx has generated from LUTs and I've stored in BRAM.... Not sure yet. But I am borrowing the following .GIF from the sine wiki so the general idea can be seen, although I currently modify the phase for ellipse. A strictly horizontal or vertical ellipse requires amplitude modulation for the sin or cos.

[ElEctric_EyE]

I still haven't made any headway into generating a sine wave with a controllable amplitude from a LUT, so I decided to tackle another portion of the project which needs attention, i.e. how to resolve video ram contention with cpu/hardware accelerator access. Also, on this projects' software side of things, I am attempting to port Bruce's take on Bob Bishop's Mandelbrot program to 65Org16. I will start a new thread on that very soon...

Now as far as the video databus contention issue, I am not pleased when the cpu only has access during non active Hsync or non active Vsync. It's very slow and old hat, quite honestly.
Even though I'm not as good as some EE's out there that have experience with scanline buffers, FIFOs and dual port FPGA blockRAMs, I think I have an idea after doing more than a little bit of research and some testing.

The GS8320Z18 SyncRAM I'm using has a 4.0ns delay time in flow through mode which I currently use with a 150MHz pixel clock. So it is within the 250MHz limit in flow though mode. In pipelined mode the delay is spec'd at 2.5ns which would allow the device to run at 400MHz. The device is not hardwired selected for this mode, however I am currently running it in flow through mode at 300MHz. And it does seem to be working for some states of my hardware graphics accelerator, which is exciting to see since the goal would be to have interleaved access to the SyncRAM and it is possible to do back to back read and/or writes with this ZBT RAM. A post by a member named PK gave me more confidence this idea would work on this thread in other electronics forum. Lots of good links to follow from this guy!

So now I have to re-test 6 of the 8 functions this hardware generator has that plots directly to the syncRAM and try to fix up timing issues:
Line Generator............: not 100%
Rectangular fill............: not 100%
Pixel Plot...................: appears working
Read Pixel Color...........: appears working
Copy & Paste Rectangle: not working
Character Plot.............: not 100%

The project has 4 clocks generated utilizing the Spartan 6 PLL_ADV from an onboard 100MHz can oscillator:
1) 75MHz 65Org16 cpu/hardware accelerator/FPGA blockRAMs used for coordinate storage
2) 300MHz 2MBx16 Synchronous RAM (videoram)
3) 150MHz pixel clock to videoDAC and H/V Sync generator
4) 150MHz main signal to next Parallel Video Board (unused presently)

EDIT:
I realize I may be violating some timing spec's of the syncRAM since it's running 50MHz above spec, so all this is just a test at 1920x1080 with the realization this may not work at this speed.
Finally added in bypass cap's for the syncRAM with no change in behavior. (I suspect the properties of the board material are providing good enough capacitance for power supply filtering)
Tried changing syncRAM pin drive strength and slew rate from the FPGA with no better performance.

[ElEctric_EyE]

This past experiment was a failure as expected...
I have reached the limitations, in my current knowledge of Verilog, with this board design. I'm still working on the Mandelbrot 65Org16 software on spare time which will take advantage of the current capabilities I've developed.

I'm currently looking at a new design where I will recycling the 4 SYncRAMs I'm currently using across 4 PVB's and put them onto 2 newly designed boards @1920x1080 @150MHz. Each board will have 2 x 2Mx18 4ns GSI SyncRAMs and a larger Spartan 6 IC's in a BGA style package. The FPGA will have 2 separate address and data buses for them. The old backplane will still be able to support up to 6 of these new PVB's because the I/O connector will be the same 96-pin 3-row style. The idea behind the 2 SyncRAM's is to make 2 possibilities an easy reality: 1) is for sprites. 2) is for buffer for the videoDAC.

I'm looking at a new PCB maunfacturer called eurocircuits. They've been mentioned here before in other threads. They're very reasonably priced and are capable of up to 16 layers. Also a great many board construction variables can be set by the maker, and finally checked by their ordering system. They favor the Eagle so I'll have to learn this program as I was used to EPCB.

What really turned me on to them is that since I'll be mounting my own BGA packages now, they offer custom stencils not only for the BGA packages but for the entire board for ~$50 for a 4"x3" board. One can order a stencil for the top and/or the bottom!

[GARTHWILSON]

ElEctric_EyE wrote:

They favor the Eagle so I'll have to learn this program as I was used to EPCB.

Almost all board houses accept industry-standard gerber files; so any CAD that produces them will work. That would be any except the ones done by board houses that want to lock you into their service alone. If you supply gerber files (gerber files for the copper layers, excellon for drill files), the board house won't care what CAD you used, and they don't have to have even heard of it.

[ElEctric_EyE]

Garth, check out their PCB guidelines on page 3 of their PCB Design PDF link.

I was about to look into OrCad until I did more research into this PCB manufacturing house.

[GARTHWILSON]

The top of page 3 is approximately what I would expect. Note that there are some board houses that won't accept gerber RS-274D anymore, only 274X; but I learned to convert it quickly by hand in the text editor a few years ago, and it made it a lot easier to get what I wanted. When we only had 274D, doing complex layers (e.g., a ground plane with splits and other cuts in it) took detailed instructions in the readme file to tell them how to put various files together to get what you want-- and then they would still mess up sometimes and use the wrong aperture file or something. 274X is so much nicer and eliminates a lot of chances for human error on their part.

We used OrCAD at the place I worked from 1985-'92, and it had more bugs than an ant hill. It was terrible! Hopefully they've improved since then; but I see they still haven't learned to draw the schematic symbol for a resistor correctly after all these years, so maybe I shouldn't expect much.

Note that they say if it's directly CAD files, it can only be from Eagle; but if your files are gerber & excellon as mentioned at the top of the page, they can be from any CAD.

[ElEctric_EyE]

Garth, 2 questions quickly come to mind regarding this new path I'm taking on this 6-layer board using the Cadsoft Eagle program for PCB design. I'm researching them, but you may already know:
1) Can one use the Eagle program for PCB development without the schematic entry first? My initial experience seems like it wants a schematic first.
2) Since I aim for 6 layers and 2 power planes, should I have slower signals on the top and bottom planes? Then the Power and Ground planes, then finally sandwiched in the middle planes are the higher speed signals which would provide a natural bypass capacitance using FR4 material?

[GARTHWILSON]

EE, I'm not familiar with Eagle, but I'm sure you don't need to use the schematic capture if you don't want to. My CAD (Easy PC Pro) has schematic capture but I don't like the way any of the schematic captures I've tried work, so I still do my schematics by hand; but in the CAD, I can still pick the devices out of the libraries I built and put the parts on a board and connect them up the way I want, without using the ratsnesting and such. Even though the CAD came with libraries with lots of parts pre-made, I've made my own for denser layouts. But regardless, there will always be various parts that won't come in any CAD's pre-made libraries, so you need to be able to make those parts. Recent examples for me have included a chip antenna, feedthru capacitor, balun, various inductors, and chip crystals.

Normally the order of the layers in a 6-layer board is approximately what you mention:

• signal 1 generally going at 0°
• power (bypassed to ground at all Vcc pins and anywhere a trace goes through a via from layer 1 to layer 6 if the signals are fast enough to matter since otherwise the return current cannot keep following the signal trace)
• signal 2 generally going 45°
• signal layer 3 generally going 135°
• ground
• and signal layer 4 generally going 90°.

Which signal layers are 0, 45, 90, and 135° is not really important except that it's good to have the 2nd and 3rd signal layers at 90° to each other to reduce inter-layer coupling, since they're next to each other. Having signal layers generally going the directions shown above will greatly reduce the number of vias you need.

Trying to keep the fastest signals on the inside may make it harder to route a dense board, and is not necessary anyway because if the power and ground layers are used correctly, you already have transmission lines that will keep from radiating or picking up noise even if they're on the outside. Just route a signal on the layer that goes the general direction that signal needs to go. When I worked in VHF/UHF power transistor applications engineering in the mid-1980's and was often building up power amplifiers for engineering testing of our parts under the conditions requested by the clients, we always used microstrip or coplanar waveguide, not stripline. In microstrip, the transmission line is always on the top, yet there's no radiation from it. In stripline, the signal lines are sandwiched between ground layers (and one could be a power plane bypassed to ground at places that are appropriate for the signal line).

[ElEctric_EyE]

Great information as always Garth. Many thanks for trying to help me!

I spent about 2 weeks trying to rise above the learning curve for Cadsoft Eagle and realized it was going to take up too much of the little time I have left to spare for this new chapter of the PVB project.

I've decided to stick with the PCB routing software I already know from Express PCB as I can route 90% of all the available signals off of the FPGA necessary with about 30 extra I/O pins left over using their ProtoPro service which limits the total board design to 21 square inches. 4 boards for $205US. The new boards will be 6"x3.5" vs the these current 3.8"x2.5" PVBs. The new boards will now have some new features, the most important being 2x 2Mx18 SyncRAMs for sprite generation or page flipping.

The following code and associated videos will show the speed of the 75 MHz 65Org16 working in unison with 8 functions of my 75MHz hardware graphics accelerator written in Verilog. The SyncRAM and Arlet's Hsync/VSync/pixel clock generator are running @150MHz. Thanks to Arlet way back, I have hsync and vsync signals from the SYNC generator going to the 65Org16 cpu through an expanded 16-bit status flag register, and also branch testing opcodes for branch on clear or set for the added status flag bits. I have yet to post this version of 65Org16 on Github.

The hardware accelerator uses the RDY flag in order to halt the cpu so the accelerator can do it's work on the videoRAM. However, there is still video interference when the software chooses not to plot without regarding the state of hvsync or vsync and so video interference results.

Since I am going to have to cannibalize the memory IC's from 4 out of the 5 boards, I thought that the best way to close this thread would be to display some basic functions, especially with different delay variables.

Each post will include the 65Org16 coding and the video. This is more for me than anyone else and I thank 6502.org for tolerating my indulgence, but hopefully some may find it interesting and inspirational since I started designing the first version of these boards back in Aug 6, 2012. I did it in order to learn the discipline of the Verilog language. I'm a slow learner, but hardware design goes in tandem with HDL, so the design must change for the little I now know.

Here are the 8 functions of the current 2D hardware accelerator:

Code: Select all

8/26/2014 smg    Programming the 16-bit 2D Graphics 'LineGen' Accelerator
                
FUNCTION...............VARIABLE = 'description'....................R(ead)/W(rite)
________________________________________________________________________________
   
Line Generator............color = pixel color..........................RW
                            lx0 = X position start.....................RW
                            ly0 = Y position start.....................RW
                            lx1 = X position end.......................RW
                            ly1 = Y position end(*)....................RW
                      
Rectangle Fill............color = pixel color..........................RW
                          fXlen = Horizontal length....................RW
                          fYlen = Vertical length......................RW
                            fXs = X position start.....................RW
                            fYs = Y position start(*)..................RW
                      
Pixel Plot................color = pixel color..........................RW
                             Xp = X position...........................RW
                             Yp = Y position(*)........................RW

Read Pixel Color.............Xr = X position...........................RW
                             Yr = Y position(*)........................RW
                      colordata = 5-6-5 16bit R-G-B color data.........R
                                              
Copy & Paste Rectangle....bXlen = Horizontal length....................RW
                          bYlen = Vertical length......................RW
                            bXc = X position start of copy.............RW
                            bYc = Y position start of copy.............RW
                            bXp = X position start of paste............RW
                            bYp = Y position start of paste(*).........RW
                        
Character Plot............color = pixel color..........................RW
                         bcolor = background pixel color...............RW
                            Att = ASCII character + font...............RW
                             (0000000axxxxxxxx. xxxxxxxx = 8bit ASCII) 
                             (a = 1 C64, a = 0 DOS. FONTS) 
                             cX = X position...........................RW
                             cY = Y position(*)........................RW
                        
SIN LUT Generator........Xblock = (0..15) BRAM LUTs for X coordinates..RW
                         Yblock = (0..15) BRAM LUTs for Y coordinates..RW
                        Xoffset = placement within 1920 Horizontal.....RW
                        Yoffset = placement within 1080 Vertical.......RW
                          xfreq = # of SIN repetition within LUT1......RW
                          yfreq = # of SIN repetition within LUT2......RW
                         Xphase = (0..1023) Phase input to LUT1........RW
                         Yphase = (0..1023) Phase input to LUT2(*).....RW
                        
Square Root Generator........Xs = 16bit number.........................RW
                             Ys = 16bit number(*)......................RW
                           Root = SQRT(Xs^2 + Ys^2)....................R
________________________________________________________________________________
NOTES:
(*) This Verilog Hardware Function is initiated after a write to these registers. 

Also, bear with me as it take longer to process video through youtube so the software will most likely be posted first.

[ElEctric_EyE]

This is software to init the 64Kx16 ZeroPage & 64Kx16 StackPage pointers within the 65Org16 (only 1Kx16 used for each utilizing FPGA BRAMs). This is because early on I allowed the cpu to directly access the videoRAM for things like plotting characters through software and screen clearing. That is before I understood how to make Verilog perform these functions in HDL... The videoRAM started at $00000000. This is no longer allowed, the accelerator now performs all the plotting, so this is legacy code (I just now realize ). Also this code initializes the H/V Sync Generator timings and then clears the videoRAM using the Rectangle Fill with color #%0000000000000000 (black):

Code: Select all

BEGIN:            LDA #$1000              
                  TAZP                    ;SET ZEROPAGE @$1000_0000
                  LDA #$1001
				      TASP                    ;SET STACKPAGE @$1001_0000                  
                                                                                                                   
                  LDA #1920               ;2430 (2200 ideal for 67.5kHz) total H cycles      @148.5MHz 16.363uS ->  61.111kHz          
                  STA hVIDEO                                                                        
                  LDA #205                 ;205                                                      
                  STA hFRONT                                                                        
                  LDA #50                  ;50                                                    
                  STA hSYNC                                                                        
                  LDA #255                 ;255                                                     
                  STA hBACK
                  
                  LDA #1080                ;1139 total     @148.5MHz ->  1139 x 16.363uS = 18.638mS = 53.654Hz
                  STA vVIDEO                                                                               
                  LDA #2                   ;2                                                              
                  STA vFRONT                                                                                
                  LDA #55                  ;55                                                            
                  STA vSYNC                                                                                 
                  LDA #2                   ;2                                                             
                  STA vBACK                                                                                                              
                  
rock              LDA #%0000000000000000                  ;16-bit 565 RGB               
                  STA color
                  
                  LDA #1919 
                  STA fXlen
                  LDA #1079
                  STA fYlen
                  LDA #0
                  STA fXs
                  STA fYs                  ;FILL!

[ElEctric_EyE]

Continuing, this final part of this software draws lines from the center of the screen to the outer edges with a significant delay. The variable 'color' is derived from a pseudo-random HDL number generator. It's not so random as every number is the same upon power-up but it does the job OK. Next board will have a true random number generator built in hardware.

Code: Select all

BEGIN:            LDA #$1000              
                  TAZP                    ;SET ZEROPAGE @$1000_0000
                  LDA #$1001
				      TASP                    ;SET STACKPAGE @$1001_0000                  
                                                                                                                                       
                  LDA #1920               ;2430 (2200 ideal for 67.5kHz) total H cycles      @148.5MHz 16.363uS ->  61.111kHz          
                  STA hVIDEO                                                                        
                  LDA #205                 ;205                                                      
                  STA hFRONT                                                                        
                  LDA #50                  ;50                                                    
                  STA hSYNC                                                                        
                  LDA #255                 ;255                                                     
                  STA hBACK
                  
                  LDA #1080                ;1139 total     @148.5MHz ->  1139 x 16.363uS = 18.638mS = 53.654Hz
                  STA vVIDEO                                                                               
                  LDA #2                   ;2                                                              
                  STA vFRONT                                                                                
                  LDA #55                  ;55                                                            
                  STA vSYNC                                                                                 
                  LDA #2                   ;2                                                             
                  STA vBACK                                                                                                              
                  
rock              LDA #%0000000000000000                  ;16-bit 565 RGB               
                  STA color
                  
                  LDA #1919 
                  STA fXlen
                  LDA #1079
                  STA fYlen
                  LDA #0
                  STA fXs
                  STA fYs                  ;FILL!

                  LDA rng
                  STA color
                  
                  LDX #0
                  LDA #1920/2
                  STA lx0
                  LDA #1080/2
                  STA ly0
liner1            TXA
                  STA lx1
                  JSR DELAY2
                  LDA #0
                  STA ly1
                  INX
                  CPX #1919
                  BNE liner1
                  
                  LDX #0
                  LDA #1920/2
                  STA lx0
                  LDA #1080/2
                  STA ly0
liner2            TXA
                  STA lx1
                  JSR DELAY2
                  LDA #1080
                  STA ly1
                  INX
                  CPX #1919
                  BNE liner2
                  
                  LDY #0
                  LDA #1920/2
                  STA lx0
                  LDA #1080/2
                  STA ly0
                  LDA #0
                  STA lx1
liner3            TYA
                  STA ly1
                  JSR DELAY2
                  INY
                  CPY #1080
                  BNE liner3
                  
                  LDY #0
                  LDA #1920/2
                  STA lx0
                  LDA #1080/2
                  STA ly0
                  LDA #1919
                  STA lx1
liner4            TYA
                  STA ly1
                  JSR DELAY2
                  INY
                  CPY #1080
                  BNE liner4

                  JMP rock

Here is the DELAY2 routine:

Code: Select all

DELAY2            PHY
                  PHX
                  LDY #$0FFF
GHJ               DEY
                  BNE GHJ
                  PLX
                  PLY
                  RTS

Quite a few cycles! I'll ry to make the video tonight.

[ElEctric_EyE]

This video is with the delay.
This video is without delay.

[BigEd]

Thanks EEye for taking us on this journey with you! It's easy enough for two years to pass without any visible progress, so congratulations for showing the determination to bring this design to a demo-worthy state and posting about your problems and progress. I'm sure you feel satisfied that you've learnt a huge amount by battling through this far: we don't often cast ourselves back to the beginning of a project and consider what used to be mysterious to us, or what we felt to be beyond our understanding.

Cheers
Ed