Showing results for 
Search instead for 
Did you mean: 

Infer dual port RAM without read during write bypass logic

I am attempting to infer a single clock, dual port RAM without any bypass logic. I am having problems with synthesis, it seems to add a lot of bypass logic that wasn't there when I used the megawizard.

// // Dual-Port RAM with different read/write addresses and single read/write clock // and with a control for writing single bytes into the memory word; byte enable // Note - bytes are VERY often not even close to 8 bits. 16 for colors, 18 for floats, etc. module byte_enabled_dual_port_ram #(parameter int RADDR_WIDTH = 8, WADDR_WIDTH = 6, BYTE_WIDTH = 8, BYTES = 4, WIDTH = BYTES * BYTE_WIDTH ) ( input [WADDR_WIDTH-1:0] waddr, input [RADDR_WIDTH-1:0] raddr, input [BYTES-1:0] be, input [WIDTH-1:0] wdata, input we, clk, output reg [WIDTH/(1<<(RADDR_WIDTH - WADDR_WIDTH))-1:0] q ); localparam int WORDS = 1 << WADDR_WIDTH; localparam int RATIO = 1 << (RADDR_WIDTH - WADDR_WIDTH); // use a multi-dimensional packed array to model individual bytes within the word logic [BYTES-1:0][BYTE_WIDTH-1:0] ram[0:WORDS-1];   logic [WADDR_WIDTH-1:0] waddr_reg; logic [RADDR_WIDTH-1:0] raddr_reg; logic [BYTES-1:0] be_reg; logic [WIDTH-1:0] wdata_reg; logic we_reg; wire [WIDTH-1:0]word = ram[raddr_reg/RATIO]; always_ff@(posedge clk) begin waddr_reg <= waddr; raddr_reg <= raddr; be_reg <= be; wdata_reg <= wdata; we_reg <= we; for(int i = 0; i < BYTES; i++) begin if(be_reg[i] && we_reg) begin ram[waddr_reg][i] <= wdata_reg[i*BYTE_WIDTH +: BYTE_WIDTH]; end end //q <= ram[raddr / RATIO][(raddr % RATIO) * (BYTES/RATIO) +: (BYTES/RATIO)]; q <= word[(raddr_reg % RATIO) * (WIDTH/RATIO) +: (WIDTH/RATIO)]; end endmodule : byte_enabled_dual_port_ram  

I'm not sure what's going on exactly - when I look at the RTL viewer, I can see that raddr is passed directly to the altsyncram module, instead of going through the raddr_reg: rtl_dual_port.png

Any ideas what's gone wrong? How to fix?

0 Kudos
2 Replies

Hi Evan,

Apologize for idle for awhile. Do you able to solve the issue? Do you need further help?


I ended up just using an altsyncram definition. I dont think quartus is able to infer such a complicated ram.