llvm / llvm/circt

[FIRRTL] Memory blocks DCE

Open
#4,193 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
2.2k
Forks
524
Avg merge
3d 2h
Merged PRs (30d)
46

Description

Currently we generally treat memories as black-box, hence we cannot optimize at all. For example, in the following example `cmem rf` is a memory whose fields are `a` and `b`. But there is no user of output `b`.
```scala
circuit DUTModule:
module DUTModule :
input clock : Clock
input reset : UInt<1>
output io : { flip addr : UInt<3>, flip dataIn : {a:UInt<2>, b:UInt<3>}, flip wen : UInt<1>, dataOut : {a:UInt<2>}}

cmem rf : {a:UInt<2>, b:UInt<3>} [8]
infer mport read = rf[io.addr], clock
io.dataOut.a <= read.a
when io.wen :
infer mport write = rf[io.addr], clock
write <= io.dataIn
```

SFC can optimize away field b because SFC lowers a memory into registers.
```verilog
reg [1:0] rf_a [0:7];
wire rf_a_read_en;
wire [2:0] rf_a_read_addr;
wire [1:0] rf_a_read_data;
wire [1:0] rf_a_write_data;
wire [2:0] rf_a_write_addr;
wire rf_a_write_mask;
wire rf_a_write_en;
assign rf_a_read_en = 1'h1;
assign rf_a_read_addr = io_addr;
assign rf_a_read_data = rf_a[rf_a_read_addr];
assign rf_a_write_data = io_dataIn_a;
assign rf_a_write_addr = io_addr;
assign rf_a_write_mask = 1'h1;
assign rf_a_write_en = io_wen;
assign io_dataOut_a = rf_a_read_data;
always @(posedge clock) begin
if (rf_a_write_en & rf_a_write_mask) begin
rf_a[rf_a_write_addr] <= rf_a_write_data;
end
end
// .. there is no register regarding `b`.
```

On the other hand, MFC:
```verilog
wire [2:0] _rf_b_ext_R0_data;
rf_a_combMem rf_a_ext (
.R0_addr (io_addr),
.R0_en (1'h1),
.R0_clk (clock),
.W0_addr (io_addr),
.W0_en (io_wen),
.W0_clk (clock),
.W0_data (io_dataIn_a),
.R0_data (io_dataOut_a)
);
rf_b_combMem rf_b_ext (
.R0_addr (io_addr),
.R0_en (1'h1),
.R0_clk (clock),
.W0_addr (io_addr),
.W0_en (io_wen),
.W0_clk (clock),
.W0_data (io_dataIn_b),
.R0_data (_rf_b_ext_R0_data)
);
```

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by examining how FIRRTL memories are represented and lowered by SFC and MFC, using the `cmem rf` example as the minimal case. Compare the generated outputs and trace where unused memory fields remain live. Done means an unused field such as `b` is eliminated without affecting the used field `a` or its read and write behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
compilers
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.