[Profiling] Small adjustments and improved timeline view (#2388)

This PR contains a series of minor changes I made over the past couple
of weeks.

I made some small changes to non-profiler parts of the repo:
- Adding `*.svg` and `*.folded` to the gitignore
- Adding a profiler-specific pass alias (`-p profiler`)
- Using `-p profiler` in fud2 so there's no inconsistency in what passes
we use.

In profiler-specific code, I fixed a bug where the flame graph recorded
one extra cycle for `main`, added colors to categorize nodes in the
aggregate tree picture (between cells, groups, and primitives), and have
the profiler emit a timeline JSON involving all cells by default.

.gitignore

@@ -61,8 +61,8 @@ tools/btor2/btor2i/build/
 tools/profiler/data
 tools/profiler/meta-logs
 tools/profiler/fg-tmp
-tools/profiler/handmade-flame-graphs/*/*.svg
-tools/profiler/inst-check-tmp
+*.svg
+*.folded
 
 temp/
 

calyx-opt/src/default_passes.rs

@@ -79,6 +79,7 @@ impl PassManager {
         pm.register_pass::<RemoveIds>()?;
         pm.register_pass::<ExternalToRef>()?;
 
+        // instrumentation pass to collect profiling information
         pm.register_pass::<ProfilerInstrumentation>()?;
 
         //add metadata
@@ -155,6 +156,20 @@ impl PassManager {
             ["validate", "pre-opt", "compile", "post-opt", "lower",]
         );
 
+        // profiler flow for pass explorer access
+        register_alias!(
+            pm,
+            "profiler",
+            [
+                StaticInliner,
+                CompileStatic,
+                CompileRepeat,
+                CompileInvoke,
+                ProfilerInstrumentation,
+                "all"
+            ]
+        );
+
         // Compilation flow with no optimizations enables
         register_alias!(
             pm,

fud2/scripts/profiler.rhai

@@ -9,10 +9,9 @@ export const flamegraph = state("flamegraph", ["svg"]);
 
 fn profiling_setup(e) {
     e.var_("cells", "cells.json");
-    e.var_("tdcc-json", "fsm.json"); // might not be necessary if we get rid of fsms?
 
     // series of passes after instrumentation?
-    e.config_var_or("passes", "profiler.compilation-passes", "all"); // set passes="no-opt" to run without optimizations
+    e.config_var_or("passes", "profiler.compilation-passes", "-p all"); // set passes="-p no-opt" to run without optimizations
 
     // rules for preprocessing
 
@@ -40,7 +39,7 @@ fn calyx_to_flamegraph(e, input, output) {
     e.build_cmd(["$cells"], "component-cells", [input], []);
     e.build_cmd([instrumented_verilog], "calyx", [input], []);
     e.arg("backend", "verilog");
-    e.arg("args", " -p static-inline -p compile-static -p compile-repeat -p compile-invoke -p profiler-instrumentation -p $passes -x tdcc:dump-fsm-json=fsm.json");
+    e.arg("args", "-p static-inline -p compile-static -p compile-repeat -p compile-invoke -p profiler-instrumentation $passes");
 
     let instrumented_sim = "instrumented.exe";
     // verilog --> sim; adapted from verilator::verilator_build()

fud2/tests/snapshots/tests__test@plan_profiler.snap

@@ -20,8 +20,7 @@ rule calyx-cider
   command = $calyx-exe -l $calyx-lib-path $cider-calyx-passes $args $in > $out
 
 cells = cells.json
-tdcc-json = fsm.json
-passes = all
+passes = -p all
 component_cells = $calyx-base/target/debug/component_cells
 rule component-cells
   command = $component_cells -l $calyx-base $in > $out
@@ -54,7 +53,7 @@ cycle-limit = 500000000
 build $cells: component-cells /input.ext
 build instrumented.sv: calyx /input.ext
   backend = verilog
-  args =  -p static-inline -p compile-static -p compile-repeat -p compile-invoke -p profiler-instrumentation -p $passes -x tdcc:dump-fsm-json=fsm.json
+  args = -p static-inline -p compile-static -p compile-repeat -p compile-invoke -p profiler-instrumentation $passes
 build verilator-out/Vtoplevel: verilator-compile-standalone-tb instrumented.sv | tb.sv
   out-dir = verilator-out
 build instrumented.exe: cp verilator-out/Vtoplevel