Merged common encode tag paths.

benchmark.py

@@ -7,7 +7,7 @@ import re
 def Run(cmd):
   subprocess.check_call(cmd, shell=True)
 
-def RunAgainstBranch(branch, outfile, runs=12):
+def RunAgainstBranch(branch, outbase, runs=12):
   tmpfile = "/tmp/bench-output.json"
   Run("rm -rf {}".format(tmpfile))
   Run("git checkout {}".format(branch))
@@ -15,10 +15,13 @@ def RunAgainstBranch(branch, outfile, runs=12):
 
   Run("./bazel-bin/benchmark --benchmark_out_format=json --benchmark_out={} --benchmark_repetitions={}".format(tmpfile, runs))
 
+  Run("bazel build -c opt --copt=-g :conformance_upb")
+  Run("cp -f bazel-bin/conformance_upb {}.bin".format(outbase))
+
   with open(tmpfile) as f:
     bench_json = json.load(f)
 
-  with open(outfile, "w") as f:
+  with open(outbase + ".txt", "w") as f:
     for run in bench_json["benchmarks"]:
       name = re.sub(r'^BM_', 'Benchmark', run["name"])
       if name.endswith("_mean") or name.endswith("_median") or name.endswith("_stddev"):
@@ -26,7 +29,17 @@ def RunAgainstBranch(branch, outfile, runs=12):
       values = (name, run["iterations"], run["cpu_time"])
       print("{} {} {} ns/op".format(*values), file=f)
 
-RunAgainstBranch("master", "/tmp/old.txt")
-RunAgainstBranch("decoder", "/tmp/new.txt")
+RunAgainstBranch("6e140c267cc9bf6a4ef89d3f9a842209d7537599", "/tmp/old")
+RunAgainstBranch("encoder", "/tmp/new")
+
+print()
+print()
 
 Run("~/go/bin/benchstat /tmp/old.txt /tmp/new.txt")
+
+print()
+print()
+
+Run("objcopy --strip-debug /tmp/old.bin /tmp/old.bin.stripped")
+Run("objcopy --strip-debug /tmp/new.bin /tmp/new.bin.stripped")
+Run("~/code/bloaty/bloaty /tmp/new.bin.stripped -- /tmp/old.bin.stripped --debug-file=/tmp/old.bin --debug-file=/tmp/new.bin -d compileunits,symbols")

upb/encode.c

@@ -154,15 +154,18 @@ static void encode_scalar(upb_encstate *e, const void *_field_mem,
                           const upb_msglayout *m, const upb_msglayout_field *f,
                           bool skip_zero_value) {
   const char *field_mem = _field_mem;
-#define CASE(ctype, type, wire_type, encodeval) do { \
-  ctype val = *(ctype*)field_mem; \
-  if (skip_zero_value && val == 0) { \
-    return; \
-  } \
-  encode_ ## type(e, encodeval); \
-  encode_tag(e, f->number, wire_type); \
-  return; \
-} while(0)
+  int wire_type;
+
+#define CASE(ctype, type, wtype, encodeval) \
+  {                                         \
+    ctype val = *(ctype *)field_mem;        \
+    if (skip_zero_value && val == 0) {      \
+      return;                               \
+    }                                       \
+    encode_##type(e, encodeval);            \
+    wire_type = wtype;                      \
+    break;                                  \
+  }
 
   switch (f->descriptortype) {
     case UPB_DESCRIPTOR_TYPE_DOUBLE:
@@ -197,8 +200,8 @@ static void encode_scalar(upb_encstate *e, const void *_field_mem,
       }
       encode_bytes(e, view.data, view.size);
       encode_varint(e, view.size);
-      encode_tag(e, f->number, UPB_WIRE_TYPE_DELIMITED);
-      return;
+      wire_type = UPB_WIRE_TYPE_DELIMITED;
+      break;
     }
     case UPB_DESCRIPTOR_TYPE_GROUP: {
       size_t size;
@@ -209,8 +212,8 @@ static void encode_scalar(upb_encstate *e, const void *_field_mem,
       }
       encode_tag(e, f->number, UPB_WIRE_TYPE_END_GROUP);
       encode_message(e, submsg, subm, &size);
-      encode_tag(e, f->number, UPB_WIRE_TYPE_START_GROUP);
-      return;
+      wire_type = UPB_WIRE_TYPE_START_GROUP;
+      break;
     }
     case UPB_DESCRIPTOR_TYPE_MESSAGE: {
       size_t size;
@@ -221,12 +224,13 @@ static void encode_scalar(upb_encstate *e, const void *_field_mem,
       }
       encode_message(e, submsg, subm, &size);
       encode_varint(e, size);
-      encode_tag(e, f->number, UPB_WIRE_TYPE_DELIMITED);
-      return;
+      wire_type = UPB_WIRE_TYPE_DELIMITED;
+      break;
     }
   }
 #undef CASE
-  UPB_UNREACHABLE();
+
+  encode_tag(e, f->number, wire_type);
 }
 
 static void encode_array(upb_encstate *e, const char *field_mem,