Optimization: \= on <symbol> is slower than it should be

[waywardmonkeys]

\= on <symbol> does a lot of work!

define sealed inline method \=
    (symbol-1 :: <symbol>, symbol-2 :: <symbol>) => (well? :: <boolean>)
  as(<string>, symbol-1) = as(<string>, symbol-2)
end method \=;

We could store a hash or something like Gwydion did, but at the least, it should check \== first, I think if it isn't going to compare hashes or something.

Right now, this expands into code that even includes some type checks:

  T29 = SLOT_VALUE_INITD(T7, 0);
  T30_0 = T29;
  T31_0 = T30_0;
  {
    MV_CHECK_TYPE_PROLOGUE(T30_0);
    primitive_type_check(T31_0, &KLbyte_stringGVKd);
    MV_CHECK_TYPE_EPILOGUE();
  }
  T32 = T31_0;
  T33 = SLOT_VALUE_INITD(T6, 0);
  T34_0 = T33;
  T35_0 = T34_0;
  {
    MV_CHECK_TYPE_PROLOGUE(T34_0);
    primitive_type_check(T35_0, &KLbyte_stringGVKd);
    MV_CHECK_TYPE_EPILOGUE();
  }
  T36 = T35_0;
  T37_0 = KEVKdMM27I(T32, T36);
  T38 = T37_0;

The type checks are an interesting anomaly. I wasn't expecting that. (I haven't dug enough to see what's up.)

[waywardmonkeys]

Just did an experimental bootstrap with this change:

-define sealed inline method \=
+define sealed method slow-=
     (symbol-1 :: <symbol>, symbol-2 :: <symbol>) => (well? :: <boolean>)
   as(<string>, symbol-1) = as(<string>, symbol-2)
+end method slow-=;
+
+define sealed inline method \=
+    (symbol-1 :: <symbol>, symbol-2 :: <symbol>) => (well? :: <boolean>)
+  (symbol-1 == symbol-2) | slow-=(symbol-1, symbol-2)
 end method \=;

And these are the locations in a build of dylan-compiler that call \= rather than \==. I know there are more when y ou build dylan-environment:

    // .../sources/dfmc/c-back-end/c-emit-object.dylan:240
    T6_0 = Kslow_EVKiMM0I(T3, T4);
--
    // .../sources/dfmc/definitions/top-level-convert.dylan:41
    T13_0 = Kslow_EVKiMM0I(T11, IKJboot_dylan_definitions_);
--
    // .../sources/environment/dfmc/database/library-objects.dylan:62
    T30_0 = Kslow_EVKiMM0I(T7, T6);
--
    // .../sources/lib/dood/table-proxy.dylan:406
    T4_0 = Kslow_EVKiMM0I(x_, y_);
--
    // .../sources/dylan/symbol.dylan:32
    T3_0 = Kslow_EVKiMM0I(symbol_1_, symbol_2_);
--
    // .../sources/environment/protocols/applications.dylan:166
    T4_0 = Kslow_EVKiMM0I(T2, IKJclosed_);
--
    // .../sources/environment/protocols/applications.dylan:161
    T4_0 = Kslow_EVKiMM0I(T2, IKJstopped_);
--
    // .../sources/environment/protocols/applications.dylan:156
    T4_0 = Kslow_EVKiMM0I(T2, IKJrunning_);
--
    // .../sources/environment/reports/library-report-rst.dylan:211
    T11_0 = Kslow_EVKiMM0I(kindF6, &KJinput_);
--
    // .../sources/lib/jam/jam-reader.dylan:255
    T16_0 = Kslow_EVKiMM0I(T14, IKJpunctuation_);
--
    // .../sources/system/file-system/unix-file-system.dylan:208
    T24_0 = Kslow_EVKiMM0I(if_existsF5, &KJsignal_);
--
    // .../sources/system/file-system/unix-file-system.dylan:181
    T18_0 = Kslow_EVKiMM0I(if_existsF5, &KJsignal_);
--
    // .../sources/lib/walker/walker.dylan:66
    T5_0 = Kslow_EVKiMM0I(T1, &KJvirtual_);

[waywardmonkeys]

Just did an experimental bootstrap with this change:

-define sealed inline method \=
+define sealed method slow-=
     (symbol-1 :: <symbol>, symbol-2 :: <symbol>) => (well? :: <boolean>)
   as(<string>, symbol-1) = as(<string>, symbol-2)
+end method slow-=;
+
+define sealed inline method \=
+    (symbol-1 :: <symbol>, symbol-2 :: <symbol>) => (well? :: <boolean>)
+  (symbol-1 == symbol-2) | slow-=(symbol-1, symbol-2)
 end method \=;

And these are the locations in a build of dylan-compiler that call \= rather than \==. I know there are more when y ou build dylan-environment:

    // .../sources/dfmc/c-back-end/c-emit-object.dylan:240
    T6_0 = Kslow_EVKiMM0I(T3, T4);
--
    // .../sources/dfmc/definitions/top-level-convert.dylan:41
    T13_0 = Kslow_EVKiMM0I(T11, IKJboot_dylan_definitions_);
--
    // .../sources/environment/dfmc/database/library-objects.dylan:62
    T30_0 = Kslow_EVKiMM0I(T7, T6);
--
    // .../sources/lib/dood/table-proxy.dylan:406
    T4_0 = Kslow_EVKiMM0I(x_, y_);
--
    // .../sources/dylan/symbol.dylan:32
    T3_0 = Kslow_EVKiMM0I(symbol_1_, symbol_2_);
--
    // .../sources/environment/protocols/applications.dylan:166
    T4_0 = Kslow_EVKiMM0I(T2, IKJclosed_);
--
    // .../sources/environment/protocols/applications.dylan:161
    T4_0 = Kslow_EVKiMM0I(T2, IKJstopped_);
--
    // .../sources/environment/protocols/applications.dylan:156
    T4_0 = Kslow_EVKiMM0I(T2, IKJrunning_);
--
    // .../sources/environment/reports/library-report-rst.dylan:211
    T11_0 = Kslow_EVKiMM0I(kindF6, &KJinput_);
--
    // .../sources/lib/jam/jam-reader.dylan:255
    T16_0 = Kslow_EVKiMM0I(T14, IKJpunctuation_);
--
    // .../sources/system/file-system/unix-file-system.dylan:208
    T24_0 = Kslow_EVKiMM0I(if_existsF5, &KJsignal_);
--
    // .../sources/system/file-system/unix-file-system.dylan:181
    T18_0 = Kslow_EVKiMM0I(if_existsF5, &KJsignal_);
--
    // .../sources/lib/walker/walker.dylan:66
    T5_0 = Kslow_EVKiMM0I(T1, &KJvirtual_);

[housel]

Redefining the method as:

define sealed inline method \=
    (symbol-1 :: <symbol>, symbol-2 :: <symbol>) => (well? :: <boolean>)
  let a = as(<string>, symbol-1) = as(<string>, symbol-2);
  let b = symbol-1 == symbol-2;
  assert(a == b);
  a
end method \=;

and doing a three-stage bootstrap succeeds, indicating that the comparing symbol-name method and the pointer equality check method are equivalent, as indeed they should be.

[housel]

Redefining the method as:

define sealed inline method \=
    (symbol-1 :: <symbol>, symbol-2 :: <symbol>) => (well? :: <boolean>)
  let a = as(<string>, symbol-1) = as(<string>, symbol-2);
  let b = symbol-1 == symbol-2;
  assert(a == b);
  a
end method \=;

and doing a three-stage bootstrap succeeds, indicating that the comparing symbol-name method and the pointer equality check method are equivalent, as indeed they should be.