Various bugs found while running OSACA on Cbrt

osaca/data/isa/x86.yml

@@ -2817,6 +2817,21 @@ instruction_forms:
           name: "xmm"
           source: true
           destination: true
+    - name: mul
+      operands:
+        - class: "register"
+          name: "gpr"
+          source: true
+          destination: false
+      hidden_operands:
+        - class: "register"
+          name: "rdx"
+          source: false
+          destination: true
+        - class: "register"
+          name: "rax"
+          source: true
+          destination: true
     - name: mulsd
       operands:
         - class: "register"

osaca/osaca.py

@@ -357,6 +357,7 @@ def inspect(args, output_file=sys.stdout):
 
     # Parse file.
     while True:
+        arch, syntax = combinations_to_try.pop()
         parser = get_asm_parser(arch, syntax)
         try:
             parsed_code = parser.parse_file(code)
@@ -366,7 +367,6 @@ def inspect(args, output_file=sys.stdout):
             if not combinations_to_try:
                 raise e
         combinations_to_try -= {(arch, syntax)}
-        arch, syntax = combinations_to_try.pop();
 
     # Reduce to marked kernel or chosen section and add semantics
     if args.lines:

osaca/parser/parser_x86intel.py

@@ -15,20 +15,15 @@
 from osaca.parser.register import RegisterOperand
 from osaca.semantics.hw_model import MachineModel
 
-# Unicode 3.0-style definition because we do not have the UCD in the Python standard library, see
-# the derivation in Table 2 of UAX #31, https://www.unicode.org/reports/tr31/#Table_Lexical_Classes_for_Identifiers.
-# See also Table 5-7 in https://www.unicode.org/versions/Unicode3.0.0/ch05.pdf#page=31, and
-# https://www.unicode.org/reports/tr55/#Evolution-Unicode-3.
-IDENTIFIER_START_CHARACTERS = "".join(
-    chr(cp)
-    for cp in range(0x10FFFF)
-    if unicodedata.category(chr(cp)).startswith("L") or unicodedata.category(chr(cp)) == "Nl"
-)
-
-IDENTIFIER_CONTINUE_CHARACTERS = IDENTIFIER_START_CHARACTERS + "".join(
-    chr(cp)
-    for cp in range(0x10FFFF)
-    if unicodedata.category(chr(cp)) in ("Mn", "Mc", "Nd", "Pc")
+# We assume any non-ASCII characters except control characters and line terminators can be part of
+# identifiers; this is based on the assumption that no assembler uses non-ASCII white space and
+# syntax characters.
+# This approach is described at the end of https://www.unicode.org/reports/tr55/#Whitespace-Syntax.
+# It is appropriate for tools, such as this one, which process source code but do not fully validate
+# it (in this case, that’s the job of the assembler).
+NON_ASCII_PRINTABLE_CHARACTERS = "".join(
+    chr(cp) for cp in range(0x80, 0x10FFFF + 1)
+    if unicodedata.category(chr(cp)) not in ("Cc", "Zl", "Zp", "Cs", "Cn")
 )
 
 # References:
@@ -206,7 +201,7 @@ def construct_parser(self):
 
         # Comment.
         self.comment = pp.Literal(";") + pp.Group(
-            pp.ZeroOrMore(pp.Word(pp.printables))
+            pp.ZeroOrMore(pp.Word(pp.printables + NON_ASCII_PRINTABLE_CHARACTERS))
         ).setResultsName(self.comment_id)
 
         # Types.
@@ -232,8 +227,8 @@ def construct_parser(self):
 
         # Identifier.  Note that $ is not mentioned in the ASM386 Assembly Language Reference,
         # but it is mentioned in the MASM syntax.  < and > apparently show up in C++ mangled names.
-        first = pp.Word(IDENTIFIER_START_CHARACTERS + "$?@_<>", exact=1)
-        rest = pp.Word(IDENTIFIER_CONTINUE_CHARACTERS + "$?@_<>")
+        first = pp.Word(pp.alphas + NON_ASCII_PRINTABLE_CHARACTERS + "$?@_<>", exact=1)
+        rest = pp.Word(pp.alphanums + NON_ASCII_PRINTABLE_CHARACTERS + "$?@_<>")
         identifier = pp.Group(
             pp.Combine(first + pp.Optional(rest)).setResultsName("name")
         ).setResultsName("identifier")
@@ -451,7 +446,7 @@ def construct_parser(self):
         directive_parameter = (
             pp.quotedString
             ^ (
-                pp.Word(pp.printables, excludeChars=",;")
+                pp.Word(pp.printables + NON_ASCII_PRINTABLE_CHARACTERS, excludeChars=",;")
                 + pp.Optional(pp.Suppress(pp.Literal(",")))
             )
             ^ pp.Suppress(pp.Literal(","))

osaca/semantics/kernel_dg.py

@@ -596,6 +596,7 @@ def export_graph(self, filepath=None):
         graph.graph["node"] = {"colorscheme" : colorscheme}
         graph.graph["edge"] = {"colorscheme" : colorscheme}
         for n, color in node_colors.items():
+            color = min(color, max_color)
             if "style" not in graph.nodes[n]:
                 graph.nodes[n]["style"] = "filled"
             else:
@@ -607,6 +608,7 @@ def export_graph(self, filepath=None):
             ):
                 graph.nodes[n]["fontcolor"] = "white"
         for (u, v), color in edge_colors.items():
+            color = min(color, max_color)
             # The backward edge of the cycle is represented as the corresponding forward
             # edge with the attribute dir=back.
             edge = graph.edges[u, v] if (u, v) in graph.edges else graph.edges[v, u]