[UpdateCCTestChecks] Implement --global-hex-value-regex

[jdenny]

Summary

For example, in OpenMP offload codegen tests, global variables like
.offload_maptypes* are much easier to read in hex.

[jdenny]

Fix the new test for the race discussed at D104714#2841465.

[jdoerfert]

I don't understand what we do before, and how this work, is [[# special in lit? Also, why i32/64 only, that seems arbitrary, no?

[jdenny]

In D104743#2889228, @jdoerfert wrote:

I don't understand what we do before, and how this work,

Maybe part of the confusion is that --global-hex-value-regex does not change how the value is expected to appear in LLVM IR: decimal is still expected. It only changes how the value is represented in the FileCheck directive: hex is the representation. The only point is to make the FileCheck directives more readable because, at least in my OpenMP use case, we're dealing with flags.

is [[# special in lit?

No, it starts a FileCheck numeric.

Also, why i32/64 only, that seems arbitrary, no?

Good point. I suppose it should be any integer. I'll work on that.

Thanks for the reviews.

[jdenny]

Applied @jdoerfert's suggestion: generalize to all integer values.

[jdoerfert]

LG, sure why not.

clang/test/utils/update_cc_test_checks/Inputs/global-hex-value-regex.c

+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -emit-llvm -o - %s | FileCheck %s
+
+void foo() {
+  static int hex = 0x10;
+  static int dec = 10;
+}
+void bar() {
+  static int hex = 0x20;
+  static int dec = 20;
+}

clang/test/utils/update_cc_test_checks/Inputs/global-hex-value-regex.c.expected

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --global-value-regex "foo\..*" "bar\..*" --global-hex-value-regex ".*\.hex"
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -emit-llvm -o - %s | FileCheck %s
+
+//.
+// CHECK: @foo.hex = internal global i32 [[#0x10]], align 4
+// CHECK: @foo.dec = internal global i32 10, align 4
+// CHECK: @bar.hex = internal global i32 [[#0x20]], align 4
+// CHECK: @bar.dec = internal global i32 20, align 4
+//.
+// CHECK-LABEL: @foo(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    ret void
+//
+void foo() {
+  static int hex = 0x10;
+  static int dec = 10;
+}
+// CHECK-LABEL: @bar(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    ret void
+//
+void bar() {
+  static int hex = 0x20;
+  static int dec = 20;
+}

clang/test/utils/update_cc_test_checks/global-hex-value-regex.test

+RUN: rm -rf %t && mkdir %t
+
+# Check --global-hex-value-regex.
+RUN: cp %S/Inputs/global-hex-value-regex.c %t/test.c
+RUN: %update_cc_test_checks %t/test.c --check-globals \
+RUN:     --global-value-regex "foo\..*" "bar\..*" \
+RUN:     --global-hex-value-regex ".*\.hex"
+RUN: diff -u %S/Inputs/global-hex-value-regex.c.expected %t/test.c
+
+# Check that the generated directives actually work correctly.
+RUN: cp %S/Inputs/lit.cfg.example %t/lit.cfg
+# Show lit failures while avoiding confusing FileCheck input dump nesting.
+RUN: %lit %t
+# Lit was successful.  Sanity-check the results with deterministic test order.
+RUN: rm %t/.lit_test_times.txt
+RUN: %lit %t 2>&1 | FileCheck %s
+
+CHECK: Testing: 1 tests
+CHECK: PASS: {{.*}} test.c

llvm/utils/UpdateTestChecks/common.py

def parse_commandline_args(parser):
   parser.add_argument('--disable', action='store_false', dest='enabled',
                       help='Deactivate CHECK line generation from this point forward')
   parser.add_argument('--replace-value-regex', nargs='+', default=[],
                       help='List of regular expressions to replace matching value names')
   parser.add_argument('--prefix-filecheck-ir-name', default='',
                       help='Add a prefix to FileCheck IR value names to avoid conflicts with scripted names')
   parser.add_argument('--global-value-regex', nargs='+', default=[],
                       help='List of regular expressions that a global value declaration must match to generate a check (has no effect if checking globals is not enabled)')
+  parser.add_argument('--global-hex-value-regex', nargs='+', default=[],
+                      help='List of regular expressions such that, for matching global value declarations, literal integer values should be encoded in hex in the associated FileCheck directives')
   args = parser.parse_args()
-  global _verbose, _global_value_regex
+  global _verbose, _global_value_regex, _global_hex_value_regex
   _verbose = args.verbose
   _global_value_regex = args.global_value_regex
+  _global_hex_value_regex = args.global_hex_value_regex
   return args
 
 
 class InputLineInfo(object):
   def __init__(self, line, line_number, args, argv):
     self.line = line
     self.line_number = line_number
     self.args = args
def transform_line_vars(match):
     # including the commas and spaces.
     return match.group(1) + rv + match.group(match.lastindex)
 
   lines_with_def = []
 
   for i, line in enumerate(lines):
     # An IR variable named '%.' matches the FileCheck regex string.
     line = line.replace('%.', '%dot')
+    for regex in _global_hex_value_regex:
+      if re.match('^@' + regex + ' = ', line):
+        line = re.sub(r'\bi([0-9]+) ([0-9]+)',
+            lambda m : 'i' + m.group(1) + ' [[#' + hex(int(m.group(2))) + ']]',
+            line)
+        break
     # Ignore any comments, since the check lines will too.
     scrubbed_line = SCRUB_IR_COMMENT_RE.sub(r'', line)
     lines[i] = scrubbed_line
     if not is_analyze:
       # It can happen that two matches are back-to-back and for some reason sub
       # will not replace both of them. For now we work around this by
       # substituting until there is no more match.
       changed = True