Improvements to seed processing script

gfauto/gfauto/process_seeds.py

@@ -28,11 +28,12 @@
 
 
 def process_chunk(  # pylint: disable=too-many-locals;
-    chunk_num: int, chunk: Set[str], log_files: List[Path], output_file: TextIO
+    chunk_num: int, chunk: Set[str], log_files: List[Path], only_show_num_signatures: bool, look_for_spirv_opt_errors: bool, look_for_llpc_errors: bool, output_file: TextIO
 ) -> None:
 
     log(f"\nChunk {chunk_num}:")
-    output_file.write(f"\nChunk {chunk_num}:\n")
+    if not only_show_num_signatures:
+        output_file.write(f"\nChunk {chunk_num}:\n")
 
     unique_signatures: Set[str] = set()
 
@@ -50,30 +51,52 @@ def process_chunk(  # pylint: disable=too-many-locals;
             end_line = 0
             found_bug = False
             for i, line in enumerate(lines):
-                match = re.fullmatch(r"STATUS (\w+)\n", line)
-                if not match:
-                    continue
-                status = match.group(1)
-                if status == "SUCCESS":
-                    start_line = i + 1
-                    continue
-                found_bug = True
-                end_line = i + 1
-                break
+                if look_for_spirv_opt_errors:
+                    match = re.match(r"Command failed: \['/usr/bin/catchsegv', .*spirv-opt'", line)
+                    if match:
+                        found_bug = True
+                        start_line = i + 1
+                        # TODO: Is there a reliable way to find the end of a spirv-opt failure?
+                        end_line = len(lines) - 1
+                        break
+                elif look_for_llpc_errors:
+                    match = re.match(r"Command failed: \['/usr/bin/catchsegv', '.*amdllpc", line)
+                    if match:
+                        found_bug = True
+                        start_line = i + 1
+                        # TODO: Is there a reliable way to find the end of an LLPC failure?
+                        end_line = len(lines) - 1
+                        break
+                else:
+                    match = re.fullmatch(r"STATUS (\w+)\n", line)
+                    if not match:
+                        continue
+                    status = match.group(1)
+                    if status == "SUCCESS":
+                        start_line = i + 1
+                        continue
+                    found_bug = True
+                    end_line = i + 1
+                    break
 
             if not found_bug:
                 continue
 
-            failure_log = "\n".join(lines[start_line:end_line])
+            failure_log = "".join(lines[start_line:end_line])
 
             signature = signature_util.get_signature_from_log_contents(failure_log)
 
             unique_signatures.add(signature)
 
-    # Print the signatures.
-    for signature in sorted(unique_signatures):
-        log(signature)
-        output_file.write(f"{signature}\n")
+    if only_show_num_signatures:
+        # Print number of unique signatures.
+        log(str(len(unique_signatures)) + '\n')
+        output_file.write(f"{len(unique_signatures)}\n")
+    else:
+        # Print the unique signatures.
+        for signature in sorted(unique_signatures):
+            log(signature)
+            output_file.write(f"{signature}\n")
 
 
 def main() -> None:
@@ -87,30 +110,53 @@ def main() -> None:
         "--out", help="Output file.", default="signatures_chunked.txt",
     )
 
+    parser.add_argument(
+        "--chunk_size", help="Chunk size.", type=int, default=1_000,
+
+    )
+
+    parser.add_argument(
+        "--only_show_num_signatures", help="Output only the number of distinct signatures per chunk, one integer per line.  This is useful for gathering statistics.", default=False, action='store_true'
+    )
+
+    parser.add_argument(
+        "--spirv_opt", help="Look for spirv-opt errors.", default=False, action='store_true'
+    )
+
+    parser.add_argument(
+        "--llpc", help="Look for LLPC errors.", default=False, action='store_true'
+    )
+
     parsed_args = parser.parse_args(sys.argv[1:])
 
     seed_file: Path = Path(parsed_args.seed_file)
     output_file: Path = Path(parsed_args.out)
+    chunk_size: int = parsed_args.chunk_size
+    only_show_num_signatures: bool = parsed_args.only_show_num_signatures
+    look_for_spirv_opt_errors: bool = parsed_args.spirv_opt
+    look_for_llpc_errors: bool = parsed_args.llpc
+
+    check(not (look_for_spirv_opt_errors and look_for_llpc_errors), AssertionError("At most one of --spirv_opt and --llpc can be used"))
 
     # Get a list of all log files.
     log_files: List[Path] = sorted(Path().glob("log_*.txt"))
 
     # Get chunks of seeds and call process_chunk.
     seeds: List[str] = util.file_read_text(seed_file).split()
 
-    check(len(seeds) == 10_000, AssertionError("Expected 10,000 seeds."))
+    check((len(seeds) % chunk_size) == 0, AssertionError("The number of seeds should be a multiple of chunk_size."))
 
     with util.file_open_text(output_file, "w") as output:
         index = 0
-        for chunk_num in range(0, 10):
+        for chunk_num in range(0, len(seeds) // chunk_size):
             chunk: Set[str] = set()
-            for _ in range(0, 1_000):
+            for _ in range(0, chunk_size):
                 chunk.add(seeds[index])
                 index += 1
-            process_chunk(chunk_num, chunk, log_files, output)
+            process_chunk(chunk_num, chunk, log_files, only_show_num_signatures, look_for_spirv_opt_errors, look_for_llpc_errors, output)
 
         check(
-            index == 10_000, AssertionError("Expected to have processed 10,000 seeds.")
+            index == len(seeds), AssertionError("Expected to have processed all seeds.")
         )