[Mlir-commits] [mlir] [mlir][utils] Add script to verify canonicalizations against Alive2 (PR #91867)

[Ivan Butygin]

https://github.com/Hardcode84 updated https://github.com/llvm/llvm-project/pull/91867

>From 2124fa47ad8308dbc0070a88c25ff37d45bec6a2 Mon Sep 17 00:00:00 2001
From: Ivan Butygin <ivan.butygin at gmail.com>
Date: Sat, 11 May 2024 22:24:37 +0200
Subject: [PATCH 1/2] [mlir][utils] Add script to verify canonicalizations
 agains Alive2

This script takes IR before and after canonicalization, translates it into llvm IR and converts to format suitablle for Alive2 https://alive2.llvm.org/ce/

This is primarily for arith canonicalizations verification, but technically it can be adapted for any dialect translatable to llvm.

Usage `python verify_canon.py canonicalize.mlir func1 func2 ...`

Example output: https://alive2.llvm.org/ce/z/KhQs4J

Initial discussion: https://github.com/llvm/llvm-project/pull/91646#pullrequestreview-2049342826
---
 mlir/utils/verify-canon/verify_canon.py | 74 +++++++++++++++++++++++++
 1 file changed, 74 insertions(+)
 create mode 100644 mlir/utils/verify-canon/verify_canon.py

diff --git a/mlir/utils/verify-canon/verify_canon.py b/mlir/utils/verify-canon/verify_canon.py
new file mode 100644
index 0000000000000..1dae30757f220
--- /dev/null
+++ b/mlir/utils/verify-canon/verify_canon.py
@@ -0,0 +1,74 @@
+# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+# Run canonicalization, convert IR to LLVM and convert to format suitable to
+# verification against Alive2 https://alive2.llvm.org/ce/.
+
+import subprocess
+import tempfile
+import sys
+from pathlib import Path
+
+
+def filter_funcs(ir, funcs):
+    if not funcs:
+        return ir
+
+    funcs_str = ",".join(funcs)
+    return subprocess.check_output(
+        ["mlir-opt", f"--symbol-privatize=exclude={funcs_str}", "--symbol-dce"],
+        input=ir,
+    )
+
+
+def add_func_prefix(src, prefix):
+    return src.replace("@", "@" + prefix)
+
+
+def merge_ir(chunks):
+    files = []
+    for chunk in chunks:
+        tmp = tempfile.NamedTemporaryFile(suffix=".ll")
+        tmp.write(chunk)
+        tmp.flush()
+        files.append(tmp)
+
+    return subprocess.check_output(["llvm-link", "-S"] + [f.name for f in files])
+
+
+if __name__ == "__main__":
+    argv = sys.argv
+    if len(argv) < 2:
+        print(f"usage: {argv[0]} canonicalize.mlir [func1] [func2] ...")
+        exit(0)
+
+    file = argv[1]
+    funcs = argv[2:]
+
+    orig_ir = Path(file).read_bytes()
+    orig_ir = filter_funcs(orig_ir, funcs)
+
+    to_llvm_args = [
+        "--convert-arith-to-llvm",
+        "--convert-func-to-llvm",
+        "--convert-ub-to-llvm",
+        "--convert-vector-to-llvm",
+    ]
+    orig_args = ["mlir-opt"] + to_llvm_args
+    canon_args = ["mlir-opt", "-canonicalize"] + to_llvm_args
+    translate_args = ["mlir-translate", "-mlir-to-llvmir"]
+
+    orig = subprocess.check_output(orig_args, input=orig_ir)
+    canonicalized = subprocess.check_output(canon_args, input=orig_ir)
+
+    orig = subprocess.check_output(translate_args, input=orig)
+    canonicalized = subprocess.check_output(translate_args, input=canonicalized)
+
+    enc = "utf-8"
+    orig = bytes(add_func_prefix(orig.decode(enc), "src_"), enc)
+    canonicalized = bytes(add_func_prefix(canonicalized.decode(enc), "tgt_"), enc)
+
+    res = merge_ir([orig, canonicalized])
+
+    print(res.decode(enc))

>From 719336566ddbfa1cb85dc55adbbf06e37302b25e Mon Sep 17 00:00:00 2001
From: Ivan Butygin <ivan.butygin at gmail.com>
Date: Sun, 12 May 2024 22:08:31 +0200
Subject: [PATCH 2/2] add example

---
 mlir/utils/verify-canon/verify_canon.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/mlir/utils/verify-canon/verify_canon.py b/mlir/utils/verify-canon/verify_canon.py
index 1dae30757f220..4198d996a8928 100644
--- a/mlir/utils/verify-canon/verify_canon.py
+++ b/mlir/utils/verify-canon/verify_canon.py
@@ -4,6 +4,7 @@
 
 # Run canonicalization, convert IR to LLVM and convert to format suitable to
 # verification against Alive2 https://alive2.llvm.org/ce/.
+# Example: `python verify_canon.py canonicalize.mlir func1 func2 func3`
 
 import subprocess
 import tempfile