[llvm] [llvm-remarkutil] Auto-detect serializer format based on file extension (PR #160545)

[Tobias Stadler via llvm-commits]

https://github.com/tobias-stadler updated https://github.com/llvm/llvm-project/pull/160545

>From 01135cd8f003ca45e329581f6fd1e91a2ff58fe2 Mon Sep 17 00:00:00 2001
From: Tobias Stadler <mail at stadler-tobias.de>
Date: Wed, 24 Sep 2025 16:45:50 +0100
Subject: [PATCH] [spr] initial version

Created using spr 1.3.7-wip
---
 llvm/test/tools/llvm-remarkutil/filter.test      | 10 ++++++++++
 llvm/tools/llvm-remarkutil/RemarkFilter.cpp      |  8 ++------
 llvm/tools/llvm-remarkutil/RemarkUtilHelpers.cpp | 14 ++++++++++++++
 llvm/tools/llvm-remarkutil/RemarkUtilHelpers.h   |  9 ++++++++-
 4 files changed, 34 insertions(+), 7 deletions(-)

diff --git a/llvm/test/tools/llvm-remarkutil/filter.test b/llvm/test/tools/llvm-remarkutil/filter.test
index 8304b9f0129a8..9fd2e94247ef7 100644
--- a/llvm/test/tools/llvm-remarkutil/filter.test
+++ b/llvm/test/tools/llvm-remarkutil/filter.test
@@ -18,9 +18,19 @@ RUN: llvm-remarkutil filter --remark-type=analysis %p/Inputs/filter.yaml | FileC
 
 RUN: llvm-remarkutil yaml2bitstream -o %t.opt.bitstream %p/Inputs/filter.yaml
 RUN: llvm-remarkutil filter --function=func1 %t.opt.bitstream | FileCheck %s --strict-whitespace --check-prefix=REMARK1
+RUN: llvm-remarkutil filter --function=func1 %t.opt.bitstream -o %t.r1.yamL
+RUN: cat %t.r1.yamL | FileCheck %s --strict-whitespace --check-prefix=REMARK1
+RUN: llvm-remarkutil filter --function=func1 %t.opt.bitstream -o %t.r1.yMl
+RUN: cat %t.r1.yMl | FileCheck %s --strict-whitespace --check-prefix=REMARK1
+RUN: llvm-remarkutil filter --function=func1 %t.opt.bitstream --serializer=yaml -o %t.r1.fake.opt.bitstream
+RUN: cat %t.r1.fake.opt.bitstream | FileCheck %s --strict-whitespace --check-prefix=REMARK1
 
 RUN: llvm-remarkutil filter --function=func1 %t.opt.bitstream -o %t.r1.opt.bitstream
 RUN: llvm-remarkutil bitstream2yaml %t.r1.opt.bitstream | FileCheck %s --strict-whitespace --check-prefix=REMARK1
+RUN: llvm-remarkutil filter --function=func1 %t.opt.bitstream -o %t.r1
+RUN: llvm-remarkutil bitstream2yaml %t.r1 | FileCheck %s --strict-whitespace --check-prefix=REMARK1
+RUN: llvm-remarkutil filter --function=func1 %p/Inputs/filter.yaml --serializer=bitstream -o %t.r1.fake.yaml
+RUN: llvm-remarkutil bitstream2yaml %t.r1.fake.yaml | FileCheck %s --strict-whitespace --check-prefix=REMARK1
 
 RUN: llvm-remarkutil filter --function=func %p/Inputs/filter.yaml | FileCheck %s --allow-empty --strict-whitespace --check-prefix=EMPTY
 
diff --git a/llvm/tools/llvm-remarkutil/RemarkFilter.cpp b/llvm/tools/llvm-remarkutil/RemarkFilter.cpp
index acfef6608677c..1fffed9b542f2 100644
--- a/llvm/tools/llvm-remarkutil/RemarkFilter.cpp
+++ b/llvm/tools/llvm-remarkutil/RemarkFilter.cpp
@@ -46,12 +46,8 @@ static Error tryFilter() {
     return MaybeParser.takeError();
   auto &Parser = **MaybeParser;
 
-  Format SerializerFormat = OutputFormat;
-  if (SerializerFormat == Format::Auto) {
-    SerializerFormat = Parser.ParserFormat;
-    if (OutputFileName.empty() || OutputFileName == "-")
-      SerializerFormat = Format::YAML;
-  }
+  Format SerializerFormat =
+      getSerializerFormat(OutputFileName, OutputFormat, Parser.ParserFormat);
 
   auto MaybeOF = getOutputFileForRemarks(OutputFileName, SerializerFormat);
   if (!MaybeOF)
diff --git a/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.cpp b/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.cpp
index be529480e7d24..b6204d00f46a4 100644
--- a/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.cpp
+++ b/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.cpp
@@ -54,6 +54,20 @@ getOutputFileForRemarks(StringRef OutputFileName, Format OutputFormat) {
                                                     : sys::fs::OF_None);
 }
 
+Format getSerializerFormat(StringRef OutputFileName, Format SelectedFormat,
+                           Format DefaultFormat) {
+  if (SelectedFormat != Format::Auto)
+    return SelectedFormat;
+  SelectedFormat = DefaultFormat;
+  if (OutputFileName.empty() || OutputFileName == "-" ||
+      OutputFileName.ends_with_insensitive(".yaml") ||
+      OutputFileName.ends_with_insensitive(".yml"))
+    SelectedFormat = Format::YAML;
+  if (OutputFileName.ends_with_insensitive(".bitstream"))
+    SelectedFormat = Format::Bitstream;
+  return SelectedFormat;
+}
+
 Expected<FilterMatcher>
 FilterMatcher::createRE(const llvm::cl::opt<std::string> &Arg) {
   return createRE(Arg.ArgStr, Arg);
diff --git a/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.h b/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.h
index 0dd550765c1c6..73867fe35f06c 100644
--- a/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.h
+++ b/llvm/tools/llvm-remarkutil/RemarkUtilHelpers.h
@@ -47,7 +47,8 @@
       "serializer", cl::init(Format::Auto),                                    \
       cl::desc("Output remark format to serialize"),                           \
       cl::values(clEnumValN(Format::Auto, "auto",                              \
-                            "Follow the parser format (default)"),             \
+                            "Automatic detection based on output file "        \
+                            "extension or parser format (default)"),           \
                  clEnumValN(Format::YAML, "yaml", "YAML"),                     \
                  clEnumValN(Format::Bitstream, "bitstream", "Bitstream")),     \
       cl::sub(SUBOPT));
@@ -151,6 +152,12 @@ getOutputFileWithFlags(StringRef OutputFileName, sys::fs::OpenFlags Flags);
 Expected<std::unique_ptr<ToolOutputFile>>
 getOutputFileForRemarks(StringRef OutputFileName, Format OutputFormat);
 
+/// Choose the serializer format. If \p SelectedFormat is Format::Auto, try to
+/// detect the format based on the extension of \p OutputFileName or fall back
+/// to \p DefaultFormat.
+Format getSerializerFormat(StringRef OutputFileName, Format SelectedFormat,
+                           Format DefaultFormat);
+
 /// Filter object which can be either a string or a regex to match with the
 /// remark properties.
 class FilterMatcher {