[RISCV] Detect empty extension name after parsing MajorVersion in parseNormalizedArchString.

[topperc]

If the string is just a version, we will end up adding an empty string as an extension which crashes in the compare function for the std::map.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Craig Topper (topperc)

Changes

If the string is just a version, we will end up adding an empty string as an extension which crashes in the compare function for the std::map.

---

Full diff: https://github.com/llvm/llvm-project/pull/90790.diff

2 Files Affected:

• (modified) llvm/lib/TargetParser/RISCVISAInfo.cpp (+4)
• (modified) llvm/unittests/TargetParser/RISCVISAInfoTest.cpp (+8)

diff --git a/llvm/lib/TargetParser/RISCVISAInfo.cpp b/llvm/lib/TargetParser/RISCVISAInfo.cpp
index e8172ebb259720..0863d9319acb4e 100644
--- a/llvm/lib/TargetParser/RISCVISAInfo.cpp
+++ b/llvm/lib/TargetParser/RISCVISAInfo.cpp
@@ -470,6 +470,10 @@ RISCVISAInfo::parseNormalizedArchString(StringRef Arch) {
       return createStringError(errc::invalid_argument,
                                "extension lacks version in expected format");
 
+    if (VersionStart == 0)
+      return createStringError(errc::invalid_argument,
+                               "missing extension name");
+
     StringRef ExtName = Prefix.slice(0, VersionStart);
     StringRef MajorVersionStr = Prefix.slice(VersionStart, StringRef::npos);
     if (MajorVersionStr.getAsInteger(10, MajorVersion))
diff --git a/llvm/unittests/TargetParser/RISCVISAInfoTest.cpp b/llvm/unittests/TargetParser/RISCVISAInfoTest.cpp
index 3aa0178100abf4..9e88b87df4e8a7 100644
--- a/llvm/unittests/TargetParser/RISCVISAInfoTest.cpp
+++ b/llvm/unittests/TargetParser/RISCVISAInfoTest.cpp
@@ -45,6 +45,14 @@ TEST(ParseNormalizedArchString, RejectsMalformedInputs) {
   }
 }
 
+TEST(ParseNormalizedArchString, OnlyVersion) {
+  for (StringRef Input : {"rv64i2p0_1p0", "rv32i2p0_1p0"}) {
+    EXPECT_EQ(
+        toString(RISCVISAInfo::parseNormalizedArchString(Input).takeError()),
+        "missing extension name");
+  }
+}
+
 TEST(ParseNormalizedArchString, AcceptsValidBaseISAsAndSetsXLen) {
   auto MaybeRV32I = RISCVISAInfo::parseNormalizedArchString("rv32i2p0");
   ASSERT_THAT_EXPECTED(MaybeRV32I, Succeeded());

[4vtomat]

I got a question, is rv64gcv_zvfh000001p0 also a valid arch string?

[topperc]

rv64gcv_zvfh000001p0

Looks like both gcc and clang accept it.

[asb]

LGTM