[llvm] e308b8e - [RISCV] Fix arch string parsing for multi-character extensions

[via llvm-commits]

Author: eopXD
Date: 2021-12-09T21:13:44-08:00
New Revision: e308b8e0c71b5ce783035d73722f59d395245cf9

URL: https://github.com/llvm/llvm-project/commit/e308b8e0c71b5ce783035d73722f59d395245cf9
DIFF: https://github.com/llvm/llvm-project/commit/e308b8e0c71b5ce783035d73722f59d395245cf9.diff

LOG: [RISCV] Fix arch string parsing for multi-character extensions

Current implementation can't parse extension names that contains digits
correctly (e.g. `zvl128b`). This patch fixes it.

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D109215

Added: 
    

Modified: 
    clang/test/Driver/riscv-arch.c
    llvm/lib/Support/RISCVISAInfo.cpp

Removed: 
    


################################################################################
diff  --git a/clang/test/Driver/riscv-arch.c b/clang/test/Driver/riscv-arch.c
index bbbc0f3ded78a..5b99643309a57 100644
--- a/clang/test/Driver/riscv-arch.c
+++ b/clang/test/Driver/riscv-arch.c
@@ -392,7 +392,7 @@
 
 // RUN: %clang -target riscv32-unknown-elf -march=rv32izbb1p0zbp0p93 -menable-experimental-extensions -### %s \
 // RUN: -fsyntax-only 2>&1 | FileCheck -check-prefix=RV32-EXPERIMENTAL-ZBB-ZBP-UNDERSCORE %s
-// RV32-EXPERIMENTAL-ZBB-ZBP-UNDERSCORE: error: invalid arch name 'rv32izbb1p0zbp0p93', multi-character extensions must be separated by underscores
+// RV32-EXPERIMENTAL-ZBB-ZBP-UNDERSCORE: error: invalid arch name 'rv32izbb1p0zbp0p93', unsupported version number 0.93 for extension 'zbb1p0zbp'
 
 // RUN: %clang -target riscv32-unknown-elf -march=rv32izba1p0 -menable-experimental-extensions -### %s \
 // RUN: -fsyntax-only 2>&1 | FileCheck -check-prefix=RV32-EXPERIMENTAL-ZBA %s

diff  --git a/llvm/lib/Support/RISCVISAInfo.cpp b/llvm/lib/Support/RISCVISAInfo.cpp
index 8e984002f90d2..1ca2120c40750 100644
--- a/llvm/lib/Support/RISCVISAInfo.cpp
+++ b/llvm/lib/Support/RISCVISAInfo.cpp
@@ -72,6 +72,28 @@ static bool stripExperimentalPrefix(StringRef &Ext) {
   return Ext.consume_front("experimental-");
 }
 
+// This function finds the first character that doesn't belong to a version
+// (e.g. zbe0p93 is extension 'zbe' of version '0p93'). So the function will
+// consume [0-9]*p[0-9]* starting from the backward. An extension name will not
+// end with a digit or the letter 'p', so this function will parse correctly.
+// NOTE: This function is NOT able to take empty strings or strings that only
+// have version numbers and no extension name. It assumes the extension name
+// will be at least more than one character.
+static size_t findFirstNonVersionCharacter(const StringRef &Ext) {
+  if (Ext.size() == 0)
+    llvm_unreachable("Already guarded by if-statement in ::parseArchString");
+
+  int Pos = Ext.size() - 1;
+  while (Pos > 0 && isDigit(Ext[Pos]))
+    Pos--;
+  if (Pos > 0 && Ext[Pos] == 'p' && isDigit(Ext[Pos - 1])) {
+    Pos--;
+    while (Pos > 0 && isDigit(Ext[Pos]))
+      Pos--;
+  }
+  return Pos;
+}
+
 struct FindByName {
   FindByName(StringRef Ext) : Ext(Ext){};
   StringRef Ext;
@@ -638,7 +660,7 @@ RISCVISAInfo::parseArchString(StringRef Arch, bool EnableExperimentalExtension,
 
     StringRef Type = getExtensionType(Ext);
     StringRef Desc = getExtensionTypeDesc(Ext);
-    auto Pos = Ext.find_if(isDigit);
+    size_t Pos = findFirstNonVersionCharacter(Ext) + 1;
     StringRef Name(Ext.substr(0, Pos));
     StringRef Vers(Ext.substr(Pos));