[LLVMbugs] [Bug 1053] NEW: Processor autodetection incorrect

bugzilla-daemon at cs.uiuc.edu bugzilla-daemon at cs.uiuc.edu
Sat Dec 16 09:57:57 PST 2006


           Summary: Processor autodetection incorrect
           Product: tools
           Version: trunk
          Platform: Macintosh
        OS/Version: MacOS X
            Status: NEW
          Severity: normal
          Priority: P2
         Component: TableGen
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: scottm at aero.org

This bug initially manifests itself while building the c frontend, but it isn't
really a cfe bug (sorta). The CFE sets the rs6000_cpu variable to
llvm::PPCSubtarget::ParseSubtargetFeatures eventually gets a string that
represents the default processor architecture, which starts with "-970",
corresponding to PROCESSOR_POWER4.

This effectively nails the processor architecture down to always being a G5/970.
tblgen emits code that will only set the specific CPU iff the CPU isn't set yet
(call setCPUIfNone())

Meanwhile, llvm has correctly identified the host CPU, which is a G4 in my case.
Because setCPUIfNone() is called and the default features string specified a
970, the processor is de-facto set to a G5 irrespective of the actual, detected

This patch is an attempt to fix/remedy the bug (but it might not be entirely

Index: utils/TableGen/SubtargetEmitter.cpp
RCS file: /var/cvs/llvm/llvm/utils/TableGen/SubtargetEmitter.cpp,v
retrieving revision 1.24
diff -u -r1.24 SubtargetEmitter.cpp
--- utils/TableGen/SubtargetEmitter.cpp 12 Dec 2006 20:55:58 -0000      1.24
+++ utils/TableGen/SubtargetEmitter.cpp 16 Dec 2006 17:50:10 -0000
@@ -461,7 +461,8 @@
   OS << "Subtarget::ParseSubtargetFeatures(const std::string &FS,\n"
         "                                  const std::string &CPU) {\n"
         "  SubtargetFeatures Features(FS);\n"
-        "  Features.setCPUIfNone(CPU);\n"
+       "  if (CPU != \"\");\n"
+        "    Features.setCPU(CPU);\n"
         "  uint32_t Bits =  Features.getBits(SubTypeKV, SubTypeKVSize,\n"
         "                                    FeatureKV, FeatureKVSize);\n";

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

More information about the llvm-bugs mailing list