[PATCH] D156963: [X86] Workaround possible CPUID bug in Sandy Bridge.

[Craig Topper via Phabricator via llvm-commits]

craig.topper created this revision.
craig.topper added reviewers: pengfei, RKSimon, FreddyYe, anna.
Herald added subscribers: Enna1, hiraditya.
Herald added a project: All.
craig.topper requested review of this revision.
Herald added projects: Sanitizers, LLVM.
Herald added a subscriber: Sanitizers.

Don't access leaf 7 subleaf 1 unless subleaf 0 says it is
supported via EAX.

Intel documentation says invalid subleaves return 0. We had been
relying on that behavior instead of checking the max sublef number.

It appears that some Sandy Bridge CPUs return at least the subleaf 0
EDX value for subleaf 1. Best guess is that this is a bug in a
microcode patch since all of the bits we're seeing set in EDX were
introduced after Sandy Bridge was originally released.

This is causing avxvnniint16 to be incorrectly enabled with -march=native
on these CPUs.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D156963

Files:
  compiler-rt/lib/builtins/cpu_model.c
  llvm/lib/TargetParser/Host.cpp


Index: llvm/lib/TargetParser/Host.cpp
===================================================================
--- llvm/lib/TargetParser/Host.cpp
+++ llvm/lib/TargetParser/Host.cpp
@@ -1768,8 +1768,11 @@
   Features["avx512fp16"] = HasLeaf7 && ((EDX >> 23) & 1) && HasAVX512Save;
   Features["amx-tile"]   = HasLeaf7 && ((EDX >> 24) & 1) && HasAMXSave;
   Features["amx-int8"]   = HasLeaf7 && ((EDX >> 25) & 1) && HasAMXSave;
+  // EAX from subleaf 0 is the maximum subleaf supported. Some CPUs don't
+  // return all 0s for invalid subleaves so check the limit.
   bool HasLeaf7Subleaf1 =
-      MaxLevel >= 7 && !getX86CpuIDAndInfoEx(0x7, 0x1, &EAX, &EBX, &ECX, &EDX);
+      HasLeaf7 && EAX >= 1 &&
+      !getX86CpuIDAndInfoEx(0x7, 0x1, &EAX, &EBX, &ECX, &EDX);
   Features["sha512"]     = HasLeaf7Subleaf1 && ((EAX >> 0) & 1);
   Features["sm3"]        = HasLeaf7Subleaf1 && ((EAX >> 1) & 1);
   Features["sm4"]        = HasLeaf7Subleaf1 && ((EAX >> 2) & 1);
Index: compiler-rt/lib/builtins/cpu_model.c
===================================================================
--- compiler-rt/lib/builtins/cpu_model.c
+++ compiler-rt/lib/builtins/cpu_model.c
@@ -771,8 +771,11 @@
   if (HasLeaf7 && ((EDX >> 8) & 1) && HasAVX512Save)
     setFeature(FEATURE_AVX512VP2INTERSECT);
 
+  // EAX from subleaf 0 is the maximum subleaf supported. Some CPUs don't
+  // return all 0s for invalid subleaves so check the limit.
   bool HasLeaf7Subleaf1 =
-      MaxLeaf >= 0x7 && !getX86CpuIDAndInfoEx(0x7, 0x1, &EAX, &EBX, &ECX, &EDX);
+      HasLeaf7 && EAX >= 1 &&
+      !getX86CpuIDAndInfoEx(0x7, 0x1, &EAX, &EBX, &ECX, &EDX);
   if (HasLeaf7Subleaf1 && ((EAX >> 5) & 1) && HasAVX512Save)
     setFeature(FEATURE_AVX512BF16);
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D156963.546689.patch
Type: text/x-patch
Size: 1717 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230803/e72a34f1/attachment.bin>