[llvm] [compiler-rt] [X86] Add CPU detection for more znver2 CPUs (PR #74955)

[Aiden Grossman via llvm-commits]

================
@@ -647,35 +647,59 @@ static const char *getAMDProcessorTypeAndSubtype(unsigned Family,
   case 23:
     CPU = "znver1";
     *Type = AMDFAM17H;
-    if ((Model >= 0x30 && Model <= 0x3f) || Model == 0x71) {
+    if ((Model >= 0x30 && Model <= 0x3f) || (Model == 0x47) ||
+        (Model >= 0x60 && Model <= 0x67) || (Model >= 0x68 && Model <= 0x6f) ||
+        (Model >= 0x70 && Model <= 0x7f) || (Model >= 0x84 && Model <= 0x87) ||
+        (Model >= 0x90 && Model <= 0x97) || (Model >= 0x98 && Model <= 0x9f) ||
+        (Model >= 0xa0 && Model <= 0xaf)) {
+      // Family 17h Models 30h-3Fh (Starship) Zen 2
+      // Family 17h Models 47h (Cardinal) Zen 2
+      // Family 17h Models 60h-67h (Renoir) Zen 2
+      // Family 17h Models 68h-6Fh (Lucienne) Zen 2
+      // Family 17h Models 70h-7Fh (Matisse) Zen 2
+      // Family 17h Models 84h-87h (ProjectX) Zen 2
+      // Family 17h Models 90h-97h (VanGogh) Zen 2
+      // Family 17h Models 98h-9Fh (Mero) Zen 2
+      // Family 17h Models A0h-AFh (Mendocino) Zen 2
       CPU = "znver2";
       *Subtype = AMDFAM17H_ZNVER2;
-      break; // 30h-3fh, 71h: Zen2
+      break;
     }
-    if (Model <= 0x0f) {
+    if ((Model >= 0x10 && Model <= 0x1f) || (Model >= 0x20 && Model <= 0x2f)) {
+      // Family 17h Models 10h-1Fh (Raven1) Zen
+      // Family 17h Models 10h-1Fh (Picasso) Zen+
+      // Family 17h Models 20h-2Fh (Raven2 x86) Zen
       *Subtype = AMDFAM17H_ZNVER1;
-      break; // 00h-0Fh: Zen1
+      break;
     }
     break;
   case 25:
     CPU = "znver3";
     *Type = AMDFAM19H;
-    if (Model <= 0x0f || (Model >= 0x20 && Model <= 0x5f)) {
-      // Family 19h Models 00h-0Fh - Zen3
-      // Family 19h Models 20h-2Fh - Zen3
-      // Family 19h Models 30h-3Fh - Zen3
-      // Family 19h Models 40h-4Fh - Zen3+
-      // Family 19h Models 50h-5Fh - Zen3+
+    if ((Model >= 0x00 && Model <= 0x0f) || (Model >= 0x20 && Model <= 0x2f) ||
----------------
boomanaiden154 wrote:

I think I'd still make the argument that representing all of the ranges consistently makes the code more readable, but I've posted 3ddf3685248a8c3ef07bf8290196c33521edb894 as it doesn't significantly impact readability and gets rid of the gcc warning. Thanks for pointing this issue out!

https://github.com/llvm/llvm-project/pull/74955