[clang] [llvm] [X86] Reduce -ffixed-r compile-time overhead (PR #184606)

[via cfe-commits]

================
@@ -366,9 +357,23 @@ void x86::getX86TargetFeatures(const Driver &D, const llvm::Triple &Triple,
   RESERVE_REG(r13)
   RESERVE_REG(r14)
   RESERVE_REG(r15)
+#undef RESERVE_REG
+
+  bool NeedDetectEGPR = Args.hasArg(
+      options::OPT_ffixed_r16, options::OPT_ffixed_r17, options::OPT_ffixed_r18,
+      options::OPT_ffixed_r19, options::OPT_ffixed_r20, options::OPT_ffixed_r21,
+      options::OPT_ffixed_r22, options::OPT_ffixed_r23, options::OPT_ffixed_r24,
+      options::OPT_ffixed_r25, options::OPT_ffixed_r26, options::OPT_ffixed_r27,
+      options::OPT_ffixed_r28, options::OPT_ffixed_r29, options::OPT_ffixed_r30,
+      options::OPT_ffixed_r31);
----------------
zhouguangyuan0718 wrote:

This point has been discussed in the previous PR, refer to here: https://github.com/llvm/llvm-project/pull/180242#discussion_r2787884240.
In general, there are two reasons, gcc does as this, and other architectures in llvm do as this.

https://github.com/llvm/llvm-project/pull/184606