[lld] [ELF] Handle and optimize x86-64 PLTOFF64 TLS sequences (PR #216263)

[Peter Smith via llvm-commits]

================
@@ -830,6 +850,18 @@ void X86_64::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
 void X86_64::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
                             uint64_t val) const {
   if (rel.type == R_X86_64_TLSGD) {
+    if (isPltOff64Tls(loc)) {
+      const uint8_t inst[] = {
+          0x64, 0x48, 0x8b, 0x04, 0x25, 0x00, 0x00,
+          0x00, 0x00,                            // mov %fs:0x0,%rax
+          0x48, 0x03, 0x05, 0,    0,    0,    0, // addq x at gottpoff(%rip),%rax
+          0x66, 0x0f, 0x1f, 0x44, 0x00, 0x00,    // nopw 0x0(%rax,%rax,1)
+      };
+      memcpy(loc - 3, inst, sizeof(inst));
+      // The new displacement is 9 bytes ahead of the original one.
----------------
smithp35 wrote:

There's a comment in the case below on line 879 that I think captures the meaning a bit better. Could we use that (with 9 instead of 8)
> // Both code sequences are PC relatives, but since we are moving the
    // constant forward by 8 bytes we have to subtract the value by 8.

https://github.com/llvm/llvm-project/pull/216263