Reland "[llvm-exegesis] Fix preservation of RDI in subprocess mode (#72458)"

This reverts commit 186db1b.

This relands commit 0718c1a.

The REQUIRES flag in the test that was added only specified that the
machine needed to have the capability to execute the snippet rather than
actually run it with performance counters. This would work with
--dummy-perf-counters, but that is not currently supported in the
subprocess execution mode. So for now, we require the ability to
actually perform measurements to prevent test failures in configurations
that don't have libpfm or access to performance counters.

Author: boomanaiden154

llvm/test/tools/llvm-exegesis/X86/latency/subprocess-preserved-registers.s

@@ -0,0 +1,30 @@
+# REQUIRES: exegesis-can-measure-latency, x86_64-linux
+
+# RUN: llvm-exegesis -mtriple=x86_64-unknown-unknown -mode=latency -snippets-file=%s -execution-mode=subprocess | FileCheck %s
+
+# See comment in ./subprocess-abnormal-exit-code.s on the transient
+# PTRACE_ATTACH failure.
+# ALLOW_RETRIES: 2
+
+# Check that the value of the registers preserved in subprocess mode while
+# making the ioctl system call are actually preserved correctly.
+
+# LLVM-EXEGESIS-DEFREG RAX 11
+# LLVM-EXEGESIS-DEFREG RDI 13
+# LLVM-EXEGESIS-DEFREG RSI 17
+# LLVM-EXEGESIS-DEFREG R13 0
+# LLVM-EXEGESIS-DEFREG R12 127
+
+cmpq $0x11, %rax
+cmovneq %r12, %r13
+cmpq $0x13, %rdi
+cmovneq %r12, %r13
+cmpq $0x17, %rsi
+cmovneq %r12, %r13
+
+movq $60, %rax
+movq %r13, %rdi
+syscall
+
+# CHECK-NOT: error:           'Child benchmarking process exited with non-zero exit code: Child process returned with unknown exit code'
+

llvm/tools/llvm-exegesis/lib/X86/Target.cpp

@@ -1205,7 +1205,7 @@ ExegesisX86Target::configurePerfCounter(long Request, bool SaveRegisters) const
   if(SaveRegisters) {
     // Restore RAX, RDI, and RSI, in reverse order.
     generateRegisterStackPop(X86::RSI, ConfigurePerfCounterCode);
-    generateRegisterStackPop(X86::RIP, ConfigurePerfCounterCode);
+    generateRegisterStackPop(X86::RDI, ConfigurePerfCounterCode);
     generateRegisterStackPop(X86::RAX, ConfigurePerfCounterCode);
   }
   return ConfigurePerfCounterCode;