[ORC][ORC-RT] Add ORC-RT based lazy compilation support for x86-64.

Adds support for the ORC-RT based lazy compilation scheme that was introduced
in 570ecdc.

Author: lhames

compiler-rt/lib/orc/CMakeLists.txt

@@ -53,6 +53,7 @@ if (APPLE)
     macho_tlv.x86-64.S
     macho_tlv.arm64.S
     sysv_reenter.arm64.S
+    sysv_reenter.x86-64.S
     )
 
   set(ORC_IMPL_HEADERS
@@ -119,6 +120,7 @@ else() # not Apple
       elfnix_tls.aarch64.S
       elfnix_tls.ppc64.S
       sysv_reenter.arm64.S
+      sysv_reenter.x86-64.S
       )
   endif()
 

compiler-rt/lib/orc/sysv_reenter.x86-64.S

@@ -0,0 +1,81 @@
+//===-- orc_rt_macho_tlv.x86-64.s -------------------------------*- ASM -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of the ORC runtime support library.
+//
+//===----------------------------------------------------------------------===//
+
+// The content of this file is x86_64-only
+#if defined(__x86_64__)
+
+// Save all GRPS except %rsp.
+// This value is also subtracted from %rsp below, despite the fact that %rbp
+// has already been pushed, because we need %rsp to stay 16-byte aligned.
+#define GPR_SAVE_SPACE_SIZE             15 * 8
+#define FXSAVE64_SAVE_SPACE_SIZE        512
+#define REGISTER_SAVE_SPACE_SIZE        \
+        GPR_SAVE_SPACE_SIZE + FXSAVE64_SAVE_SPACE_SIZE
+
+        .text
+
+        // returns address of TLV in %rax, all other registers preserved
+        .globl __orc_rt_sysv_reenter
+__orc_rt_sysv_reenter:
+        pushq           %rbp
+        movq            %rsp,        %rbp
+        subq            $REGISTER_SAVE_SPACE_SIZE, %rsp
+        movq            %rax,     -8(%rbp)
+        movq            %rbx,    -16(%rbp)
+        movq            %rcx,    -24(%rbp)
+        movq            %rdx,    -32(%rbp)
+        movq            %rsi,    -40(%rbp)
+        movq            %rdi,    -48(%rbp)
+        movq            %r8,     -56(%rbp)
+        movq            %r9,     -64(%rbp)
+        movq            %r10,    -72(%rbp)
+        movq            %r11,    -80(%rbp)
+        movq            %r12,    -88(%rbp)
+        movq            %r13,    -96(%rbp)
+        movq            %r14,   -104(%rbp)
+        movq            %r15,   -112(%rbp)
+        fxsave64        (%rsp)
+        movq            8(%rbp), %rdi
+
+        // Load return address and subtract five from it (on the assumption
+        // that it's a call instruction).
+        subq            $5, %rdi
+
+        // Call __orc_rt_resolve to look up the implementation corresponding to
+        // the calling stub, then store this in x17 (which we'll return to
+        // below).
+#if !defined(__APPLE__)
+        call            __orc_rt_resolve
+#else
+        call            ___orc_rt_resolve
+#endif
+        movq            %rax,   8(%rbp)
+        fxrstor64       (%rsp)
+        movq            -112(%rbp),     %r15
+        movq            -104(%rbp),     %r14
+        movq            -96(%rbp),      %r13
+        movq            -88(%rbp),      %r12
+        movq            -80(%rbp),      %r11
+        movq            -72(%rbp),      %r10
+        movq            -64(%rbp),      %r9
+        movq            -56(%rbp),      %r8
+        movq            -48(%rbp),      %rdi
+        movq            -40(%rbp),      %rsi
+        movq            -32(%rbp),      %rdx
+        movq            -24(%rbp),      %rcx
+        movq            -16(%rbp),      %rbx
+        movq            -8(%rbp),       %rax
+        addq            $REGISTER_SAVE_SPACE_SIZE, %rsp
+        popq            %rbp
+        ret
+
+#endif // defined(__x86_64__)

compiler-rt/test/orc/TestCases/Generic/lazy-link.ll

@@ -11,7 +11,7 @@
 ; RUN:     -lazy %t/x.o | FileCheck %s
 ;
 ; UNSUPPORTED: system-windows
-; REQUIRES: target={{(arm|aarch)64.*}}
+; REQUIRES: target={{(arm|aarch|x86_)64.*}}
 ;
 ; CHECK: Linking {{.*}}main.o
 ; CHECK-DAG: Linking <indirect stubs graph #1>

llvm/include/llvm/ExecutionEngine/JITLink/x86_64.h

@@ -641,6 +641,31 @@ inline Symbol &createAnonymousPointerJumpStub(LinkGraph &G,
       false);
 }
 
+/// x86-64 reentry trampoline.
+///
+/// Contains the instruction sequence for a trampoline that stores its return
+/// address on the stack and calls <reentry-symbol>:
+///   call  <reentry-symbol>
+extern const char ReentryTrampolineContent[5];
+
+/// Create a block of N reentry trampolines.
+inline Block &createReentryTrampolineBlock(LinkGraph &G,
+                                           Section &TrampolineSection,
+                                           Symbol &ReentrySymbol) {
+  auto &B = G.createContentBlock(TrampolineSection, ReentryTrampolineContent,
+                                 orc::ExecutorAddr(~uint64_t(7)), 1, 0);
+  B.addEdge(BranchPCRel32, 1, ReentrySymbol, 0);
+  return B;
+}
+
+inline Symbol &createAnonymousReentryTrampoline(LinkGraph &G,
+                                                Section &TrampolineSection,
+                                                Symbol &ReentrySymbol) {
+  return G.addAnonymousSymbol(
+      createReentryTrampolineBlock(G, TrampolineSection, ReentrySymbol), 0,
+      sizeof(ReentryTrampolineContent), true, false);
+}
+
 /// Global Offset Table Builder.
 class GOTTableManager : public TableManager<GOTTableManager> {
 public:

llvm/lib/ExecutionEngine/JITLink/x86_64.cpp

@@ -85,6 +85,10 @@ const char NullPointerContent[PointerSize] = {0x00, 0x00, 0x00, 0x00,
 const char PointerJumpStubContent[6] = {
     static_cast<char>(0xFFu), 0x25, 0x00, 0x00, 0x00, 0x00};
 
+const char ReentryTrampolineContent[5] = {
+  static_cast<char>(0xe8), 0x00, 0x00, 0x00, 0x00
+};
+
 Error optimizeGOTAndStubAccesses(LinkGraph &G) {
   LLVM_DEBUG(dbgs() << "Optimizing GOT entries and stubs:\n");
 

llvm/lib/ExecutionEngine/Orc/JITLinkReentryTrampolines.cpp

@@ -9,6 +9,7 @@
 #include "llvm/ExecutionEngine/Orc/JITLinkReentryTrampolines.h"
 
 #include "llvm/ExecutionEngine/JITLink/aarch64.h"
+#include "llvm/ExecutionEngine/JITLink/x86_64.h"
 #include "llvm/ExecutionEngine/Orc/ObjectLinkingLayer.h"
 
 #include <memory>
@@ -91,6 +92,9 @@ JITLinkReentryTrampolines::Create(ObjectLinkingLayer &ObjLinkingLayer) {
   case Triple::aarch64:
     EmitTrampoline = aarch64::createAnonymousReentryTrampoline;
     break;
+  case Triple::x86_64:
+    EmitTrampoline = x86_64::createAnonymousReentryTrampoline;
+    break;
   default:
     return make_error<StringError>("JITLinkReentryTrampolines: architecture " +
 				   TT.getArchName() + " not supported",