Add epilogues for arm64 epilogue sharing logic

[MatzeB]

This is the compiler-rt part of http://reviews.llvm.org/D15600

lib/builtins/CMakeLists.txt

set(arm_SOURCES
   arm/udivmodsi4.S
   arm/udivsi3.S
   arm/umodsi3.S
   arm/unorddf2vfp.S
   arm/unordsf2vfp.S
   ${GENERIC_SOURCES})
 
 set(aarch64_SOURCES
+  arm64/epilogues.S
   comparetf2.c
   extenddftf2.c
   extendsftf2.c
   fixtfdi.c
   fixtfsi.c
   fixtfti.c
   fixunstfdi.c
   fixunstfsi.c

lib/builtins/arm64/epilogues.S

+#include "../assembly.h"
+
+  .text
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_X19_X20)
+  ldp x20, x19, [sp], #16
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_X19_X20)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP)
+  ldp x29, x30, [sp], #16
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_X27_X28)
+  ldp x28, x27, [sp], #16
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_X27_X28)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22)
+  ldp x20, x19, [sp, #16]
+  ldp x22, x21, [sp], #32
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20)
+  ldp x29, x30, [sp, #16]
+  ldp x20, x19, [sp], #32
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X27_X28)
+  ldp x29, x30, [sp, #16]
+  ldp x28, x27, [sp], #32
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X27_X28)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22_X23_X24)
+  ldp x20, x19, [sp, #32]
+  ldp x22, x21, [sp, #16]
+  ldp x24, x23, [sp], #48
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22_X23_X24)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22)
+  ldp x29, x30, [sp, #32]
+  ldp x20, x19, [sp, #16]
+  ldp x22, x21, [sp], #48
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X27_X28)
+  ldp x29, x30, [sp, #32]
+  ldp x20, x19, [sp, #16]
+  ldp x28, x27, [sp], #48
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X27_X28)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22_X23_X24_X25_X26)
+  ldp x20, x19, [sp, #48]
+  ldp x22, x21, [sp, #32]
+  ldp x24, x23, [sp, #16]
+  ldp x26, x25, [sp], #64
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22_X23_X24_X25_X26)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24)
+  ldp x29, x30, [sp, #48]
+  ldp x20, x19, [sp, #32]
+  ldp x22, x21, [sp, #16]
+  ldp x24, x23, [sp], #64
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24)
+
+  .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X27_X28)
+  ldp x29, x30, [sp, #48]
+  ldp x20, x19, [sp, #32]
+  ldp x22, x21, [sp, #16]
+  ldp x28, x27, [sp], #64
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X27_X28)
+
+.p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22_X23_X24_X25_X26_X27_X28)
+  ldp x20, x19, [sp, #64]
+  ldp x22, x21, [sp, #48]
+  ldp x24, x23, [sp, #32]
+  ldp x26, x25, [sp, #16]
+  ldp x28, x27, [sp], #80
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_X19_X20_X21_X22_X23_X24_X25_X26_X27_X28)
+
+.p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24_X25_X26)
+  ldp x29, x30, [sp, #64]
+  ldp x20, x19, [sp, #48]
+  ldp x22, x21, [sp, #32]
+  ldp x24, x23, [sp, #16]
+  ldp x26, x25, [sp], #80
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24_X25_X26)
+
+.p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24_X27_X28)
+  ldp x29, x30, [sp, #64]
+  ldp x20, x19, [sp, #48]
+  ldp x22, x21, [sp, #32]
+  ldp x24, x23, [sp, #16]
+  ldp x28, x27, [sp], #80
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24_X27_X28)
+
+.p2align 2
+DEFINE_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24_X25_X26_X27_X28)
+  ldp x29, x30, [sp, #80]
+  ldp x20, x19, [sp, #64]
+  ldp x22, x21, [sp, #48]
+  ldp x24, x23, [sp, #32]
+  ldp x26, x25, [sp, #16]
+  ldp x28, x27, [sp], #96
+  ret
+END_COMPILERRT_FUNCTION(__epilogue_LR_FP_X19_X20_X21_X22_X23_X24_X25_X26_X27_X28)

lib/builtins/assembly.h

  * ===----------------------------------------------------------------------===
  */
 
 #ifndef COMPILERRT_ASSEMBLY_H
 #define COMPILERRT_ASSEMBLY_H
 
 #if defined(__POWERPC__) || defined(__powerpc__) || defined(__ppc__)
 #define SEPARATOR @
+#elif defined(__APPLE__) && defined(__aarch64__)
+#define SEPARATOR %%
 #else
 #define SEPARATOR ;
 #endif
 
 #if defined(__APPLE__)
 #define HIDDEN(name) .private_extern name
 #define LOCAL_LABEL(name) L_##name
 // tell linker it can break up file at label boundaries