link in memcpy after opts and lto if it shows up only at that time

3 files changed, 73 insertions, 3 deletions

diff --git a/emcc b/emcc
index d6b6ed17..e927fbd8 100755
--- a/emcc
+++ b/emcc
@@ -90,6 +90,8 @@ LLVM_OPT_LEVEL = {
   3: 3,
 }
 
+MEMCPY_ALIASES = ['memcpy', 'llvm.memcpy.i32', 'llvm.memcpy.i64', 'llvm.memcpy.p0i8.p0i8.i32', 'llvm.memcpy.p0i8.p0i8.i64']
+
 DEBUG = int(os.environ.get('EMCC_DEBUG') or 0)
 TEMP_DIR = os.environ.get('EMCC_TEMP_DIR')
 LEAVE_INPUTS_RAW = os.environ.get('EMCC_LEAVE_INPUTS_RAW') # Do not compile .ll files into .bc, just compile them with emscripten directly
@@ -947,7 +949,7 @@ try:
 
     def fix_libc(need):
       # If an intrinsic alias of memcpy is used, we need memcpy
-      for memcpy_alias in ['llvm.memcpy.i32', 'llvm.memcpy.i64', 'llvm.memcpy.p0i8.p0i8.i32', 'llvm.memcpy.p0i8.p0i8.i64']:
+      for memcpy_alias in MEMCPY_ALIASES:
         if memcpy_alias in need:
           if '_memcpy' not in shared.Settings.EXPORTED_FUNCTIONS:
             shared.Settings.EXPORTED_FUNCTIONS.append('_memcpy')
@@ -1087,6 +1089,30 @@ try:
       shared.Building.llvm_opt(in_temp(target_basename + '.bc'), link_opts)
       if DEBUG: save_intermediate('linktime', 'bc')
 
+    # Optimization and lto can add new intrinsics like memcpy that were not present before. We
+    # are now *after* linking in libc, so we missed our chance to get memcpy - check and add it now
+    # if necessary
+    final_symbols = shared.Building.llvm_nm(final)
+    need_memcpy = False
+    for symbol in final_symbols.undefs:
+      if symbol in MEMCPY_ALIASES:
+        need_memcpy = True
+        break
+    has_memcpy = False
+    for symbol in final_symbols.defs:
+      if symbol in MEMCPY_ALIASES:
+        has_memcpy = True
+        break
+    if need_memcpy and not has_memcpy:
+      if DEBUG: print >> sys.stderr, 'memcpy intrinsic added in optimizations, linking in optimized memcpy'
+      memcpy = in_temp('memcpy.bc')
+      execute([shared.PYTHON, shared.EMCC, shared.path_from_root('system', 'lib', 'libc', 'musl', 'memcpy.c'), '-o', memcpy], stdout=stdout, stderr=stderr)
+      shared.Building.llvm_opt(memcpy, llvm_opts) # optimize it just like normal code; no point in lto though
+      next = final + '.postrinsics.bc'
+      shared.Building.link([final, memcpy], next)
+      final = next
+      if DEBUG: save_intermediate('postrinsics', 'bc')
+
   # Prepare .ll for Emscripten
   if not LEAVE_INPUTS_RAW:
     final = shared.Building.llvm_dis(final, final + '.ll')
diff --git a/src/library.js b/src/library.js
index 74425e4f..f9cab641 100644
--- a/src/library.js
+++ b/src/library.js
@@ -4202,8 +4202,10 @@ LibraryManager.library = {
 
   memcpy: function (dest, src, num) {
     // simple version, in general it should not be used - we should pull it in from libc
-#if ASSERTIONS
-    Module.printErr('warning: library.js memcpy should not be running!');
+    if (!_memcpy.shown) {
+      _memcpy.shown = true;
+      Module.printErr('warning: library.js memcpy should not be running, it is only for testing!');
+    }
 #endif
     while (num--) {
       HEAP8[dest++] = HEAP8[src++];
diff --git a/tests/runner.py b/tests/runner.py
index 35a484be..ea124c30 100755
--- a/tests/runner.py
+++ b/tests/runner.py
@@ -8754,6 +8754,48 @@ f.close()
       Popen([PYTHON, EMCC, os.path.join(self.get_dir(), 'main.cpp')]).communicate()
       self.assertContained('1234, 1234, 4321\n', run_js(os.path.join(self.get_dir(), 'a.out.js')))
 
+    def test_link_memcpy(self):
+      # memcpy can show up *after* optimizations, so after our opportunity to link in libc, so it must be special-cased
+      open(os.path.join(self.get_dir(), 'main.cpp'), 'w').write(r'''
+        #include <stdio.h>
+
+        int main(int argc, char **argv) {
+          int num = argc + 10;
+          char buf[num], buf2[num];
+          for (int i = 0; i < num; i++) {
+            buf[i] = i*i+i/3;
+          }
+          for (int i = 1; i < num; i++) {
+            buf[i] += buf[i-1];
+          }
+          for (int i = 0; i < num; i++) {
+            buf2[i] = buf[i];
+          }          
+          for (int i = 1; i < num; i++) {
+            buf2[i] += buf2[i-1];
+          }
+          for (int i = 0; i < num; i++) {
+            printf("%d:%d\n", i, buf2[i]);
+          }
+          return 0;
+        }
+      ''')
+      Popen([PYTHON, EMCC, '-O2', '--closure', '-0', os.path.join(self.get_dir(), 'main.cpp')]).communicate()
+      output = run_js(os.path.join(self.get_dir(), 'a.out.js'), full_output=True, stderr=PIPE)
+      self.assertContained('''0:0
+1:1
+2:6
+3:21
+4:53
+5:111
+6:-49
+7:98
+8:55
+9:96
+10:-16
+''', output)
+      self.assertNotContained('warning: library.js memcpy should not be running, it is only for testing!', output)
+
     def test_warn_undefined(self):
       open(os.path.join(self.get_dir(), 'main.cpp'), 'w').write(r'''
         #include <stdio.h>