8136524: aarch64: test/compiler/runtime/7196199/Test7196199.java fails
authorenevill
Tue, 15 Sep 2015 12:59:51 +0000
changeset 33061 69a83b5ce390
parent 32733 0982a7e7eb15
child 33062 f6b082ab565e
child 33063 9e222337a81e
8136524: aarch64: test/compiler/runtime/7196199/Test7196199.java fails Summary: Fix safepoint handlers to save 128 bits on vector poll Reviewed-by: kvn Contributed-by: felix.yang@linaro.org
hotspot/src/cpu/aarch64/vm/macroAssembler_aarch64.cpp
hotspot/src/cpu/aarch64/vm/macroAssembler_aarch64.hpp
hotspot/src/cpu/aarch64/vm/sharedRuntime_aarch64.cpp
--- a/hotspot/src/cpu/aarch64/vm/macroAssembler_aarch64.cpp	Tue Sep 08 19:19:08 2015 +0200
+++ b/hotspot/src/cpu/aarch64/vm/macroAssembler_aarch64.cpp	Tue Sep 15 12:59:51 2015 +0000
@@ -2286,18 +2286,30 @@
 }
 #endif
 
-void MacroAssembler::push_CPU_state() {
-    push(0x3fffffff, sp);         // integer registers except lr & sp
-
+void MacroAssembler::push_CPU_state(bool save_vectors) {
+  push(0x3fffffff, sp);         // integer registers except lr & sp
+
+  if (!save_vectors) {
     for (int i = 30; i >= 0; i -= 2)
       stpd(as_FloatRegister(i), as_FloatRegister(i+1),
            Address(pre(sp, -2 * wordSize)));
+  } else {
+    for (int i = 30; i >= 0; i -= 2)
+      stpq(as_FloatRegister(i), as_FloatRegister(i+1),
+           Address(pre(sp, -4 * wordSize)));
+  }
 }
 
-void MacroAssembler::pop_CPU_state() {
-  for (int i = 0; i < 32; i += 2)
-    ldpd(as_FloatRegister(i), as_FloatRegister(i+1),
-         Address(post(sp, 2 * wordSize)));
+void MacroAssembler::pop_CPU_state(bool restore_vectors) {
+  if (!restore_vectors) {
+    for (int i = 0; i < 32; i += 2)
+      ldpd(as_FloatRegister(i), as_FloatRegister(i+1),
+           Address(post(sp, 2 * wordSize)));
+  } else {
+    for (int i = 0; i < 32; i += 2)
+      ldpq(as_FloatRegister(i), as_FloatRegister(i+1),
+           Address(post(sp, 4 * wordSize)));
+  }
 
   pop(0x3fffffff, sp);         // integer registers except lr & sp
 }
--- a/hotspot/src/cpu/aarch64/vm/macroAssembler_aarch64.hpp	Tue Sep 08 19:19:08 2015 +0200
+++ b/hotspot/src/cpu/aarch64/vm/macroAssembler_aarch64.hpp	Tue Sep 15 12:59:51 2015 +0000
@@ -777,8 +777,8 @@
 
   DEBUG_ONLY(void verify_heapbase(const char* msg);)
 
-  void push_CPU_state();
-  void pop_CPU_state() ;
+  void push_CPU_state(bool save_vectors = false);
+  void pop_CPU_state(bool restore_vectors = false) ;
 
   // Round up to a power of two
   void round_to(Register reg, int modulus);
--- a/hotspot/src/cpu/aarch64/vm/sharedRuntime_aarch64.cpp	Tue Sep 08 19:19:08 2015 +0200
+++ b/hotspot/src/cpu/aarch64/vm/sharedRuntime_aarch64.cpp	Tue Sep 15 12:59:51 2015 +0000
@@ -75,8 +75,8 @@
 // FIXME -- this is used by C1
 class RegisterSaver {
  public:
-  static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
-  static void restore_live_registers(MacroAssembler* masm);
+  static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors = false);
+  static void restore_live_registers(MacroAssembler* masm, bool restore_vectors = false);
 
   // Offsets into the register save area
   // Used by deoptimization when it is managing result register
@@ -108,7 +108,17 @@
 
 };
 
-OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
+OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors) {
+#ifdef COMPILER2
+  if (save_vectors) {
+    // Save upper half of vector registers
+    int vect_words = 32 * 8 / wordSize;
+    additional_frame_words += vect_words;
+  }
+#else
+  assert(!save_vectors, "vectors are generated only by C2");
+#endif
+
   int frame_size_in_bytes = round_to(additional_frame_words*wordSize +
                                      reg_save_size*BytesPerInt, 16);
   // OopMap frame size is in compiler stack slots (jint's) not bytes or words
@@ -122,7 +132,7 @@
   // Save registers, fpu state, and flags.
 
   __ enter();
-  __ push_CPU_state();
+  __ push_CPU_state(save_vectors);
 
   // Set an oopmap for the call site.  This oopmap will map all
   // oop-registers and debug-info registers as callee-saved.  This
@@ -139,14 +149,14 @@
                                     // register slots are 8 bytes
                                     // wide, 32 floating-point
                                     // registers
-      oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
+      oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset + additional_frame_slots),
                                 r->as_VMReg());
     }
   }
 
   for (int i = 0; i < FloatRegisterImpl::number_of_registers; i++) {
     FloatRegister r = as_FloatRegister(i);
-    int sp_offset = 2 * i;
+    int sp_offset = save_vectors ? (4 * i) : (2 * i);
     oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
                               r->as_VMReg());
   }
@@ -154,8 +164,11 @@
   return oop_map;
 }
 
-void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
-  __ pop_CPU_state();
+void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) {
+#ifndef COMPILER2
+  assert(!restore_vectors, "vectors are generated only by C2");
+#endif
+  __ pop_CPU_state(restore_vectors);
   __ leave();
 }
 
@@ -177,9 +190,9 @@
 }
 
 // Is vector's size (in bytes) bigger than a size saved by default?
-// 16 bytes XMM registers are saved by default using fxsave/fxrstor instructions.
+// 8 bytes vector registers are saved by default on AArch64.
 bool SharedRuntime::is_wide_vector(int size) {
-  return size > 16;
+  return size > 8;
 }
 // The java_calling_convention describes stack locations as ideal slots on
 // a frame with no abi restrictions. Since we must observe abi restrictions
@@ -2742,7 +2755,7 @@
   bool save_vectors = (poll_type == POLL_AT_VECTOR_LOOP);
 
   // Save registers, fpu state, and flags
-  map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
+  map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words, save_vectors);
 
   // The following is basically a call_VM.  However, we need the precise
   // address of the call in order to generate an oopmap. Hence, we do all the
@@ -2793,7 +2806,7 @@
   __ bind(noException);
 
   // Normal exit, restore registers and exit.
-  RegisterSaver::restore_live_registers(masm);
+  RegisterSaver::restore_live_registers(masm, save_vectors);
 
   __ ret(lr);