--- a/hotspot/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Mon Sep 17 17:02:10 2012 -0700
+++ b/hotspot/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Mon Sep 17 19:39:07 2012 -0700
@@ -116,8 +116,8 @@
};
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
@@ -134,7 +134,19 @@
static void restore_result_registers(MacroAssembler* masm);
};
-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) {
+ int vect_words = 0;
+#ifdef COMPILER2
+ if (save_vectors) {
+ assert(UseAVX > 0, "256bit vectors are supported only with AVX");
+ assert(MaxVectorSize == 32, "only 256bit vectors are supported now");
+ // Save upper half of YMM registes
+ vect_words = 16 * 16 / wordSize;
+ additional_frame_words += vect_words;
+ }
+#else
+ assert(!save_vectors, "vectors are generated only by C2");
+#endif
// Always make the frame size 16-byte aligned
int frame_size_in_bytes = round_to(additional_frame_words*wordSize +
@@ -155,6 +167,27 @@
__ enter(); // rsp becomes 16-byte aligned here
__ push_CPU_state(); // Push a multiple of 16 bytes
+
+ if (vect_words > 0) {
+ assert(vect_words*wordSize == 256, "");
+ __ subptr(rsp, 256); // Save upper half of YMM registes
+ __ vextractf128h(Address(rsp, 0),xmm0);
+ __ vextractf128h(Address(rsp, 16),xmm1);
+ __ vextractf128h(Address(rsp, 32),xmm2);
+ __ vextractf128h(Address(rsp, 48),xmm3);
+ __ vextractf128h(Address(rsp, 64),xmm4);
+ __ vextractf128h(Address(rsp, 80),xmm5);
+ __ vextractf128h(Address(rsp, 96),xmm6);
+ __ vextractf128h(Address(rsp,112),xmm7);
+ __ vextractf128h(Address(rsp,128),xmm8);
+ __ vextractf128h(Address(rsp,144),xmm9);
+ __ vextractf128h(Address(rsp,160),xmm10);
+ __ vextractf128h(Address(rsp,176),xmm11);
+ __ vextractf128h(Address(rsp,192),xmm12);
+ __ vextractf128h(Address(rsp,208),xmm13);
+ __ vextractf128h(Address(rsp,224),xmm14);
+ __ vextractf128h(Address(rsp,240),xmm15);
+ }
if (frame::arg_reg_save_area_bytes != 0) {
// Allocate argument register save area
__ subptr(rsp, frame::arg_reg_save_area_bytes);
@@ -167,112 +200,111 @@
OopMapSet *oop_maps = new OopMapSet();
OopMap* map = new OopMap(frame_size_in_slots, 0);
- map->set_callee_saved(VMRegImpl::stack2reg( rax_off + additional_frame_slots), rax->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( rcx_off + additional_frame_slots), rcx->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( rdx_off + additional_frame_slots), rdx->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( rbx_off + additional_frame_slots), rbx->as_VMReg());
+
+#define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_slots)
+
+ map->set_callee_saved(STACK_OFFSET( rax_off ), rax->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( rcx_off ), rcx->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( rdx_off ), rdx->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( rbx_off ), rbx->as_VMReg());
// rbp location is known implicitly by the frame sender code, needs no oopmap
// and the location where rbp was saved by is ignored
- map->set_callee_saved(VMRegImpl::stack2reg( rsi_off + additional_frame_slots), rsi->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( rdi_off + additional_frame_slots), rdi->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r8_off + additional_frame_slots), r8->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r9_off + additional_frame_slots), r9->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r10_off + additional_frame_slots), r10->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r11_off + additional_frame_slots), r11->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r12_off + additional_frame_slots), r12->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r13_off + additional_frame_slots), r13->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r14_off + additional_frame_slots), r14->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg( r15_off + additional_frame_slots), r15->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm0_off + additional_frame_slots), xmm0->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm1_off + additional_frame_slots), xmm1->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm2_off + additional_frame_slots), xmm2->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm3_off + additional_frame_slots), xmm3->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm4_off + additional_frame_slots), xmm4->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm5_off + additional_frame_slots), xmm5->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm6_off + additional_frame_slots), xmm6->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm7_off + additional_frame_slots), xmm7->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm8_off + additional_frame_slots), xmm8->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm9_off + additional_frame_slots), xmm9->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm10_off + additional_frame_slots), xmm10->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm11_off + additional_frame_slots), xmm11->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm12_off + additional_frame_slots), xmm12->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm13_off + additional_frame_slots), xmm13->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm14_off + additional_frame_slots), xmm14->as_VMReg());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm15_off + additional_frame_slots), xmm15->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( rsi_off ), rsi->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( rdi_off ), rdi->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r8_off ), r8->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r9_off ), r9->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r10_off ), r10->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r11_off ), r11->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r12_off ), r12->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r13_off ), r13->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r14_off ), r14->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET( r15_off ), r15->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm0_off ), xmm0->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm1_off ), xmm1->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm2_off ), xmm2->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm3_off ), xmm3->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm4_off ), xmm4->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm5_off ), xmm5->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm6_off ), xmm6->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm7_off ), xmm7->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm8_off ), xmm8->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm9_off ), xmm9->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm10_off), xmm10->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm11_off), xmm11->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm12_off), xmm12->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm13_off), xmm13->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm14_off), xmm14->as_VMReg());
+ map->set_callee_saved(STACK_OFFSET(xmm15_off), xmm15->as_VMReg());
// %%% These should all be a waste but we'll keep things as they were for now
if (true) {
- map->set_callee_saved(VMRegImpl::stack2reg( raxH_off + additional_frame_slots),
- rax->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( rcxH_off + additional_frame_slots),
- rcx->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( rdxH_off + additional_frame_slots),
- rdx->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( rbxH_off + additional_frame_slots),
- rbx->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( raxH_off ), rax->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( rcxH_off ), rcx->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( rdxH_off ), rdx->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( rbxH_off ), rbx->as_VMReg()->next());
// rbp location is known implicitly by the frame sender code, needs no oopmap
- map->set_callee_saved(VMRegImpl::stack2reg( rsiH_off + additional_frame_slots),
- rsi->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( rdiH_off + additional_frame_slots),
- rdi->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r8H_off + additional_frame_slots),
- r8->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r9H_off + additional_frame_slots),
- r9->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r10H_off + additional_frame_slots),
- r10->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r11H_off + additional_frame_slots),
- r11->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r12H_off + additional_frame_slots),
- r12->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r13H_off + additional_frame_slots),
- r13->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r14H_off + additional_frame_slots),
- r14->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg( r15H_off + additional_frame_slots),
- r15->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm0H_off + additional_frame_slots),
- xmm0->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm1H_off + additional_frame_slots),
- xmm1->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm2H_off + additional_frame_slots),
- xmm2->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm3H_off + additional_frame_slots),
- xmm3->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm4H_off + additional_frame_slots),
- xmm4->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm5H_off + additional_frame_slots),
- xmm5->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm6H_off + additional_frame_slots),
- xmm6->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm7H_off + additional_frame_slots),
- xmm7->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm8H_off + additional_frame_slots),
- xmm8->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm9H_off + additional_frame_slots),
- xmm9->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm10H_off + additional_frame_slots),
- xmm10->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm11H_off + additional_frame_slots),
- xmm11->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm12H_off + additional_frame_slots),
- xmm12->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm13H_off + additional_frame_slots),
- xmm13->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm14H_off + additional_frame_slots),
- xmm14->as_VMReg()->next());
- map->set_callee_saved(VMRegImpl::stack2reg(xmm15H_off + additional_frame_slots),
- xmm15->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( rsiH_off ), rsi->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( rdiH_off ), rdi->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r8H_off ), r8->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r9H_off ), r9->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r10H_off ), r10->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r11H_off ), r11->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r12H_off ), r12->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r13H_off ), r13->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r14H_off ), r14->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET( r15H_off ), r15->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm0H_off ), xmm0->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm1H_off ), xmm1->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm2H_off ), xmm2->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm3H_off ), xmm3->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm4H_off ), xmm4->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm5H_off ), xmm5->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm6H_off ), xmm6->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm7H_off ), xmm7->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm8H_off ), xmm8->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm9H_off ), xmm9->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm10H_off), xmm10->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm11H_off), xmm11->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm12H_off), xmm12->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm13H_off), xmm13->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm14H_off), xmm14->as_VMReg()->next());
+ map->set_callee_saved(STACK_OFFSET(xmm15H_off), xmm15->as_VMReg()->next());
}
return map;
}
-void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
+void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) {
if (frame::arg_reg_save_area_bytes != 0) {
// Pop arg register save area
__ addptr(rsp, frame::arg_reg_save_area_bytes);
}
+#ifdef COMPILER2
+ if (restore_vectors) {
+ // Restore upper half of YMM registes.
+ assert(UseAVX > 0, "256bit vectors are supported only with AVX");
+ assert(MaxVectorSize == 32, "only 256bit vectors are supported now");
+ __ vinsertf128h(xmm0, Address(rsp, 0));
+ __ vinsertf128h(xmm1, Address(rsp, 16));
+ __ vinsertf128h(xmm2, Address(rsp, 32));
+ __ vinsertf128h(xmm3, Address(rsp, 48));
+ __ vinsertf128h(xmm4, Address(rsp, 64));
+ __ vinsertf128h(xmm5, Address(rsp, 80));
+ __ vinsertf128h(xmm6, Address(rsp, 96));
+ __ vinsertf128h(xmm7, Address(rsp,112));
+ __ vinsertf128h(xmm8, Address(rsp,128));
+ __ vinsertf128h(xmm9, Address(rsp,144));
+ __ vinsertf128h(xmm10, Address(rsp,160));
+ __ vinsertf128h(xmm11, Address(rsp,176));
+ __ vinsertf128h(xmm12, Address(rsp,192));
+ __ vinsertf128h(xmm13, Address(rsp,208));
+ __ vinsertf128h(xmm14, Address(rsp,224));
+ __ vinsertf128h(xmm15, Address(rsp,240));
+ __ addptr(rsp, 256);
+ }
+#else
+ assert(!restore_vectors, "vectors are generated only by C2");
+#endif
// Recover CPU state
__ pop_CPU_state();
// Get the rbp described implicitly by the calling convention (no oopMap)
@@ -297,6 +329,12 @@
__ addptr(rsp, return_offset_in_bytes());
}
+// 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.
+bool SharedRuntime::is_wide_vector(int size) {
+ return size > 16;
+}
+
// The java_calling_convention describes stack locations as ideal slots on
// a frame with no abi restrictions. Since we must observe abi restrictions
// (like the placement of the register window) the slots must be biased by
@@ -3235,7 +3273,6 @@
return 0;
}
-
//------------------------------generate_deopt_blob----------------------------
void SharedRuntime::generate_deopt_blob() {
// Allocate space for the code
@@ -3740,7 +3777,7 @@
// Generate a special Compile2Runtime blob that saves all registers,
// and setup oopmap.
//
-SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, bool cause_return) {
+SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
assert(StubRoutines::forward_exception_entry() != NULL,
"must be generated before");
@@ -3755,6 +3792,8 @@
address start = __ pc();
address call_pc = NULL;
int frame_size_in_words;
+ bool cause_return = (poll_type == POLL_AT_RETURN);
+ bool save_vectors = (poll_type == POLL_AT_VECTOR_LOOP);
// Make room for return address (or push it again)
if (!cause_return) {
@@ -3762,7 +3801,7 @@
}
// 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
@@ -3799,7 +3838,7 @@
// Exception pending
- RegisterSaver::restore_live_registers(masm);
+ RegisterSaver::restore_live_registers(masm, save_vectors);
__ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
@@ -3807,7 +3846,7 @@
__ bind(noException);
// Normal exit, restore registers and exit.
- RegisterSaver::restore_live_registers(masm);
+ RegisterSaver::restore_live_registers(masm, save_vectors);
__ ret(0);