Mercurial > hg > truffle

comparison src/cpu/x86/vm/sharedRuntime_x86_64.cpp @ 21559:be896a1983c0

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
recast all Graal native code as JVMCI code (JBS:GRAAL-53)
author Doug Simon <doug.simon@oracle.com>
date Thu, 28 May 2015 15:36:48 +0200
parents 2e35a4ea22ac
children b3d5463a8362
comparison
equal deleted inserted replaced
21558:d563baeca9df 21559:be896a1983c0
38 #include "c1/c1_Runtime1.hpp" 38 #include "c1/c1_Runtime1.hpp"
39 #endif 39 #endif
40 #ifdef COMPILER2 40 #ifdef COMPILER2
41 #include "opto/runtime.hpp" 41 #include "opto/runtime.hpp"
42 #endif 42 #endif
43 #ifdef GRAAL 43 #ifdef JVMCI
44 #include "graal/graalJavaAccess.hpp" 44 #include "jvmci/jvmciJavaAccess.hpp"
45 #endif 45 #endif
46 46
47 #define __ masm-> 47 #define __ masm->
48 48
49 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size; 49 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
71 // units because compiler frame slots are jints. 71 // units because compiler frame slots are jints.
72 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off 72 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
73 #define DEF_YMM_HI_OFFS(regnum) ymm_hi ## regnum ## _off = ymm_off + (regnum)*16/BytesPerInt 73 #define DEF_YMM_HI_OFFS(regnum) ymm_hi ## regnum ## _off = ymm_off + (regnum)*16/BytesPerInt
74 enum layout { 74 enum layout {
75 fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area 75 fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area
76 #if defined(COMPILER2) || defined(GRAAL) 76 #if defined(COMPILER2) || defined(JVMCI)
77 ymm_off = fpu_state_off, // offset in fxsave save area 77 ymm_off = fpu_state_off, // offset in fxsave save area
78 DEF_YMM_HI_OFFS(0), 78 DEF_YMM_HI_OFFS(0),
79 DEF_YMM_HI_OFFS(1), 79 DEF_YMM_HI_OFFS(1),
80 DEF_YMM_HI_OFFS(2), 80 DEF_YMM_HI_OFFS(2),
81 DEF_YMM_HI_OFFS(3), 81 DEF_YMM_HI_OFFS(3),
180 // to be under the return like a normal enter. 180 // to be under the return like a normal enter.
181 181
182 __ enter(); // rsp becomes 16-byte aligned here 182 __ enter(); // rsp becomes 16-byte aligned here
183 __ push_CPU_state(); // Push a multiple of 16 bytes 183 __ push_CPU_state(); // Push a multiple of 16 bytes
184 184
185 #if defined(COMPILER2) || defined(GRAAL) 185 #if defined(COMPILER2) || defined(JVMCI)
186 __ subptr(rsp, 256); // Save upper half of YMM registers 186 __ subptr(rsp, 256); // Save upper half of YMM registers
187 if (save_vectors) { 187 if (save_vectors) {
188 assert(UseAVX > 0, "256bit vectors are supported only with AVX"); 188 assert(UseAVX > 0, "256bit vectors are supported only with AVX");
189 assert(MaxVectorSize == 32, "only 256bit vectors are supported now"); 189 assert(MaxVectorSize == 32, "only 256bit vectors are supported now");
190 // Save upper half of YMM registers 190 // Save upper half of YMM registers
204 __ vextractf128h(Address(rsp, ymm_hi13_off * BytesPerInt), xmm13); 204 __ vextractf128h(Address(rsp, ymm_hi13_off * BytesPerInt), xmm13);
205 __ vextractf128h(Address(rsp, ymm_hi14_off * BytesPerInt), xmm14); 205 __ vextractf128h(Address(rsp, ymm_hi14_off * BytesPerInt), xmm14);
206 __ vextractf128h(Address(rsp, ymm_hi15_off * BytesPerInt), xmm15); 206 __ vextractf128h(Address(rsp, ymm_hi15_off * BytesPerInt), xmm15);
207 } 207 }
208 #else 208 #else
209 assert(!save_vectors, "vectors are generated only by C2 and Graal"); 209 assert(!save_vectors, "vectors are generated only by C2 and JVMCI");
210 #endif 210 #endif
211 if (frame::arg_reg_save_area_bytes != 0) { 211 if (frame::arg_reg_save_area_bytes != 0) {
212 // Allocate argument register save area 212 // Allocate argument register save area
213 __ subptr(rsp, frame::arg_reg_save_area_bytes); 213 __ subptr(rsp, frame::arg_reg_save_area_bytes);
214 } 214 }
255 map->set_callee_saved(STACK_OFFSET(xmm13_off), xmm13->as_VMReg()); 255 map->set_callee_saved(STACK_OFFSET(xmm13_off), xmm13->as_VMReg());
256 map->set_callee_saved(STACK_OFFSET(xmm14_off), xmm14->as_VMReg()); 256 map->set_callee_saved(STACK_OFFSET(xmm14_off), xmm14->as_VMReg());
257 map->set_callee_saved(STACK_OFFSET(xmm15_off), xmm15->as_VMReg()); 257 map->set_callee_saved(STACK_OFFSET(xmm15_off), xmm15->as_VMReg());
258 258
259 259
260 #if defined(COMPILER2) || defined(GRAAL) 260 #if defined(COMPILER2) || defined(JVMCI)
261 if (save_vectors) { 261 if (save_vectors) {
262 map->set_callee_saved(STACK_OFFSET(ymm_hi0_off ), xmm0->as_VMReg()->next()->next()->next()->next()); 262 map->set_callee_saved(STACK_OFFSET(ymm_hi0_off ), xmm0->as_VMReg()->next()->next()->next()->next());
263 map->set_callee_saved(STACK_OFFSET(ymm_hi1_off ), xmm1->as_VMReg()->next()->next()->next()->next()); 263 map->set_callee_saved(STACK_OFFSET(ymm_hi1_off ), xmm1->as_VMReg()->next()->next()->next()->next());
264 map->set_callee_saved(STACK_OFFSET(ymm_hi2_off ), xmm2->as_VMReg()->next()->next()->next()->next()); 264 map->set_callee_saved(STACK_OFFSET(ymm_hi2_off ), xmm2->as_VMReg()->next()->next()->next()->next());
265 map->set_callee_saved(STACK_OFFSET(ymm_hi3_off ), xmm3->as_VMReg()->next()->next()->next()->next()); 265 map->set_callee_saved(STACK_OFFSET(ymm_hi3_off ), xmm3->as_VMReg()->next()->next()->next()->next());
319 void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) { 319 void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) {
320 if (frame::arg_reg_save_area_bytes != 0) { 320 if (frame::arg_reg_save_area_bytes != 0) {
321 // Pop arg register save area 321 // Pop arg register save area
322 __ addptr(rsp, frame::arg_reg_save_area_bytes); 322 __ addptr(rsp, frame::arg_reg_save_area_bytes);
323 } 323 }
324 #if defined(COMPILER2) || defined(GRAAL) 324 #if defined(COMPILER2) || defined(JVMCI)
325 if (restore_vectors) { 325 if (restore_vectors) {
326 // Restore upper half of YMM registes. 326 // Restore upper half of YMM registes.
327 assert(UseAVX > 0, "256bit vectors are supported only with AVX"); 327 assert(UseAVX > 0, "256bit vectors are supported only with AVX");
328 assert(MaxVectorSize == 32, "only 256bit vectors are supported now"); 328 assert(MaxVectorSize == 32, "only 256bit vectors are supported now");
329 __ vinsertf128h(xmm0, Address(rsp, ymm_hi0_off * BytesPerInt)); 329 __ vinsertf128h(xmm0, Address(rsp, ymm_hi0_off * BytesPerInt));
343 __ vinsertf128h(xmm14, Address(rsp, ymm_hi14_off * BytesPerInt)); 343 __ vinsertf128h(xmm14, Address(rsp, ymm_hi14_off * BytesPerInt));
344 __ vinsertf128h(xmm15, Address(rsp, ymm_hi15_off * BytesPerInt)); 344 __ vinsertf128h(xmm15, Address(rsp, ymm_hi15_off * BytesPerInt));
345 } 345 }
346 __ addptr(rsp, 256); 346 __ addptr(rsp, 256);
347 #else 347 #else
348 assert(!restore_vectors, "vectors are generated only by C2 and Graal"); 348 assert(!restore_vectors, "vectors are generated only by C2 and JVMCI");
349 #endif 349 #endif
350 // Recover CPU state 350 // Recover CPU state
351 __ pop_CPU_state(); 351 __ pop_CPU_state();
352 // Get the rbp described implicitly by the calling convention (no oopMap) 352 // Get the rbp described implicitly by the calling convention (no oopMap)
353 __ pop(rbp); 353 __ pop(rbp);
741 __ stop(msg); 741 __ stop(msg);
742 __ bind(L_ok); 742 __ bind(L_ok);
743 __ block_comment("} verify_i2ce "); 743 __ block_comment("} verify_i2ce ");
744 } 744 }
745 745
746 #ifdef GRAAL 746 #ifdef JVMCI
747 if (frame_extension_argument != -1) { 747 if (frame_extension_argument != -1) {
748 // The frame_extension_argument is an int that describes the 748 // The frame_extension_argument is an int that describes the
749 // expected amount of argument space in the caller frame. If that 749 // expected amount of argument space in the caller frame. If that
750 // is greater than total_args_passed then enlarge the caller frame 750 // is greater than total_args_passed then enlarge the caller frame
751 // by that amount to ensure deopt works correctly. 751 // by that amount to ensure deopt works correctly.
812 812
813 // Will jump to the compiled code just as if compiled code was doing it. 813 // Will jump to the compiled code just as if compiled code was doing it.
814 // Pre-load the register-jump target early, to schedule it better. 814 // Pre-load the register-jump target early, to schedule it better.
815 __ movptr(r11, Address(rbx, in_bytes(Method::from_compiled_offset()))); 815 __ movptr(r11, Address(rbx, in_bytes(Method::from_compiled_offset())));
816 816
817 #ifdef GRAAL 817 #ifdef JVMCI
818 // check if this call should be routed towards a specific entry point 818 // check if this call should be routed towards a specific entry point
819 __ cmpptr(Address(r15_thread, in_bytes(JavaThread::graal_alternate_call_target_offset())), 0); 819 __ cmpptr(Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())), 0);
820 Label no_alternative_target; 820 Label no_alternative_target;
821 __ jcc(Assembler::equal, no_alternative_target); 821 __ jcc(Assembler::equal, no_alternative_target);
822 __ movptr(r11, Address(r15_thread, in_bytes(JavaThread::graal_alternate_call_target_offset()))); 822 __ movptr(r11, Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
823 __ movptr(Address(r15_thread, in_bytes(JavaThread::graal_alternate_call_target_offset())), 0); 823 __ movptr(Address(r15_thread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())), 0);
824 __ bind(no_alternative_target); 824 __ bind(no_alternative_target);
825 #endif 825 #endif
826 826
827 // Now generate the shuffle code. Pick up all register args and move the 827 // Now generate the shuffle code. Pick up all register args and move the
828 // rest through the floating point stack top. 828 // rest through the floating point stack top.
3426 // Normal deoptimization. Save exec mode for unpack_frames. 3426 // Normal deoptimization. Save exec mode for unpack_frames.
3427 __ movl(r14, Deoptimization::Unpack_deopt); // callee-saved 3427 __ movl(r14, Deoptimization::Unpack_deopt); // callee-saved
3428 __ jmp(cont); 3428 __ jmp(cont);
3429 3429
3430 int reexecute_offset = __ pc() - start; 3430 int reexecute_offset = __ pc() - start;
3431 #if defined(COMPILERGRAAL) && !defined(COMPILER1) 3431 #if defined(COMPILERJVMCI) && !defined(COMPILER1)
3432 // Graal does not use this kind of deoptimization 3432 // JVMCI does not use this kind of deoptimization
3433 __ should_not_reach_here(); 3433 __ should_not_reach_here();
3434 #endif 3434 #endif
3435 3435
3436 // Reexecute case 3436 // Reexecute case
3437 // return address is the pc describes what bci to do re-execute at 3437 // return address is the pc describes what bci to do re-execute at
3440 (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); 3440 (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3441 3441
3442 __ movl(r14, Deoptimization::Unpack_reexecute); // callee-saved 3442 __ movl(r14, Deoptimization::Unpack_reexecute); // callee-saved
3443 __ jmp(cont); 3443 __ jmp(cont);
3444 3444
3445 #ifdef GRAAL 3445 #ifdef JVMCI
3446 int implicit_exception_uncommon_trap_offset = __ pc() - start; 3446 int implicit_exception_uncommon_trap_offset = __ pc() - start;
3447 3447
3448 __ pushptr(Address(r15_thread, in_bytes(JavaThread::graal_implicit_exception_pc_offset()))); 3448 __ pushptr(Address(r15_thread, in_bytes(JavaThread::jvmci_implicit_exception_pc_offset())));
3449 3449
3450 int uncommon_trap_offset = __ pc() - start; 3450 int uncommon_trap_offset = __ pc() - start;
3451 3451
3452 // Save everything in sight. 3452 // Save everything in sight.
3453 RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words); 3453 RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3464 3464
3465 __ reset_last_Java_frame(false, false); 3465 __ reset_last_Java_frame(false, false);
3466 3466
3467 Label after_fetch_unroll_info_call; 3467 Label after_fetch_unroll_info_call;
3468 __ jmp(after_fetch_unroll_info_call); 3468 __ jmp(after_fetch_unroll_info_call);
3469 #endif // GRAAL 3469 #endif // JVMCI
3470 3470
3471 int exception_offset = __ pc() - start; 3471 int exception_offset = __ pc() - start;
3472 3472
3473 // Prolog for exception case 3473 // Prolog for exception case
3474 3474
3551 // find any register it might need. 3551 // find any register it might need.
3552 oop_maps->add_gc_map(__ pc() - start, map); 3552 oop_maps->add_gc_map(__ pc() - start, map);
3553 3553
3554 __ reset_last_Java_frame(false, false); 3554 __ reset_last_Java_frame(false, false);
3555 3555
3556 #ifdef GRAAL 3556 #ifdef JVMCI
3557 __ bind(after_fetch_unroll_info_call); 3557 __ bind(after_fetch_unroll_info_call);
3558 #endif 3558 #endif
3559 3559
3560 // Load UnrollBlock* into rdi 3560 // Load UnrollBlock* into rdi
3561 __ mov(rdi, rax); 3561 __ mov(rdi, rax);
3729 // Make sure all code is generated 3729 // Make sure all code is generated
3730 masm->flush(); 3730 masm->flush();
3731 3731
3732 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words); 3732 _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
3733 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset); 3733 _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
3734 #ifdef GRAAL 3734 #ifdef JVMCI
3735 _deopt_blob->set_uncommon_trap_offset(uncommon_trap_offset); 3735 _deopt_blob->set_uncommon_trap_offset(uncommon_trap_offset);
3736 _deopt_blob->set_implicit_exception_uncommon_trap_offset(implicit_exception_uncommon_trap_offset); 3736 _deopt_blob->set_implicit_exception_uncommon_trap_offset(implicit_exception_uncommon_trap_offset);
3737 #endif 3737 #endif
3738 } 3738 }
3739 3739