175 }
176
177 int off = st0_off;
178 int delta = st1_off - off;
179
180 // Save the FPU registers in de-opt-able form
181 for (int n = 0; n < FloatRegisterImpl::number_of_registers; n++) {
182 __ fstp_d(Address(rsp, off*wordSize));
183 off += delta;
184 }
185
186 off = xmm0_off;
187 delta = xmm1_off - off;
188 if(UseSSE == 1) { // Save the XMM state
189 for (int n = 0; n < num_xmm_regs; n++) {
190 __ movflt(Address(rsp, off*wordSize), as_XMMRegister(n));
191 off += delta;
192 }
193 } else if(UseSSE >= 2) {
194 // Save whole 128bit (16 bytes) XMM regiters
195 if (VM_Version::supports_avx512novl()) {
196 for (int n = 0; n < num_xmm_regs; n++) {
197 __ vextractf32x4h(Address(rsp, off*wordSize), as_XMMRegister(n), 0);
198 off += delta;
199 }
200 } else {
201 for (int n = 0; n < num_xmm_regs; n++) {
202 __ movdqu(Address(rsp, off*wordSize), as_XMMRegister(n));
203 off += delta;
204 }
205 }
206 }
207
208 if (vect_words > 0) {
209 assert(vect_words*wordSize == 128, "");
210 __ subptr(rsp, 128); // Save upper half of YMM registes
211 off = 0;
212 for (int n = 0; n < num_xmm_regs; n++) {
213 __ vextractf128h(Address(rsp, off++*16), as_XMMRegister(n));
214 }
215 if (UseAVX > 2) {
216 __ subptr(rsp, 256); // Save upper half of ZMM registes
217 off = 0;
218 for (int n = 0; n < num_xmm_regs; n++) {
219 __ vextractf64x4h(Address(rsp, off++*32), as_XMMRegister(n));
220 }
221 }
222 }
223
224 // Set an oopmap for the call site. This oopmap will map all
225 // oop-registers and debug-info registers as callee-saved. This
226 // will allow deoptimization at this safepoint to find all possible
227 // debug-info recordings, as well as let GC find all oops.
228
229 OopMapSet *oop_maps = new OopMapSet();
230 OopMap* map = new OopMap( frame_words, 0 );
231
232 #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words)
233 #define NEXTREG(x) (x)->as_VMReg()->next()
234
235 map->set_callee_saved(STACK_OFFSET(rax_off), rax->as_VMReg());
236 map->set_callee_saved(STACK_OFFSET(rcx_off), rcx->as_VMReg());
237 map->set_callee_saved(STACK_OFFSET(rdx_off), rdx->as_VMReg());
238 map->set_callee_saved(STACK_OFFSET(rbx_off), rbx->as_VMReg());
239 // rbp, location is known implicitly, no oopMap
268 int additional_frame_bytes = 0;
269 #ifdef COMPILER2
270 if (restore_vectors) {
271 assert(UseAVX > 0, "512bit vectors are supported only with EVEX");
272 assert(MaxVectorSize == 64, "only 512bit vectors are supported now");
273 additional_frame_bytes = 128;
274 }
275 #else
276 assert(!restore_vectors, "vectors are generated only by C2");
277 #endif
278 int off = xmm0_off;
279 int delta = xmm1_off - off;
280
281 if (UseSSE == 1) {
282 assert(additional_frame_bytes == 0, "");
283 for (int n = 0; n < num_xmm_regs; n++) {
284 __ movflt(as_XMMRegister(n), Address(rsp, off*wordSize));
285 off += delta;
286 }
287 } else if (UseSSE >= 2) {
288 if (VM_Version::supports_avx512novl()) {
289 for (int n = 0; n < num_xmm_regs; n++) {
290 __ vinsertf32x4h(as_XMMRegister(n), Address(rsp, off*wordSize+additional_frame_bytes), 0);
291 off += delta;
292 }
293 } else {
294 for (int n = 0; n < num_xmm_regs; n++) {
295 __ movdqu(as_XMMRegister(n), Address(rsp, off*wordSize+additional_frame_bytes));
296 off += delta;
297 }
298 }
299 }
300 if (restore_vectors) {
301 if (UseAVX > 2) {
302 off = 0;
303 for (int n = 0; n < num_xmm_regs; n++) {
304 __ vinsertf64x4h(as_XMMRegister(n), Address(rsp, off++*32));
305 }
306 __ addptr(rsp, additional_frame_bytes*2); // Save upper half of ZMM registes
307 }
308 // Restore upper half of YMM registes.
309 assert(additional_frame_bytes == 128, "");
310 off = 0;
311 for (int n = 0; n < num_xmm_regs; n++) {
312 __ vinsertf128h(as_XMMRegister(n), Address(rsp, off++*16));
313 }
314 __ addptr(rsp, additional_frame_bytes); // Save upper half of YMM registes
315 }
316 __ pop_FPU_state();
317 __ addptr(rsp, FPU_regs_live*wordSize); // Pop FPU registers
318
319 __ popf();
320 __ popa();
321 // Get the rbp, described implicitly by the frame sender code (no oopMap)
322 __ pop(rbp);
323
324 }
325
326 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
327
328 // Just restore result register. Only used by deoptimization. By
329 // now any callee save register that needs to be restore to a c2
330 // caller of the deoptee has been extracted into the vframeArray
331 // and will be stuffed into the c2i adapter we create for later
332 // restoration so only result registers need to be restored here.
|
175 }
176
177 int off = st0_off;
178 int delta = st1_off - off;
179
180 // Save the FPU registers in de-opt-able form
181 for (int n = 0; n < FloatRegisterImpl::number_of_registers; n++) {
182 __ fstp_d(Address(rsp, off*wordSize));
183 off += delta;
184 }
185
186 off = xmm0_off;
187 delta = xmm1_off - off;
188 if(UseSSE == 1) { // Save the XMM state
189 for (int n = 0; n < num_xmm_regs; n++) {
190 __ movflt(Address(rsp, off*wordSize), as_XMMRegister(n));
191 off += delta;
192 }
193 } else if(UseSSE >= 2) {
194 // Save whole 128bit (16 bytes) XMM regiters
195 for (int n = 0; n < num_xmm_regs; n++) {
196 __ movdqu(Address(rsp, off*wordSize), as_XMMRegister(n));
197 off += delta;
198 }
199 }
200
201 if (vect_words > 0) {
202 assert(vect_words*wordSize == 128, "");
203 __ subptr(rsp, 128); // Save upper half of YMM registes
204 for (int n = 0; n < num_xmm_regs; n++) {
205 __ vextractf128h(Address(rsp, n*16), as_XMMRegister(n));
206 }
207 if (UseAVX > 2) {
208 __ subptr(rsp, 256); // Save upper half of ZMM registes
209 for (int n = 0; n < num_xmm_regs; n++) {
210 __ vextractf64x4h(Address(rsp, n*32), as_XMMRegister(n), 1);
211 }
212 }
213 }
214
215 // Set an oopmap for the call site. This oopmap will map all
216 // oop-registers and debug-info registers as callee-saved. This
217 // will allow deoptimization at this safepoint to find all possible
218 // debug-info recordings, as well as let GC find all oops.
219
220 OopMapSet *oop_maps = new OopMapSet();
221 OopMap* map = new OopMap( frame_words, 0 );
222
223 #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words)
224 #define NEXTREG(x) (x)->as_VMReg()->next()
225
226 map->set_callee_saved(STACK_OFFSET(rax_off), rax->as_VMReg());
227 map->set_callee_saved(STACK_OFFSET(rcx_off), rcx->as_VMReg());
228 map->set_callee_saved(STACK_OFFSET(rdx_off), rdx->as_VMReg());
229 map->set_callee_saved(STACK_OFFSET(rbx_off), rbx->as_VMReg());
230 // rbp, location is known implicitly, no oopMap
259 int additional_frame_bytes = 0;
260 #ifdef COMPILER2
261 if (restore_vectors) {
262 assert(UseAVX > 0, "512bit vectors are supported only with EVEX");
263 assert(MaxVectorSize == 64, "only 512bit vectors are supported now");
264 additional_frame_bytes = 128;
265 }
266 #else
267 assert(!restore_vectors, "vectors are generated only by C2");
268 #endif
269 int off = xmm0_off;
270 int delta = xmm1_off - off;
271
272 if (UseSSE == 1) {
273 assert(additional_frame_bytes == 0, "");
274 for (int n = 0; n < num_xmm_regs; n++) {
275 __ movflt(as_XMMRegister(n), Address(rsp, off*wordSize));
276 off += delta;
277 }
278 } else if (UseSSE >= 2) {
279 for (int n = 0; n < num_xmm_regs; n++) {
280 __ movdqu(as_XMMRegister(n), Address(rsp, off*wordSize+additional_frame_bytes));
281 off += delta;
282 }
283 }
284 if (restore_vectors) {
285 assert(additional_frame_bytes == 128, "");
286 if (UseAVX > 2) {
287 // Restore upper half of ZMM registers.
288 for (int n = 0; n < num_xmm_regs; n++) {
289 __ vinsertf64x4h(as_XMMRegister(n), Address(rsp, n*32), 1);
290 }
291 __ addptr(rsp, additional_frame_bytes*2); // Save upper half of ZMM registes
292 }
293 // Restore upper half of YMM registes.
294 for (int n = 0; n < num_xmm_regs; n++) {
295 __ vinsertf128h(as_XMMRegister(n), Address(rsp, n*16));
296 }
297 __ addptr(rsp, additional_frame_bytes); // Save upper half of YMM registes
298 }
299 __ pop_FPU_state();
300 __ addptr(rsp, FPU_regs_live*wordSize); // Pop FPU registers
301
302 __ popf();
303 __ popa();
304 // Get the rbp, described implicitly by the frame sender code (no oopMap)
305 __ pop(rbp);
306
307 }
308
309 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
310
311 // Just restore result register. Only used by deoptimization. By
312 // now any callee save register that needs to be restore to a c2
313 // caller of the deoptee has been extracted into the vframeArray
314 // and will be stuffed into the c2i adapter we create for later
315 // restoration so only result registers need to be restored here.
|