Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(77)

Side by Side Diff: src/x64/code-stubs-x64.cc

Issue 1416053010: [x64] Replace addsd, subsd, mulsd, divsd with AVX versions under AVX. (Closed) Base URL: https://chromium.googlesource.com/v8/v8.git@master
Patch Set: Created 5 years, 2 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
« no previous file with comments | « src/crankshaft/x64/lithium-codegen-x64.cc ('k') | src/x64/codegen-x64.cc » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 // Copyright 2013 the V8 project authors. All rights reserved. 1 // Copyright 2013 the V8 project authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be 2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file. 3 // found in the LICENSE file.
4 4
5 #if V8_TARGET_ARCH_X64 5 #if V8_TARGET_ARCH_X64
6 6
7 #include "src/bootstrapper.h" 7 #include "src/bootstrapper.h"
8 #include "src/code-stubs.h" 8 #include "src/code-stubs.h"
9 #include "src/codegen.h" 9 #include "src/codegen.h"
10 #include "src/ic/handler-compiler.h" 10 #include "src/ic/handler-compiler.h"
(...skipping 338 matching lines...) Expand 10 before | Expand all | Expand 10 after
349 __ movq(scratch, V8_UINT64_C(0xFFF0000000000000)); 349 __ movq(scratch, V8_UINT64_C(0xFFF0000000000000));
350 __ Movq(double_scratch, scratch); 350 __ Movq(double_scratch, scratch);
351 __ Ucomisd(double_scratch, double_base); 351 __ Ucomisd(double_scratch, double_base);
352 // Comparing -Infinity with NaN results in "unordered", which sets the 352 // Comparing -Infinity with NaN results in "unordered", which sets the
353 // zero flag as if both were equal. However, it also sets the carry flag. 353 // zero flag as if both were equal. However, it also sets the carry flag.
354 __ j(not_equal, &continue_sqrt, Label::kNear); 354 __ j(not_equal, &continue_sqrt, Label::kNear);
355 __ j(carry, &continue_sqrt, Label::kNear); 355 __ j(carry, &continue_sqrt, Label::kNear);
356 356
357 // Set result to Infinity in the special case. 357 // Set result to Infinity in the special case.
358 __ Xorpd(double_result, double_result); 358 __ Xorpd(double_result, double_result);
359 __ subsd(double_result, double_scratch); 359 __ Subsd(double_result, double_scratch);
360 __ jmp(&done); 360 __ jmp(&done);
361 361
362 __ bind(&continue_sqrt); 362 __ bind(&continue_sqrt);
363 // sqrtsd returns -0 when input is -0. ECMA spec requires +0. 363 // sqrtsd returns -0 when input is -0. ECMA spec requires +0.
364 __ Xorpd(double_scratch, double_scratch); 364 __ Xorpd(double_scratch, double_scratch);
365 __ addsd(double_scratch, double_base); // Convert -0 to 0. 365 __ Addsd(double_scratch, double_base); // Convert -0 to 0.
366 __ Sqrtsd(double_result, double_scratch); 366 __ Sqrtsd(double_result, double_scratch);
367 __ jmp(&done); 367 __ jmp(&done);
368 368
369 // Test for -0.5. 369 // Test for -0.5.
370 __ bind(&not_plus_half); 370 __ bind(&not_plus_half);
371 // Load double_scratch with -0.5 by substracting 1. 371 // Load double_scratch with -0.5 by substracting 1.
372 __ subsd(double_scratch, double_result); 372 __ Subsd(double_scratch, double_result);
373 // Already ruled out NaNs for exponent. 373 // Already ruled out NaNs for exponent.
374 __ Ucomisd(double_scratch, double_exponent); 374 __ Ucomisd(double_scratch, double_exponent);
375 __ j(not_equal, &fast_power, Label::kNear); 375 __ j(not_equal, &fast_power, Label::kNear);
376 376
377 // Calculates reciprocal of square root of base. Check for the special 377 // Calculates reciprocal of square root of base. Check for the special
378 // case of Math.pow(-Infinity, -0.5) == 0 (ECMA spec, 15.8.2.13). 378 // case of Math.pow(-Infinity, -0.5) == 0 (ECMA spec, 15.8.2.13).
379 // According to IEEE-754, double-precision -Infinity has the highest 379 // According to IEEE-754, double-precision -Infinity has the highest
380 // 12 bits set and the lowest 52 bits cleared. 380 // 12 bits set and the lowest 52 bits cleared.
381 __ movq(scratch, V8_UINT64_C(0xFFF0000000000000)); 381 __ movq(scratch, V8_UINT64_C(0xFFF0000000000000));
382 __ Movq(double_scratch, scratch); 382 __ Movq(double_scratch, scratch);
383 __ Ucomisd(double_scratch, double_base); 383 __ Ucomisd(double_scratch, double_base);
384 // Comparing -Infinity with NaN results in "unordered", which sets the 384 // Comparing -Infinity with NaN results in "unordered", which sets the
385 // zero flag as if both were equal. However, it also sets the carry flag. 385 // zero flag as if both were equal. However, it also sets the carry flag.
386 __ j(not_equal, &continue_rsqrt, Label::kNear); 386 __ j(not_equal, &continue_rsqrt, Label::kNear);
387 __ j(carry, &continue_rsqrt, Label::kNear); 387 __ j(carry, &continue_rsqrt, Label::kNear);
388 388
389 // Set result to 0 in the special case. 389 // Set result to 0 in the special case.
390 __ Xorpd(double_result, double_result); 390 __ Xorpd(double_result, double_result);
391 __ jmp(&done); 391 __ jmp(&done);
392 392
393 __ bind(&continue_rsqrt); 393 __ bind(&continue_rsqrt);
394 // sqrtsd returns -0 when input is -0. ECMA spec requires +0. 394 // sqrtsd returns -0 when input is -0. ECMA spec requires +0.
395 __ Xorpd(double_exponent, double_exponent); 395 __ Xorpd(double_exponent, double_exponent);
396 __ addsd(double_exponent, double_base); // Convert -0 to +0. 396 __ Addsd(double_exponent, double_base); // Convert -0 to +0.
397 __ Sqrtsd(double_exponent, double_exponent); 397 __ Sqrtsd(double_exponent, double_exponent);
398 __ divsd(double_result, double_exponent); 398 __ Divsd(double_result, double_exponent);
399 __ jmp(&done); 399 __ jmp(&done);
400 } 400 }
401 401
402 // Using FPU instructions to calculate power. 402 // Using FPU instructions to calculate power.
403 Label fast_power_failed; 403 Label fast_power_failed;
404 __ bind(&fast_power); 404 __ bind(&fast_power);
405 __ fnclex(); // Clear flags to catch exceptions later. 405 __ fnclex(); // Clear flags to catch exceptions later.
406 // Transfer (B)ase and (E)xponent onto the FPU register stack. 406 // Transfer (B)ase and (E)xponent onto the FPU register stack.
407 __ subp(rsp, Immediate(kDoubleSize)); 407 __ subp(rsp, Immediate(kDoubleSize));
408 __ Movsd(Operand(rsp, 0), double_exponent); 408 __ Movsd(Operand(rsp, 0), double_exponent);
(...skipping 49 matching lines...) Expand 10 before | Expand all | Expand 10 after
458 __ j(zero, &while_false, Label::kNear); 458 __ j(zero, &while_false, Label::kNear);
459 __ shrl(scratch, Immediate(1)); 459 __ shrl(scratch, Immediate(1));
460 // Above condition means CF==0 && ZF==0. This means that the 460 // Above condition means CF==0 && ZF==0. This means that the
461 // bit that has been shifted out is 0 and the result is not 0. 461 // bit that has been shifted out is 0 and the result is not 0.
462 __ j(above, &while_true, Label::kNear); 462 __ j(above, &while_true, Label::kNear);
463 __ Movsd(double_result, double_scratch); 463 __ Movsd(double_result, double_scratch);
464 __ j(zero, &while_false, Label::kNear); 464 __ j(zero, &while_false, Label::kNear);
465 465
466 __ bind(&while_true); 466 __ bind(&while_true);
467 __ shrl(scratch, Immediate(1)); 467 __ shrl(scratch, Immediate(1));
468 __ mulsd(double_scratch, double_scratch); 468 __ Mulsd(double_scratch, double_scratch);
469 __ j(above, &while_true, Label::kNear); 469 __ j(above, &while_true, Label::kNear);
470 __ mulsd(double_result, double_scratch); 470 __ Mulsd(double_result, double_scratch);
471 __ j(not_zero, &while_true); 471 __ j(not_zero, &while_true);
472 472
473 __ bind(&while_false); 473 __ bind(&while_false);
474 // If the exponent is negative, return 1/result. 474 // If the exponent is negative, return 1/result.
475 __ testl(exponent, exponent); 475 __ testl(exponent, exponent);
476 __ j(greater, &done); 476 __ j(greater, &done);
477 __ divsd(double_scratch2, double_result); 477 __ Divsd(double_scratch2, double_result);
478 __ Movsd(double_result, double_scratch2); 478 __ Movsd(double_result, double_scratch2);
479 // Test whether result is zero. Bail out to check for subnormal result. 479 // Test whether result is zero. Bail out to check for subnormal result.
480 // Due to subnormals, x^-y == (1/x)^y does not hold in all cases. 480 // Due to subnormals, x^-y == (1/x)^y does not hold in all cases.
481 __ Xorpd(double_scratch2, double_scratch2); 481 __ Xorpd(double_scratch2, double_scratch2);
482 __ Ucomisd(double_scratch2, double_result); 482 __ Ucomisd(double_scratch2, double_result);
483 // double_exponent aliased as double_scratch2 has already been overwritten 483 // double_exponent aliased as double_scratch2 has already been overwritten
484 // and may not have contained the exponent value in the first place when the 484 // and may not have contained the exponent value in the first place when the
485 // input was a smi. We reset it with exponent value before bailing out. 485 // input was a smi. We reset it with exponent value before bailing out.
486 __ j(not_equal, &done); 486 __ j(not_equal, &done);
487 __ Cvtlsi2sd(double_exponent, exponent); 487 __ Cvtlsi2sd(double_exponent, exponent);
(...skipping 5107 matching lines...) Expand 10 before | Expand all | Expand 10 after
5595 kStackSpace, nullptr, return_value_operand, NULL); 5595 kStackSpace, nullptr, return_value_operand, NULL);
5596 } 5596 }
5597 5597
5598 5598
5599 #undef __ 5599 #undef __
5600 5600
5601 } // namespace internal 5601 } // namespace internal
5602 } // namespace v8 5602 } // namespace v8
5603 5603
5604 #endif // V8_TARGET_ARCH_X64 5604 #endif // V8_TARGET_ARCH_X64
OLDNEW
« no previous file with comments | « src/crankshaft/x64/lithium-codegen-x64.cc ('k') | src/x64/codegen-x64.cc » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698