2014-07-09 11:08:26 +00:00
|
|
|
// Copyright 2013 the V8 project authors. All rights reserved.
|
|
|
|
// Redistribution and use in source and binary forms, with or without
|
|
|
|
// modification, are permitted provided that the following conditions are
|
|
|
|
// met:
|
|
|
|
//
|
|
|
|
// * Redistributions of source code must retain the above copyright
|
|
|
|
// notice, this list of conditions and the following disclaimer.
|
|
|
|
// * Redistributions in binary form must reproduce the above
|
|
|
|
// copyright notice, this list of conditions and the following
|
|
|
|
// disclaimer in the documentation and/or other materials provided
|
|
|
|
// with the distribution.
|
|
|
|
// * Neither the name of Google Inc. nor the names of its
|
|
|
|
// contributors may be used to endorse or promote products derived
|
|
|
|
// from this software without specific prior written permission.
|
|
|
|
//
|
|
|
|
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
|
|
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
|
|
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
|
|
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
|
|
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
|
|
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
|
|
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
|
|
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
|
|
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
|
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
|
|
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
|
|
|
|
#include <stdlib.h>
|
2015-07-13 18:26:17 +00:00
|
|
|
#include <iostream> // NOLINT(readability/streams)
|
2014-07-09 11:08:26 +00:00
|
|
|
|
2019-05-24 13:51:59 +00:00
|
|
|
#include "src/init/v8.h"
|
2014-07-09 11:08:26 +00:00
|
|
|
#include "test/cctest/cctest.h"
|
|
|
|
|
2015-07-13 18:26:17 +00:00
|
|
|
#include "src/base/utils/random-number-generator.h"
|
2019-05-21 09:30:15 +00:00
|
|
|
#include "src/codegen/macro-assembler.h"
|
2019-05-22 07:55:37 +00:00
|
|
|
#include "src/execution/simulator.h"
|
2018-12-17 17:01:48 +00:00
|
|
|
#include "src/objects/heap-number.h"
|
2019-05-23 13:27:57 +00:00
|
|
|
#include "src/utils/ostreams.h"
|
2019-05-23 08:51:46 +00:00
|
|
|
#include "src/objects/objects-inl.h"
|
2014-07-09 11:08:26 +00:00
|
|
|
|
2017-08-31 12:34:55 +00:00
|
|
|
namespace v8 {
|
|
|
|
namespace internal {
|
2014-07-09 11:08:26 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
// TODO(mips64): Refine these signatures per test case.
|
|
|
|
using FV = void*(int64_t x, int64_t y, int p2, int p3, int p4);
|
2018-12-25 00:19:47 +00:00
|
|
|
using F1 = void*(int x, int p1, int p2, int p3, int p4);
|
|
|
|
using F3 = void*(void* p, int p1, int p2, int p3, int p4);
|
|
|
|
using F4 = void*(void* p0, void* p1, int p2, int p3, int p4);
|
2014-07-09 11:08:26 +00:00
|
|
|
|
|
|
|
#define __ masm->
|
|
|
|
|
2016-06-27 14:35:01 +00:00
|
|
|
TEST(BYTESWAP) {
|
|
|
|
DCHECK(kArchVariant == kMips64r6 || kArchVariant == kMips64r2);
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
|
|
|
|
|
|
|
struct T {
|
2018-06-21 09:11:13 +00:00
|
|
|
uint64_t s8;
|
|
|
|
uint64_t s4;
|
|
|
|
uint64_t s2;
|
|
|
|
uint64_t u4;
|
|
|
|
uint64_t u2;
|
2016-06-27 14:35:01 +00:00
|
|
|
};
|
2018-06-21 09:11:13 +00:00
|
|
|
|
2016-06-27 14:35:01 +00:00
|
|
|
T t;
|
2018-06-21 09:11:13 +00:00
|
|
|
uint64_t test_values[] = {0x5612FFCD9D327ACC,
|
|
|
|
0x781A15C3,
|
|
|
|
0xFCDE,
|
|
|
|
0x9F,
|
|
|
|
0xC81A15C3,
|
|
|
|
0x8000000000000000,
|
|
|
|
0xFFFFFFFFFFFFFFFF,
|
|
|
|
0x0000000080000000,
|
|
|
|
0x0000000000008000};
|
2016-06-27 14:35:01 +00:00
|
|
|
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-06-27 14:35:01 +00:00
|
|
|
|
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Ld(a4, MemOperand(a0, offsetof(T, s8)));
|
2016-06-27 14:35:01 +00:00
|
|
|
__ nop();
|
2016-08-16 13:07:18 +00:00
|
|
|
__ ByteSwapSigned(a4, a4, 8);
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Sd(a4, MemOperand(a0, offsetof(T, s8)));
|
2016-06-27 14:35:01 +00:00
|
|
|
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Ld(a4, MemOperand(a0, offsetof(T, s4)));
|
2016-06-27 14:35:01 +00:00
|
|
|
__ nop();
|
2016-08-16 13:07:18 +00:00
|
|
|
__ ByteSwapSigned(a4, a4, 4);
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Sd(a4, MemOperand(a0, offsetof(T, s4)));
|
2016-06-27 14:35:01 +00:00
|
|
|
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Ld(a4, MemOperand(a0, offsetof(T, s2)));
|
2016-06-27 14:35:01 +00:00
|
|
|
__ nop();
|
2016-08-16 13:07:18 +00:00
|
|
|
__ ByteSwapSigned(a4, a4, 2);
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Sd(a4, MemOperand(a0, offsetof(T, s2)));
|
2016-06-27 14:35:01 +00:00
|
|
|
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Ld(a4, MemOperand(a0, offsetof(T, u4)));
|
2016-06-27 14:35:01 +00:00
|
|
|
__ nop();
|
2018-06-21 09:11:13 +00:00
|
|
|
__ ByteSwapUnsigned(a4, a4, 4);
|
|
|
|
__ Sd(a4, MemOperand(a0, offsetof(T, u4)));
|
2016-06-27 14:35:01 +00:00
|
|
|
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Ld(a4, MemOperand(a0, offsetof(T, u2)));
|
2016-06-27 14:35:01 +00:00
|
|
|
__ nop();
|
2016-08-16 13:07:18 +00:00
|
|
|
__ ByteSwapUnsigned(a4, a4, 2);
|
2018-06-21 09:11:13 +00:00
|
|
|
__ Sd(a4, MemOperand(a0, offsetof(T, u2)));
|
2016-06-27 14:35:01 +00:00
|
|
|
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F3>::FromCode(*code);
|
2018-06-21 09:11:13 +00:00
|
|
|
|
|
|
|
for (size_t i = 0; i < arraysize(test_values); i++) {
|
|
|
|
int32_t in_s4 = static_cast<int32_t>(test_values[i]);
|
|
|
|
int16_t in_s2 = static_cast<int16_t>(test_values[i]);
|
|
|
|
uint32_t in_u4 = static_cast<uint32_t>(test_values[i]);
|
|
|
|
uint16_t in_u2 = static_cast<uint16_t>(test_values[i]);
|
|
|
|
|
|
|
|
t.s8 = test_values[i];
|
|
|
|
t.s4 = static_cast<uint64_t>(in_s4);
|
|
|
|
t.s2 = static_cast<uint64_t>(in_s2);
|
|
|
|
t.u4 = static_cast<uint64_t>(in_u4);
|
|
|
|
t.u2 = static_cast<uint64_t>(in_u2);
|
|
|
|
|
|
|
|
f.Call(&t, 0, 0, 0, 0);
|
|
|
|
|
|
|
|
CHECK_EQ(ByteReverse<uint64_t>(test_values[i]), t.s8);
|
|
|
|
CHECK_EQ(ByteReverse<int32_t>(in_s4), static_cast<int32_t>(t.s4));
|
|
|
|
CHECK_EQ(ByteReverse<int16_t>(in_s2), static_cast<int16_t>(t.s2));
|
|
|
|
CHECK_EQ(ByteReverse<uint32_t>(in_u4), static_cast<uint32_t>(t.u4));
|
|
|
|
CHECK_EQ(ByteReverse<uint16_t>(in_u2), static_cast<uint16_t>(t.u2));
|
|
|
|
}
|
2016-06-27 14:35:01 +00:00
|
|
|
}
|
2014-07-09 11:08:26 +00:00
|
|
|
|
|
|
|
TEST(LoadConstants) {
|
|
|
|
CcTest::InitializeVM();
|
2015-11-23 13:48:05 +00:00
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
2014-07-09 11:08:26 +00:00
|
|
|
HandleScope handles(isolate);
|
|
|
|
|
|
|
|
int64_t refConstants[64];
|
|
|
|
int64_t result[64];
|
|
|
|
|
|
|
|
int64_t mask = 1;
|
|
|
|
for (int i = 0; i < 64; i++) {
|
|
|
|
refConstants[i] = ~(mask << i);
|
|
|
|
}
|
|
|
|
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2014-07-09 11:08:26 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
__ mov(a4, a0);
|
|
|
|
for (int i = 0; i < 64; i++) {
|
|
|
|
// Load constant.
|
|
|
|
__ li(a5, Operand(refConstants[i]));
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Sd(a5, MemOperand(a4));
|
2014-07-09 11:08:26 +00:00
|
|
|
__ Daddu(a4, a4, Operand(kPointerSize));
|
|
|
|
}
|
|
|
|
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2014-07-09 11:08:26 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<FV>::FromCode(*code);
|
|
|
|
(void)f.Call(reinterpret_cast<int64_t>(result), 0, 0, 0, 0);
|
2014-07-09 11:08:26 +00:00
|
|
|
// Check results.
|
|
|
|
for (int i = 0; i < 64; i++) {
|
|
|
|
CHECK(refConstants[i] == result[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
TEST(LoadAddress) {
|
|
|
|
CcTest::InitializeVM();
|
2015-11-23 13:48:05 +00:00
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
2014-07-09 11:08:26 +00:00
|
|
|
HandleScope handles(isolate);
|
|
|
|
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2014-07-09 11:08:26 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
Label to_jump, skip;
|
|
|
|
__ mov(a4, a0);
|
|
|
|
|
|
|
|
__ Branch(&skip);
|
|
|
|
__ bind(&to_jump);
|
|
|
|
__ nop();
|
|
|
|
__ nop();
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
__ bind(&skip);
|
|
|
|
__ li(a4, Operand(masm->jump_address(&to_jump)), ADDRESS_LOAD);
|
|
|
|
int check_size = masm->InstructionsGeneratedSince(&skip);
|
2017-02-07 14:20:02 +00:00
|
|
|
CHECK_EQ(4, check_size);
|
2014-07-09 11:08:26 +00:00
|
|
|
__ jr(a4);
|
|
|
|
__ nop();
|
2019-07-12 07:13:01 +00:00
|
|
|
__ stop();
|
|
|
|
__ stop();
|
|
|
|
__ stop();
|
|
|
|
__ stop();
|
|
|
|
__ stop();
|
2014-07-09 11:08:26 +00:00
|
|
|
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2014-07-09 11:08:26 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<FV>::FromCode(*code);
|
|
|
|
(void)f.Call(0, 0, 0, 0, 0);
|
2014-07-09 11:08:26 +00:00
|
|
|
// Check results.
|
|
|
|
}
|
|
|
|
|
2015-07-13 18:26:17 +00:00
|
|
|
|
|
|
|
TEST(jump_tables4) {
|
|
|
|
// Similar to test-assembler-mips jump_tables1, with extra test for branch
|
|
|
|
// trampoline required before emission of the dd table (where trampolines are
|
|
|
|
// blocked), and proper transition to long-branch mode.
|
|
|
|
// Regression test for v8:4294.
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2015-07-13 18:26:17 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
const int kNumCases = 512;
|
|
|
|
int values[kNumCases];
|
|
|
|
isolate->random_number_generator()->NextBytes(values, sizeof(values));
|
|
|
|
Label labels[kNumCases];
|
2016-02-03 13:30:57 +00:00
|
|
|
Label near_start, end, done;
|
2015-07-13 18:26:17 +00:00
|
|
|
|
2016-02-03 13:30:57 +00:00
|
|
|
__ Push(ra);
|
2015-07-13 18:26:17 +00:00
|
|
|
__ mov(v0, zero_reg);
|
|
|
|
|
|
|
|
__ Branch(&end);
|
|
|
|
__ bind(&near_start);
|
|
|
|
|
|
|
|
// Generate slightly less than 32K instructions, which will soon require
|
|
|
|
// trampoline for branch distance fixup.
|
|
|
|
for (int i = 0; i < 32768 - 256; ++i) {
|
|
|
|
__ addiu(v0, v0, 1);
|
|
|
|
}
|
|
|
|
|
2016-02-03 13:30:57 +00:00
|
|
|
__ GenerateSwitchTable(a0, kNumCases,
|
|
|
|
[&labels](size_t i) { return labels + i; });
|
2015-07-13 18:26:17 +00:00
|
|
|
|
|
|
|
for (int i = 0; i < kNumCases; ++i) {
|
|
|
|
__ bind(&labels[i]);
|
2016-02-03 13:30:57 +00:00
|
|
|
__ li(v0, values[i]);
|
2015-07-13 18:26:17 +00:00
|
|
|
__ Branch(&done);
|
|
|
|
}
|
|
|
|
|
|
|
|
__ bind(&done);
|
2016-02-03 13:30:57 +00:00
|
|
|
__ Pop(ra);
|
2015-07-13 18:26:17 +00:00
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
__ bind(&end);
|
|
|
|
__ Branch(&near_start);
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2015-07-13 18:26:17 +00:00
|
|
|
#ifdef OBJECT_PRINT
|
2018-07-12 10:06:42 +00:00
|
|
|
code->Print(std::cout);
|
2015-07-13 18:26:17 +00:00
|
|
|
#endif
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F1>::FromCode(*code);
|
2015-07-13 18:26:17 +00:00
|
|
|
for (int i = 0; i < kNumCases; ++i) {
|
2018-01-09 09:50:34 +00:00
|
|
|
int64_t res = reinterpret_cast<int64_t>(f.Call(i, 0, 0, 0, 0));
|
2015-07-13 18:26:17 +00:00
|
|
|
::printf("f(%d) = %" PRId64 "\n", i, res);
|
|
|
|
CHECK_EQ(values[i], res);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-01-05 19:45:59 +00:00
|
|
|
|
2016-01-12 19:48:51 +00:00
|
|
|
TEST(jump_tables5) {
|
|
|
|
if (kArchVariant != kMips64r6) return;
|
|
|
|
|
|
|
|
// Similar to test-assembler-mips jump_tables1, with extra test for emitting a
|
|
|
|
// compact branch instruction before emission of the dd table.
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-01-12 19:48:51 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
const int kNumCases = 512;
|
|
|
|
int values[kNumCases];
|
|
|
|
isolate->random_number_generator()->NextBytes(values, sizeof(values));
|
|
|
|
Label labels[kNumCases];
|
|
|
|
Label done;
|
|
|
|
|
2016-02-03 13:30:57 +00:00
|
|
|
__ Push(ra);
|
|
|
|
|
|
|
|
// Opposite of Align(8) as we have unaligned number of instructions in the
|
|
|
|
// following block before the first dd().
|
|
|
|
if ((masm->pc_offset() & 7) == 0) {
|
|
|
|
__ nop();
|
|
|
|
}
|
2016-01-12 19:48:51 +00:00
|
|
|
|
|
|
|
{
|
2016-02-03 13:30:57 +00:00
|
|
|
__ BlockTrampolinePoolFor(kNumCases * 2 + 6 + 1);
|
|
|
|
|
|
|
|
__ addiupc(at, 6 + 1);
|
2016-03-10 13:39:56 +00:00
|
|
|
__ Dlsa(at, at, a0, 3);
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Ld(at, MemOperand(at));
|
2016-01-12 19:48:51 +00:00
|
|
|
__ jalr(at);
|
|
|
|
__ nop(); // Branch delay slot nop.
|
|
|
|
__ bc(&done);
|
2016-01-13 15:15:26 +00:00
|
|
|
// A nop instruction must be generated by the forbidden slot guard
|
|
|
|
// (Assembler::dd(Label*)) so the first label goes to an 8 bytes aligned
|
|
|
|
// location.
|
2016-01-12 19:48:51 +00:00
|
|
|
for (int i = 0; i < kNumCases; ++i) {
|
|
|
|
__ dd(&labels[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int i = 0; i < kNumCases; ++i) {
|
|
|
|
__ bind(&labels[i]);
|
2016-02-03 13:30:57 +00:00
|
|
|
__ li(v0, values[i]);
|
2016-01-12 19:48:51 +00:00
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
}
|
|
|
|
|
|
|
|
__ bind(&done);
|
2016-02-03 13:30:57 +00:00
|
|
|
__ Pop(ra);
|
2016-01-12 19:48:51 +00:00
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2016-01-12 19:48:51 +00:00
|
|
|
#ifdef OBJECT_PRINT
|
2018-07-12 10:06:42 +00:00
|
|
|
code->Print(std::cout);
|
2016-01-12 19:48:51 +00:00
|
|
|
#endif
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F1>::FromCode(*code);
|
2016-01-12 19:48:51 +00:00
|
|
|
for (int i = 0; i < kNumCases; ++i) {
|
2018-01-09 09:50:34 +00:00
|
|
|
int64_t res = reinterpret_cast<int64_t>(f.Call(i, 0, 0, 0, 0));
|
2016-01-12 19:48:51 +00:00
|
|
|
::printf("f(%d) = %" PRId64 "\n", i, res);
|
|
|
|
CHECK_EQ(values[i], res);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-12-01 13:03:00 +00:00
|
|
|
TEST(jump_tables6) {
|
|
|
|
// Similar to test-assembler-mips jump_tables1, with extra test for branch
|
|
|
|
// trampoline required after emission of the dd table (where trampolines are
|
|
|
|
// blocked). This test checks if number of really generated instructions is
|
|
|
|
// greater than number of counted instructions from code, as we are expecting
|
|
|
|
// generation of trampoline in this case (when number of kFillInstr
|
|
|
|
// instructions is close to 32K)
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-12-01 13:03:00 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
2016-12-07 09:13:22 +00:00
|
|
|
const int kSwitchTableCases = 40;
|
|
|
|
|
|
|
|
const int kMaxBranchOffset = Assembler::kMaxBranchOffset;
|
|
|
|
const int kTrampolineSlotsSize = Assembler::kTrampolineSlotsSize;
|
|
|
|
const int kSwitchTablePrologueSize = MacroAssembler::kSwitchTablePrologueSize;
|
|
|
|
|
2016-12-01 13:03:00 +00:00
|
|
|
const int kMaxOffsetForTrampolineStart =
|
|
|
|
kMaxBranchOffset - 16 * kTrampolineSlotsSize;
|
2016-12-07 09:13:22 +00:00
|
|
|
const int kFillInstr = (kMaxOffsetForTrampolineStart / kInstrSize) -
|
|
|
|
(kSwitchTablePrologueSize + 2 * kSwitchTableCases) -
|
|
|
|
20;
|
2016-12-01 13:03:00 +00:00
|
|
|
|
2016-12-07 09:13:22 +00:00
|
|
|
int values[kSwitchTableCases];
|
2016-12-01 13:03:00 +00:00
|
|
|
isolate->random_number_generator()->NextBytes(values, sizeof(values));
|
2016-12-07 09:13:22 +00:00
|
|
|
Label labels[kSwitchTableCases];
|
2016-12-01 13:03:00 +00:00
|
|
|
Label near_start, end, done;
|
|
|
|
|
|
|
|
__ Push(ra);
|
|
|
|
__ mov(v0, zero_reg);
|
|
|
|
|
|
|
|
int offs1 = masm->pc_offset();
|
|
|
|
int gen_insn = 0;
|
|
|
|
|
|
|
|
__ Branch(&end);
|
2016-12-07 09:13:22 +00:00
|
|
|
gen_insn += Assembler::IsCompactBranchSupported() ? 1 : 2;
|
2016-12-01 13:03:00 +00:00
|
|
|
__ bind(&near_start);
|
|
|
|
|
|
|
|
// Generate slightly less than 32K instructions, which will soon require
|
|
|
|
// trampoline for branch distance fixup.
|
|
|
|
for (int i = 0; i < kFillInstr; ++i) {
|
|
|
|
__ addiu(v0, v0, 1);
|
|
|
|
}
|
|
|
|
gen_insn += kFillInstr;
|
|
|
|
|
2016-12-07 09:13:22 +00:00
|
|
|
__ GenerateSwitchTable(a0, kSwitchTableCases,
|
2016-12-01 13:03:00 +00:00
|
|
|
[&labels](size_t i) { return labels + i; });
|
2016-12-07 09:13:22 +00:00
|
|
|
gen_insn += (kSwitchTablePrologueSize + 2 * kSwitchTableCases);
|
2016-12-01 13:03:00 +00:00
|
|
|
|
2016-12-07 09:13:22 +00:00
|
|
|
for (int i = 0; i < kSwitchTableCases; ++i) {
|
2016-12-01 13:03:00 +00:00
|
|
|
__ bind(&labels[i]);
|
|
|
|
__ li(v0, values[i]);
|
|
|
|
__ Branch(&done);
|
|
|
|
}
|
2016-12-07 09:13:22 +00:00
|
|
|
gen_insn +=
|
|
|
|
((Assembler::IsCompactBranchSupported() ? 3 : 4) * kSwitchTableCases);
|
2016-12-01 13:03:00 +00:00
|
|
|
|
|
|
|
// If offset from here to first branch instr is greater than max allowed
|
|
|
|
// offset for trampoline ...
|
|
|
|
CHECK_LT(kMaxOffsetForTrampolineStart, masm->pc_offset() - offs1);
|
|
|
|
// ... number of generated instructions must be greater then "gen_insn",
|
|
|
|
// as we are expecting trampoline generation
|
2016-12-07 09:13:22 +00:00
|
|
|
CHECK_LT(gen_insn, (masm->pc_offset() - offs1) / kInstrSize);
|
2016-12-01 13:03:00 +00:00
|
|
|
|
|
|
|
__ bind(&done);
|
|
|
|
__ Pop(ra);
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
__ bind(&end);
|
|
|
|
__ Branch(&near_start);
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2016-12-01 13:03:00 +00:00
|
|
|
#ifdef OBJECT_PRINT
|
2018-07-12 10:06:42 +00:00
|
|
|
code->Print(std::cout);
|
2016-12-01 13:03:00 +00:00
|
|
|
#endif
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F1>::FromCode(*code);
|
2016-12-07 09:13:22 +00:00
|
|
|
for (int i = 0; i < kSwitchTableCases; ++i) {
|
2018-01-09 09:50:34 +00:00
|
|
|
int64_t res = reinterpret_cast<int64_t>(f.Call(i, 0, 0, 0, 0));
|
2016-12-01 13:03:00 +00:00
|
|
|
::printf("f(%d) = %" PRId64 "\n", i, res);
|
|
|
|
CHECK_EQ(values[i], res);
|
|
|
|
}
|
|
|
|
}
|
2016-01-12 19:48:51 +00:00
|
|
|
|
2016-01-05 19:45:59 +00:00
|
|
|
static uint64_t run_lsa(uint32_t rt, uint32_t rs, int8_t sa) {
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-01-05 19:45:59 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
__ Lsa(v0, a0, a1, sa);
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
assembler.GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2016-01-05 19:45:59 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F1>::FromCode(*code);
|
2016-01-05 19:45:59 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
uint64_t res = reinterpret_cast<uint64_t>(f.Call(rt, rs, 0, 0, 0));
|
2016-01-05 19:45:59 +00:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
TEST(Lsa) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
struct TestCaseLsa {
|
|
|
|
int32_t rt;
|
|
|
|
int32_t rs;
|
|
|
|
uint8_t sa;
|
|
|
|
uint64_t expected_res;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct TestCaseLsa tc[] = {// rt, rs, sa, expected_res
|
|
|
|
{0x4, 0x1, 1, 0x6},
|
|
|
|
{0x4, 0x1, 2, 0x8},
|
2017-12-02 00:30:37 +00:00
|
|
|
{0x4, 0x1, 3, 0xC},
|
2016-01-05 19:45:59 +00:00
|
|
|
{0x4, 0x1, 4, 0x14},
|
|
|
|
{0x4, 0x1, 5, 0x24},
|
|
|
|
{0x0, 0x1, 1, 0x2},
|
|
|
|
{0x0, 0x1, 2, 0x4},
|
|
|
|
{0x0, 0x1, 3, 0x8},
|
|
|
|
{0x0, 0x1, 4, 0x10},
|
|
|
|
{0x0, 0x1, 5, 0x20},
|
|
|
|
{0x4, 0x0, 1, 0x4},
|
|
|
|
{0x4, 0x0, 2, 0x4},
|
|
|
|
{0x4, 0x0, 3, 0x4},
|
|
|
|
{0x4, 0x0, 4, 0x4},
|
|
|
|
{0x4, 0x0, 5, 0x4},
|
|
|
|
|
|
|
|
// Shift overflow.
|
|
|
|
{0x4, INT32_MAX, 1, 0x2},
|
|
|
|
{0x4, INT32_MAX >> 1, 2, 0x0},
|
2017-12-02 00:30:37 +00:00
|
|
|
{0x4, INT32_MAX >> 2, 3, 0xFFFFFFFFFFFFFFFC},
|
|
|
|
{0x4, INT32_MAX >> 3, 4, 0xFFFFFFFFFFFFFFF4},
|
|
|
|
{0x4, INT32_MAX >> 4, 5, 0xFFFFFFFFFFFFFFE4},
|
2016-01-05 19:45:59 +00:00
|
|
|
|
|
|
|
// Signed addition overflow.
|
2017-12-02 00:30:37 +00:00
|
|
|
{INT32_MAX - 1, 0x1, 1, 0xFFFFFFFF80000000},
|
|
|
|
{INT32_MAX - 3, 0x1, 2, 0xFFFFFFFF80000000},
|
|
|
|
{INT32_MAX - 7, 0x1, 3, 0xFFFFFFFF80000000},
|
|
|
|
{INT32_MAX - 15, 0x1, 4, 0xFFFFFFFF80000000},
|
|
|
|
{INT32_MAX - 31, 0x1, 5, 0xFFFFFFFF80000000},
|
2016-01-05 19:45:59 +00:00
|
|
|
|
|
|
|
// Addition overflow.
|
|
|
|
{-2, 0x1, 1, 0x0},
|
|
|
|
{-4, 0x1, 2, 0x0},
|
|
|
|
{-8, 0x1, 3, 0x0},
|
|
|
|
{-16, 0x1, 4, 0x0},
|
|
|
|
{-32, 0x1, 5, 0x0}};
|
|
|
|
|
|
|
|
size_t nr_test_cases = sizeof(tc) / sizeof(TestCaseLsa);
|
|
|
|
for (size_t i = 0; i < nr_test_cases; ++i) {
|
|
|
|
uint64_t res = run_lsa(tc[i].rt, tc[i].rs, tc[i].sa);
|
|
|
|
PrintF("0x%" PRIx64 " =? 0x%" PRIx64 " == Lsa(v0, %x, %x, %hhu)\n",
|
|
|
|
tc[i].expected_res, res, tc[i].rt, tc[i].rs, tc[i].sa);
|
|
|
|
CHECK_EQ(tc[i].expected_res, res);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static uint64_t run_dlsa(uint64_t rt, uint64_t rs, int8_t sa) {
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-01-05 19:45:59 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
__ Dlsa(v0, a0, a1, sa);
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
assembler.GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2016-01-05 19:45:59 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<FV>::FromCode(*code);
|
2016-01-05 19:45:59 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
uint64_t res = reinterpret_cast<uint64_t>(f.Call(rt, rs, 0, 0, 0));
|
2016-01-05 19:45:59 +00:00
|
|
|
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
TEST(Dlsa) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
struct TestCaseLsa {
|
|
|
|
int64_t rt;
|
|
|
|
int64_t rs;
|
|
|
|
uint8_t sa;
|
|
|
|
uint64_t expected_res;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct TestCaseLsa tc[] = {// rt, rs, sa, expected_res
|
|
|
|
{0x4, 0x1, 1, 0x6},
|
|
|
|
{0x4, 0x1, 2, 0x8},
|
2017-12-02 00:30:37 +00:00
|
|
|
{0x4, 0x1, 3, 0xC},
|
2016-01-05 19:45:59 +00:00
|
|
|
{0x4, 0x1, 4, 0x14},
|
|
|
|
{0x4, 0x1, 5, 0x24},
|
|
|
|
{0x0, 0x1, 1, 0x2},
|
|
|
|
{0x0, 0x1, 2, 0x4},
|
|
|
|
{0x0, 0x1, 3, 0x8},
|
|
|
|
{0x0, 0x1, 4, 0x10},
|
|
|
|
{0x0, 0x1, 5, 0x20},
|
|
|
|
{0x4, 0x0, 1, 0x4},
|
|
|
|
{0x4, 0x0, 2, 0x4},
|
|
|
|
{0x4, 0x0, 3, 0x4},
|
|
|
|
{0x4, 0x0, 4, 0x4},
|
|
|
|
{0x4, 0x0, 5, 0x4},
|
|
|
|
|
|
|
|
// Shift overflow.
|
|
|
|
{0x4, INT64_MAX, 1, 0x2},
|
|
|
|
{0x4, INT64_MAX >> 1, 2, 0x0},
|
2017-12-02 00:30:37 +00:00
|
|
|
{0x4, INT64_MAX >> 2, 3, 0xFFFFFFFFFFFFFFFC},
|
|
|
|
{0x4, INT64_MAX >> 3, 4, 0xFFFFFFFFFFFFFFF4},
|
|
|
|
{0x4, INT64_MAX >> 4, 5, 0xFFFFFFFFFFFFFFE4},
|
2016-01-05 19:45:59 +00:00
|
|
|
|
|
|
|
// Signed addition overflow.
|
|
|
|
{INT64_MAX - 1, 0x1, 1, 0x8000000000000000},
|
|
|
|
{INT64_MAX - 3, 0x1, 2, 0x8000000000000000},
|
|
|
|
{INT64_MAX - 7, 0x1, 3, 0x8000000000000000},
|
|
|
|
{INT64_MAX - 15, 0x1, 4, 0x8000000000000000},
|
|
|
|
{INT64_MAX - 31, 0x1, 5, 0x8000000000000000},
|
|
|
|
|
|
|
|
// Addition overflow.
|
|
|
|
{-2, 0x1, 1, 0x0},
|
|
|
|
{-4, 0x1, 2, 0x0},
|
|
|
|
{-8, 0x1, 3, 0x0},
|
|
|
|
{-16, 0x1, 4, 0x0},
|
|
|
|
{-32, 0x1, 5, 0x0}};
|
|
|
|
|
|
|
|
size_t nr_test_cases = sizeof(tc) / sizeof(TestCaseLsa);
|
|
|
|
for (size_t i = 0; i < nr_test_cases; ++i) {
|
|
|
|
uint64_t res = run_dlsa(tc[i].rt, tc[i].rs, tc[i].sa);
|
|
|
|
PrintF("0x%" PRIx64 " =? 0x%" PRIx64 " == Dlsa(v0, %" PRIx64 ", %" PRIx64
|
|
|
|
", %hhu)\n",
|
|
|
|
tc[i].expected_res, res, tc[i].rt, tc[i].rs, tc[i].sa);
|
|
|
|
CHECK_EQ(tc[i].expected_res, res);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-04-07 07:15:25 +00:00
|
|
|
static const std::vector<uint32_t> cvt_trunc_uint32_test_values() {
|
2017-12-02 00:30:37 +00:00
|
|
|
static const uint32_t kValues[] = {0x00000000, 0x00000001, 0x00FFFF00,
|
|
|
|
0x7FFFFFFF, 0x80000000, 0x80000001,
|
|
|
|
0x80FFFF00, 0x8FFFFFFF, 0xFFFFFFFF};
|
2016-03-09 12:43:54 +00:00
|
|
|
return std::vector<uint32_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
2016-04-07 07:15:25 +00:00
|
|
|
static const std::vector<int32_t> cvt_trunc_int32_test_values() {
|
2016-03-09 12:43:54 +00:00
|
|
|
static const int32_t kValues[] = {
|
|
|
|
static_cast<int32_t>(0x00000000), static_cast<int32_t>(0x00000001),
|
2017-12-02 00:30:37 +00:00
|
|
|
static_cast<int32_t>(0x00FFFF00), static_cast<int32_t>(0x7FFFFFFF),
|
2016-03-09 12:43:54 +00:00
|
|
|
static_cast<int32_t>(0x80000000), static_cast<int32_t>(0x80000001),
|
2017-12-02 00:30:37 +00:00
|
|
|
static_cast<int32_t>(0x80FFFF00), static_cast<int32_t>(0x8FFFFFFF),
|
|
|
|
static_cast<int32_t>(0xFFFFFFFF)};
|
2016-03-09 12:43:54 +00:00
|
|
|
return std::vector<int32_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
2016-04-07 07:15:25 +00:00
|
|
|
static const std::vector<uint64_t> cvt_trunc_uint64_test_values() {
|
2016-03-09 12:43:54 +00:00
|
|
|
static const uint64_t kValues[] = {
|
2017-12-02 00:30:37 +00:00
|
|
|
0x0000000000000000, 0x0000000000000001, 0x0000FFFFFFFF0000,
|
|
|
|
0x7FFFFFFFFFFFFFFF, 0x8000000000000000, 0x8000000000000001,
|
|
|
|
0x8000FFFFFFFF0000, 0x8FFFFFFFFFFFFFFF, 0xFFFFFFFFFFFFFFFF};
|
2016-03-09 12:43:54 +00:00
|
|
|
return std::vector<uint64_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
2016-04-07 07:15:25 +00:00
|
|
|
static const std::vector<int64_t> cvt_trunc_int64_test_values() {
|
2016-03-09 12:43:54 +00:00
|
|
|
static const int64_t kValues[] = {static_cast<int64_t>(0x0000000000000000),
|
|
|
|
static_cast<int64_t>(0x0000000000000001),
|
2017-12-02 00:30:37 +00:00
|
|
|
static_cast<int64_t>(0x0000FFFFFFFF0000),
|
|
|
|
static_cast<int64_t>(0x7FFFFFFFFFFFFFFF),
|
2016-03-09 12:43:54 +00:00
|
|
|
static_cast<int64_t>(0x8000000000000000),
|
|
|
|
static_cast<int64_t>(0x8000000000000001),
|
2017-12-02 00:30:37 +00:00
|
|
|
static_cast<int64_t>(0x8000FFFFFFFF0000),
|
|
|
|
static_cast<int64_t>(0x8FFFFFFFFFFFFFFF),
|
|
|
|
static_cast<int64_t>(0xFFFFFFFFFFFFFFFF)};
|
2016-03-09 12:43:54 +00:00
|
|
|
return std::vector<int64_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Helper macros that can be used in FOR_INT32_INPUTS(i) { ... *i ... }
|
2016-04-07 07:15:25 +00:00
|
|
|
#define FOR_INPUTS(ctype, itype, var, test_vector) \
|
|
|
|
std::vector<ctype> var##_vec = test_vector(); \
|
2016-03-09 12:43:54 +00:00
|
|
|
for (std::vector<ctype>::iterator var = var##_vec.begin(); \
|
|
|
|
var != var##_vec.end(); ++var)
|
|
|
|
|
2016-04-28 15:09:27 +00:00
|
|
|
#define FOR_INPUTS2(ctype, itype, var, var2, test_vector) \
|
|
|
|
std::vector<ctype> var##_vec = test_vector(); \
|
|
|
|
std::vector<ctype>::iterator var; \
|
|
|
|
std::vector<ctype>::reverse_iterator var2; \
|
|
|
|
for (var = var##_vec.begin(), var2 = var##_vec.rbegin(); \
|
|
|
|
var != var##_vec.end(); ++var, ++var2)
|
|
|
|
|
2016-04-07 07:15:25 +00:00
|
|
|
#define FOR_ENUM_INPUTS(var, type, test_vector) \
|
|
|
|
FOR_INPUTS(enum type, type, var, test_vector)
|
|
|
|
#define FOR_STRUCT_INPUTS(var, type, test_vector) \
|
|
|
|
FOR_INPUTS(struct type, type, var, test_vector)
|
|
|
|
#define FOR_INT32_INPUTS(var, test_vector) \
|
|
|
|
FOR_INPUTS(int32_t, int32, var, test_vector)
|
2016-04-28 15:09:27 +00:00
|
|
|
#define FOR_INT32_INPUTS2(var, var2, test_vector) \
|
|
|
|
FOR_INPUTS2(int32_t, int32, var, var2, test_vector)
|
2016-04-07 07:15:25 +00:00
|
|
|
#define FOR_INT64_INPUTS(var, test_vector) \
|
|
|
|
FOR_INPUTS(int64_t, int64, var, test_vector)
|
|
|
|
#define FOR_UINT32_INPUTS(var, test_vector) \
|
|
|
|
FOR_INPUTS(uint32_t, uint32, var, test_vector)
|
|
|
|
#define FOR_UINT64_INPUTS(var, test_vector) \
|
|
|
|
FOR_INPUTS(uint64_t, uint64, var, test_vector)
|
2016-03-09 12:43:54 +00:00
|
|
|
|
|
|
|
template <typename RET_TYPE, typename IN_TYPE, typename Func>
|
|
|
|
RET_TYPE run_Cvt(IN_TYPE x, Func GenerateConvertInstructionFunc) {
|
2019-05-28 08:28:04 +00:00
|
|
|
using F_CVT = RET_TYPE(IN_TYPE x0, int x1, int x2, int x3, int x4);
|
2016-03-09 12:43:54 +00:00
|
|
|
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assm(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-03-09 12:43:54 +00:00
|
|
|
MacroAssembler* masm = &assm;
|
|
|
|
|
|
|
|
GenerateConvertInstructionFunc(masm);
|
|
|
|
__ dmfc1(v0, f2);
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
assm.GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2016-03-09 12:43:54 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F_CVT>::FromCode(*code);
|
2016-03-09 12:43:54 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
return reinterpret_cast<RET_TYPE>(f.Call(x, 0, 0, 0, 0));
|
2016-03-09 12:43:54 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Cvt_s_uw_Trunc_uw_s) {
|
|
|
|
CcTest::InitializeVM();
|
2016-04-07 07:15:25 +00:00
|
|
|
FOR_UINT32_INPUTS(i, cvt_trunc_uint32_test_values) {
|
2016-03-09 12:43:54 +00:00
|
|
|
uint32_t input = *i;
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm) {
|
|
|
|
__ Cvt_s_uw(f0, a0);
|
|
|
|
__ mthc1(zero_reg, f2);
|
|
|
|
__ Trunc_uw_s(f2, f0, f1);
|
|
|
|
};
|
|
|
|
CHECK_EQ(static_cast<float>(input), run_Cvt<uint64_t>(input, fn));
|
2016-03-09 12:43:54 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Cvt_s_ul_Trunc_ul_s) {
|
|
|
|
CcTest::InitializeVM();
|
2016-04-07 07:15:25 +00:00
|
|
|
FOR_UINT64_INPUTS(i, cvt_trunc_uint64_test_values) {
|
2016-03-09 12:43:54 +00:00
|
|
|
uint64_t input = *i;
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm) {
|
|
|
|
__ Cvt_s_ul(f0, a0);
|
|
|
|
__ Trunc_ul_s(f2, f0, f1, v0);
|
|
|
|
};
|
|
|
|
CHECK_EQ(static_cast<float>(input), run_Cvt<uint64_t>(input, fn));
|
2016-03-09 12:43:54 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Cvt_d_ul_Trunc_ul_d) {
|
|
|
|
CcTest::InitializeVM();
|
2016-04-07 07:15:25 +00:00
|
|
|
FOR_UINT64_INPUTS(i, cvt_trunc_uint64_test_values) {
|
2016-03-09 12:43:54 +00:00
|
|
|
uint64_t input = *i;
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm) {
|
|
|
|
__ Cvt_d_ul(f0, a0);
|
|
|
|
__ Trunc_ul_d(f2, f0, f1, v0);
|
|
|
|
};
|
|
|
|
CHECK_EQ(static_cast<double>(input), run_Cvt<uint64_t>(input, fn));
|
2016-03-09 12:43:54 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(cvt_d_l_Trunc_l_d) {
|
|
|
|
CcTest::InitializeVM();
|
2016-04-07 07:15:25 +00:00
|
|
|
FOR_INT64_INPUTS(i, cvt_trunc_int64_test_values) {
|
2016-03-09 12:43:54 +00:00
|
|
|
int64_t input = *i;
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm) {
|
|
|
|
__ dmtc1(a0, f4);
|
|
|
|
__ cvt_d_l(f0, f4);
|
|
|
|
__ Trunc_l_d(f2, f0);
|
|
|
|
};
|
|
|
|
CHECK_EQ(static_cast<double>(input), run_Cvt<int64_t>(input, fn));
|
2016-03-09 12:43:54 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(cvt_d_l_Trunc_l_ud) {
|
|
|
|
CcTest::InitializeVM();
|
2016-04-07 07:15:25 +00:00
|
|
|
FOR_INT64_INPUTS(i, cvt_trunc_int64_test_values) {
|
2016-03-09 12:43:54 +00:00
|
|
|
int64_t input = *i;
|
|
|
|
uint64_t abs_input = (input < 0) ? -input : input;
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm) {
|
|
|
|
__ dmtc1(a0, f4);
|
|
|
|
__ cvt_d_l(f0, f4);
|
|
|
|
__ Trunc_l_ud(f2, f0, f6);
|
|
|
|
};
|
|
|
|
CHECK_EQ(static_cast<double>(abs_input), run_Cvt<uint64_t>(input, fn));
|
2016-03-09 12:43:54 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(cvt_d_w_Trunc_w_d) {
|
|
|
|
CcTest::InitializeVM();
|
2016-04-07 07:15:25 +00:00
|
|
|
FOR_INT32_INPUTS(i, cvt_trunc_int32_test_values) {
|
2016-03-09 12:43:54 +00:00
|
|
|
int32_t input = *i;
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm) {
|
|
|
|
__ mtc1(a0, f4);
|
|
|
|
__ cvt_d_w(f0, f4);
|
|
|
|
__ Trunc_w_d(f2, f0);
|
|
|
|
__ mfc1(v1, f2);
|
|
|
|
__ dmtc1(v1, f2);
|
|
|
|
};
|
|
|
|
CHECK_EQ(static_cast<double>(input), run_Cvt<int64_t>(input, fn));
|
2016-03-09 12:43:54 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-04-07 07:15:25 +00:00
|
|
|
static const std::vector<int64_t> overflow_int64_test_values() {
|
2017-12-02 00:30:37 +00:00
|
|
|
static const int64_t kValues[] = {static_cast<int64_t>(0xF000000000000000),
|
2016-04-07 07:15:25 +00:00
|
|
|
static_cast<int64_t>(0x0000000000000001),
|
2017-12-02 00:30:37 +00:00
|
|
|
static_cast<int64_t>(0xFF00000000000000),
|
|
|
|
static_cast<int64_t>(0x0000F00111111110),
|
|
|
|
static_cast<int64_t>(0x0F00001000000000),
|
|
|
|
static_cast<int64_t>(0x991234AB12A96731),
|
|
|
|
static_cast<int64_t>(0xB0FFFF0F0F0F0F01),
|
|
|
|
static_cast<int64_t>(0x00006FFFFFFFFFFF),
|
|
|
|
static_cast<int64_t>(0xFFFFFFFFFFFFFFFF)};
|
2016-04-07 07:15:25 +00:00
|
|
|
return std::vector<int64_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
2018-03-16 09:41:18 +00:00
|
|
|
TEST(OverflowInstructions) {
|
|
|
|
CcTest::InitializeVM();
|
2016-04-07 07:15:25 +00:00
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
2018-03-16 09:41:18 +00:00
|
|
|
HandleScope handles(isolate);
|
2016-04-07 07:15:25 +00:00
|
|
|
|
2018-03-16 09:41:18 +00:00
|
|
|
struct T {
|
|
|
|
int64_t lhs;
|
|
|
|
int64_t rhs;
|
|
|
|
int64_t output_add;
|
|
|
|
int64_t output_add2;
|
|
|
|
int64_t output_sub;
|
|
|
|
int64_t output_sub2;
|
|
|
|
int64_t output_mul;
|
|
|
|
int64_t output_mul2;
|
|
|
|
int64_t overflow_add;
|
|
|
|
int64_t overflow_add2;
|
|
|
|
int64_t overflow_sub;
|
|
|
|
int64_t overflow_sub2;
|
|
|
|
int64_t overflow_mul;
|
|
|
|
int64_t overflow_mul2;
|
|
|
|
};
|
|
|
|
T t;
|
2016-04-07 07:15:25 +00:00
|
|
|
|
|
|
|
FOR_INT64_INPUTS(i, overflow_int64_test_values) {
|
|
|
|
FOR_INT64_INPUTS(j, overflow_int64_test_values) {
|
2018-03-16 09:41:18 +00:00
|
|
|
int64_t ii = *i;
|
|
|
|
int64_t jj = *j;
|
|
|
|
int64_t expected_add, expected_sub;
|
|
|
|
int32_t ii32 = static_cast<int32_t>(ii);
|
|
|
|
int32_t jj32 = static_cast<int32_t>(jj);
|
|
|
|
int32_t expected_mul;
|
|
|
|
int64_t expected_add_ovf, expected_sub_ovf, expected_mul_ovf;
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2018-03-16 09:41:18 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
__ ld(t0, MemOperand(a0, offsetof(T, lhs)));
|
|
|
|
__ ld(t1, MemOperand(a0, offsetof(T, rhs)));
|
|
|
|
|
|
|
|
__ DaddOverflow(t2, t0, Operand(t1), t3);
|
|
|
|
__ sd(t2, MemOperand(a0, offsetof(T, output_add)));
|
|
|
|
__ sd(t3, MemOperand(a0, offsetof(T, overflow_add)));
|
|
|
|
__ mov(t3, zero_reg);
|
|
|
|
__ DaddOverflow(t0, t0, Operand(t1), t3);
|
|
|
|
__ sd(t0, MemOperand(a0, offsetof(T, output_add2)));
|
|
|
|
__ sd(t3, MemOperand(a0, offsetof(T, overflow_add2)));
|
|
|
|
|
|
|
|
__ ld(t0, MemOperand(a0, offsetof(T, lhs)));
|
|
|
|
__ ld(t1, MemOperand(a0, offsetof(T, rhs)));
|
|
|
|
|
|
|
|
__ DsubOverflow(t2, t0, Operand(t1), t3);
|
|
|
|
__ sd(t2, MemOperand(a0, offsetof(T, output_sub)));
|
|
|
|
__ sd(t3, MemOperand(a0, offsetof(T, overflow_sub)));
|
|
|
|
__ mov(t3, zero_reg);
|
|
|
|
__ DsubOverflow(t0, t0, Operand(t1), t3);
|
|
|
|
__ sd(t0, MemOperand(a0, offsetof(T, output_sub2)));
|
|
|
|
__ sd(t3, MemOperand(a0, offsetof(T, overflow_sub2)));
|
|
|
|
|
|
|
|
__ ld(t0, MemOperand(a0, offsetof(T, lhs)));
|
|
|
|
__ ld(t1, MemOperand(a0, offsetof(T, rhs)));
|
|
|
|
__ sll(t0, t0, 0);
|
|
|
|
__ sll(t1, t1, 0);
|
|
|
|
|
|
|
|
__ MulOverflow(t2, t0, Operand(t1), t3);
|
|
|
|
__ sd(t2, MemOperand(a0, offsetof(T, output_mul)));
|
|
|
|
__ sd(t3, MemOperand(a0, offsetof(T, overflow_mul)));
|
|
|
|
__ mov(t3, zero_reg);
|
|
|
|
__ MulOverflow(t0, t0, Operand(t1), t3);
|
|
|
|
__ sd(t0, MemOperand(a0, offsetof(T, output_mul2)));
|
|
|
|
__ sd(t3, MemOperand(a0, offsetof(T, overflow_mul2)));
|
|
|
|
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
|
|
|
masm->GetCode(isolate, &desc);
|
|
|
|
Handle<Code> code =
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2018-03-16 09:41:18 +00:00
|
|
|
auto f = GeneratedCode<F3>::FromCode(*code);
|
|
|
|
t.lhs = ii;
|
|
|
|
t.rhs = jj;
|
|
|
|
f.Call(&t, 0, 0, 0, 0);
|
|
|
|
|
|
|
|
expected_add_ovf = base::bits::SignedAddOverflow64(ii, jj, &expected_add);
|
|
|
|
expected_sub_ovf = base::bits::SignedSubOverflow64(ii, jj, &expected_sub);
|
|
|
|
expected_mul_ovf =
|
|
|
|
base::bits::SignedMulOverflow32(ii32, jj32, &expected_mul);
|
|
|
|
|
|
|
|
CHECK_EQ(expected_add_ovf, t.overflow_add < 0);
|
|
|
|
CHECK_EQ(expected_sub_ovf, t.overflow_sub < 0);
|
|
|
|
CHECK_EQ(expected_mul_ovf, t.overflow_mul != 0);
|
|
|
|
|
|
|
|
CHECK_EQ(t.overflow_add, t.overflow_add2);
|
|
|
|
CHECK_EQ(t.overflow_sub, t.overflow_sub2);
|
|
|
|
CHECK_EQ(t.overflow_mul, t.overflow_mul2);
|
|
|
|
|
|
|
|
CHECK_EQ(expected_add, t.output_add);
|
|
|
|
CHECK_EQ(expected_add, t.output_add2);
|
|
|
|
CHECK_EQ(expected_sub, t.output_sub);
|
|
|
|
CHECK_EQ(expected_sub, t.output_sub2);
|
|
|
|
if (!expected_mul_ovf) {
|
|
|
|
CHECK_EQ(expected_mul, t.output_mul);
|
|
|
|
CHECK_EQ(expected_mul, t.output_mul2);
|
2016-04-07 07:15:25 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-03-25 16:51:11 +00:00
|
|
|
TEST(min_max_nan) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-03-25 16:51:11 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
struct TestFloat {
|
|
|
|
double a;
|
|
|
|
double b;
|
|
|
|
double c;
|
|
|
|
double d;
|
|
|
|
float e;
|
|
|
|
float f;
|
|
|
|
float g;
|
|
|
|
float h;
|
|
|
|
};
|
|
|
|
|
|
|
|
TestFloat test;
|
|
|
|
const double dnan = std::numeric_limits<double>::quiet_NaN();
|
|
|
|
const double dinf = std::numeric_limits<double>::infinity();
|
|
|
|
const double dminf = -std::numeric_limits<double>::infinity();
|
|
|
|
const float fnan = std::numeric_limits<float>::quiet_NaN();
|
|
|
|
const float finf = std::numeric_limits<float>::infinity();
|
|
|
|
const float fminf = std::numeric_limits<float>::infinity();
|
|
|
|
const int kTableLength = 13;
|
|
|
|
|
|
|
|
double inputsa[kTableLength] = {2.0, 3.0, -0.0, 0.0, 42.0, dinf, dminf,
|
|
|
|
dinf, dnan, 3.0, dinf, dnan, dnan};
|
|
|
|
double inputsb[kTableLength] = {3.0, 2.0, 0.0, -0.0, dinf, 42.0, dinf,
|
|
|
|
dminf, 3.0, dnan, dnan, dinf, dnan};
|
|
|
|
double outputsdmin[kTableLength] = {2.0, 2.0, -0.0, -0.0, 42.0,
|
|
|
|
42.0, dminf, dminf, dnan, dnan,
|
|
|
|
dnan, dnan, dnan};
|
|
|
|
double outputsdmax[kTableLength] = {3.0, 3.0, 0.0, 0.0, dinf, dinf, dinf,
|
|
|
|
dinf, dnan, dnan, dnan, dnan, dnan};
|
|
|
|
|
|
|
|
float inputse[kTableLength] = {2.0, 3.0, -0.0, 0.0, 42.0, finf, fminf,
|
|
|
|
finf, fnan, 3.0, finf, fnan, fnan};
|
|
|
|
float inputsf[kTableLength] = {3.0, 2.0, 0.0, -0.0, finf, 42.0, finf,
|
|
|
|
fminf, 3.0, fnan, fnan, finf, fnan};
|
|
|
|
float outputsfmin[kTableLength] = {2.0, 2.0, -0.0, -0.0, 42.0, 42.0, fminf,
|
|
|
|
fminf, fnan, fnan, fnan, fnan, fnan};
|
|
|
|
float outputsfmax[kTableLength] = {3.0, 3.0, 0.0, 0.0, finf, finf, finf,
|
|
|
|
finf, fnan, fnan, fnan, fnan, fnan};
|
|
|
|
|
|
|
|
auto handle_dnan = [masm](FPURegister dst, Label* nan, Label* back) {
|
|
|
|
__ bind(nan);
|
2018-09-20 10:14:46 +00:00
|
|
|
__ LoadRoot(t8, RootIndex::kNanValue);
|
2017-07-28 10:34:10 +00:00
|
|
|
__ Ldc1(dst, FieldMemOperand(t8, HeapNumber::kValueOffset));
|
2016-03-25 16:51:11 +00:00
|
|
|
__ Branch(back);
|
|
|
|
};
|
|
|
|
|
|
|
|
auto handle_snan = [masm, fnan](FPURegister dst, Label* nan, Label* back) {
|
|
|
|
__ bind(nan);
|
|
|
|
__ Move(dst, fnan);
|
|
|
|
__ Branch(back);
|
|
|
|
};
|
|
|
|
|
|
|
|
Label handle_mind_nan, handle_maxd_nan, handle_mins_nan, handle_maxs_nan;
|
|
|
|
Label back_mind_nan, back_maxd_nan, back_mins_nan, back_maxs_nan;
|
|
|
|
|
|
|
|
__ push(s6);
|
|
|
|
__ InitializeRootRegister();
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Ldc1(f4, MemOperand(a0, offsetof(TestFloat, a)));
|
|
|
|
__ Ldc1(f8, MemOperand(a0, offsetof(TestFloat, b)));
|
|
|
|
__ Lwc1(f2, MemOperand(a0, offsetof(TestFloat, e)));
|
|
|
|
__ Lwc1(f6, MemOperand(a0, offsetof(TestFloat, f)));
|
2016-12-06 10:16:13 +00:00
|
|
|
__ Float64Min(f10, f4, f8, &handle_mind_nan);
|
2016-03-25 16:51:11 +00:00
|
|
|
__ bind(&back_mind_nan);
|
2016-12-06 10:16:13 +00:00
|
|
|
__ Float64Max(f12, f4, f8, &handle_maxd_nan);
|
2016-03-25 16:51:11 +00:00
|
|
|
__ bind(&back_maxd_nan);
|
2016-12-06 10:16:13 +00:00
|
|
|
__ Float32Min(f14, f2, f6, &handle_mins_nan);
|
2016-03-25 16:51:11 +00:00
|
|
|
__ bind(&back_mins_nan);
|
2016-12-06 10:16:13 +00:00
|
|
|
__ Float32Max(f16, f2, f6, &handle_maxs_nan);
|
2016-03-25 16:51:11 +00:00
|
|
|
__ bind(&back_maxs_nan);
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Sdc1(f10, MemOperand(a0, offsetof(TestFloat, c)));
|
|
|
|
__ Sdc1(f12, MemOperand(a0, offsetof(TestFloat, d)));
|
|
|
|
__ Swc1(f14, MemOperand(a0, offsetof(TestFloat, g)));
|
|
|
|
__ Swc1(f16, MemOperand(a0, offsetof(TestFloat, h)));
|
2016-03-25 16:51:11 +00:00
|
|
|
__ pop(s6);
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
handle_dnan(f10, &handle_mind_nan, &back_mind_nan);
|
|
|
|
handle_dnan(f12, &handle_maxd_nan, &back_maxd_nan);
|
|
|
|
handle_snan(f14, &handle_mins_nan, &back_mins_nan);
|
|
|
|
handle_snan(f16, &handle_maxs_nan, &back_maxs_nan);
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F3>::FromCode(*code);
|
2016-03-25 16:51:11 +00:00
|
|
|
for (int i = 0; i < kTableLength; i++) {
|
|
|
|
test.a = inputsa[i];
|
|
|
|
test.b = inputsb[i];
|
|
|
|
test.e = inputse[i];
|
|
|
|
test.f = inputsf[i];
|
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
f.Call(&test, 0, 0, 0, 0);
|
2016-03-25 16:51:11 +00:00
|
|
|
|
|
|
|
CHECK_EQ(0, memcmp(&test.c, &outputsdmin[i], sizeof(test.c)));
|
|
|
|
CHECK_EQ(0, memcmp(&test.d, &outputsdmax[i], sizeof(test.d)));
|
|
|
|
CHECK_EQ(0, memcmp(&test.g, &outputsfmin[i], sizeof(test.g)));
|
|
|
|
CHECK_EQ(0, memcmp(&test.h, &outputsfmax[i], sizeof(test.h)));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-04-28 15:09:27 +00:00
|
|
|
template <typename IN_TYPE, typename Func>
|
|
|
|
bool run_Unaligned(char* memory_buffer, int32_t in_offset, int32_t out_offset,
|
|
|
|
IN_TYPE value, Func GenerateUnalignedInstructionFunc) {
|
2019-05-28 08:28:04 +00:00
|
|
|
using F_CVT = int32_t(char* x0, int x1, int x2, int x3, int x4);
|
2016-04-28 15:09:27 +00:00
|
|
|
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assm(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-04-28 15:09:27 +00:00
|
|
|
MacroAssembler* masm = &assm;
|
|
|
|
IN_TYPE res;
|
|
|
|
|
|
|
|
GenerateUnalignedInstructionFunc(masm, in_offset, out_offset);
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
assm.GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2016-04-28 15:09:27 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F_CVT>::FromCode(*code);
|
2016-04-28 15:09:27 +00:00
|
|
|
|
|
|
|
MemCopy(memory_buffer + in_offset, &value, sizeof(IN_TYPE));
|
2018-01-09 09:50:34 +00:00
|
|
|
f.Call(memory_buffer, 0, 0, 0, 0);
|
2016-04-28 15:09:27 +00:00
|
|
|
MemCopy(&res, memory_buffer + out_offset, sizeof(IN_TYPE));
|
|
|
|
|
|
|
|
return res == value;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const std::vector<uint64_t> unsigned_test_values() {
|
|
|
|
static const uint64_t kValues[] = {
|
2017-12-02 00:30:37 +00:00
|
|
|
0x2180F18A06384414, 0x000A714532102277, 0xBC1ACCCF180649F0,
|
|
|
|
0x8000000080008000, 0x0000000000000001, 0xFFFFFFFFFFFFFFFF,
|
2016-04-28 15:09:27 +00:00
|
|
|
};
|
|
|
|
return std::vector<uint64_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
|
|
|
static const std::vector<int32_t> unsigned_test_offset() {
|
|
|
|
static const int32_t kValues[] = {// value, offset
|
|
|
|
-132 * KB, -21 * KB, 0, 19 * KB, 135 * KB};
|
|
|
|
return std::vector<int32_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
|
|
|
static const std::vector<int32_t> unsigned_test_offset_increment() {
|
|
|
|
static const int32_t kValues[] = {-5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5};
|
|
|
|
return std::vector<int32_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Ulh) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
static const int kBufferSize = 300 * KB;
|
|
|
|
char memory_buffer[kBufferSize];
|
|
|
|
char* buffer_middle = memory_buffer + (kBufferSize / 2);
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, unsigned_test_values) {
|
|
|
|
FOR_INT32_INPUTS2(j1, j2, unsigned_test_offset) {
|
|
|
|
FOR_INT32_INPUTS2(k1, k2, unsigned_test_offset_increment) {
|
|
|
|
uint16_t value = static_cast<uint64_t>(*i & 0xFFFF);
|
|
|
|
int32_t in_offset = *j1 + *k1;
|
|
|
|
int32_t out_offset = *j2 + *k2;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn_1 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ Ulh(v0, MemOperand(a0, in_offset));
|
|
|
|
__ Ush(v0, MemOperand(a0, out_offset), v0);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint16_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn_1));
|
|
|
|
|
|
|
|
auto fn_2 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ mov(t0, a0);
|
|
|
|
__ Ulh(a0, MemOperand(a0, in_offset));
|
|
|
|
__ Ush(a0, MemOperand(t0, out_offset), v0);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint16_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn_2));
|
|
|
|
|
|
|
|
auto fn_3 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ mov(t0, a0);
|
|
|
|
__ Ulhu(a0, MemOperand(a0, in_offset));
|
|
|
|
__ Ush(a0, MemOperand(t0, out_offset), t1);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint16_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn_3));
|
|
|
|
|
|
|
|
auto fn_4 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ Ulhu(v0, MemOperand(a0, in_offset));
|
|
|
|
__ Ush(v0, MemOperand(a0, out_offset), t1);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint16_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn_4));
|
2016-04-28 15:09:27 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Ulh_bitextension) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
static const int kBufferSize = 300 * KB;
|
|
|
|
char memory_buffer[kBufferSize];
|
|
|
|
char* buffer_middle = memory_buffer + (kBufferSize / 2);
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, unsigned_test_values) {
|
|
|
|
FOR_INT32_INPUTS2(j1, j2, unsigned_test_offset) {
|
|
|
|
FOR_INT32_INPUTS2(k1, k2, unsigned_test_offset_increment) {
|
|
|
|
uint16_t value = static_cast<uint64_t>(*i & 0xFFFF);
|
|
|
|
int32_t in_offset = *j1 + *k1;
|
|
|
|
int32_t out_offset = *j2 + *k2;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
Label success, fail, end, different;
|
|
|
|
__ Ulh(t0, MemOperand(a0, in_offset));
|
|
|
|
__ Ulhu(t1, MemOperand(a0, in_offset));
|
|
|
|
__ Branch(&different, ne, t0, Operand(t1));
|
|
|
|
|
|
|
|
// If signed and unsigned values are same, check
|
|
|
|
// the upper bits to see if they are zero
|
|
|
|
__ sra(t0, t0, 15);
|
|
|
|
__ Branch(&success, eq, t0, Operand(zero_reg));
|
|
|
|
__ Branch(&fail);
|
|
|
|
|
|
|
|
// If signed and unsigned values are different,
|
|
|
|
// check that the upper bits are complementary
|
|
|
|
__ bind(&different);
|
|
|
|
__ sra(t1, t1, 15);
|
|
|
|
__ Branch(&fail, ne, t1, Operand(1));
|
|
|
|
__ sra(t0, t0, 15);
|
|
|
|
__ addiu(t0, t0, 1);
|
|
|
|
__ Branch(&fail, ne, t0, Operand(zero_reg));
|
|
|
|
// Fall through to success
|
|
|
|
|
|
|
|
__ bind(&success);
|
|
|
|
__ Ulh(t0, MemOperand(a0, in_offset));
|
|
|
|
__ Ush(t0, MemOperand(a0, out_offset), v0);
|
|
|
|
__ Branch(&end);
|
|
|
|
__ bind(&fail);
|
|
|
|
__ Ush(zero_reg, MemOperand(a0, out_offset), v0);
|
|
|
|
__ bind(&end);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint16_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn));
|
2016-04-28 15:09:27 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Ulw) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
static const int kBufferSize = 300 * KB;
|
|
|
|
char memory_buffer[kBufferSize];
|
|
|
|
char* buffer_middle = memory_buffer + (kBufferSize / 2);
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, unsigned_test_values) {
|
|
|
|
FOR_INT32_INPUTS2(j1, j2, unsigned_test_offset) {
|
|
|
|
FOR_INT32_INPUTS2(k1, k2, unsigned_test_offset_increment) {
|
|
|
|
uint32_t value = static_cast<uint32_t>(*i & 0xFFFFFFFF);
|
|
|
|
int32_t in_offset = *j1 + *k1;
|
|
|
|
int32_t out_offset = *j2 + *k2;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn_1 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ Ulw(v0, MemOperand(a0, in_offset));
|
|
|
|
__ Usw(v0, MemOperand(a0, out_offset));
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint32_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn_1));
|
|
|
|
|
|
|
|
auto fn_2 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ mov(t0, a0);
|
|
|
|
__ Ulw(a0, MemOperand(a0, in_offset));
|
|
|
|
__ Usw(a0, MemOperand(t0, out_offset));
|
|
|
|
};
|
2016-04-28 15:09:27 +00:00
|
|
|
CHECK_EQ(true,
|
2016-12-01 16:15:33 +00:00
|
|
|
run_Unaligned<uint32_t>(buffer_middle, in_offset, out_offset,
|
|
|
|
(uint32_t)value, fn_2));
|
|
|
|
|
|
|
|
auto fn_3 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ Ulwu(v0, MemOperand(a0, in_offset));
|
|
|
|
__ Usw(v0, MemOperand(a0, out_offset));
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint32_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn_3));
|
|
|
|
|
|
|
|
auto fn_4 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ mov(t0, a0);
|
|
|
|
__ Ulwu(a0, MemOperand(a0, in_offset));
|
|
|
|
__ Usw(a0, MemOperand(t0, out_offset));
|
|
|
|
};
|
2016-04-28 15:09:27 +00:00
|
|
|
CHECK_EQ(true,
|
2016-12-01 16:15:33 +00:00
|
|
|
run_Unaligned<uint32_t>(buffer_middle, in_offset, out_offset,
|
|
|
|
(uint32_t)value, fn_4));
|
2016-04-28 15:09:27 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Ulw_extension) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
static const int kBufferSize = 300 * KB;
|
|
|
|
char memory_buffer[kBufferSize];
|
|
|
|
char* buffer_middle = memory_buffer + (kBufferSize / 2);
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, unsigned_test_values) {
|
|
|
|
FOR_INT32_INPUTS2(j1, j2, unsigned_test_offset) {
|
|
|
|
FOR_INT32_INPUTS2(k1, k2, unsigned_test_offset_increment) {
|
|
|
|
uint32_t value = static_cast<uint32_t>(*i & 0xFFFFFFFF);
|
|
|
|
int32_t in_offset = *j1 + *k1;
|
|
|
|
int32_t out_offset = *j2 + *k2;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
Label success, fail, end, different;
|
|
|
|
__ Ulw(t0, MemOperand(a0, in_offset));
|
|
|
|
__ Ulwu(t1, MemOperand(a0, in_offset));
|
|
|
|
__ Branch(&different, ne, t0, Operand(t1));
|
|
|
|
|
|
|
|
// If signed and unsigned values are same, check
|
|
|
|
// the upper bits to see if they are zero
|
|
|
|
__ dsra(t0, t0, 31);
|
|
|
|
__ Branch(&success, eq, t0, Operand(zero_reg));
|
|
|
|
__ Branch(&fail);
|
|
|
|
|
|
|
|
// If signed and unsigned values are different,
|
|
|
|
// check that the upper bits are complementary
|
|
|
|
__ bind(&different);
|
|
|
|
__ dsra(t1, t1, 31);
|
|
|
|
__ Branch(&fail, ne, t1, Operand(1));
|
|
|
|
__ dsra(t0, t0, 31);
|
|
|
|
__ daddiu(t0, t0, 1);
|
|
|
|
__ Branch(&fail, ne, t0, Operand(zero_reg));
|
|
|
|
// Fall through to success
|
|
|
|
|
|
|
|
__ bind(&success);
|
|
|
|
__ Ulw(t0, MemOperand(a0, in_offset));
|
|
|
|
__ Usw(t0, MemOperand(a0, out_offset));
|
|
|
|
__ Branch(&end);
|
|
|
|
__ bind(&fail);
|
|
|
|
__ Usw(zero_reg, MemOperand(a0, out_offset));
|
|
|
|
__ bind(&end);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint32_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn));
|
2016-04-28 15:09:27 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Uld) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
static const int kBufferSize = 300 * KB;
|
|
|
|
char memory_buffer[kBufferSize];
|
|
|
|
char* buffer_middle = memory_buffer + (kBufferSize / 2);
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, unsigned_test_values) {
|
|
|
|
FOR_INT32_INPUTS2(j1, j2, unsigned_test_offset) {
|
|
|
|
FOR_INT32_INPUTS2(k1, k2, unsigned_test_offset_increment) {
|
|
|
|
uint64_t value = *i;
|
|
|
|
int32_t in_offset = *j1 + *k1;
|
|
|
|
int32_t out_offset = *j2 + *k2;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn_1 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ Uld(v0, MemOperand(a0, in_offset));
|
|
|
|
__ Usd(v0, MemOperand(a0, out_offset));
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<uint64_t>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn_1));
|
|
|
|
|
|
|
|
auto fn_2 = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ mov(t0, a0);
|
|
|
|
__ Uld(a0, MemOperand(a0, in_offset));
|
|
|
|
__ Usd(a0, MemOperand(t0, out_offset));
|
|
|
|
};
|
2016-04-28 15:09:27 +00:00
|
|
|
CHECK_EQ(true,
|
2016-12-01 16:15:33 +00:00
|
|
|
run_Unaligned<uint64_t>(buffer_middle, in_offset, out_offset,
|
|
|
|
(uint32_t)value, fn_2));
|
2016-04-28 15:09:27 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Ulwc1) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
static const int kBufferSize = 300 * KB;
|
|
|
|
char memory_buffer[kBufferSize];
|
|
|
|
char* buffer_middle = memory_buffer + (kBufferSize / 2);
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, unsigned_test_values) {
|
|
|
|
FOR_INT32_INPUTS2(j1, j2, unsigned_test_offset) {
|
|
|
|
FOR_INT32_INPUTS2(k1, k2, unsigned_test_offset_increment) {
|
|
|
|
float value = static_cast<float>(*i & 0xFFFFFFFF);
|
|
|
|
int32_t in_offset = *j1 + *k1;
|
|
|
|
int32_t out_offset = *j2 + *k2;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ Ulwc1(f0, MemOperand(a0, in_offset), t0);
|
|
|
|
__ Uswc1(f0, MemOperand(a0, out_offset), t0);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<float>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn));
|
2016-04-28 15:09:27 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Uldc1) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
static const int kBufferSize = 300 * KB;
|
|
|
|
char memory_buffer[kBufferSize];
|
|
|
|
char* buffer_middle = memory_buffer + (kBufferSize / 2);
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, unsigned_test_values) {
|
|
|
|
FOR_INT32_INPUTS2(j1, j2, unsigned_test_offset) {
|
|
|
|
FOR_INT32_INPUTS2(k1, k2, unsigned_test_offset_increment) {
|
|
|
|
double value = static_cast<double>(*i);
|
|
|
|
int32_t in_offset = *j1 + *k1;
|
|
|
|
int32_t out_offset = *j2 + *k2;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn = [](MacroAssembler* masm, int32_t in_offset,
|
|
|
|
int32_t out_offset) {
|
|
|
|
__ Uldc1(f0, MemOperand(a0, in_offset), t0);
|
|
|
|
__ Usdc1(f0, MemOperand(a0, out_offset), t0);
|
|
|
|
};
|
|
|
|
CHECK_EQ(true, run_Unaligned<double>(buffer_middle, in_offset,
|
|
|
|
out_offset, value, fn));
|
2016-04-28 15:09:27 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-09-30 18:28:33 +00:00
|
|
|
static const std::vector<uint64_t> sltu_test_values() {
|
|
|
|
static const uint64_t kValues[] = {
|
|
|
|
0,
|
|
|
|
1,
|
2017-12-02 00:30:37 +00:00
|
|
|
0x7FFE,
|
|
|
|
0x7FFF,
|
2016-09-30 18:28:33 +00:00
|
|
|
0x8000,
|
|
|
|
0x8001,
|
2017-12-02 00:30:37 +00:00
|
|
|
0xFFFE,
|
|
|
|
0xFFFF,
|
|
|
|
0xFFFFFFFFFFFF7FFE,
|
|
|
|
0xFFFFFFFFFFFF7FFF,
|
|
|
|
0xFFFFFFFFFFFF8000,
|
|
|
|
0xFFFFFFFFFFFF8001,
|
|
|
|
0xFFFFFFFFFFFFFFFE,
|
|
|
|
0xFFFFFFFFFFFFFFFF,
|
2016-09-30 18:28:33 +00:00
|
|
|
};
|
|
|
|
return std::vector<uint64_t>(&kValues[0], &kValues[arraysize(kValues)]);
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename Func>
|
|
|
|
bool run_Sltu(uint64_t rs, uint64_t rd, Func GenerateSltuInstructionFunc) {
|
2019-05-28 08:28:04 +00:00
|
|
|
using F_CVT = int64_t(uint64_t x0, uint64_t x1, int x2, int x3, int x4);
|
2016-09-30 18:28:33 +00:00
|
|
|
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assm(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-09-30 18:28:33 +00:00
|
|
|
MacroAssembler* masm = &assm;
|
|
|
|
|
|
|
|
GenerateSltuInstructionFunc(masm, rd);
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
assm.GetCode(isolate, &desc);
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Handle<Code> code =
|
|
|
|
Factory::CodeBuilder(isolate, desc, CodeKind::FOR_TESTING).Build();
|
2016-09-30 18:28:33 +00:00
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
auto f = GeneratedCode<F_CVT>::FromCode(*code);
|
|
|
|
int64_t res = reinterpret_cast<int64_t>(f.Call(rs, rd, 0, 0, 0));
|
2016-09-30 18:28:33 +00:00
|
|
|
return res == 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST(Sltu) {
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
|
|
|
|
FOR_UINT64_INPUTS(i, sltu_test_values) {
|
|
|
|
FOR_UINT64_INPUTS(j, sltu_test_values) {
|
|
|
|
uint64_t rs = *i;
|
|
|
|
uint64_t rd = *j;
|
|
|
|
|
2016-12-01 16:15:33 +00:00
|
|
|
auto fn_1 = [](MacroAssembler* masm, uint64_t imm) {
|
|
|
|
__ Sltu(v0, a0, Operand(imm));
|
|
|
|
};
|
|
|
|
CHECK_EQ(rs < rd, run_Sltu(rs, rd, fn_1));
|
|
|
|
|
|
|
|
auto fn_2 = [](MacroAssembler* masm, uint64_t imm) {
|
|
|
|
__ Sltu(v0, a0, a1);
|
|
|
|
};
|
|
|
|
CHECK_EQ(rs < rd, run_Sltu(rs, rd, fn_2));
|
2016-09-30 18:28:33 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-12-06 10:16:13 +00:00
|
|
|
template <typename T, typename Inputs, typename Results>
|
2018-01-09 09:50:34 +00:00
|
|
|
static GeneratedCode<F4> GenerateMacroFloat32MinMax(MacroAssembler* masm) {
|
2016-12-06 10:16:13 +00:00
|
|
|
T a = T::from_code(4); // f4
|
|
|
|
T b = T::from_code(6); // f6
|
|
|
|
T c = T::from_code(8); // f8
|
|
|
|
|
|
|
|
Label ool_min_abc, ool_min_aab, ool_min_aba;
|
|
|
|
Label ool_max_abc, ool_max_aab, ool_max_aba;
|
|
|
|
|
|
|
|
Label done_min_abc, done_min_aab, done_min_aba;
|
|
|
|
Label done_max_abc, done_max_aab, done_max_aba;
|
|
|
|
|
|
|
|
#define FLOAT_MIN_MAX(fminmax, res, x, y, done, ool, res_field) \
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Lwc1(x, MemOperand(a0, offsetof(Inputs, src1_))); \
|
|
|
|
__ Lwc1(y, MemOperand(a0, offsetof(Inputs, src2_))); \
|
2016-12-06 10:16:13 +00:00
|
|
|
__ fminmax(res, x, y, &ool); \
|
|
|
|
__ bind(&done); \
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Swc1(a, MemOperand(a1, offsetof(Results, res_field)))
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
// a = min(b, c);
|
|
|
|
FLOAT_MIN_MAX(Float32Min, a, b, c, done_min_abc, ool_min_abc, min_abc_);
|
|
|
|
// a = min(a, b);
|
|
|
|
FLOAT_MIN_MAX(Float32Min, a, a, b, done_min_aab, ool_min_aab, min_aab_);
|
|
|
|
// a = min(b, a);
|
|
|
|
FLOAT_MIN_MAX(Float32Min, a, b, a, done_min_aba, ool_min_aba, min_aba_);
|
|
|
|
|
|
|
|
// a = max(b, c);
|
|
|
|
FLOAT_MIN_MAX(Float32Max, a, b, c, done_max_abc, ool_max_abc, max_abc_);
|
|
|
|
// a = max(a, b);
|
|
|
|
FLOAT_MIN_MAX(Float32Max, a, a, b, done_max_aab, ool_max_aab, max_aab_);
|
|
|
|
// a = max(b, a);
|
|
|
|
FLOAT_MIN_MAX(Float32Max, a, b, a, done_max_aba, ool_max_aba, max_aba_);
|
|
|
|
|
|
|
|
#undef FLOAT_MIN_MAX
|
|
|
|
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
// Generate out-of-line cases.
|
|
|
|
__ bind(&ool_min_abc);
|
|
|
|
__ Float32MinOutOfLine(a, b, c);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_min_abc);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_min_aab);
|
|
|
|
__ Float32MinOutOfLine(a, a, b);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_min_aab);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_min_aba);
|
|
|
|
__ Float32MinOutOfLine(a, b, a);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_min_aba);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_max_abc);
|
|
|
|
__ Float32MaxOutOfLine(a, b, c);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_max_abc);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_max_aab);
|
|
|
|
__ Float32MaxOutOfLine(a, a, b);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_max_aab);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_max_aba);
|
|
|
|
__ Float32MaxOutOfLine(a, b, a);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_max_aba);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(masm->isolate(), &desc);
|
2017-09-29 14:59:24 +00:00
|
|
|
Handle<Code> code =
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Factory::CodeBuilder(masm->isolate(), desc, CodeKind::FOR_TESTING)
|
2020-09-30 11:28:28 +00:00
|
|
|
.Build();
|
2016-12-06 10:16:13 +00:00
|
|
|
#ifdef DEBUG
|
2018-06-14 12:46:07 +00:00
|
|
|
StdoutStream os;
|
2018-07-12 10:06:42 +00:00
|
|
|
code->Print(os);
|
2016-12-06 10:16:13 +00:00
|
|
|
#endif
|
2018-01-09 09:50:34 +00:00
|
|
|
return GeneratedCode<F4>::FromCode(*code);
|
2016-12-06 10:16:13 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
TEST(macro_float_minmax_f32) {
|
|
|
|
// Test the Float32Min and Float32Max macros.
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
|
|
|
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-12-06 10:16:13 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
struct Inputs {
|
|
|
|
float src1_;
|
|
|
|
float src2_;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct Results {
|
|
|
|
// Check all register aliasing possibilities in order to exercise all
|
|
|
|
// code-paths in the macro assembler.
|
|
|
|
float min_abc_;
|
|
|
|
float min_aab_;
|
|
|
|
float min_aba_;
|
|
|
|
float max_abc_;
|
|
|
|
float max_aab_;
|
|
|
|
float max_aba_;
|
|
|
|
};
|
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
GeneratedCode<F4> f =
|
|
|
|
GenerateMacroFloat32MinMax<FPURegister, Inputs, Results>(masm);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
#define CHECK_MINMAX(src1, src2, min, max) \
|
|
|
|
do { \
|
|
|
|
Inputs inputs = {src1, src2}; \
|
|
|
|
Results results; \
|
2018-01-09 09:50:34 +00:00
|
|
|
f.Call(&inputs, &results, 0, 0, 0); \
|
2016-12-06 10:16:13 +00:00
|
|
|
CHECK_EQ(bit_cast<uint32_t>(min), bit_cast<uint32_t>(results.min_abc_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint32_t>(min), bit_cast<uint32_t>(results.min_aab_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint32_t>(min), bit_cast<uint32_t>(results.min_aba_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint32_t>(max), bit_cast<uint32_t>(results.max_abc_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint32_t>(max), bit_cast<uint32_t>(results.max_aab_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint32_t>(max), bit_cast<uint32_t>(results.max_aba_)); \
|
|
|
|
/* Use a bit_cast to correctly identify -0.0 and NaNs. */ \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
float nan_a = std::numeric_limits<float>::quiet_NaN();
|
|
|
|
float nan_b = std::numeric_limits<float>::quiet_NaN();
|
|
|
|
|
|
|
|
CHECK_MINMAX(1.0f, -1.0f, -1.0f, 1.0f);
|
|
|
|
CHECK_MINMAX(-1.0f, 1.0f, -1.0f, 1.0f);
|
|
|
|
CHECK_MINMAX(0.0f, -1.0f, -1.0f, 0.0f);
|
|
|
|
CHECK_MINMAX(-1.0f, 0.0f, -1.0f, 0.0f);
|
|
|
|
CHECK_MINMAX(-0.0f, -1.0f, -1.0f, -0.0f);
|
|
|
|
CHECK_MINMAX(-1.0f, -0.0f, -1.0f, -0.0f);
|
|
|
|
CHECK_MINMAX(0.0f, 1.0f, 0.0f, 1.0f);
|
|
|
|
CHECK_MINMAX(1.0f, 0.0f, 0.0f, 1.0f);
|
|
|
|
|
|
|
|
CHECK_MINMAX(0.0f, 0.0f, 0.0f, 0.0f);
|
|
|
|
CHECK_MINMAX(-0.0f, -0.0f, -0.0f, -0.0f);
|
|
|
|
CHECK_MINMAX(-0.0f, 0.0f, -0.0f, 0.0f);
|
|
|
|
CHECK_MINMAX(0.0f, -0.0f, -0.0f, 0.0f);
|
|
|
|
|
|
|
|
CHECK_MINMAX(0.0f, nan_a, nan_a, nan_a);
|
|
|
|
CHECK_MINMAX(nan_a, 0.0f, nan_a, nan_a);
|
|
|
|
CHECK_MINMAX(nan_a, nan_b, nan_a, nan_a);
|
|
|
|
CHECK_MINMAX(nan_b, nan_a, nan_b, nan_b);
|
|
|
|
|
|
|
|
#undef CHECK_MINMAX
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename T, typename Inputs, typename Results>
|
2018-01-09 09:50:34 +00:00
|
|
|
static GeneratedCode<F4> GenerateMacroFloat64MinMax(MacroAssembler* masm) {
|
2016-12-06 10:16:13 +00:00
|
|
|
T a = T::from_code(4); // f4
|
|
|
|
T b = T::from_code(6); // f6
|
|
|
|
T c = T::from_code(8); // f8
|
|
|
|
|
|
|
|
Label ool_min_abc, ool_min_aab, ool_min_aba;
|
|
|
|
Label ool_max_abc, ool_max_aab, ool_max_aba;
|
|
|
|
|
|
|
|
Label done_min_abc, done_min_aab, done_min_aba;
|
|
|
|
Label done_max_abc, done_max_aab, done_max_aba;
|
|
|
|
|
|
|
|
#define FLOAT_MIN_MAX(fminmax, res, x, y, done, ool, res_field) \
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Ldc1(x, MemOperand(a0, offsetof(Inputs, src1_))); \
|
|
|
|
__ Ldc1(y, MemOperand(a0, offsetof(Inputs, src2_))); \
|
2016-12-06 10:16:13 +00:00
|
|
|
__ fminmax(res, x, y, &ool); \
|
|
|
|
__ bind(&done); \
|
2017-04-20 13:45:05 +00:00
|
|
|
__ Sdc1(a, MemOperand(a1, offsetof(Results, res_field)))
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
// a = min(b, c);
|
|
|
|
FLOAT_MIN_MAX(Float64Min, a, b, c, done_min_abc, ool_min_abc, min_abc_);
|
|
|
|
// a = min(a, b);
|
|
|
|
FLOAT_MIN_MAX(Float64Min, a, a, b, done_min_aab, ool_min_aab, min_aab_);
|
|
|
|
// a = min(b, a);
|
|
|
|
FLOAT_MIN_MAX(Float64Min, a, b, a, done_min_aba, ool_min_aba, min_aba_);
|
|
|
|
|
|
|
|
// a = max(b, c);
|
|
|
|
FLOAT_MIN_MAX(Float64Max, a, b, c, done_max_abc, ool_max_abc, max_abc_);
|
|
|
|
// a = max(a, b);
|
|
|
|
FLOAT_MIN_MAX(Float64Max, a, a, b, done_max_aab, ool_max_aab, max_aab_);
|
|
|
|
// a = max(b, a);
|
|
|
|
FLOAT_MIN_MAX(Float64Max, a, b, a, done_max_aba, ool_max_aba, max_aba_);
|
|
|
|
|
|
|
|
#undef FLOAT_MIN_MAX
|
|
|
|
|
|
|
|
__ jr(ra);
|
|
|
|
__ nop();
|
|
|
|
|
|
|
|
// Generate out-of-line cases.
|
|
|
|
__ bind(&ool_min_abc);
|
|
|
|
__ Float64MinOutOfLine(a, b, c);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_min_abc);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_min_aab);
|
|
|
|
__ Float64MinOutOfLine(a, a, b);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_min_aab);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_min_aba);
|
|
|
|
__ Float64MinOutOfLine(a, b, a);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_min_aba);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_max_abc);
|
|
|
|
__ Float64MaxOutOfLine(a, b, c);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_max_abc);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_max_aab);
|
|
|
|
__ Float64MaxOutOfLine(a, a, b);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_max_aab);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
__ bind(&ool_max_aba);
|
|
|
|
__ Float64MaxOutOfLine(a, b, a);
|
2016-12-08 14:56:55 +00:00
|
|
|
__ Branch(&done_max_aba);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
CodeDesc desc;
|
2017-05-31 14:00:11 +00:00
|
|
|
masm->GetCode(masm->isolate(), &desc);
|
2019-04-30 10:41:40 +00:00
|
|
|
Handle<Code> code =
|
Reland "Reland "[deoptimizer] Change deopt entries into builtins""
This is a reland of fbfa9bf4ec72b1b73a96b70ccb68cd98c321511b
The arm64 was missing proper codegen for CFI, thus sizes were off.
Original change's description:
> Reland "[deoptimizer] Change deopt entries into builtins"
>
> This is a reland of 7f58ced72eb65b6b5530ccabaf2eaebe45bf9d33
>
> It fixes the different exit size emitted on x64/Atom CPUs due to
> performance tuning in TurboAssembler::Call. Additionally, add
> cctests to verify the fixed size exits.
>
> Original change's description:
> > [deoptimizer] Change deopt entries into builtins
> >
> > While the overall goal of this commit is to change deoptimization
> > entries into builtins, there are multiple related things happening:
> >
> > - Deoptimization entries, formerly stubs (i.e. Code objects generated
> > at runtime, guaranteed to be immovable), have been converted into
> > builtins. The major restriction is that we now need to preserve the
> > kRootRegister, which was formerly used on most architectures to pass
> > the deoptimization id. The solution differs based on platform.
> > - Renamed DEOPT_ENTRIES_OR_FOR_TESTING code kind to FOR_TESTING.
> > - Removed heap/ support for immovable Code generation.
> > - Removed the DeserializerData class (no longer needed).
> > - arm64: to preserve 4-byte deopt exits, introduced a new optimization
> > in which the final jump to the deoptimization entry is generated
> > once per Code object, and deopt exits can continue to emit a
> > near-call.
> > - arm,ia32,x64: change to fixed-size deopt exits. This reduces exit
> > sizes by 4/8, 5, and 5 bytes, respectively.
> >
> > On arm the deopt exit size is reduced from 12 (or 16) bytes to 8 bytes
> > by using the same strategy as on arm64 (recalc deopt id from return
> > address). Before:
> >
> > e300a002 movw r10, <id>
> > e59fc024 ldr ip, [pc, <entry offset>]
> > e12fff3c blx ip
> >
> > After:
> >
> > e59acb35 ldr ip, [r10, <entry offset>]
> > e12fff3c blx ip
> >
> > On arm64 the deopt exit size remains 4 bytes (or 8 bytes in same cases
> > with CFI). Additionally, up to 4 builtin jumps are emitted per Code
> > object (max 32 bytes added overhead per Code object). Before:
> >
> > 9401cdae bl <entry offset>
> >
> > After:
> >
> > # eager deoptimization entry jump.
> > f95b1f50 ldr x16, [x26, <eager entry offset>]
> > d61f0200 br x16
> > # lazy deoptimization entry jump.
> > f95b2b50 ldr x16, [x26, <lazy entry offset>]
> > d61f0200 br x16
> > # the deopt exit.
> > 97fffffc bl <eager deoptimization entry jump offset>
> >
> > On ia32 the deopt exit size is reduced from 10 to 5 bytes. Before:
> >
> > bb00000000 mov ebx,<id>
> > e825f5372b call <entry>
> >
> > After:
> >
> > e8ea2256ba call <entry>
> >
> > On x64 the deopt exit size is reduced from 12 to 7 bytes. Before:
> >
> > 49c7c511000000 REX.W movq r13,<id>
> > e8ea2f0700 call <entry>
> >
> > After:
> >
> > 41ff9560360000 call [r13+<entry offset>]
> >
> > Bug: v8:8661,v8:8768
> > Change-Id: I13e30aedc360474dc818fecc528ce87c3bfeed42
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2465834
> > Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
> > Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#70597}
>
> Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
> Bug: v8:8661,v8:8768,chromium:1140165
> Change-Id: Ibcd5c39c58a70bf2b2ac221aa375fc68d495e144
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2485506
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
> Commit-Queue: Jakob Gruber <jgruber@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#70655}
Tbr: ulan@chromium.org, tebbi@chromium.org, rmcilroy@chromium.org
Bug: v8:8661
Bug: v8:8768
Bug: chromium:1140165
Change-Id: I471cc94fc085e527dc9bfb5a84b96bd907c2333f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488682
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70672}
2020-10-21 05:12:25 +00:00
|
|
|
Factory::CodeBuilder(masm->isolate(), desc, CodeKind::FOR_TESTING)
|
2020-09-30 11:28:28 +00:00
|
|
|
.Build();
|
2016-12-06 10:16:13 +00:00
|
|
|
#ifdef DEBUG
|
2018-06-14 12:46:07 +00:00
|
|
|
StdoutStream os;
|
2018-07-12 10:06:42 +00:00
|
|
|
code->Print(os);
|
2016-12-06 10:16:13 +00:00
|
|
|
#endif
|
2018-01-09 09:50:34 +00:00
|
|
|
return GeneratedCode<F4>::FromCode(*code);
|
2016-12-06 10:16:13 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
TEST(macro_float_minmax_f64) {
|
|
|
|
// Test the Float64Min and Float64Max macros.
|
|
|
|
CcTest::InitializeVM();
|
|
|
|
Isolate* isolate = CcTest::i_isolate();
|
|
|
|
HandleScope scope(isolate);
|
|
|
|
|
2019-01-16 18:29:52 +00:00
|
|
|
MacroAssembler assembler(isolate, v8::internal::CodeObjectRequired::kYes);
|
2016-12-06 10:16:13 +00:00
|
|
|
MacroAssembler* masm = &assembler;
|
|
|
|
|
|
|
|
struct Inputs {
|
|
|
|
double src1_;
|
|
|
|
double src2_;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct Results {
|
|
|
|
// Check all register aliasing possibilities in order to exercise all
|
|
|
|
// code-paths in the macro assembler.
|
|
|
|
double min_abc_;
|
|
|
|
double min_aab_;
|
|
|
|
double min_aba_;
|
|
|
|
double max_abc_;
|
|
|
|
double max_aab_;
|
|
|
|
double max_aba_;
|
|
|
|
};
|
|
|
|
|
2018-01-09 09:50:34 +00:00
|
|
|
GeneratedCode<F4> f =
|
|
|
|
GenerateMacroFloat64MinMax<DoubleRegister, Inputs, Results>(masm);
|
2016-12-06 10:16:13 +00:00
|
|
|
|
|
|
|
#define CHECK_MINMAX(src1, src2, min, max) \
|
|
|
|
do { \
|
|
|
|
Inputs inputs = {src1, src2}; \
|
|
|
|
Results results; \
|
2018-01-09 09:50:34 +00:00
|
|
|
f.Call(&inputs, &results, 0, 0, 0); \
|
2016-12-06 10:16:13 +00:00
|
|
|
CHECK_EQ(bit_cast<uint64_t>(min), bit_cast<uint64_t>(results.min_abc_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint64_t>(min), bit_cast<uint64_t>(results.min_aab_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint64_t>(min), bit_cast<uint64_t>(results.min_aba_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint64_t>(max), bit_cast<uint64_t>(results.max_abc_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint64_t>(max), bit_cast<uint64_t>(results.max_aab_)); \
|
|
|
|
CHECK_EQ(bit_cast<uint64_t>(max), bit_cast<uint64_t>(results.max_aba_)); \
|
|
|
|
/* Use a bit_cast to correctly identify -0.0 and NaNs. */ \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
double nan_a = std::numeric_limits<double>::quiet_NaN();
|
|
|
|
double nan_b = std::numeric_limits<double>::quiet_NaN();
|
|
|
|
|
|
|
|
CHECK_MINMAX(1.0, -1.0, -1.0, 1.0);
|
|
|
|
CHECK_MINMAX(-1.0, 1.0, -1.0, 1.0);
|
|
|
|
CHECK_MINMAX(0.0, -1.0, -1.0, 0.0);
|
|
|
|
CHECK_MINMAX(-1.0, 0.0, -1.0, 0.0);
|
|
|
|
CHECK_MINMAX(-0.0, -1.0, -1.0, -0.0);
|
|
|
|
CHECK_MINMAX(-1.0, -0.0, -1.0, -0.0);
|
|
|
|
CHECK_MINMAX(0.0, 1.0, 0.0, 1.0);
|
|
|
|
CHECK_MINMAX(1.0, 0.0, 0.0, 1.0);
|
|
|
|
|
|
|
|
CHECK_MINMAX(0.0, 0.0, 0.0, 0.0);
|
|
|
|
CHECK_MINMAX(-0.0, -0.0, -0.0, -0.0);
|
|
|
|
CHECK_MINMAX(-0.0, 0.0, -0.0, 0.0);
|
|
|
|
CHECK_MINMAX(0.0, -0.0, -0.0, 0.0);
|
|
|
|
|
|
|
|
CHECK_MINMAX(0.0, nan_a, nan_a, nan_a);
|
|
|
|
CHECK_MINMAX(nan_a, 0.0, nan_a, nan_a);
|
|
|
|
CHECK_MINMAX(nan_a, nan_b, nan_a, nan_a);
|
|
|
|
CHECK_MINMAX(nan_b, nan_a, nan_b, nan_b);
|
|
|
|
|
|
|
|
#undef CHECK_MINMAX
|
|
|
|
}
|
|
|
|
|
2014-07-09 11:08:26 +00:00
|
|
|
#undef __
|
2017-08-31 12:34:55 +00:00
|
|
|
|
|
|
|
} // namespace internal
|
|
|
|
} // namespace v8
|