[turbofan] Implement rounding of floats on x64 and ia32 without sse4.1.

[turbofan] Implement rounding of floats on x64 and ia32 without sse4.1.

The implementation sets the rounding mode flag and then uses the
cvtsd2si and cvtsi2sd instructions (convert between float and int) to do
the rounding. Input values outside int range either don't have to be
rounded anyways, or are rounded by calculating input + 2^52 - 2^52 for
positive inputs, or input -2^52 + 2^52 for negative inputs. The original
rounding mode is restored afterwards.

R=titzer@chromium.org

B=575379

Committed: https://crrev.com/fa5d09e547abe79a8c82f780deb980c53ad78beb
Cr-Commit-Position: refs/heads/master@{#33367}

[ahaas, 2016-01-14 15:16:07]

Description was changed from

==========
[turbofan] Implement rounding of floats on x64 and ia32 without sse4.1.

The implementation sets the rounding mode flag and then uses the
cvtsd2si and cvtsi2sd instructions (convert between float and int) to do
the rounding. Input values outside int range either don't have to be
rounded anyways, or are rounded by calculating input + 2^52 - 2^52 for
positive inputs, or input -2^52 + 2^52 for negative inputs. The original
rounding mode is restored afterwards.

R=titzer@chromium.org
==========

to

==========
[turbofan] Implement rounding of floats on x64 and ia32 without sse4.1.

The implementation sets the rounding mode flag and then uses the
cvtsd2si and cvtsi2sd instructions (convert between float and int) to do
the rounding. Input values outside int range either don't have to be
rounded anyways, or are rounded by calculating input + 2^52 - 2^52 for
positive inputs, or input -2^52 + 2^52 for negative inputs. The original
rounding mode is restored afterwards.

R=titzer@chromium.org

B=575379
==========

[titzer, 2016-01-18 10:44:29]

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
File src/compiler/ia32/code-generator-ia32.cc (right):

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:663: Register kScratchRegister =
i.TempRegister(0);
Don't call this kScratchRegister, it's not a constant.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:665: // We have to store the original
rounding mode to restore it later.
This code sucks so much I think it should go to macro assembler hell.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:676: __ cvtss2si(kScratchRegister,
i.InputDoubleRegister(0));
If you require the input to be in a register, you can get rid of some of this.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:692: // The input value is already an
integer number, no rounding necessary.
If you do a DefineSameAsFirst, you can get rid of these explicit moves.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:791: if
(instr->InputAt(0)->IsDoubleRegister()) {
Same for this one. Require the input in a register, since we use it so many
times.

[ahaas, 2016-01-18 15:30:30]

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
File src/compiler/ia32/code-generator-ia32.cc (right):

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:663: Register kScratchRegister =
i.TempRegister(0);
On 2016/01/18 at 10:44:28, titzer wrote:
> Don't call this kScratchRegister, it's not a constant.

Done.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:665: // We have to store the original
rounding mode to restore it later.
On 2016/01/18 at 10:44:28, titzer wrote:
> This code sucks so much I think it should go to macro assembler hell.

Done.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:676: __ cvtss2si(kScratchRegister,
i.InputDoubleRegister(0));
On 2016/01/18 at 10:44:28, titzer wrote:
> If you require the input to be in a register, you can get rid of some of this.

Done.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:692: // The input value is already an
integer number, no rounding necessary.
On 2016/01/18 at 10:44:28, titzer wrote:
> If you do a DefineSameAsFirst, you can get rid of these explicit moves.

Done.

https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
src/compiler/ia32/code-generator-ia32.cc:791: if
(instr->InputAt(0)->IsDoubleRegister()) {
On 2016/01/18 at 10:44:28, titzer wrote:
> Same for this one. Require the input in a register, since we use it so many
times.

Done.

[titzer, 2016-01-18 15:57:41]

On 2016/01/18 15:30:30, ahaas wrote:
>
https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
> File src/compiler/ia32/code-generator-ia32.cc (right):
> 
>
https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
> src/compiler/ia32/code-generator-ia32.cc:663: Register kScratchRegister =
> i.TempRegister(0);
> On 2016/01/18 at 10:44:28, titzer wrote:
> > Don't call this kScratchRegister, it's not a constant.
> 
> Done.
> 
>
https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
> src/compiler/ia32/code-generator-ia32.cc:665: // We have to store the original
> rounding mode to restore it later.
> On 2016/01/18 at 10:44:28, titzer wrote:
> > This code sucks so much I think it should go to macro assembler hell.
> 
> Done.
> 
>
https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
> src/compiler/ia32/code-generator-ia32.cc:676: __ cvtss2si(kScratchRegister,
> i.InputDoubleRegister(0));
> On 2016/01/18 at 10:44:28, titzer wrote:
> > If you require the input to be in a register, you can get rid of some of
this.
> 
> Done.
> 
>
https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
> src/compiler/ia32/code-generator-ia32.cc:692: // The input value is already an
> integer number, no rounding necessary.
> On 2016/01/18 at 10:44:28, titzer wrote:
> > If you do a DefineSameAsFirst, you can get rid of these explicit moves.
> 
> Done.
> 
>
https://codereview.chromium.org/1584663007/diff/1/src/compiler/ia32/code-gene...
> src/compiler/ia32/code-generator-ia32.cc:791: if
> (instr->InputAt(0)->IsDoubleRegister()) {
> On 2016/01/18 at 10:44:28, titzer wrote:
> > Same for this one. Require the input in a register, since we use it so many
> times.
> 
> Done.

lgtm

[ahaas, 2016-01-18 16:05:38]

The CQ bit was checked by ahaas@chromium.org

[commit-bot: I haz the power, 2016-01-18 16:05:51]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1584663007/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1584663007/20001

[commit-bot: I haz the power, 2016-01-18 16:09:26]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2016-01-18 16:10:18]

Description was changed from

==========
[turbofan] Implement rounding of floats on x64 and ia32 without sse4.1.

The implementation sets the rounding mode flag and then uses the
cvtsd2si and cvtsi2sd instructions (convert between float and int) to do
the rounding. Input values outside int range either don't have to be
rounded anyways, or are rounded by calculating input + 2^52 - 2^52 for
positive inputs, or input -2^52 + 2^52 for negative inputs. The original
rounding mode is restored afterwards.

R=titzer@chromium.org

B=575379
==========

to

==========
[turbofan] Implement rounding of floats on x64 and ia32 without sse4.1.

The implementation sets the rounding mode flag and then uses the
cvtsd2si and cvtsi2sd instructions (convert between float and int) to do
the rounding. Input values outside int range either don't have to be
rounded anyways, or are rounded by calculating input + 2^52 - 2^52 for
positive inputs, or input -2^52 + 2^52 for negative inputs. The original
rounding mode is restored afterwards.

R=titzer@chromium.org

B=575379

Committed: https://crrev.com/fa5d09e547abe79a8c82f780deb980c53ad78beb
Cr-Commit-Position: refs/heads/master@{#33367}
==========

[commit-bot: I haz the power, 2016-01-18 16:10:18]

Patchset 2 (id:??) landed as
https://crrev.com/fa5d09e547abe79a8c82f780deb980c53ad78beb
Cr-Commit-Position: refs/heads/master@{#33367}

[ahaas, 2016-01-18 16:26:31]

A revert of this CL (patchset #2 id:20001) has been created in
https://codereview.chromium.org/1593313010/ by ahaas@chromium.org.

The reason for reverting is: Code is incorrect for -0..

[ahaas, 2016-01-19 12:06:00]

PTAL. 

-0.0 is supported properly now. This is achieved by storing the sign of the
input in a temporary register and restoring it after the conversion.

https://codereview.chromium.org/1584663007/diff/40001/src/x64/macro-assembler...
File src/x64/macro-assembler-x64.cc (right):

https://codereview.chromium.org/1584663007/diff/40001/src/x64/macro-assembler...
src/x64/macro-assembler-x64.cc:2888: // Saving the sign bit.
The idea to support -0 is to store the sign of the input in xtmp and then
restore it after the conversion.