Page MenuHomePhabricator

kevin.qin (Kevin Qin)
User

Projects

User does not belong to any projects.

User Details

User Since
Sep 12 2013, 7:01 PM (314 w, 1 d)

Recent Activity

May 13 2015

kevin.qin added a comment to D9655: [AArch64 NEON]Allow to define poly64_t as 'unsigned long long' on LLP64 system..

2015-05-11 16:55 GMT+08:00 Kevin Qin <kevinqindev@gmail.com>:

May 13 2015, 7:36 PM

May 11 2015

kevin.qin retitled D9655: [AArch64 NEON]Allow to define poly64_t as 'unsigned long long' on LLP64 system. from to [AArch64 NEON]Allow to define poly64_t as 'unsigned long long' on LLP64 system..
May 11 2015, 1:55 AM

Oct 9 2014

kevin.qin added a comment to D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.

Committed at r219401.

Oct 9 2014, 3:30 AM

Oct 8 2014

kevin.qin updated the diff for D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.

Hi Eric,

Oct 8 2014, 6:01 AM

Oct 6 2014

kevin.qin updated the diff for D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.

After some performance investigation on smaller granular, I proposed to set 16 as loop buffer size for A57. I will post some experiments result in following comments. Thanks Chandler and Jiangning for their advices.

Oct 6 2014, 3:09 AM

Sep 24 2014

kevin.qin updated D5147: [Runtime Unrolling] use a loop to simplify the runtime unrolling prologue..
Sep 24 2014, 8:45 AM

Sep 23 2014

kevin.qin updated the diff for D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.

Hi Renato,

Sep 23 2014, 4:09 AM

Sep 22 2014

kevin.qin updated the diff for D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.

I proposed to set 20 as loop buffer size for A57 after a lot of benchmarking experiments. Then this number will be used as a threshold to allow partial & runtime unrolling on small loops. The effect on spec2006 is,

Sep 22 2014, 3:15 AM

Sep 9 2014

kevin.qin added a comment to D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.

I can give more details on the performance regressions here. Basically, partial unrolling contributes small performance improvement, regressions and code size changes. Most of the regressions are caused by the runtime unrolling prologue. This prologue will do some extra work on checking the loop iterations and execute the reminder times of loop bodies. If the runtime unrolled loop is inside another loop, and inner loop count for each running is quite small, then there's a overhead happened in the prologue and caused the regession.

Sep 9 2014, 3:19 AM
kevin.qin added a comment to D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.

Hi Eric,

Sep 9 2014, 2:23 AM

Sep 3 2014

kevin.qin updated the diff for D5147: [Runtime Unrolling] use a loop to simplify the runtime unrolling prologue..

Hi Hal,

Sep 3 2014, 6:06 AM

Sep 2 2014

kevin.qin retitled D5148: [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target from to [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target.
Sep 2 2014, 3:09 AM
kevin.qin retitled D5147: [Runtime Unrolling] use a loop to simplify the runtime unrolling prologue. from to [Runtime Unrolling] use a loop to simplify the runtime unrolling prologue..
Sep 2 2014, 2:55 AM

Jul 23 2014

kevin.qin added a comment to D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector..

Hi Tim,

Jul 23 2014, 7:17 PM · deleted
kevin.qin updated the diff for D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector..

Hi Tim,

Jul 23 2014, 12:47 AM · deleted

Jul 22 2014

kevin.qin added a comment to D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector..

See comments below.

Jul 22 2014, 1:25 AM · deleted

Jul 21 2014

kevin.qin updated the diff for D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector..

This is the updated patch. Please review.

Jul 21 2014, 10:15 PM · deleted
kevin.qin added a comment to D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector..

Hi Tim,

Jul 21 2014, 8:18 PM · deleted

Jul 20 2014

kevin.qin added a comment to D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector..

2014-07-12 0:13 GMT+08:00 Chad Rosier <mcrosier@codeaurora.org>:

Jul 20 2014, 6:31 PM · deleted

Jul 18 2014

kevin.qin added a comment to D4346: [AArch64] Implement Clang CLI interface proposal about "-march"..

Committed in r213353.

Jul 18 2014, 12:14 AM

Jul 17 2014

kevin.qin updated the diff for D4346: [AArch64] Implement Clang CLI interface proposal about "-march"..

Fix another typo missing in previous patch.

Jul 17 2014, 11:24 PM
kevin.qin updated the diff for D4346: [AArch64] Implement Clang CLI interface proposal about "-march"..

Hi Kristof,

Jul 17 2014, 11:12 PM

Jul 16 2014

kevin.qin updated the diff for D4346: [AArch64] Implement Clang CLI interface proposal about "-march"..

Parts of codes are refactored to match this behavior.
"Where this option is used in conjunction with -march or -mtune, those options take precedence over the appropriate part of this option."

Jul 16 2014, 10:47 PM

Jul 15 2014

kevin.qin updated the diff for D4346: [AArch64] Implement Clang CLI interface proposal about "-march"..

This patch is to change variable name from "Splited" to "Split".

Jul 15 2014, 11:04 PM
kevin.qin updated the diff for D4346: [AArch64] Implement Clang CLI interface proposal about "-march"..

Minor changes in test file to get check-all pass.

Jul 15 2014, 3:39 AM
kevin.qin updated the diff for D4346: [AArch64] Implement Clang CLI interface proposal about "-march"..

Hi all,

Jul 15 2014, 3:32 AM

Jul 7 2014

kevin.qin updated the diff for D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector..

Small changes according to Chad's comments.

Jul 7 2014, 7:19 PM · deleted

Jul 4 2014

kevin.qin retitled D4385: [AArch64] Fix a bug generating incorrect instruction when building small vector. from to [AArch64] Fix a bug generating incorrect instruction when building small vector..
Jul 4 2014, 2:08 AM · deleted

Jul 1 2014

kevin.qin updated the diff for D4228: [AArch64] Fix a build_vector pattern fail caused by defect in isBuildVectorAllZeros(). .

Sorry for following up this issue late. This patch is to normalize all constants building a vector. The value of constant nodes will be truncated to fit element width.

Jul 1 2014, 2:05 AM

Jun 30 2014

kevin.qin retitled D4346: [AArch64] Implement Clang CLI interface proposal about "-march". from to [AArch64] Implement Clang CLI interface proposal about "-march"..
Jun 30 2014, 3:34 AM

Jun 23 2014

kevin.qin added a comment to D4228: [AArch64] Fix a build_vector pattern fail caused by defect in isBuildVectorAllZeros(). .

Hi Tim,

Jun 23 2014, 10:52 PM
kevin.qin added a comment to D4228: [AArch64] Fix a build_vector pattern fail caused by defect in isBuildVectorAllZeros(). .

Hi Tim,

Jun 23 2014, 8:22 PM

Jun 20 2014

kevin.qin retitled D4228: [AArch64] Fix a build_vector pattern fail caused by defect in isBuildVectorAllZeros(). from to [AArch64] Fix a build_vector pattern fail caused by defect in isBuildVectorAllZeros(). .
Jun 20 2014, 3:10 AM

Jun 17 2014

kevin.qin added a comment to D4080: [AArch64] Fix a pattern match failure caused by creating improper CONCAT_VECTOR..

Hi Tim,

Jun 17 2014, 11:04 PM
kevin.qin added a comment to D4080: [AArch64] Fix a pattern match failure caused by creating improper CONCAT_VECTOR..

Hi Tim,

Jun 17 2014, 1:33 AM

Jun 16 2014

kevin.qin updated the diff for D4080: [AArch64] Fix a pattern match failure caused by creating improper CONCAT_VECTOR..

Here is updated patch. Please review again. Thanks.

Jun 16 2014, 10:49 PM
kevin.qin added a comment to D4080: [AArch64] Fix a pattern match failure caused by creating improper CONCAT_VECTOR..

Hi Tim,

Jun 16 2014, 10:44 PM

Jun 10 2014

kevin.qin retitled D4082: [AArch64] Add default feature for CPUs on AArch64 target in Clang. from to [AArch64] Add default feature for CPUs on AArch64 target in Clang..
Jun 10 2014, 1:01 AM

Jun 9 2014

kevin.qin retitled D4080: [AArch64] Fix a pattern match failure caused by creating improper CONCAT_VECTOR. from to [AArch64] Fix a pattern match failure caused by creating improper CONCAT_VECTOR..
Jun 9 2014, 7:06 PM

May 13 2014

kevin.qin updated the diff for D3662: [ARM64] Fix the misleading diagnostic on bad extend amount of reg+reg addressing mode..

Changed to the vague diagnostic as Tim suggested.

May 13 2014, 12:03 AM

May 11 2014

kevin.qin added a comment to D3662: [ARM64] Fix the misleading diagnostic on bad extend amount of reg+reg addressing mode..

Hi Tim,

May 11 2014, 11:34 PM

May 8 2014

kevin.qin retitled D3662: [ARM64] Fix the misleading diagnostic on bad extend amount of reg+reg addressing mode. from to [ARM64] Fix the misleading diagnostic on bad extend amount of reg+reg addressing mode..
May 8 2014, 3:11 AM

Apr 25 2014

kevin.qin retitled D3504: [ARM64] Add RUN lines for "–target arm64 –mattr=-fp-armv8" on AArch64 no-fp test. from to [ARM64] Add RUN lines for "–target arm64 –mattr=-fp-armv8" on AArch64 no-fp test. .
Apr 25 2014, 2:29 AM
kevin.qin retitled D3502: [ARM64] Support crc predicate on ARM64. from to [ARM64] Support crc predicate on ARM64..
Apr 25 2014, 2:22 AM

Apr 23 2014

kevin.qin updated the diff for D3467: [ARM64] Enable alignment control option in front-end for ARM64 - LLVM..

Add new run line to test "-arm64-no-strict-align".

Apr 23 2014, 8:03 PM
kevin.qin added a comment to D3467: [ARM64] Enable alignment control option in front-end for ARM64 - LLVM..

Hi Tim,

Apr 23 2014, 3:13 AM
kevin.qin updated subscribers of D3466: [ARM64] Enable alignment control option in front-end for ARM64 - Clang..
Apr 23 2014, 3:08 AM
kevin.qin retitled D3467: [ARM64] Enable alignment control option in front-end for ARM64 - LLVM. from to [ARM64] Enable alignment control option in front-end for ARM64 - LLVM..
Apr 23 2014, 3:07 AM
kevin.qin retitled D3466: [ARM64] Enable alignment control option in front-end for ARM64 - Clang. from to [ARM64] Enable alignment control option in front-end for ARM64 - Clang..
Apr 23 2014, 3:05 AM