Improve the string descriptor. #1036

robertsipka · 2016-05-02T14:49:25Z

Separate the utf-8 and ascii strings.
In case of ascii strings the size is equal to the length, so we able to store this information
in the string descriptor instead of using the string header to store it.

JerryScript-DCO-1.0-Signed-off-by: Robert Sipka rsipka.uszeged@partner.samsung.com

zherczeg · 2016-05-03T10:44:09Z

jerry-core/ecma/base/ecma-globals.h

+      mem_cpointer_t ascii_collection_cp;
+      /** Size of ascii string in bytes */
+      uint16_t size;
+    } ascii_string;


ascii_string_t, and missing comments

zherczeg · 2016-05-04T09:18:32Z

jerry-core/ecma/base/ecma-globals.h

-  ECMA_STRING_CONTAINER_HEAP_CHUNKS, /**< actual data is on the heap
-                                          in a ecma_collection_chunk_t chain */
+  ECMA_STRING_CONTAINER_HEAP_UTF8_STRING, /**< actual data is on the heap as an utf-8 string */
+  ECMA_STRING_CONTAINER_HEAP_ASCII_STRING, /**< actual data is on the heap as an ascii string */


I would put ASCII_STRING first.

zherczeg · 2016-05-05T09:06:39Z

jerry-core/ecma/base/ecma-globals.h

  ECMA_STRING_CONTAINER_HEAP_NUMBER, /**< actual data is on the heap as a ecma_number_t */
+  ECMA_STRING_CONTAINER_HEAP_UTF8_STRING, /**< actual data is on the heap as an utf-8 string */


Please move this right after the ECMA_STRING_CONTAINER_HEAP_ASCII_STRING.

robertsipka · 2016-05-05T13:40:54Z

Thanks for the comments, I've updated and rebased with master.

LaszloLango · 2016-05-05T13:58:21Z

jerry-core/ecma/base/ecma-helpers-string.c

+  {
+    string_desc_p->refs_and_container = ECMA_STRING_CONTAINER_HEAP_ASCII_STRING | ECMA_STRING_REF_ONE;
+    const size_t data_size = new_size;
+    void *data_p = (void *) mem_heap_alloc_block (data_size);


Why don't you use lit_utf8_byte_t * here, if you cast void* to it anyway?

Good point, I've updated these pointers.

zherczeg · 2016-05-06T06:57:52Z

What is the effect on memory of this patch?

robertsipka · 2016-05-06T08:56:45Z

                  Benchmark |          RSS(+ is better) |                   Perf(+ is better)
                 3d-cube.js |   84k ->   84k :  +0.000% |   2.170s ->   2.156s :  +0.645%  (+-1.079%) : [≈]
     access-binary-trees.js |   72k ->   72k :  +0.000% |   1.192s ->   1.176s :  +1.342%  (+-1.215%) : [+]
         access-fannkuch.js |   32k ->   32k :  +0.000% |   6.300s ->   6.332s :  -0.508%  (+-0.908%) : [≈]
            access-nbody.js |   36k ->   36k :  +0.000% |   2.374s ->   2.348s :  +1.095%  (+-0.609%) : [+]
bitops-3bit-bits-in-byte.js |   24k ->   24k :  +0.000% |   1.364s ->   1.360s :  +0.293%  (+-0.824%) : [≈]
     bitops-bits-in-byte.js |   24k ->   24k :  +0.000% |   2.036s ->   2.030s :  +0.295%  (+-0.552%) : [≈]
      bitops-bitwise-and.js |   24k ->   24k :  +0.000% |   2.798s ->   2.858s :  -2.144%  (+-1.475%) : [-]
      bitops-nsieve-bits.js |  164k ->  164k :  +0.000% |   3.932s ->   3.906s :  +0.661%  (+-1.164%) : [≈]
   controlflow-recursive.js |  124k ->  124k :  +0.000% |   0.802s ->   0.800s :  +0.249%  (+-1.145%) : [≈]
              crypto-aes.js |  104k ->  104k :  +0.000% |   2.874s ->   2.868s :  +0.209%  (+-0.716%) : [≈]
              crypto-md5.js |  168k ->  168k :  +0.000% |   1.180s ->   1.180s :  +0.000%  (+-0.000%) : [≈]
             crypto-sha1.js |  116k ->  116k :  +0.000% |   1.116s ->   1.118s :  -0.179%  (+-1.306%) : [≈]
       date-format-tofte.js |   56k ->   56k :  +0.000% |   1.716s ->   1.716s :  +0.000%  (+-0.929%) : [≈]
       date-format-xparb.js |   52k ->   52k :  +0.000% |   0.722s ->   0.720s :  +0.277%  (+-1.272%) : [≈]
             math-cordic.js |   28k ->   28k :  +0.000% |   2.446s ->   2.442s :  +0.164%  (+-1.031%) : [≈]
       math-partial-sums.js |   24k ->   24k :  +0.000% |   1.324s ->   1.312s :  +0.906%  (+-1.094%) : [≈]
      math-spectral-norm.js |   32k ->   32k :  +0.000% |   1.196s ->   1.188s :  +0.669%  (+-1.212%) : [≈]
           string-base64.js |  156k ->  148k :  +5.128% |   5.234s ->   5.278s :  -0.841%  (+-0.737%) : [-]
            string-fasta.js |   40k ->   40k :  +0.000% |   2.662s ->   2.676s :  -0.526%  (+-0.951%) : [≈]
            Geometric mean: |     RSS reduction: 0.277% |                 Speed up: 0.140% (+-0.233%) : [≈]

zherczeg · 2016-05-06T09:00:33Z

Cool a string based test improved!

LaszloLango · 2016-05-06T09:02:42Z

@robertsipka, could you also compare with '--mem-stats'? @bzsolt can help if you don't know how to.

robertsipka · 2016-05-06T11:24:54Z

Benchmark	Mem heap (bytes)
3d-cube.js	32760 -> 32752
access-binary-trees.js	49144 -> 49144
access-fannkuch.js	8184 -> 8184
access-nbody.js	8184 -> 8184
bitops-3bit-bits-in-byte.js	928 -> 920
bitops-bits-in-byte.js	872 -> 864
bitops-bitwise-and.js	560 -> 552
bitops-nsieve-bits.js	142736 -> 142728
controlflow-recursive.js	11112 -> 11104
crypto-aes.js	65544 -> 65544
crypto-md5.js	140920 -> 141048
crypto-sha1.js	91680 -> 91648
date-format-tofte.js	24568 -> 24568
date-format-xparb.js	16400 -> 16384
math-cordic.js	2384 -> 2376
math-partial-sums.js	1864 -> 1856
math-spectral-norm.js	8184 -> 8176
string-base64.js	93664 -> 93624
string-fasta.js	16376 -> 16376

LaszloLango · 2016-05-06T11:26:18Z

jerry-core/ecma/base/ecma-helpers-string.c

+    string_desc_p->refs_and_container = ECMA_STRING_CONTAINER_HEAP_UTF8_STRING | ECMA_STRING_REF_ONE;
+    const size_t data_size = new_size + sizeof (ecma_string_heap_header_t);
+    ecma_string_heap_header_t *data_p = (ecma_string_heap_header_t *) mem_heap_alloc_block (data_size);
+    lit_utf8_size_t bytes_copied = ecma_string_to_utf8_string (string1_p,
                                                             (lit_utf8_byte_t *) (data_p + 1),


wrong indentation

LaszloLango · 2016-05-06T12:16:39Z

jerry-core/ecma/base/ecma-globals.h

+      mem_cpointer_t ascii_collection_cp;
+      /** Size of ascii string in bytes */
+      uint16_t size;
+    } ascii_string_t;


Rename to ascii_string, this is not a typedef

LaszloLango · 2016-05-06T12:30:37Z

LGTM

zherczeg · 2016-05-09T07:56:43Z

LGTM

Separate the utf-8 and ascii strings. In case of ascii strings the size is equal to the length, so we able to store this information in the string descriptor instead of using the string header to store it. JerryScript-DCO-1.0-Signed-off-by: Robert Sipka rsipka.uszeged@partner.samsung.com

robertsipka · 2016-05-09T07:59:05Z

Thanks, I've rebased it with master.

zherczeg reviewed May 3, 2016
View reviewed changes

LaszloLango added the enhancement An improvement label May 3, 2016

LaszloLango added this to the Engine optimization & enhancement milestone May 3, 2016

robertsipka force-pushed the ascii_string branch from 43f6440 to b5ad328 Compare May 3, 2016 13:44

zherczeg reviewed May 4, 2016
View reviewed changes

robertsipka force-pushed the ascii_string branch from b5ad328 to 24a9813 Compare May 4, 2016 12:17

zherczeg reviewed May 5, 2016
View reviewed changes

robertsipka force-pushed the ascii_string branch from 24a9813 to 656da04 Compare May 5, 2016 11:56

robertsipka force-pushed the ascii_string branch from 656da04 to 85de4e0 Compare May 5, 2016 13:40

LaszloLango reviewed May 5, 2016
View reviewed changes

robertsipka force-pushed the ascii_string branch 2 times, most recently from 448d2ee to cee537e Compare May 6, 2016 11:22

LaszloLango reviewed May 6, 2016
View reviewed changes

robertsipka force-pushed the ascii_string branch from cee537e to f90d9a3 Compare May 6, 2016 12:16

LaszloLango reviewed May 6, 2016
View reviewed changes

robertsipka force-pushed the ascii_string branch from f90d9a3 to d4ecbd4 Compare May 6, 2016 12:27

robertsipka force-pushed the ascii_string branch from d4ecbd4 to de1912f Compare May 9, 2016 07:58

LaszloLango merged commit de1912f into jerryscript-project:master May 9, 2016

zherczeg mentioned this pull request Sep 5, 2016

Nominating Robert Sipka (@robertsipka) to become a JerryScript committer #1321

Closed

robertsipka deleted the ascii_string branch November 24, 2016 08:50

		ECMA_STRING_CONTAINER_HEAP_NUMBER, /*< actual data is on the heap as a ecma_number_t /
		ECMA_STRING_CONTAINER_HEAP_UTF8_STRING, /*< actual data is on the heap as an utf-8 string /

Improve the string descriptor. #1036

Improve the string descriptor. #1036

Uh oh!

Conversation

robertsipka commented May 2, 2016

Uh oh!

zherczeg May 3, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

robertsipka commented May 5, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

zherczeg commented May 6, 2016

Uh oh!

robertsipka commented May 6, 2016

Uh oh!

zherczeg commented May 6, 2016

Uh oh!

LaszloLango commented May 6, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

robertsipka commented May 6, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

LaszloLango commented May 6, 2016

Uh oh!

zherczeg commented May 9, 2016

Uh oh!

robertsipka commented May 9, 2016

Uh oh!

Uh oh!

zherczeg May 3, 2016 •

edited

Loading

LaszloLango commented May 6, 2016 •

edited

Loading