Differences

This shows you the differences between two versions of the page.

--- rfc:tsrmls-fetch-reduction [2010/09/09 23:52] – update felipe
+++ rfc:tsrmls-fetch-reduction [2010/09/22 13:51] – update felipe
@@ Line 2: / Line 2: @@
   * Version: 1.0
   * Date: 2010-09-09
-  * Author: Felipe Pena <felipe@php.net>
+  * Author: Felipe Pena <felipe@php.net>, Pierre Alain Joye <pajoye@php.net>
-  * Status: Under Discussion
+  * Status: Under implementation
 ===== Introduction =====
-The purpose of RFC is to show a way to reduce several function call by avoiding useless TSRMLS_FETCH() usage on ZTS.
+The purpose of RFC is to show a way to reduce several function call by avoiding useless TSRMLS_FETCH() usage on ZTS. Adding TSRMLS_CC to the memory allocation macros/function, hash api, zend_llist and streams.
 ===== Changes ======
-The main goal was to create a new version of each memory allocation function/macro where TSRMLS_FETCH() is not used, thus using these new functions/macros only when suitable avoiding to call the function that does thread id fetch.
 See below some macros/functions will be added by the patch.
 ^ Current function/macro ^ New function/macro ^
-| emalloc          | emallocts    |
+| emalloc          | emalloc2    |
-| ecalloc          | ecallocts    |
+| ecalloc          | ecalloc2    |
-| pemalloc         | pemallocts   |
+| pemalloc         | pemalloc2   |
-| erealloc         | ereallocts   |
+| erealloc         | erealloc2   |
-| efree            | efreets      |
+| efree            | efree2      |
-| estrdup          | estrdupts    |
+| estrdup          | estrdup2    |
-| estrndup         | estrndupts   |
+| estrndup         | estrndup2   |
-| ALLOC_INIT_ZVAL  | ALLOC_INIT_ZVAL_TS |
-| SEPARATE_ZVAL    | SEPARATE_ZVAL_TS   |
+===== Profiler =====
-| SEPARATE_ZVAL_IF_NOT_REF | SEPARATE_ZVAL_IF_NOT_REF_TS |
-| SEPARATE_ZVAL_TO_MAKE_IS_REF | SEPARATE_ZVAL_TO_MAKE_IS_REF_TS |
+Using cachegrind (//valgrind --tool=cachegrind --cachegrind-out-file=profile sapi/cli/php Zend/bench.php//) to see the improvement, we can see the follow differences:
+^ Function ^ Ir (Patched) ^ Ir (Current) ^
+| pthread_getspecific | 498,129,021 | 1,269,764,085 |
+| ts_resource_ex | 394,353,075 | 1,005,230,834 |
+| efree | 459,940,754 | 840,217,602 |
+^ Ir Total (Patched) ^ Ir Total (Current) ^
+| 42,537,064,394 | 44,080,331,320 |
+// Ir - Instruction read //
 ===== Benchmark =====
@@ Line 92: / Line 100: @@
 </code>
-**Non-patched**
+**Current (trunk)**
 <code>
 empty_loop         0.798
@@ Line 129: / Line 137: @@
 </code>
-===== Proposal and Patch =====
+===== Patch =====
-Comming soon
+http://felipe.ath.cx/diff/tsrmls-fetch-reduction.diff
 ===== Changelog =====

Differences

Page Tools