From: Anton Blanchard <anton@samba.org>
Date: Tue, 26 May 2015 13:41:42 +0000 (+1000)
Subject: Improve performance of PowerPC64 fast_long_add_function
X-Git-Tag: php-7.0.0beta3~5^2~70
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=fd8fb17be8773a34989347140ce4cb25867f720f;p=php

Improve performance of PowerPC64 fast_long_add_function

Detecting overflow with the XER is slow, partially because we have to
clear it before use.

PHP already has a fast way of detecting overflow in its fallback
c implementation. Overflow only occurs if the signs of the two
operands are the same and the sign of the result is different.
Furthermore, leaving it in c allows gcc to schedule the instructions
better.

This is 9% faster on a POWER8 running a simple testcase:

<?php
        function testcase($count = 100000000) {
		$x = 1;
                for ($i = 0; $i < $count; $i++) {
                        $x = $x + 1;
                        $x = $x + 1;
                        $x = $x + 1;
                        $x = $x + 1;
                        $x = $x + 1;
                }
        }

        testcase();
?>
---

diff --git a/Zend/zend_operators.h b/Zend/zend_operators.h
index 7e4fdccb93..f46a729045 100644
--- a/Zend/zend_operators.h
+++ b/Zend/zend_operators.h
@@ -556,36 +556,6 @@ static zend_always_inline void fast_long_add_function(zval *result, zval *op1, z
 		  "n"(IS_DOUBLE),
 		  "n"(ZVAL_OFFSETOF_TYPE)
 		: "rax","cc");
-#elif defined(__GNUC__) && defined(__powerpc64__)
-	__asm__(
-		"ld 14, 0(%1)\n\t"
-		"ld 15, 0(%2)\n\t"
-		"li 16, 0 \n\t"
-		"mtxer 16\n\t"
-		"addo. 14, 14, 15\n\t"
-		"bso- 0f\n\t"
-		"std 14, 0(%0)\n\t"
-		"li 14, %3\n\t"
-		"stw 14, %c5(%0)\n\t"
-		"b 1f\n"
-		"0:\n\t"
-		"lfd 0, 0(%1)\n\t"
-		"lfd 1, 0(%2)\n\t"
-		"fcfid 0, 0\n\t"
-		"fcfid 1, 1\n\t"
-		"fadd 0, 0, 1\n\t"
-		"li 14, %4\n\t"
-		"stw 14, %c5(%0)\n\t"
-		"stfd 0, 0(%0)\n"
-		"1:"
-		:
-		: "r"(&result->value),
-		  "r"(&op1->value),
-		  "r"(&op2->value),
-		  "n"(IS_LONG),
-		  "n"(IS_DOUBLE),
-		  "n"(ZVAL_OFFSETOF_TYPE)
-		: "r14","r15","r16","fr0","fr1","cc");
 #else
 	/*
 	 * 'result' may alias with op1 or op2, so we need to
@@ -680,36 +650,6 @@ static zend_always_inline void fast_long_sub_function(zval *result, zval *op1, z
 		  "n"(IS_DOUBLE),
 		  "n"(ZVAL_OFFSETOF_TYPE)
 		: "rax","cc");
-#elif defined(__GNUC__) && defined(__powerpc64__)
-	__asm__(
-		"ld 14, 0(%1)\n\t"
-		"ld 15, 0(%2)\n\t"
-		"li 16, 0\n\t"
-		"mtxer 16\n\t"
-		"subo. 14, 14, 15\n\t"
-		"bso- 0f\n\t"
-		"std 14, 0(%0)\n\t"
-		"li 14, %3\n\t"
-		"stw 14, %c5(%0)\n\t"
-		"b 1f\n"
-		"0:\n\t"
-		"lfd 0, 0(%1)\n\t"
-		"lfd 1, 0(%2)\n\t"
-		"fcfid 0, 0\n\t"
-		"fcfid 1, 1\n\t"
-		"fsub 0, 0, 1\n\t"
-		"li 14, %4\n\t"
-		"stw 14, %c5(%0)\n\t"
-		"stfd 0, 0(%0)\n"
-		"1:"
-		:
-		: "r"(&result->value),
-		  "r"(&op1->value),
-		  "r"(&op2->value),
-		  "n"(IS_LONG),
-		  "n"(IS_DOUBLE),
-		  "n"(ZVAL_OFFSETOF_TYPE)
-		: "r14","r15","r16","fr0","fr1","cc");
 #else
 	ZVAL_LONG(result, Z_LVAL_P(op1) - Z_LVAL_P(op2));