[webkit-reviews] review requested: [Bug 50295] Regexp JIT'ed Code Contains Branches to Branches for Backtracking : [Attachment 75232] Patch to remove most paren related branches to branches

[bugzilla-daemon at webkit.org]

Michael Saboff <msaboff at apple.com> has asked  for review:
Bug 50295: Regexp JIT'ed Code Contains Branches to Branches for Backtracking
https://bugs.webkit.org/show_bug.cgi?id=50295

Attachment 75232: Patch to remove most paren related branches to branches
https://bugs.webkit.org/attachment.cgi?id=75232&action=review

------- Additional Comments from Michael Saboff <msaboff at apple.com>
This patch propagates branches from within parentheses backtracking code to
later alternatives.  It also emits the parentheses failure code at the end of
the routine.  It handles indirect branches as backtracking destinations and
emits a trampoline to an indirect branch when needed.

Some cleanups were done as part of this patch.	Specifically the
invertOrCapture flag was separated into two.

The code generated by this patch for the same regular expression in the
description is:

	push   %rbp	
	mov    %rsp,%rbp
	push   %rbx	
	mov    %esi,(%rcx)
	sub    $0x10,%rsp
	add    $0x2,%esi
	cmp    %edx,%esi
	ja     nextChar
loop:
	mov    %rsi,%rax
	add    $0xfffffffffffffffe,%eax
	mov    %eax,0x8(%rcx)  
	cmpw   $0x61,-0x4(%rdi,%rsi,2)
	jne    tryAlt1A 
	mov    $tryAlt1A,%r11  
	mov    %r11,(%rsp)
	jmpq   term1Match
tryAlt1A:
	cmpw   $0x41,-0x4(%rdi,%rsi,2)
	jne    term1Fail
	mov    $term1Fail,%r11 
	mov    %r11,(%rsp)
	jmpq   term1Match
term1Match:
	mov    %rsi,%rax
	add    $0xffffffffffffffff,%eax
	mov    %eax,0xc(%rcx)  
	mov    %rsi,%rax
	add    $0xffffffffffffffff,%eax
	mov    %eax,0x10(%rcx) 
	cmpw   $0x62,-0x2(%rdi,%rsi,2)
	jne    tryAlt2B 
	mov    $tryAlt2B,%r11  
	mov    %r11,0x8(%rsp)  
	jmpq   term2Match
tryAlt2B:
	cmpw   $0x42,-0x2(%rdi,%rsi,2)
	jne    term2Fail
	mov    $term2Fail,%r11 
	mov    %r11,0x8(%rsp)  
	jmpq   term2Match
term2Match:
	mov    %esi,0x14(%rcx) 
	add    $0x10,%rsp
	mov    (%rcx),%eax
	mov    %esi,0x4(%rcx)  
	pop    %rbx	
	pop    %rbp	
	retq   
nextChar:
	mov    %rsi,%rax
	sub    $0x1,%eax
	mov    %eax,(%rcx)
	add    $0x1,%esi
	cmp    %edx,%esi
	jbe    loop:	
	add    $0x10,%rsp
	mov    $0xffffffff,%eax
	pop    %rbx	
	pop    %rbp	
	retq   
term2Fail:
	movl   $0xffffffff,0x10(%rcx)
	jmpq   *(%rsp)	
term1Fail:
	movl   $0xffffffff,0x8(%rcx)
	jmpq   nextChar