236753 - use __builtin_expect to provide branch prediction hints

Status: RESOLVED FIXED

(SeaMonkey :: General, defect)

Description

[Brian Ryner (not reading)]

We should use gcc's __builtin_expect() to provide branch prediction hints to the
compiler.  I'd propose adding macros NS_LIKELY and NS_UNLIKELY that wrap the
expression to be tested and give the expected value as 1 and 0 respectively.

And a good place to make use of this would be NS_SUCCEEDED and NS_FAILED.  We
should expect success.

Comment 1

[Brian Ryner (not reading)]

Attachment: patch

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 143229 [details] [diff] [review]
patch

r+sr=dbaron, but feel free to drop the outer set of parentheses on the
definitions of NS_FAILED and NS_SUCCEEDED.

Comment 3

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 143229 [details] [diff] [review]
patch

Also, your definitions of NS_LIKELY and NS_UNLIKELY should parenthesize the x
inside the macro.

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

(And you actually should be ok without the outer paretheses in the definitions
of NS_LIKELY and NS_UNLIKELY as well -- for the gcc3 definitions -- which is
what the previous comment was referring to as well.)

Comment 5

[Brian Ryner (not reading)]

checked in.

Comment 6

[Brian Ryner (not reading)]

Attachment: same for NS_ENSURE_TRUE

mark NS_ENSURE_TRUE's test as unlikely.  The only concern I have with this is
that you will get two __builtin_expect()s with NS_ENSURE_SUCCESS:

if (__builtin_expect(__builtin_expect(!(rv & 0x80000000), 0), 0))

but I don't think this causes a problem.