Bug status and Stage 3

[jvdelisle@verizon.net]

Tobias Burnus wrote:

> 37754 - [4.4 Reg.] READ I/O Performance
>   13% slower formatted reads than 4.3. I think Jerry has a look.
>   However, for performance regressions, having more eyes can never
>   harm as everyone finds different bottlenecks
>

As an update on this, I would appreciate more eyes.

Here is a summary of the delta in time for selected functions:

Function		         4.3	4.4	Delta
------------------              ---	---	-----
data_transfer_init	.42	.54	 .12 **
fflush			.20	.15	-.05
get_external_unit	.10	.14	 .03
finalize_transfer	.15	.12	-.03
next_char		.10	.12	 .02
fd_read				.11	  *
fd_alloc_r_at		.05		  *
memcpy				.06	  *

* It appears that removing fd_alloc_r_at in hopes of relying on the OS 
buffering has resulted in more costly use of fd_read and memcpy.

** The increase in data_transfer_init is suspicious, possibly due to the 
added checks for all the f2003 I/O features.

I will be looking more closely at data_tansfer_init first.

Regards,

Jerry