#52 MIPS DSPr2 optimization for quantize_float and convsamp_float routines

[Teodora Novkovic]

The attached archive file includes the following:
1. patch with MIPS DSPR2 optimization for quantize_float and convsamp_float routines
2. txt file with cjpeg profiling test results
This code is bit-exact and 'make test' passes on linux.

I was not able to call these float routine from tjbench (dotest script) by using available options, so I don't have spreadsheet with results. Could you please give me additional information how to test these two routines with tjbench? Thanks in advance.

[DRC]

Patch does not apply correctly. Please reformat.

Note that floating point DCT/IDCT is not exercised by tjbench, but very few people use those routines, so I don't really track the performance of them.

[Teodora Novkovic]

Here is the new patch for revision 1057.

[DRC]

Committed. Thanks.