[patch] SSAR opts

Jules Bergmann jules at codesourcery.com
Wed Sep 17 21:12:42 UTC 2008


This makes a couple of optimizations to the by-function digital 
spotlighting:

  - avoid copying the result of an in-place FFT.

	If fft_ is a by-reference Fft, then

		B = fft_(A)
		use(B)

	can be replaced by

		fft_(A)
		use(A)

	Likewise, A = fft_(A) can be simplified to fft_(A)

  - avoid mmul between matrices with different dimension order

	If RM is a row-major matrix, and CM is a col-major matrix,
	the following cannot be dispatched easily to the SPUs:

		RM_Z = CM_A * RM_B;

	Instead, its better to transpose first, the multiply:

		RM_A = CM_A;	// transpose
		RM_Z = RM_A * RM_B;

It also adds a few more scope events.

Ok to apply?

				-- Jules


-- 
Jules Bergmann
CodeSourcery
jules at codesourcery.com
(650) 331-3385 x705
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: ssar.diff
URL: <http://sourcerytools.com/pipermail/vsipl++/attachments/20080917/b403fd56/attachment.ksh>


More information about the vsipl++ mailing list