[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Sc-devel] [BUG] 1.0 + 0.0 = 0.0?

On Dec 5, 2007, at 6:25 AM, Fredrik Olofsson wrote:

i just tested both.  ppc fine, intel signal disappears.
denormals issue again it seems.

I don't think it's denormals. If you trace the synth (which is the first thing you should do whenever you get strange behavior), you see that the problem is in MulAdd.

  unit 6 MulAdd
    in  0.00817199 1 0
    out 0

... where the output should be 0.00817199 * 1 + 0 == 0.00817199.

I'm inclined to suspect ampmix_kk in the source code, since it has a number of branches and one of those branches clears the output (sets all output samples to 0). PPC is probably using the vectorized function while Intel uses the non vector code.

Might be interesting to tweak a PPC build not to use vector opcodes and see if the same behavior occurs.


: H. James Harkins

: jamshark70@xxxxxxxxxxxxxxxxx

: http://www.dewdrop-world.net


"Come said the Muse,

Sing me a song no poet has yet chanted,

Sing me the universal."  -- Whitman