michael@0: If you are reading this, it means you think you may be interested in using the SIMD extensions in kissfft 
michael@0: to do 4 *separate* FFTs at once.
michael@0: 
michael@0: Beware! Beyond here there be dragons!
michael@0: 
michael@0: This API is not easy to use, is not well documented, and breaks the KISS principle.  
michael@0: 
michael@0: 
michael@0: Still reading? Okay, you may get rewarded for your patience with a considerable speedup 
michael@0: (2-3x) on intel x86 machines with SSE if you are willing to jump through some hoops.
michael@0: 
michael@0: The basic idea is to use the packed 4 float __m128 data type as a scalar element.  
michael@0: This means that the format is pretty convoluted. It performs 4 FFTs per fft call on signals A,B,C,D.
michael@0: 
michael@0: For complex data, the data is interlaced as follows:
michael@0: rA0,rB0,rC0,rD0,      iA0,iB0,iC0,iD0,   rA1,rB1,rC1,rD1, iA1,iB1,iC1,iD1 ...
michael@0: where "rA0" is the real part of the zeroth sample for signal A
michael@0: 
michael@0: Real-only data is laid out:
michael@0: rA0,rB0,rC0,rD0,     rA1,rB1,rC1,rD1,      ... 
michael@0: 
michael@0: Compile with gcc flags something like
michael@0: -O3 -mpreferred-stack-boundary=4  -DUSE_SIMD=1 -msse 
michael@0: 
michael@0: Be aware of SIMD alignment.  This is the most likely cause of segfaults.  
michael@0: The code within kissfft uses scratch variables on the stack.  
michael@0: With SIMD, these must have addresses on 16 byte boundaries.  
michael@0: Search on "SIMD alignment" for more info.
michael@0: 
michael@0: 
michael@0: 
michael@0: Robin at Divide Concept was kind enough to share his code for formatting to/from the SIMD kissfft.  
michael@0: I have not run it -- use it at your own risk.  It appears to do 4xN and Nx4 transpositions 
michael@0: (out of place).
michael@0: 
michael@0: void SSETools::pack128(float* target, float* source, unsigned long size128)
michael@0: {
michael@0:    __m128* pDest = (__m128*)target;
michael@0:    __m128* pDestEnd = pDest+size128;
michael@0:    float* source0=source;
michael@0:    float* source1=source0+size128;
michael@0:    float* source2=source1+size128;
michael@0:    float* source3=source2+size128;
michael@0: 
michael@0:    while(pDest<pDestEnd)
michael@0:    {
michael@0:        *pDest=_mm_set_ps(*source3,*source2,*source1,*source0);
michael@0:        source0++;
michael@0:        source1++;
michael@0:        source2++;
michael@0:        source3++;
michael@0:        pDest++;
michael@0:    }
michael@0: }
michael@0: 
michael@0: void SSETools::unpack128(float* target, float* source, unsigned long size128)
michael@0: {
michael@0: 
michael@0:    float* pSrc = source;
michael@0:    float* pSrcEnd = pSrc+size128*4;
michael@0:    float* target0=target;
michael@0:    float* target1=target0+size128;
michael@0:    float* target2=target1+size128;
michael@0:    float* target3=target2+size128;
michael@0: 
michael@0:    while(pSrc<pSrcEnd)
michael@0:    {
michael@0:        *target0=pSrc[0];
michael@0:        *target1=pSrc[1];
michael@0:        *target2=pSrc[2];
michael@0:        *target3=pSrc[3];
michael@0:        target0++;
michael@0:        target1++;
michael@0:        target2++;
michael@0:        target3++;
michael@0:        pSrc+=4;
michael@0:    }
michael@0: }