OMAP-L138 LCDK, DSPLIB DSPF_sp_cfftr2_dit with Inverse

When using the assembly-optimized dsplib library on the OMAP-L138 LCDK, I encounter a quirk with subsequent calls to DSPF_sp_cfftr2_dit and DSPF_sp_icfftr2_dif. The API states that the twiddle factors should be bit-reversed; however upon performing this action to both the transform and inverse twiddle factors, there is distortion present in an N=32 sized transform. If the twiddle factors are passed non-bit-reversed in back-to-back transform/inverse calls, this distortion is not present. Why might this be the case? Code snippet is as follows, for context:

FFT operations within ISR:

    // Extract data (complex) from input framebuffer
    for (i=0; i < BUFFER_COUNT; i++) {
    	// Populate input sample arrays
    	inpLcpx[2*i] = Left[i];
    	inpLcpx[2*i+1] = 0.0;
    	inpRcpx[2*i] = Right[i];
    	inpRcpx[2*i+1] = 0.0;
    }

    // Calculate FFT using call to optimized-assembly DSPF_sp_cffrt2_dit
    DSPF_sp_cfftr2_dit(inpLcpx, twiddle, BUFFER_COUNT);
    DSPF_sp_cfftr2_dit(inpRcpx, twiddle, BUFFER_COUNT);

    // Calculate IFFT using call to optimized-assembly DSPF_sp_icfftr2_dif
    DSPF_sp_icfftr2_dif(inpLcpx, twiddleInv, BUFFER_COUNT);
    DSPF_sp_icfftr2_dif(inpRcpx, twiddleInv, BUFFER_COUNT);

    // Then pack real data into output framebuffer

Twiddle factors are generated as follows. Call is to method defined within ISR, from main:

void twidInit()
{
	int i=0;
	float a = 2.0f*PI/BUFFER_COUNT;
	for (i=0; i < BUFFER_COUNT; i+=2) {
    	     // Populate twiddle factors
             twiddle[i] = cosf(-(i/2)*a);
             twiddle[i+1] = -sinf(-(i/2)*a);
             twiddleInv[i] = cosf((i/2)*a);
             twiddleInv[i+1] = -sinf(-(i/2)*a);
	}

}