amf_delayntapoffchip.c

ADI SHARC DSP 音频算法标准模块库

// Copyright(c) 2005 Analog Devices, Inc. All Rights Reserved. ADI Confidential.

#include "AMF_DelayNTapOffChip.h"

void AMF_DelayNTapOffChip_Render(AMF_DelayNTapOffChip * restrict instance,float * restrict * buffers,int tickSize);

#pragma optimize_for_speed
SEG_MOD_FAST_CODE  void AMF_DelayNTapOffChip_Render(AMF_DelayNTapOffChip * restrict instance,float * restrict * buffers,int tickSize) {
    int delaySize = instance->delaySize; // length in samples of delay buffer
    float *delayBuffer = &instance->delayBuffer[0];		 // delay buffer
    int writeIndex = instance->delayWritePtr-delayBuffer;	// delay index
    int *delays = &instance->delays[0];
    float *amps = &instance->amps[0];
    int numTaps = instance->numTaps;
    int i, j;
    float *in = buffers[0]; // Don't use restrict because we say they can alias (see .h file)
    float *out = buffers[1];
	float pm *bufPtr;
	float *readBufPtr;
    float pm *tmp = ((float pm *)((int)buffers[2]));
	
    //// write loop
/*   	// weird kludge to get the compiler to actually set the B0 register...
	bufPtr = (float pm *)((int)delayBuffer);
	bufPtr = (float pm *)__builtin_circptr(bufPtr, writeIndex, (float pm *)((int)delayBuffer), delaySize);		
#pragma loop_count(8,128,8)
#pragma no_alias
	for (i=0; i<tickSize; i++) {
		*bufPtr = in[i];
		bufPtr = (float pm *)__builtin_circptr(bufPtr, 1, (float pm *)((int)delayBuffer), delaySize);		
    } */

asm("	R0 = dm(I0,M6);\
	LCNTR=%4, do (pc,1) until lce;\
		R0 = dm(I0,M6), pm(I12,M14)=R0;"\
	:"+I12"(instance->delayWritePtr) \
	:"B12"(delayBuffer),"L12"(delaySize),"#I0"(in),"d"(tickSize) \
	:"R0","I0");
		
	// clear output vector
#pragma loop_count(8,128,8)
	for(i=0;i<tickSize;i++) 
		out[i] = 0.0;
		
    //// tap loop
#pragma loop_count(1,256,1)
	for(j=0;j<numTaps;j++) {
        float amp;
   
        //// read from external memory
/*		readBufPtr = (float *)((int)delayBuffer);
		readBufPtr = (float *)__builtin_circptr(readBufPtr, writeIndex-delays[j], (float *)((int)delayBuffer), delaySize);
#pragma loop_count(8,128,8)
#pragma no_alias
        for (i=0; i<tickSize; i++)
        {
            tmp[i] = *readBufPtr;
			readBufPtr = (float *)__builtin_circptr(readBufPtr, 1, (float *)((int)delayBuffer), delaySize);		
        }*/

asm("modify(i11,%5);\
		R0=pm(I11,M14);\
	LCNTR=%4, do (pc,1) until lce;\
		dm(I0,M6)=R0, R0=pm(I11,M14);"\
	: \
	:"#I11"(delayBuffer), "B11"(delayBuffer),"L11"(delaySize),"#I0"(tmp),"d"(tickSize),"z"(writeIndex-delays[j]) \
	:"R0","I0","I11");
          
		//// accumulate
        amp = amps[j];
#pragma loop_count(8,128,8)
#pragma vector_for
#pragma no_alias
		for (i=0; i<tickSize; i++) {
			out[i] += tmp[i] * amp;
    	} 
    }
    
//    instance->delayWritePtr = bufPtr;
}




SEG_MOD_SLOW_CONST const AMF_ModuleClass AMFClassDelayNTapOffChip = {
	/* Flags */
	0,

	/** Reference to render function. */
    (AMF_RenderFunction)AMF_DelayNTapOffChip_Render,  // render function 

    /* Default bypass */
	(void *)0,
    
	/* Input descriptor - 1 input, and it is mono. */
	1, 0,

	/* Output descriptor - 1 output, and it is mono. */
    1, 0
};