yuv2rgb_altivec.c

video linux conference

/*
  [e-mail]    March 8, 2004

  Altivec Acceleration for Color Space Conversion revision 0.2

  convert I420 YV12 to RGB in various formats,
    it rejects images that are not in 420 formats
    it rejects images that don't have widths of multiples of 16
    it rejects images that don't have heights of multiples of 2
  reject defers to C simulation codes.

  lots of optimizations to be done here

  1. need to fix saturation code, I just couldn't get it to fly with packs and adds.
     so we currently use max min to clip

  2. the inefficient use of chroma loading needs a bit of brushing up

  3. analysis of pipeline stalls needs to be done, use shark to identify pipeline stalls


  MODIFIED to calculate coeffs from currently selected color space.
  MODIFIED core to be a macro which you spec the output format.
  ADDED UYVY conversion which is never called due to some thing in SWSCALE.
  CORRECTED algorithim selection to be strict on input formats.
  ADDED runtime detection of altivec.

  ADDED altivec_yuv2packedX vertical scl + RGB converter

  March 27,2004
  PERFORMANCE ANALYSIS

  The C version use 25% of the processor or ~250Mips for D1 video rawvideo used as test
  The ALTIVEC version uses 10% of the processor or ~100Mips for D1 video same sequence

  720*480*30  ~10MPS

  so we have roughly 10clocks per pixel this is too high something has to be wrong.

  OPTIMIZED clip codes to utilize vec_max and vec_packs removing the need for vec_min.

  OPTIMIZED DST OUTPUT cache/dma controls. we are pretty much
  guaranteed to have the input video frame it was just decompressed so
  it probably resides in L1 caches.  However we are creating the
  output video stream this needs to use the DSTST instruction to
  optimize for the cache.  We couple this with the fact that we are
  not going to be visiting the input buffer again so we mark it Least
  Recently Used.  This shaves 25% of the processor cycles off.

  Now MEMCPY is the largest mips consumer in the system, probably due
  to the inefficient X11 stuff.

  GL libraries seem to be very slow on this machine 1.33Ghz PB running
  Jaguar, this is not the case for my 1Ghz PB.  I thought it might be
  a versioning issues, however i have libGL.1.2.dylib for both
  machines. ((We need to figure this out now))

  GL2 libraries work now with patch for RGB32

  NOTE quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor

  Integrated luma prescaling adjustment for saturation/contrast/brightness adjustment. 

*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <inttypes.h>
#include <assert.h>
#include "config.h"
#include "rgb2rgb.h"
#include "swscale.h"
#include "swscale_internal.h"
#include "mangle.h"
#include "img_format.h" //FIXME try to reduce dependency of such stuff

#undef PROFILE_THE_BEAST
#undef INC_SCALING

typedef unsigned char ubyte;
typedef signed char   sbyte;


/* RGB interleaver, 16 planar pels 8-bit samples per channel in
   homogeneous vector registers x0,x1,x2 are interleaved with the
   following technique:

      o0 = vec_mergeh (x0,x1);
      o1 = vec_perm (o0, x2, perm_rgb_0);
      o2 = vec_perm (o0, x2, perm_rgb_1);
      o3 = vec_mergel (x0,x1);
      o4 = vec_perm (o3,o2,perm_rgb_2);
      o5 = vec_perm (o3,o2,perm_rgb_3);

  perm_rgb_0:   o0(RG).h v1(B) --> o1*
              0   1  2   3   4
             rgbr|gbrg|brgb|rgbr
             0010 0100 1001 0010
             0102 3145 2673 894A

  perm_rgb_1:   o0(RG).h v1(B) --> o2
              0   1  2   3   4
             gbrg|brgb|bbbb|bbbb
             0100 1001 1111 1111
             B5CD 6EF7 89AB CDEF

  perm_rgb_2:   o3(RG).l o2(rgbB.l) --> o4*
              0   1  2   3   4
             gbrg|brgb|rgbr|gbrg
             1111 1111 0010 0100
             89AB CDEF 0182 3945

  perm_rgb_2:   o3(RG).l o2(rgbB.l) ---> o5*
              0   1  2   3   4
             brgb|rgbr|gbrg|brgb
             1001 0010 0100 1001
             a67b 89cA BdCD eEFf

*/
static
const vector unsigned char
  perm_rgb_0 = (vector unsigned char)(0x00,0x01,0x10,0x02,0x03,0x11,0x04,0x05,
				      0x12,0x06,0x07,0x13,0x08,0x09,0x14,0x0a),
  perm_rgb_1 = (vector unsigned char)(0x0b,0x15,0x0c,0x0d,0x16,0x0e,0x0f,0x17,
				      0x18,0x19,0x1a,0x1b,0x1c,0x1d,0x1e,0x1f),
  perm_rgb_2 = (vector unsigned char)(0x10,0x11,0x12,0x13,0x14,0x15,0x16,0x17,
				      0x00,0x01,0x18,0x02,0x03,0x19,0x04,0x05),
  perm_rgb_3 = (vector unsigned char)(0x1a,0x06,0x07,0x1b,0x08,0x09,0x1c,0x0a,
				      0x0b,0x1d,0x0c,0x0d,0x1e,0x0e,0x0f,0x1f);

#define vec_merge3(x2,x1,x0,y0,y1,y2)    \
do {					 \
  typeof(x0) o0,o2,o3;			 \
      o0 = vec_mergeh (x0,x1);		 \
      y0 = vec_perm (o0, x2, perm_rgb_0);\
      o2 = vec_perm (o0, x2, perm_rgb_1);\
      o3 = vec_mergel (x0,x1);		 \
      y1 = vec_perm (o3,o2,perm_rgb_2);	 \
      y2 = vec_perm (o3,o2,perm_rgb_3);	 \
} while(0)

#define vec_mstrgb24(x0,x1,x2,ptr)        \
do {					 \
  typeof(x0) _0,_1,_2;			 \
  vec_merge3 (x0,x1,x2,_0,_1,_2);	 \
  vec_st (_0, 0, ptr++);		 \
  vec_st (_1, 0, ptr++);		 \
  vec_st (_2, 0, ptr++);		 \
}  while (0);

#define vec_mstbgr24(x0,x1,x2,ptr)       \
do {					 \
  typeof(x0) _0,_1,_2;			 \
  vec_merge3 (x2,x1,x0,_0,_1,_2);	 \
  vec_st (_0, 0, ptr++);		 \
  vec_st (_1, 0, ptr++);		 \
  vec_st (_2, 0, ptr++);		 \
}  while (0);

/* pack the pixels in rgb0 format
   msb R
   lsb 0
*/
#define vec_mstrgb32(T,x0,x1,x2,x3,ptr)						       \
do {										       \
  T _0,_1,_2,_3;								       \
  _0 = vec_mergeh (x0,x1);							       \
  _1 = vec_mergeh (x2,x3);					    	  	       \
  _2 = (T)vec_mergeh ((vector unsigned short)_0,(vector unsigned short)_1);            \
  _3 = (T)vec_mergel ((vector unsigned short)_0,(vector unsigned short)_1);            \
  vec_st (_2, 0*16, (T *)ptr);						               \
  vec_st (_3, 1*16, (T *)ptr);						       	       \
  _0 = vec_mergel (x0,x1);							       \
  _1 = vec_mergel (x2,x3);						       	       \
  _2 = (T)vec_mergeh ((vector unsigned short)_0,(vector unsigned short)_1); 	       \
  _3 = (T)vec_mergel ((vector unsigned short)_0,(vector unsigned short)_1); 	       \
  vec_st (_2, 2*16, (T *)ptr);						       	       \
  vec_st (_3, 3*16, (T *)ptr);						       	       \
  ptr += 4;									       \
}  while (0);

/*

  | 1     0       1.4021   | | Y |
  | 1    -0.3441 -0.7142   |x| Cb|
  | 1     1.7718  0	   | | Cr|


  Y:      [-128 127]
  Cb/Cr : [-128 127]

  typical yuv conversion work on Y: 0-255 this version has been optimized for jpeg decode.

*/




#define vec_unh(x) \
  (vector signed short) \
    vec_perm(x,(typeof(x))(0),\
             (vector unsigned char)(0x10,0x00,0x10,0x01,0x10,0x02,0x10,0x03,\
                                    0x10,0x04,0x10,0x05,0x10,0x06,0x10,0x07))
#define vec_unl(x) \
  (vector signed short) \
    vec_perm(x,(typeof(x))(0),\
             (vector unsigned char)(0x10,0x08,0x10,0x09,0x10,0x0A,0x10,0x0B,\
                                    0x10,0x0C,0x10,0x0D,0x10,0x0E,0x10,0x0F))

#define vec_clip(x) \
  vec_max (vec_min (x, (typeof(x))(255)), (typeof(x))(0))

#define vec_packclp_a(x,y) \
  (vector unsigned char)vec_pack (vec_clip (x), vec_clip (y))

#define vec_packclp(x,y) \
  (vector unsigned char)vec_packs \
      ((vector unsigned short)vec_max (x,(vector signed short) (0)), \
       (vector unsigned short)vec_max (y,(vector signed short) (0)))

//#define out_pixels(a,b,c,ptr) vec_mstrgb32(typeof(a),((typeof (a))(0)),a,a,a,ptr)


static inline void cvtyuvtoRGB (SwsContext *c,
			   vector signed short Y, vector signed short U, vector signed short V,
			   vector signed short *R, vector signed short *G, vector signed short *B)
{
  vector signed   short vx,ux,uvx;

  Y = vec_mradds (Y, c->CY, c->OY);

  U = vec_sub (U,(vector signed short)(128));
  V = vec_sub (V,(vector signed short)(128));

  //   ux  = (CBU*(u<<c->CSHIFT)+0x4000)>>15;
  ux = vec_sl (U, c->CSHIFT);
  *B = vec_mradds (ux, c->CBU, Y);

  // vx  = (CRV*(v<<c->CSHIFT)+0x4000)>>15;
  vx = vec_sl (V, c->CSHIFT);
  *R = vec_mradds (vx, c->CRV, Y);

  // uvx = ((CGU*u) + (CGV*v))>>15;
  uvx = vec_mradds (U, c->CGU, Y);
  *G = vec_mradds (V, c->CGV, uvx);
}


/*
  ------------------------------------------------------------------------------
  CS converters
  ------------------------------------------------------------------------------
*/


#define DEFCSP420_CVT(name,out_pixels)                                     \
static int altivec_##name (SwsContext *c,                                  \
				unsigned char **in, int *instrides,	   \
				int srcSliceY,	int srcSliceH,		   \
				unsigned char **oplanes, int *outstrides)  \
{									   \
  int w = c->srcW;							   \
  int h = srcSliceH;							   \
  int i,j;								   \
  int instrides_scl[3];							   \
  vector unsigned char y0,y1;						   \
									   \
  vector signed char  u,v;						   \
									   \
  vector signed short Y0,Y1,Y2,Y3;					   \
  vector signed short U,V;						   \
  vector signed short vx,ux,uvx;					   \
  vector signed short vx0,ux0,uvx0;					   \
  vector signed short vx1,ux1,uvx1;					   \
  vector signed short R0,G0,B0;						   \
  vector signed short R1,G1,B1;						   \
  vector unsigned char R,G,B;						   \
									   \
  vector unsigned char *uivP, *vivP;			   		   \
  vector unsigned char align_perm;					   \
									   \
  vector signed short 							   \
    lCY  = c->CY,							   \
    lOY  = c->OY,							   \
    lCRV = c->CRV,							   \
    lCBU = c->CBU,							   \
    lCGU = c->CGU,							   \
    lCGV = c->CGV;							   \
									   \
  vector unsigned short lCSHIFT = c->CSHIFT;				   \
									   \
  ubyte *y1i   = in[0];							   \
  ubyte *y2i   = in[0]+w;						   \
  ubyte *ui    = in[1];							   \
  ubyte *vi    = in[2];							   \
									   \
  vector unsigned char *oute						   \
    = (vector unsigned char *)						   \
        (oplanes[0]+srcSliceY*outstrides[0]);				   \
  vector unsigned char *outo						   \
    = (vector unsigned char *)						   \
        (oplanes[0]+srcSliceY*outstrides[0]+outstrides[0]);		   \
									   \
									   \
  instrides_scl[0] = instrides[0];					   \
  instrides_scl[1] = instrides[1]-w/2;  /* the loop moves ui by w/2 */	   \
  instrides_scl[2] = instrides[2]-w/2;  /* the loop moves vi by w/2 */	   \
									   \
									   \
  for (i=0;i<h/2;i++) {							   \
    vec_dstst (outo, (0x02000002|(((w*3+32)/32)<<16)), 0);                 \
    vec_dstst (oute, (0x02000002|(((w*3+32)/32)<<16)), 1);                 \
									   \
    for (j=0;j<w/16;j++) {						   \
									   \
      y0 = vec_ldl (0,y1i);						   \
      y1 = vec_ldl (0,y2i);						   \
      uivP = (vector unsigned char *)ui;				   \
      vivP = (vector unsigned char *)vi;				   \
									   \
      align_perm = vec_lvsl (0, ui);					   \
      u = (vector signed char)vec_perm (uivP[0], uivP[1], align_perm);	   \
									   \
      align_perm = vec_lvsl (0, vi);					   \
      v = (vector signed char)vec_perm (vivP[0], vivP[1], align_perm);	   \
									   \
      u  = (vector signed char)vec_sub (u, (vector signed char)(128));	   \
      v  = (vector signed char)vec_sub (v, (vector signed char)(128));	   \
      U  = vec_unpackh (u);						   \
      V  = vec_unpackh (v);						   \
									   \
									   \
	Y0 = vec_unh (y0);						   \
	Y1 = vec_unl (y0);						   \
	Y2 = vec_unh (y1);						   \
	Y3 = vec_unl (y1);						   \
									   \
        Y0 = vec_mradds (Y0, lCY, lOY);					   \
        Y1 = vec_mradds (Y1, lCY, lOY);					   \
        Y2 = vec_mradds (Y2, lCY, lOY);					   \
        Y3 = vec_mradds (Y3, lCY, lOY);					   \
									   \
	/*   ux  = (CBU*(u<<CSHIFT)+0x4000)>>15 */			   \
	ux = vec_sl (U, lCSHIFT);					   \
	ux = vec_mradds (ux, lCBU, (vector signed short)(0));		   \
	ux0  = vec_mergeh (ux,ux);					   \
	ux1  = vec_mergel (ux,ux);					   \
									   \
	/* vx  = (CRV*(v<<CSHIFT)+0x4000)>>15;	*/			   \
	vx = vec_sl (V, lCSHIFT);					   \
	vx = vec_mradds (vx, lCRV, (vector signed short)(0));		   \
	vx0  = vec_mergeh (vx,vx);					   \
	vx1  = vec_mergel (vx,vx);					   \
									   \
	/* uvx = ((CGU*u) + (CGV*v))>>15 */				   \
	uvx = vec_mradds (U, lCGU, (vector signed short)(0));		   \
	uvx = vec_mradds (V, lCGV, uvx);				   \
	uvx0 = vec_mergeh (uvx,uvx);					   \
	uvx1 = vec_mergel (uvx,uvx);					   \
									   \
	R0 = vec_add (Y0,vx0);						   \
	G0 = vec_add (Y0,uvx0);						   \
	B0 = vec_add (Y0,ux0);						   \
	R1 = vec_add (Y1,vx1);						   \
	G1 = vec_add (Y1,uvx1);						   \
	B1 = vec_add (Y1,ux1);						   \
									   \
	R  = vec_packclp (R0,R1);					   \
	G  = vec_packclp (G0,G1);					   \
	B  = vec_packclp (B0,B1);					   \
									   \
	out_pixels(R,G,B,oute);						   \
									   \
	R0 = vec_add (Y2,vx0);						   \
	G0 = vec_add (Y2,uvx0);						   \
	B0 = vec_add (Y2,ux0);						   \
	R1 = vec_add (Y3,vx1);						   \
	G1 = vec_add (Y3,uvx1);						   \
	B1 = vec_add (Y3,ux1);						   \
	R  = vec_packclp (R0,R1);					   \
	G  = vec_packclp (G0,G1);					   \
	B  = vec_packclp (B0,B1);					   \
									   \
									   \
	out_pixels(R,G,B,outo);						   \
									   \
      y1i  += 16;							   \
      y2i  += 16;							   \
      ui   += 8;							   \
      vi   += 8;							   \
									   \
    }									   \
									   \
    outo += (outstrides[0])>>4;					           \
    oute += (outstrides[0])>>4;					           \
									   \
    ui    += instrides_scl[1];						   \
    vi    += instrides_scl[2];						   \
    y1i   += instrides_scl[0];						   \
    y2i   += instrides_scl[0];						   \