Commit 9f38e9b1 authored by thomas dettbarn's avatar thomas dettbarn
Browse files

broken down the headroom shifts a little bit more.

parent bab21b8c

lib_dec/ivas_svd_dec_fx.c

+15 −13
Original line number Diff line number Diff line
@@ -1003,9 +1003,10 @@ static void biDiagonalReductionLeft_64( 
    Word16 *g_e )

{



#define HEADROOM_LEFT_1 1

#define HEADROOM_LEFT_1 2

#define HEADROOM_LEFT_2 2

#define HEADROOM_LEFT_3 2

#define HEADROOM_LEFT_3 15

#define HEADROOM_LEFT_4 15



    Word16 iCh, jCh;

    Word32 norm_x;
@@ -1075,17 +1076,17 @@ static void biDiagonalReductionLeft_64( 
            norm_64 = 0;

            for ( jCh = currChannel; jCh < nChannelsL; jCh++ )

            {

                factor1 = W_extract_l( W_shr( singularVectors_Left_64[jCh][currChannel], HEADROOM_LEFT_2 ) );

                factor2 = W_extract_l( W_shr( singularVectors_Left_64[jCh][iCh], HEADROOM_LEFT_2 ) );

                factor1 = W_extract_l( W_shr( singularVectors_Left_64[jCh][currChannel], HEADROOM_LEFT_3 ) );

                factor2 = W_extract_l( W_shr( singularVectors_Left_64[jCh][iCh], HEADROOM_LEFT_3 ) );

                norm_64 = W_add( norm_64, W_mult0_32_32( factor1, factor2 ) );

            }

            norm_x_e = W_norm( norm_64 );

            norm_x = W_extract_h( W_shl( norm_64, norm_x_e ) );

            f = Mpy_32_32( norm_x, invVal );

            magic_shift = norm_x_e - r_e + ( 28 - HEADROOM_LEFT_3 ); // works with headroom_left: 1,2,2

            magic_shift = norm_x_e - r_e + ( 32 + 2*HEADROOM_LEFT_1 - HEADROOM_LEFT_2 -2*HEADROOM_LEFT_3 - HEADROOM_LEFT_4 );

            FOR( jCh = currChannel; jCh < nChannelsL; jCh++ )

            {

                factor1 = W_extract_h( W_shl( singularVectors_Left_64[jCh][currChannel], 32 - HEADROOM_LEFT_3 ) );

                factor1 = W_extract_h( W_shl( singularVectors_Left_64[jCh][currChannel], 32 - HEADROOM_LEFT_4 ) );

                singularVectors_Left_64[jCh][iCh] = W_add( singularVectors_Left_64[jCh][iCh], W_shr( W_mult0_32_32( f, factor1 ), magic_shift ) );

            }

        }
@@ -1114,7 +1115,8 @@ static void biDiagonalReductionRight_64( 


#define HEADROOM_RIGHT_1 2

#define HEADROOM_RIGHT_2 2

#define HEADROOM_RIGHT_3 2

#define HEADROOM_RIGHT_3 15

#define HEADROOM_RIGHT_4 15





    ( *g ) = 0;
@@ -1159,8 +1161,8 @@ static void biDiagonalReductionRight_64( 
            }

            move32();

            move16();

            factor2 = W_extract_l( W_shr( singularVectors_Left_64[currChannel][idx], HEADROOM_RIGHT_1 ) );

            tmp_e = sub( HEADROOM_RIGHT_1, *g_e );

            factor2 = W_extract_l( W_shr( singularVectors_Left_64[currChannel][idx], HEADROOM_RIGHT_2 ) );

            tmp_e = sub( HEADROOM_RIGHT_2, *g_e );

            tmpmul = W_mult0_32_32( *g, factor2 );

            tmpmul = W_shr( tmpmul, tmp_e );

            r_64 = W_sub( tmpmul, norm_64 );
@@ -1181,19 +1183,19 @@ static void biDiagonalReductionRight_64( 
                move64();

                FOR( jCh = idx; jCh < nChannelsC; jCh++ )

                {

                    factor1 = W_extract_l( W_shr( singularVectors_Left_64[iCh][jCh], HEADROOM_RIGHT_2 ) );

                    factor2 = W_extract_l( W_shr( singularVectors_Left_64[currChannel][jCh], HEADROOM_RIGHT_2 ) );

                    factor1 = W_extract_l( W_shr( singularVectors_Left_64[iCh][jCh], HEADROOM_RIGHT_3 ) );

                    factor2 = W_extract_l( W_shr( singularVectors_Left_64[currChannel][jCh], HEADROOM_RIGHT_3 ) );

                    norm_64 = W_add( norm_64, W_mult0_32_32( factor1, factor2 ) );

                }



                norm_x_e = W_norm( norm_64 );

                norm_x = W_extract_h( W_shl( norm_64, norm_x_e ) );

                f = Mpy_32_32( norm_x, invVal );

                magic_shift = norm_x_e - r_e + 29 - ( HEADROOM_RIGHT_3 ); // works with headroom_right: 2,2,2

                magic_shift = norm_x_e - r_e + ( 32 + 2*HEADROOM_RIGHT_1 - HEADROOM_RIGHT_2 -2*HEADROOM_RIGHT_3 - HEADROOM_RIGHT_4 );



                FOR( jCh = idx; jCh < nChannelsC; jCh++ )

                {

                    factor2 = W_extract_h( W_shl( singularVectors_Left_64[currChannel][jCh], 32 - HEADROOM_RIGHT_3 ) );

                    factor2 = W_extract_h( W_shl( singularVectors_Left_64[currChannel][jCh], 32 - HEADROOM_RIGHT_4 ) );

                    singularVectors_Left_64[iCh][jCh] = W_add( singularVectors_Left_64[iCh][jCh], W_shr( W_mult0_32_32( f, factor2 ), magic_shift ) );

                }

            }