/* Store a vector to an unaligned location in memory */
static inline void
StoreUnaligned (vector unsigned char v,
const guchar *where)
{
if ((unsigned long)where & 0x0f)
{
/* Load the surrounding area */
vector unsigned char low = vec_ld(0, where);
vector unsigned char high = vec_ld(16, where);
/* Prepare the constants that we need */
vector unsigned char permuteVector = vec_lvsr(0, where);
vector signed char oxFF = vec_splat_s8(-1);
vector signed char ox00 = vec_splat_s8(0);
/* Make a mask for which parts of the vectors to swap out */
vector unsigned char mask = (vector unsigned char)vec_perm(ox00, oxFF, permuteVector);
v = vec_perm(v, v, permuteVector);
/* Insert our data into the low and high vectors */
low = vec_sel(low, v, mask);
high = vec_sel(v, high, mask);
/* Store the two aligned result vectors */
vec_st(low, 0, CONST_BUFFER(where));
vec_st(high, 16, CONST_BUFFER(where));
}
else
{ /* prevent overflow */
vec_st(v, 0, CONST_BUFFER(where));
}
}
static void ProjectDlightTexture_altivec( void ) {
int i, l;
vec_t origin0, origin1, origin2;
float texCoords0, texCoords1;
vector float floatColorVec0, floatColorVec1;
vector float modulateVec, colorVec, zero;
vector short colorShort;
vector signed int colorInt;
vector unsigned char floatColorVecPerm, modulatePerm, colorChar;
vector unsigned char vSel = VECCONST_UINT8(0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff);
float *texCoords;
byte *colors;
byte clipBits[SHADER_MAX_VERTEXES];
float texCoordsArray[SHADER_MAX_VERTEXES][2];
byte colorArray[SHADER_MAX_VERTEXES][4];
unsigned hitIndexes[SHADER_MAX_INDEXES];
int numIndexes;
float scale;
float radius;
vec3_t floatColor;
float modulate = 0.0f;
if ( !backEnd.refdef.num_dlights ) {
return;
}
// There has to be a better way to do this so that floatColor
// and/or modulate are already 16-byte aligned.
floatColorVecPerm = vec_lvsl(0,(float *)floatColor);
modulatePerm = vec_lvsl(0,(float *)&modulate);
modulatePerm = (vector unsigned char)vec_splat((vector unsigned int)modulatePerm,0);
zero = (vector float)vec_splat_s8(0);
for ( l = 0 ; l < backEnd.refdef.num_dlights ; l++ ) {
dlight_t *dl;
if ( !( tess.dlightBits & ( 1 << l ) ) ) {
continue; // this surface definately doesn't have any of this light
}
texCoords = texCoordsArray[0];
colors = colorArray[0];
dl = &backEnd.refdef.dlights[l];
origin0 = dl->transformed[0];
origin1 = dl->transformed[1];
origin2 = dl->transformed[2];
radius = dl->radius;
scale = 1.0f / radius;
if(r_greyscale->integer)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = floatColor[1] = floatColor[2] = luminance;
}
else if(r_greyscale->value)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
}
else
{
floatColor[0] = dl->color[0] * 255.0f;
floatColor[1] = dl->color[1] * 255.0f;
floatColor[2] = dl->color[2] * 255.0f;
}
floatColorVec0 = vec_ld(0, floatColor);
floatColorVec1 = vec_ld(11, floatColor);
floatColorVec0 = vec_perm(floatColorVec0,floatColorVec0,floatColorVecPerm);
for ( i = 0 ; i < tess.numVertexes ; i++, texCoords += 2, colors += 4 ) {
int clip = 0;
vec_t dist0, dist1, dist2;
dist0 = origin0 - tess.xyz[i][0];
dist1 = origin1 - tess.xyz[i][1];
dist2 = origin2 - tess.xyz[i][2];
backEnd.pc.c_dlightVertexes++;
texCoords0 = 0.5f + dist0 * scale;
texCoords1 = 0.5f + dist1 * scale;
if( !r_dlightBacks->integer &&
// dist . tess.normal[i]
( dist0 * tess.normal[i][0] +
dist1 * tess.normal[i][1] +
dist2 * tess.normal[i][2] ) < 0.0f ) {
clip = 63;
} else {
if ( texCoords0 < 0.0f ) {
clip |= 1;
} else if ( texCoords0 > 1.0f ) {
clip |= 2;
}
if ( texCoords1 < 0.0f ) {
clip |= 4;
} else if ( texCoords1 > 1.0f ) {
clip |= 8;
}
texCoords[0] = texCoords0;
texCoords[1] = texCoords1;
// modulate the strength based on the height and color
if ( dist2 > radius ) {
clip |= 16;
modulate = 0.0f;
} else if ( dist2 < -radius ) {
clip |= 32;
modulate = 0.0f;
} else {
dist2 = Q_fabs(dist2);
if ( dist2 < radius * 0.5f ) {
modulate = 1.0f;
} else {
modulate = 2.0f * (radius - dist2) * scale;
}
}
}
clipBits[i] = clip;
modulateVec = vec_ld(0,(float *)&modulate);
modulateVec = vec_perm(modulateVec,modulateVec,modulatePerm);
colorVec = vec_madd(floatColorVec0,modulateVec,zero);
colorInt = vec_cts(colorVec,0); // RGBx
colorShort = vec_pack(colorInt,colorInt); // RGBxRGBx
colorChar = vec_packsu(colorShort,colorShort); // RGBxRGBxRGBxRGBx
colorChar = vec_sel(colorChar,vSel,vSel); // RGBARGBARGBARGBA replace alpha with 255
vec_ste((vector unsigned int)colorChar,0,(unsigned int *)colors); // store color
}
// build a list of triangles that need light
numIndexes = 0;
for ( i = 0 ; i < tess.numIndexes ; i += 3 ) {
int a, b, c;
a = tess.indexes[i];
b = tess.indexes[i+1];
c = tess.indexes[i+2];
if ( clipBits[a] & clipBits[b] & clipBits[c] ) {
continue; // not lighted
}
hitIndexes[numIndexes] = a;
hitIndexes[numIndexes+1] = b;
hitIndexes[numIndexes+2] = c;
numIndexes += 3;
}
if ( !numIndexes ) {
continue;
}
qglEnableClientState( GL_TEXTURE_COORD_ARRAY );
qglTexCoordPointer( 2, GL_FLOAT, 0, texCoordsArray[0] );
qglEnableClientState( GL_COLOR_ARRAY );
qglColorPointer( 4, GL_UNSIGNED_BYTE, 0, colorArray );
GL_Bind( tr.dlightImage );
// include GLS_DEPTHFUNC_EQUAL so alpha tested surfaces don't add light
// where they aren't rendered
if ( dl->additive ) {
GL_State( GLS_SRCBLEND_ONE | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
else {
GL_State( GLS_SRCBLEND_DST_COLOR | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
}
}
vector signed char
testsc_1 ()
{
return vec_splat_s8 (5);
}
vector signed char
testsc_3 ()
{
return vec_splat_s8 (15);
}
vector signed char
testsc_2 ()
{
return vec_splat_s8 (-5);
}
static int dct_quantize_altivec(MpegEncContext* s,
DCTELEM* data, int n,
int qscale, int* overflow)
{
int lastNonZero;
vector float row0, row1, row2, row3, row4, row5, row6, row7;
vector float alt0, alt1, alt2, alt3, alt4, alt5, alt6, alt7;
const vector float zero = (const vector float)FOUROF(0.);
// used after quantize step
int oldBaseValue = 0;
// Load the data into the row/alt vectors
{
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
data0 = vec_ld(0, data);
data1 = vec_ld(16, data);
data2 = vec_ld(32, data);
data3 = vec_ld(48, data);
data4 = vec_ld(64, data);
data5 = vec_ld(80, data);
data6 = vec_ld(96, data);
data7 = vec_ld(112, data);
// Transpose the data before we start
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
// load the data into floating point vectors. We load
// the high half of each row into the main row vectors
// and the low half into the alt vectors.
row0 = vec_ctf(vec_unpackh(data0), 0);
alt0 = vec_ctf(vec_unpackl(data0), 0);
row1 = vec_ctf(vec_unpackh(data1), 0);
alt1 = vec_ctf(vec_unpackl(data1), 0);
row2 = vec_ctf(vec_unpackh(data2), 0);
alt2 = vec_ctf(vec_unpackl(data2), 0);
row3 = vec_ctf(vec_unpackh(data3), 0);
alt3 = vec_ctf(vec_unpackl(data3), 0);
row4 = vec_ctf(vec_unpackh(data4), 0);
alt4 = vec_ctf(vec_unpackl(data4), 0);
row5 = vec_ctf(vec_unpackh(data5), 0);
alt5 = vec_ctf(vec_unpackl(data5), 0);
row6 = vec_ctf(vec_unpackh(data6), 0);
alt6 = vec_ctf(vec_unpackl(data6), 0);
row7 = vec_ctf(vec_unpackh(data7), 0);
alt7 = vec_ctf(vec_unpackl(data7), 0);
}
// The following block could exist as a separate an altivec dct
// function. However, if we put it inline, the DCT data can remain
// in the vector local variables, as floats, which we'll use during the
// quantize step...
{
const vector float vec_0_298631336 = (vector float)FOUROF(0.298631336f);
const vector float vec_0_390180644 = (vector float)FOUROF(-0.390180644f);
const vector float vec_0_541196100 = (vector float)FOUROF(0.541196100f);
const vector float vec_0_765366865 = (vector float)FOUROF(0.765366865f);
const vector float vec_0_899976223 = (vector float)FOUROF(-0.899976223f);
const vector float vec_1_175875602 = (vector float)FOUROF(1.175875602f);
const vector float vec_1_501321110 = (vector float)FOUROF(1.501321110f);
const vector float vec_1_847759065 = (vector float)FOUROF(-1.847759065f);
const vector float vec_1_961570560 = (vector float)FOUROF(-1.961570560f);
const vector float vec_2_053119869 = (vector float)FOUROF(2.053119869f);
const vector float vec_2_562915447 = (vector float)FOUROF(-2.562915447f);
const vector float vec_3_072711026 = (vector float)FOUROF(3.072711026f);
int whichPass, whichHalf;
for(whichPass = 1; whichPass<=2; whichPass++) {
for(whichHalf = 1; whichHalf<=2; whichHalf++) {
vector float tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
vector float tmp10, tmp11, tmp12, tmp13;
vector float z1, z2, z3, z4, z5;
tmp0 = vec_add(row0, row7); // tmp0 = dataptr[0] + dataptr[7];
tmp7 = vec_sub(row0, row7); // tmp7 = dataptr[0] - dataptr[7];
tmp3 = vec_add(row3, row4); // tmp3 = dataptr[3] + dataptr[4];
tmp4 = vec_sub(row3, row4); // tmp4 = dataptr[3] - dataptr[4];
tmp1 = vec_add(row1, row6); // tmp1 = dataptr[1] + dataptr[6];
tmp6 = vec_sub(row1, row6); // tmp6 = dataptr[1] - dataptr[6];
tmp2 = vec_add(row2, row5); // tmp2 = dataptr[2] + dataptr[5];
tmp5 = vec_sub(row2, row5); // tmp5 = dataptr[2] - dataptr[5];
tmp10 = vec_add(tmp0, tmp3); // tmp10 = tmp0 + tmp3;
tmp13 = vec_sub(tmp0, tmp3); // tmp13 = tmp0 - tmp3;
tmp11 = vec_add(tmp1, tmp2); // tmp11 = tmp1 + tmp2;
tmp12 = vec_sub(tmp1, tmp2); // tmp12 = tmp1 - tmp2;
// dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
row0 = vec_add(tmp10, tmp11);
// dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
row4 = vec_sub(tmp10, tmp11);
// z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
z1 = vec_madd(vec_add(tmp12, tmp13), vec_0_541196100, (vector float)zero);
// dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
// CONST_BITS-PASS1_BITS);
row2 = vec_madd(tmp13, vec_0_765366865, z1);
// dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
// CONST_BITS-PASS1_BITS);
row6 = vec_madd(tmp12, vec_1_847759065, z1);
z1 = vec_add(tmp4, tmp7); // z1 = tmp4 + tmp7;
z2 = vec_add(tmp5, tmp6); // z2 = tmp5 + tmp6;
z3 = vec_add(tmp4, tmp6); // z3 = tmp4 + tmp6;
z4 = vec_add(tmp5, tmp7); // z4 = tmp5 + tmp7;
// z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
z5 = vec_madd(vec_add(z3, z4), vec_1_175875602, (vector float)zero);
// z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z3 = vec_madd(z3, vec_1_961570560, z5);
// z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z4 = vec_madd(z4, vec_0_390180644, z5);
// The following adds are rolled into the multiplies above
// z3 = vec_add(z3, z5); // z3 += z5;
// z4 = vec_add(z4, z5); // z4 += z5;
// z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
// Wow! It's actually more efficient to roll this multiply
// into the adds below, even thought the multiply gets done twice!
// z2 = vec_madd(z2, vec_2_562915447, (vector float)zero);
// z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
// Same with this one...
// z1 = vec_madd(z1, vec_0_899976223, (vector float)zero);
// tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
// dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
row7 = vec_madd(tmp4, vec_0_298631336, vec_madd(z1, vec_0_899976223, z3));
// tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
// dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
row5 = vec_madd(tmp5, vec_2_053119869, vec_madd(z2, vec_2_562915447, z4));
// tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
// dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
row3 = vec_madd(tmp6, vec_3_072711026, vec_madd(z2, vec_2_562915447, z3));
// tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
// dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
row1 = vec_madd(z1, vec_0_899976223, vec_madd(tmp7, vec_1_501321110, z4));
// Swap the row values with the alts. If this is the first half,
// this sets up the low values to be acted on in the second half.
// If this is the second half, it puts the high values back in
// the row values where they are expected to be when we're done.
SWAP(row0, alt0);
SWAP(row1, alt1);
SWAP(row2, alt2);
SWAP(row3, alt3);
SWAP(row4, alt4);
SWAP(row5, alt5);
SWAP(row6, alt6);
SWAP(row7, alt7);
}
if (whichPass == 1) {
// transpose the data for the second pass
// First, block transpose the upper right with lower left.
SWAP(row4, alt0);
SWAP(row5, alt1);
SWAP(row6, alt2);
SWAP(row7, alt3);
// Now, transpose each block of four
TRANSPOSE4(row0, row1, row2, row3);
TRANSPOSE4(row4, row5, row6, row7);
TRANSPOSE4(alt0, alt1, alt2, alt3);
TRANSPOSE4(alt4, alt5, alt6, alt7);
}
}
}
// perform the quantize step, using the floating point data
// still in the row/alt registers
{
const int* biasAddr;
const vector signed int* qmat;
vector float bias, negBias;
if (s->mb_intra) {
vector signed int baseVector;
// We must cache element 0 in the intra case
// (it needs special handling).
baseVector = vec_cts(vec_splat(row0, 0), 0);
vec_ste(baseVector, 0, &oldBaseValue);
qmat = (vector signed int*)s->q_intra_matrix[qscale];
biasAddr = &(s->intra_quant_bias);
} else {
qmat = (vector signed int*)s->q_inter_matrix[qscale];
biasAddr = &(s->inter_quant_bias);
}
// Load the bias vector (We add 0.5 to the bias so that we're
// rounding when we convert to int, instead of flooring.)
{
vector signed int biasInt;
const vector float negOneFloat = (vector float)FOUROF(-1.0f);
LOAD4(biasInt, biasAddr);
bias = vec_ctf(biasInt, QUANT_BIAS_SHIFT);
negBias = vec_madd(bias, negOneFloat, zero);
}
{
vector float q0, q1, q2, q3, q4, q5, q6, q7;
q0 = vec_ctf(qmat[0], QMAT_SHIFT);
q1 = vec_ctf(qmat[2], QMAT_SHIFT);
q2 = vec_ctf(qmat[4], QMAT_SHIFT);
q3 = vec_ctf(qmat[6], QMAT_SHIFT);
q4 = vec_ctf(qmat[8], QMAT_SHIFT);
q5 = vec_ctf(qmat[10], QMAT_SHIFT);
q6 = vec_ctf(qmat[12], QMAT_SHIFT);
q7 = vec_ctf(qmat[14], QMAT_SHIFT);
row0 = vec_sel(vec_madd(row0, q0, negBias), vec_madd(row0, q0, bias),
vec_cmpgt(row0, zero));
row1 = vec_sel(vec_madd(row1, q1, negBias), vec_madd(row1, q1, bias),
vec_cmpgt(row1, zero));
row2 = vec_sel(vec_madd(row2, q2, negBias), vec_madd(row2, q2, bias),
vec_cmpgt(row2, zero));
row3 = vec_sel(vec_madd(row3, q3, negBias), vec_madd(row3, q3, bias),
vec_cmpgt(row3, zero));
row4 = vec_sel(vec_madd(row4, q4, negBias), vec_madd(row4, q4, bias),
vec_cmpgt(row4, zero));
row5 = vec_sel(vec_madd(row5, q5, negBias), vec_madd(row5, q5, bias),
vec_cmpgt(row5, zero));
row6 = vec_sel(vec_madd(row6, q6, negBias), vec_madd(row6, q6, bias),
vec_cmpgt(row6, zero));
row7 = vec_sel(vec_madd(row7, q7, negBias), vec_madd(row7, q7, bias),
vec_cmpgt(row7, zero));
q0 = vec_ctf(qmat[1], QMAT_SHIFT);
q1 = vec_ctf(qmat[3], QMAT_SHIFT);
q2 = vec_ctf(qmat[5], QMAT_SHIFT);
q3 = vec_ctf(qmat[7], QMAT_SHIFT);
q4 = vec_ctf(qmat[9], QMAT_SHIFT);
q5 = vec_ctf(qmat[11], QMAT_SHIFT);
q6 = vec_ctf(qmat[13], QMAT_SHIFT);
q7 = vec_ctf(qmat[15], QMAT_SHIFT);
alt0 = vec_sel(vec_madd(alt0, q0, negBias), vec_madd(alt0, q0, bias),
vec_cmpgt(alt0, zero));
alt1 = vec_sel(vec_madd(alt1, q1, negBias), vec_madd(alt1, q1, bias),
vec_cmpgt(alt1, zero));
alt2 = vec_sel(vec_madd(alt2, q2, negBias), vec_madd(alt2, q2, bias),
vec_cmpgt(alt2, zero));
alt3 = vec_sel(vec_madd(alt3, q3, negBias), vec_madd(alt3, q3, bias),
vec_cmpgt(alt3, zero));
alt4 = vec_sel(vec_madd(alt4, q4, negBias), vec_madd(alt4, q4, bias),
vec_cmpgt(alt4, zero));
alt5 = vec_sel(vec_madd(alt5, q5, negBias), vec_madd(alt5, q5, bias),
vec_cmpgt(alt5, zero));
alt6 = vec_sel(vec_madd(alt6, q6, negBias), vec_madd(alt6, q6, bias),
vec_cmpgt(alt6, zero));
alt7 = vec_sel(vec_madd(alt7, q7, negBias), vec_madd(alt7, q7, bias),
vec_cmpgt(alt7, zero));
}
}
// Store the data back into the original block
{
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
data0 = vec_pack(vec_cts(row0, 0), vec_cts(alt0, 0));
data1 = vec_pack(vec_cts(row1, 0), vec_cts(alt1, 0));
data2 = vec_pack(vec_cts(row2, 0), vec_cts(alt2, 0));
data3 = vec_pack(vec_cts(row3, 0), vec_cts(alt3, 0));
data4 = vec_pack(vec_cts(row4, 0), vec_cts(alt4, 0));
data5 = vec_pack(vec_cts(row5, 0), vec_cts(alt5, 0));
data6 = vec_pack(vec_cts(row6, 0), vec_cts(alt6, 0));
data7 = vec_pack(vec_cts(row7, 0), vec_cts(alt7, 0));
{
// Clamp for overflow
vector signed int max_q_int, min_q_int;
vector signed short max_q, min_q;
LOAD4(max_q_int, &(s->max_qcoeff));
LOAD4(min_q_int, &(s->min_qcoeff));
max_q = vec_pack(max_q_int, max_q_int);
min_q = vec_pack(min_q_int, min_q_int);
data0 = vec_max(vec_min(data0, max_q), min_q);
data1 = vec_max(vec_min(data1, max_q), min_q);
data2 = vec_max(vec_min(data2, max_q), min_q);
data4 = vec_max(vec_min(data4, max_q), min_q);
data5 = vec_max(vec_min(data5, max_q), min_q);
data6 = vec_max(vec_min(data6, max_q), min_q);
data7 = vec_max(vec_min(data7, max_q), min_q);
}
{
vector bool char zero_01, zero_23, zero_45, zero_67;
vector signed char scanIndexes_01, scanIndexes_23, scanIndexes_45, scanIndexes_67;
vector signed char negOne = vec_splat_s8(-1);
vector signed char* scanPtr =
(vector signed char*)(s->intra_scantable.inverse);
signed char lastNonZeroChar;
// Determine the largest non-zero index.
zero_01 = vec_pack(vec_cmpeq(data0, (vector signed short)zero),
vec_cmpeq(data1, (vector signed short)zero));
zero_23 = vec_pack(vec_cmpeq(data2, (vector signed short)zero),
vec_cmpeq(data3, (vector signed short)zero));
zero_45 = vec_pack(vec_cmpeq(data4, (vector signed short)zero),
vec_cmpeq(data5, (vector signed short)zero));
zero_67 = vec_pack(vec_cmpeq(data6, (vector signed short)zero),
vec_cmpeq(data7, (vector signed short)zero));
// 64 biggest values
scanIndexes_01 = vec_sel(scanPtr[0], negOne, zero_01);
scanIndexes_23 = vec_sel(scanPtr[1], negOne, zero_23);
scanIndexes_45 = vec_sel(scanPtr[2], negOne, zero_45);
scanIndexes_67 = vec_sel(scanPtr[3], negOne, zero_67);
// 32 largest values
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_23);
scanIndexes_45 = vec_max(scanIndexes_45, scanIndexes_67);
// 16 largest values
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_45);
// 8 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// 4 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// 2 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// largest value
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
scanIndexes_01 = vec_splat(scanIndexes_01, 0);
vec_ste(scanIndexes_01, 0, &lastNonZeroChar);
lastNonZero = lastNonZeroChar;
// While the data is still in vectors we check for the transpose IDCT permute
// and handle it using the vector unit if we can. This is the permute used
// by the altivec idct, so it is common when using the altivec dct.
if ((lastNonZero > 0) && (s->dsp.idct_permutation_type == FF_TRANSPOSE_IDCT_PERM)) {
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
}
vec_st(data0, 0, data);
vec_st(data1, 16, data);
vec_st(data2, 32, data);
vec_st(data3, 48, data);
vec_st(data4, 64, data);
vec_st(data5, 80, data);
vec_st(data6, 96, data);
vec_st(data7, 112, data);
}
}
// special handling of block[0]
if (s->mb_intra) {
if (!s->h263_aic) {
if (n < 4)
oldBaseValue /= s->y_dc_scale;
else
oldBaseValue /= s->c_dc_scale;
}
// Divide by 8, rounding the result
data[0] = (oldBaseValue + 4) >> 3;
}
// We handled the transpose permutation above and we don't
// need to permute the "no" permutation case.
if ((lastNonZero > 0) &&
(s->dsp.idct_permutation_type != FF_TRANSPOSE_IDCT_PERM) &&
(s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)) {
ff_block_permute(data, s->dsp.idct_permutation,
s->intra_scantable.scantable, lastNonZero);
}
return lastNonZero;
}
/* this code assume stride % 16 == 0 *and* tmp is properly aligned */
static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
register int i;
const vector signed int vzero = vec_splat_s32(0);
const vector unsigned char permM2 = vec_lvsl(-2, src);
const vector unsigned char permM1 = vec_lvsl(-1, src);
const vector unsigned char permP0 = vec_lvsl(+0, src);
const vector unsigned char permP1 = vec_lvsl(+1, src);
const vector unsigned char permP2 = vec_lvsl(+2, src);
const vector unsigned char permP3 = vec_lvsl(+3, src);
const vector signed short v20ss = (const vector signed short)AVV(20);
const vector unsigned int v10ui = vec_splat_u32(10);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v1ss = vec_splat_s16(1);
const vector signed int v512si = (const vector signed int)AVV(512);
const vector unsigned int v16ui = (const vector unsigned int)AVV(16);
register int align = ((((unsigned long)src) - 2) % 16);
src -= (2 * srcStride);
for (i = 0 ; i < 21 ; i ++) {
vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
vector unsigned char srcR1 = vec_ld(-2, src);
vector unsigned char srcR2 = vec_ld(14, src);
switch (align) {
default: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = vec_perm(srcR1, srcR2, permP3);
} break;
case 11: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = srcR2;
} break;
case 12: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = srcR2;
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = srcR2;
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
}
const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
const vector signed short pp1A = vec_mladd(sum1A, v20ss, sum3A);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, sum3B);
const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
const vector signed short psumA = vec_sub(pp1A, pp2A);
const vector signed short psumB = vec_sub(pp1B, pp2B);
vec_st(psumA, 0, tmp);
vec_st(psumB, 16, tmp);
src += srcStride;
tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */
}
const vector unsigned char dstperm = vec_lvsr(0, dst);
const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
const vector unsigned char mperm = (const vector unsigned char)
AVV(0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B,
0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F);
int16_t *tmpbis = tmp - (tmpStride * 21);
vector signed short tmpM2ssA = vec_ld(0, tmpbis);
vector signed short tmpM2ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpM1ssA = vec_ld(0, tmpbis);
vector signed short tmpM1ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpP0ssA = vec_ld(0, tmpbis);
vector signed short tmpP0ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpP1ssA = vec_ld(0, tmpbis);
vector signed short tmpP1ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
vector signed short tmpP2ssA = vec_ld(0, tmpbis);
vector signed short tmpP2ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
for (i = 0 ; i < 16 ; i++) {
const vector signed short tmpP3ssA = vec_ld(0, tmpbis);
const vector signed short tmpP3ssB = vec_ld(16, tmpbis);
tmpbis += tmpStride;
const vector signed short sum1A = vec_adds(tmpP0ssA, tmpP1ssA);
const vector signed short sum1B = vec_adds(tmpP0ssB, tmpP1ssB);
const vector signed short sum2A = vec_adds(tmpM1ssA, tmpP2ssA);
const vector signed short sum2B = vec_adds(tmpM1ssB, tmpP2ssB);
const vector signed short sum3A = vec_adds(tmpM2ssA, tmpP3ssA);
const vector signed short sum3B = vec_adds(tmpM2ssB, tmpP3ssB);
tmpM2ssA = tmpM1ssA;
tmpM2ssB = tmpM1ssB;
tmpM1ssA = tmpP0ssA;
tmpM1ssB = tmpP0ssB;
tmpP0ssA = tmpP1ssA;
tmpP0ssB = tmpP1ssB;
tmpP1ssA = tmpP2ssA;
tmpP1ssB = tmpP2ssB;
tmpP2ssA = tmpP3ssA;
tmpP2ssB = tmpP3ssB;
const vector signed int pp1Ae = vec_mule(sum1A, v20ss);
const vector signed int pp1Ao = vec_mulo(sum1A, v20ss);
const vector signed int pp1Be = vec_mule(sum1B, v20ss);
const vector signed int pp1Bo = vec_mulo(sum1B, v20ss);
const vector signed int pp2Ae = vec_mule(sum2A, v5ss);
const vector signed int pp2Ao = vec_mulo(sum2A, v5ss);
const vector signed int pp2Be = vec_mule(sum2B, v5ss);
const vector signed int pp2Bo = vec_mulo(sum2B, v5ss);
const vector signed int pp3Ae = vec_sra((vector signed int)sum3A, v16ui);
const vector signed int pp3Ao = vec_mulo(sum3A, v1ss);
const vector signed int pp3Be = vec_sra((vector signed int)sum3B, v16ui);
const vector signed int pp3Bo = vec_mulo(sum3B, v1ss);
const vector signed int pp1cAe = vec_add(pp1Ae, v512si);
const vector signed int pp1cAo = vec_add(pp1Ao, v512si);
const vector signed int pp1cBe = vec_add(pp1Be, v512si);
const vector signed int pp1cBo = vec_add(pp1Bo, v512si);
const vector signed int pp32Ae = vec_sub(pp3Ae, pp2Ae);
const vector signed int pp32Ao = vec_sub(pp3Ao, pp2Ao);
const vector signed int pp32Be = vec_sub(pp3Be, pp2Be);
const vector signed int pp32Bo = vec_sub(pp3Bo, pp2Bo);
const vector signed int sumAe = vec_add(pp1cAe, pp32Ae);
const vector signed int sumAo = vec_add(pp1cAo, pp32Ao);
const vector signed int sumBe = vec_add(pp1cBe, pp32Be);
const vector signed int sumBo = vec_add(pp1cBo, pp32Bo);
const vector signed int ssumAe = vec_sra(sumAe, v10ui);
const vector signed int ssumAo = vec_sra(sumAo, v10ui);
const vector signed int ssumBe = vec_sra(sumBe, v10ui);
const vector signed int ssumBo = vec_sra(sumBo, v10ui);
const vector signed short ssume = vec_packs(ssumAe, ssumBe);
const vector signed short ssumo = vec_packs(ssumAo, ssumBo);
const vector unsigned char sumv = vec_packsu(ssume, ssumo);
const vector unsigned char sum = vec_perm(sumv, sumv, mperm);
const vector unsigned char dst1 = vec_ld(0, dst);
const vector unsigned char dst2 = vec_ld(16, dst);
const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, sum, vdst);
const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
vec_st(fdst1, 0, dst);
vec_st(fdst2, 16, dst);
dst += dstStride;
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
}
/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
register int i;
const vector signed int vzero = vec_splat_s32(0);
const vector unsigned char perm = vec_lvsl(0, src);
const vector signed short v20ss = (const vector signed short)AVV(20);
const vector unsigned short v5us = vec_splat_u16(5);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v16ss = (const vector signed short)AVV(16);
const vector unsigned char dstperm = vec_lvsr(0, dst);
const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
uint8_t *srcbis = src - (srcStride * 2);
const vector unsigned char srcM2a = vec_ld(0, srcbis);
const vector unsigned char srcM2b = vec_ld(16, srcbis);
const vector unsigned char srcM2 = vec_perm(srcM2a, srcM2b, perm);
srcbis += srcStride;
const vector unsigned char srcM1a = vec_ld(0, srcbis);
const vector unsigned char srcM1b = vec_ld(16, srcbis);
const vector unsigned char srcM1 = vec_perm(srcM1a, srcM1b, perm);
srcbis += srcStride;
const vector unsigned char srcP0a = vec_ld(0, srcbis);
const vector unsigned char srcP0b = vec_ld(16, srcbis);
const vector unsigned char srcP0 = vec_perm(srcP0a, srcP0b, perm);
srcbis += srcStride;
const vector unsigned char srcP1a = vec_ld(0, srcbis);
const vector unsigned char srcP1b = vec_ld(16, srcbis);
const vector unsigned char srcP1 = vec_perm(srcP1a, srcP1b, perm);
srcbis += srcStride;
const vector unsigned char srcP2a = vec_ld(0, srcbis);
const vector unsigned char srcP2b = vec_ld(16, srcbis);
const vector unsigned char srcP2 = vec_perm(srcP2a, srcP2b, perm);
srcbis += srcStride;
vector signed short srcM2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
vector signed short srcM2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
vector signed short srcM1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
vector signed short srcM1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
vector signed short srcP0ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
vector signed short srcP0ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
vector signed short srcP1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
vector signed short srcP1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
vector signed short srcP2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
vector signed short srcP2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
for (i = 0 ; i < 16 ; i++) {
const vector unsigned char srcP3a = vec_ld(0, srcbis);
const vector unsigned char srcP3b = vec_ld(16, srcbis);
const vector unsigned char srcP3 = vec_perm(srcP3a, srcP3b, perm);
const vector signed short srcP3ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
const vector signed short srcP3ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
srcbis += srcStride;
const vector signed short sum1A = vec_adds(srcP0ssA, srcP1ssA);
const vector signed short sum1B = vec_adds(srcP0ssB, srcP1ssB);
const vector signed short sum2A = vec_adds(srcM1ssA, srcP2ssA);
const vector signed short sum2B = vec_adds(srcM1ssB, srcP2ssB);
const vector signed short sum3A = vec_adds(srcM2ssA, srcP3ssA);
const vector signed short sum3B = vec_adds(srcM2ssB, srcP3ssB);
srcM2ssA = srcM1ssA;
srcM2ssB = srcM1ssB;
srcM1ssA = srcP0ssA;
srcM1ssB = srcP0ssB;
srcP0ssA = srcP1ssA;
srcP0ssB = srcP1ssB;
srcP1ssA = srcP2ssA;
srcP1ssB = srcP2ssB;
srcP2ssA = srcP3ssA;
srcP2ssB = srcP3ssB;
const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
const vector signed short pp3A = vec_add(sum3A, pp1A);
const vector signed short pp3B = vec_add(sum3B, pp1B);
const vector signed short psumA = vec_sub(pp3A, pp2A);
const vector signed short psumB = vec_sub(pp3B, pp2B);
const vector signed short sumA = vec_sra(psumA, v5us);
const vector signed short sumB = vec_sra(psumB, v5us);
const vector unsigned char sum = vec_packsu(sumA, sumB);
const vector unsigned char dst1 = vec_ld(0, dst);
const vector unsigned char dst2 = vec_ld(16, dst);
const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, sum, vdst);
const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
vec_st(fdst1, 0, dst);
vec_st(fdst2, 16, dst);
dst += dstStride;
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
}
/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
register int i;
const vector signed int vzero = vec_splat_s32(0);
const vector unsigned char permM2 = vec_lvsl(-2, src);
const vector unsigned char permM1 = vec_lvsl(-1, src);
const vector unsigned char permP0 = vec_lvsl(+0, src);
const vector unsigned char permP1 = vec_lvsl(+1, src);
const vector unsigned char permP2 = vec_lvsl(+2, src);
const vector unsigned char permP3 = vec_lvsl(+3, src);
const vector signed short v20ss = (const vector signed short)AVV(20);
const vector unsigned short v5us = vec_splat_u16(5);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v16ss = (const vector signed short)AVV(16);
const vector unsigned char dstperm = vec_lvsr(0, dst);
const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
register int align = ((((unsigned long)src) - 2) % 16);
for (i = 0 ; i < 16 ; i ++) {
vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
vector unsigned char srcR1 = vec_ld(-2, src);
vector unsigned char srcR2 = vec_ld(14, src);
switch (align) {
default: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = vec_perm(srcR1, srcR2, permP3);
} break;
case 11: {
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = vec_perm(srcR1, srcR2, permP2);
srcP3 = srcR2;
} break;
case 12: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = vec_perm(srcR1, srcR2, permP1);
srcP2 = srcR2;
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 13: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = vec_perm(srcR1, srcR2, permP0);
srcP1 = srcR2;
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 14: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = vec_perm(srcR1, srcR2, permM1);
srcP0 = srcR2;
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
case 15: {
vector unsigned char srcR3 = vec_ld(30, src);
srcM2 = vec_perm(srcR1, srcR2, permM2);
srcM1 = srcR2;
srcP0 = vec_perm(srcR2, srcR3, permP0);
srcP1 = vec_perm(srcR2, srcR3, permP1);
srcP2 = vec_perm(srcR2, srcR3, permP2);
srcP3 = vec_perm(srcR2, srcR3, permP3);
} break;
}
const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
const vector signed short pp3A = vec_add(sum3A, pp1A);
const vector signed short pp3B = vec_add(sum3B, pp1B);
const vector signed short psumA = vec_sub(pp3A, pp2A);
const vector signed short psumB = vec_sub(pp3B, pp2B);
const vector signed short sumA = vec_sra(psumA, v5us);
const vector signed short sumB = vec_sra(psumB, v5us);
const vector unsigned char sum = vec_packsu(sumA, sumB);
const vector unsigned char dst1 = vec_ld(0, dst);
const vector unsigned char dst2 = vec_ld(16, dst);
const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
vector unsigned char fsum;
OP_U8_ALTIVEC(fsum, sum, vdst);
const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
vec_st(fdst1, 0, dst);
vec_st(fdst2, 16, dst);
src += srcStride;
dst += dstStride;
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
}
static void ProjectDlightTexture_altivec( void ) {
int i, l;
vec_t origin0, origin1, origin2;
float texCoords0, texCoords1;
vector float floatColorVec0, floatColorVec1;
vector float modulateVec, colorVec, zero;
vector short colorShort;
vector signed int colorInt;
vector unsigned char floatColorVecPerm, modulatePerm, colorChar;
vector unsigned char vSel = VECCONST_UINT8(0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff);
float *texCoords;
byte *colors;
int *intColors;
byte clipBits[SHADER_MAX_VERTEXES];
float texCoordsArray[SHADER_MAX_VERTEXES][2];
byte colorArray[SHADER_MAX_VERTEXES][4];
glIndex_t hitIndexes[SHADER_MAX_INDEXES];
int numIndexes;
float scale;
float radius;
float radiusInverseCubed;
float intensity, remainder;
vec3_t floatColor;
float modulate = 0.0f;
qboolean vertexLight;
if ( !backEnd.refdef.num_dlights ) {
return;
}
// There has to be a better way to do this so that floatColor
// and/or modulate are already 16-byte aligned.
floatColorVecPerm = vec_lvsl(0,(float *)floatColor);
modulatePerm = vec_lvsl(0,(float *)&modulate);
modulatePerm = (vector unsigned char)vec_splat((vector unsigned int)modulatePerm,0);
zero = (vector float)vec_splat_s8(0);
for ( l = 0 ; l < backEnd.refdef.num_dlights ; l++ ) {
dlight_t *dl;
if ( !( tess.dlightBits & ( 1 << l ) ) ) {
continue; // this surface definately doesn't have any of this light
}
// clear colors
Com_Memset( colorArray, 0, sizeof( colorArray ) );
texCoords = texCoordsArray[0];
colors = colorArray[0];
dl = &backEnd.refdef.dlights[l];
origin0 = dl->transformed[0];
origin1 = dl->transformed[1];
origin2 = dl->transformed[2];
radius = dl->radius;
scale = 1.0f / radius;
radiusInverseCubed = dl->radiusInverseCubed;
intensity = dl->intensity;
vertexLight = ( ( dl->flags & REF_DIRECTED_DLIGHT ) || ( dl->flags & REF_VERTEX_DLIGHT ) );
// directional lights have max intensity and washout remainder intensity
if ( dl->flags & REF_DIRECTED_DLIGHT ) {
remainder = intensity * 0.125;
} else {
remainder = 0.0f;
}
if(r_greyscale->integer)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = floatColor[1] = floatColor[2] = luminance;
}
else if(r_greyscale->value)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
}
else
{
floatColor[0] = dl->color[0] * 255.0f;
floatColor[1] = dl->color[1] * 255.0f;
floatColor[2] = dl->color[2] * 255.0f;
}
floatColorVec0 = vec_ld(0, floatColor);
floatColorVec1 = vec_ld(11, floatColor);
floatColorVec0 = vec_perm(floatColorVec0,floatColorVec0,floatColorVecPerm);
for ( i = 0 ; i < tess.numVertexes ; i++, texCoords += 2, colors += 4 ) {
int clip = 0;
vec_t dist0, dist1, dist2;
dist0 = origin0 - tess.xyz[i][0];
dist1 = origin1 - tess.xyz[i][1];
dist2 = origin2 - tess.xyz[i][2];
backEnd.pc.c_dlightVertexes++;
// directional dlight, origin is a directional normal
if ( dl->flags & REF_DIRECTED_DLIGHT ) {
// twosided surfaces use absolute value of the calculated lighting
modulate = intensity * DotProduct( dl->origin, tess.normal[ i ] );
if ( tess.shader->cullType == CT_TWO_SIDED ) {
modulate = fabs( modulate );
}
modulate += remainder;
}
// spherical vertex lit dlight
else if ( dl->flags & REF_VERTEX_DLIGHT )
{
vec3_t dir;
dir[ 0 ] = radius - fabs( dist0 );
if ( dir[ 0 ] <= 0.0f ) {
continue;
}
dir[ 1 ] = radius - fabs( dist1 );
if ( dir[ 1 ] <= 0.0f ) {
continue;
}
dir[ 2 ] = radius - fabs( dist2 );
if ( dir[ 2 ] <= 0.0f ) {
continue;
}
modulate = intensity * dir[ 0 ] * dir[ 1 ] * dir[ 2 ] * radiusInverseCubed;
}
// vertical cylinder dlight
else
{
texCoords0 = 0.5f + dist0 * scale;
texCoords1 = 0.5f + dist1 * scale;
if( !r_dlightBacks->integer &&
// dist . tess.normal[i]
( dist0 * tess.normal[i][0] +
dist1 * tess.normal[i][1] +
dist2 * tess.normal[i][2] ) < 0.0f ) {
clip = 63;
} else {
if ( texCoords0 < 0.0f ) {
clip |= 1;
} else if ( texCoords0 > 1.0f ) {
clip |= 2;
}
if ( texCoords1 < 0.0f ) {
clip |= 4;
} else if ( texCoords1 > 1.0f ) {
clip |= 8;
}
texCoords[0] = texCoords0;
texCoords[1] = texCoords1;
// modulate the strength based on the height and color
if ( dist2 > radius ) {
clip |= 16;
modulate = 0.0f;
} else if ( dist2 < -radius ) {
clip |= 32;
modulate = 0.0f;
} else {
dist2 = Q_fabs(dist2);
if ( dist2 < radius * 0.5f ) {
modulate = intensity;
} else {
modulate = intensity * 2.0f * (radius - dist2) * scale;
}
}
}
}
clipBits[i] = clip;
// optimizations
if ( vertexLight && modulate < ( 1.0f / 128.0f ) ) {
continue;
} else if ( modulate > 1.0f ) {
modulate = 1.0f;
}
// ZTM: FIXME: should probably clamp to 0-255 range before converting to char,
// but I don't know how to do altvec stuff or if it's even used anymore
modulateVec = vec_ld(0,(float *)&modulate);
modulateVec = vec_perm(modulateVec,modulateVec,modulatePerm);
colorVec = vec_madd(floatColorVec0,modulateVec,zero);
colorInt = vec_cts(colorVec,0); // RGBx
colorShort = vec_pack(colorInt,colorInt); // RGBxRGBx
colorChar = vec_packsu(colorShort,colorShort); // RGBxRGBxRGBxRGBx
colorChar = vec_sel(colorChar,vSel,vSel); // RGBARGBARGBARGBA replace alpha with 255
vec_ste((vector unsigned int)colorChar,0,(unsigned int *)colors); // store color
}
// build a list of triangles that need light
intColors = (int*) colorArray;
numIndexes = 0;
for ( i = 0 ; i < tess.numIndexes ; i += 3 ) {
int a, b, c;
a = tess.indexes[i];
b = tess.indexes[i+1];
c = tess.indexes[i+2];
if ( vertexLight ) {
if ( !( intColors[ a ] | intColors[ b ] | intColors[ c ] ) ) {
continue;
}
} else {
if ( clipBits[a] & clipBits[b] & clipBits[c] ) {
continue; // not lighted
}
}
hitIndexes[numIndexes] = a;
hitIndexes[numIndexes+1] = b;
hitIndexes[numIndexes+2] = c;
numIndexes += 3;
}
if ( !numIndexes ) {
continue;
}
if ( !vertexLight ) {
qglEnableClientState( GL_TEXTURE_COORD_ARRAY );
qglTexCoordPointer( 2, GL_FLOAT, 0, texCoordsArray[0] );
} else {
qglDisableClientState( GL_TEXTURE_COORD_ARRAY );
}
qglEnableClientState( GL_COLOR_ARRAY );
qglColorPointer( 4, GL_UNSIGNED_BYTE, 0, colorArray );
if ( dl->dlshader ) {
shader_t *dls = dl->dlshader;
for ( i = 0; i < dls->numUnfoggedPasses; i++ ) {
shaderStage_t *stage = dls->stages[i];
R_BindAnimatedImage( &dls->stages[i]->bundle[0] );
GL_State( stage->stateBits | GLS_DEPTHFUNC_EQUAL );
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
}
} else {
R_FogOff();
if ( !vertexLight ) {
GL_Bind( tr.dlightImage );
} else {
GL_Bind( tr.whiteImage );
}
// include GLS_DEPTHFUNC_EQUAL so alpha tested surfaces don't add light
// where they aren't rendered
if ( dl->flags & REF_ADDITIVE_DLIGHT ) {
GL_State( GLS_SRCBLEND_ONE | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
else {
GL_State( GLS_SRCBLEND_DST_COLOR | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
RB_FogOn();
}
}
}
