/**
* @v
* @vth
*/
inline vec_float4 updateF(vec_float4 v,vec_float4 vth)
{
vec_float4 updateF_rtn;
updateF_rtn = spu_sel(updateF_rtn,1.0,spu_cmpgt(spu_sub(v,vth),0.0));
updateF_rtn = spu_sel(updateF_rtn,0.0,spu_nand(spu_cmpgt(spu_sub(v,vth),0.0),spu_cmpgt(spu_sub(v,vth),0.0)));
return updateF_rtn;
}
void compute(
in0_type const& in,
out0_type const& out,
Pinfo const& p_in,
Pinfo const& p_out)
{
#if DEBUG
printf("uk_ccfft_f(%d): compute -- start %05x %05x\n", size, in, out);
#endif
// Handle inverse FFT explicitly so that shuffle and scale can happen
// in single step.
cml_core_ccfft1d_op_mi_f(fft, (float*)in, (float*)out, CML_FFT_FWD);
if (dir == -1)
{
if (scale != 1.f)
cml_core_rcsvmul1_f(scale, (float*)out, (float*)out, size);
}
else
{
// Code for the inverse FFT taken from the CBE SDK Libraries
// Overview and Users Guide, sec. 8.1.
int const vec_size = 4;
vector float* start = (vector float*)out;
vector float* end = start + 2 * size / vec_size;
vector float s0, s1, e0, e1;
vector unsigned int mask = (vector unsigned int){-1, -1, 0, 0};
vector float vscale = spu_splats(scale);
unsigned int i;
// Scale the output vector and swap the order of the outputs.
// Note: there are two float values for each of 'n' complex values.
s0 = e1 = *start;
for (i = 0; i < size / vec_size; ++i)
{
s1 = *(start + 1);
e0 = *(--end);
*start++ = spu_mul(spu_sel(e0, e1, mask), vscale);
*end = spu_mul(spu_sel(s0, s1, mask), vscale);
s0 = s1;
e1 = e0;
}
}
}
// Member data
size_t size;
int dir;
float scale;
fft1d_f* fft;
static char buf1[FFT_BUF1_SIZE_BYTES];
static char buf2[FFT_BUF2_SIZE_BYTES];
};
static btVector3 convexHullSupport (const btVector3& localDirOrg, const btVector3* points, int numPoints, const btVector3& localScaling)
{
btVector3 vec = localDirOrg * localScaling;
#if defined (__CELLOS_LV2__) && defined (__SPU__)
btVector3 localDir = vec;
vec_float4 v_distMax = {-FLT_MAX,0,0,0};
vec_int4 v_idxMax = {-999,0,0,0};
int v=0;
int numverts = numPoints;
for(;v<(int)numverts-4;v+=4) {
vec_float4 p0 = vec_dot3(points[v ].get128(),localDir.get128());
vec_float4 p1 = vec_dot3(points[v+1].get128(),localDir.get128());
vec_float4 p2 = vec_dot3(points[v+2].get128(),localDir.get128());
vec_float4 p3 = vec_dot3(points[v+3].get128(),localDir.get128());
const vec_int4 i0 = {v ,0,0,0};
const vec_int4 i1 = {v+1,0,0,0};
const vec_int4 i2 = {v+2,0,0,0};
const vec_int4 i3 = {v+3,0,0,0};
vec_uint4 retGt01 = spu_cmpgt(p0,p1);
vec_float4 pmax01 = spu_sel(p1,p0,retGt01);
vec_int4 imax01 = spu_sel(i1,i0,retGt01);
vec_uint4 retGt23 = spu_cmpgt(p2,p3);
vec_float4 pmax23 = spu_sel(p3,p2,retGt23);
vec_int4 imax23 = spu_sel(i3,i2,retGt23);
vec_uint4 retGt0123 = spu_cmpgt(pmax01,pmax23);
vec_float4 pmax0123 = spu_sel(pmax23,pmax01,retGt0123);
vec_int4 imax0123 = spu_sel(imax23,imax01,retGt0123);
vec_uint4 retGtMax = spu_cmpgt(v_distMax,pmax0123);
v_distMax = spu_sel(pmax0123,v_distMax,retGtMax);
v_idxMax = spu_sel(imax0123,v_idxMax,retGtMax);
}
for(;v<(int)numverts;v++) {
vec_float4 p = vec_dot3(points[v].get128(),localDir.get128());
const vec_int4 i = {v,0,0,0};
vec_uint4 retGtMax = spu_cmpgt(v_distMax,p);
v_distMax = spu_sel(p,v_distMax,retGtMax);
v_idxMax = spu_sel(i,v_idxMax,retGtMax);
}
int ptIndex = spu_extract(v_idxMax,0);
const btVector3& supVec= points[ptIndex] * localScaling;
return supVec;
#else
btScalar maxDot;
long ptIndex = vec.maxDot( points, numPoints, maxDot);
btAssert(ptIndex >= 0);
btVector3 supVec = points[ptIndex] * localScaling;
return supVec;
#endif //__SPU__
}
void check_pull_dma(int side){
// Check left
if(md[am].held_tag[side] < 32){
mfc_write_tag_mask( 1 << md[am].held_tag[side] );
int status = mfc_read_tag_status_immediate();
if(status){
// Update idx
md[am].idx[side][HEAD] = spu_add(md[am].idx[side][HEAD], md[am].num_waiting[side]);
vector signed int buffer_size = spu_splats(mcb[am].buffer_size[side] -1);
vector unsigned int cmp_v = spu_cmpgt(md[am].idx[side][HEAD], buffer_size);
vector signed int zeros = {0,0,0,0};
buffer_size = spu_add(buffer_size,1);
zeros = spu_sel(zeros,buffer_size,cmp_v);
md[am].idx[side][HEAD] = spu_sub(md[am].idx[side][HEAD],zeros);
md[am].num_pulled[side] += md[am].num_waiting[side];
md[am].num_waiting[side] = 0;
if(md[am].num_pulled[side] == mcb[am].data_size[side]){
md[am].mm_depleted[side] = 1;
}
// Release tag
mfc_tag_release( md[am].held_tag[side] );
md[am].held_tag[side] = 32;
}
}
}
vec_uint4 ulpDiff_f4(vec_float4 ref, vec_float4 vals) {
vec_int4 refi = (vec_int4)ref;
vec_int4 valsi = (vec_int4)vals;
vec_int4 diff = spu_sub(refi, valsi);
vec_int4 negdiff = spu_sub(spu_splats((int)0), diff);
return (vec_uint4)(spu_sel(negdiff, diff, spu_cmpgt(diff, 0)));
}
unsigned int
__mfc_multi_tag_reserve (unsigned int number_of_tags)
{
vector unsigned int table_copy;
vector unsigned int one = (vector unsigned int)
{ 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF };
vector unsigned int count_busy, is_valid;
vector unsigned int count_total;
vector unsigned int count_avail = (vector unsigned int) { 0, 0, 0, 0 };
vector unsigned int index = (vector unsigned int) { 0, 0, 0, 0 };
table_copy = __mfc_tag_table;
/* count_busy: number of consecutive busy tags
count_avail: number of consecutive free tags
table_copy: temporary copy of the tag table
count_total: sum of count_busy and count_avail
index: index of the current working tag */
do
{
table_copy = spu_sl (table_copy, count_avail);
count_busy = spu_cntlz (table_copy);
table_copy = spu_sl (table_copy, count_busy);
count_avail = spu_cntlz (spu_xor(table_copy, -1));
count_total = spu_add (count_busy, count_avail);
index = spu_add (index, count_total);
}
while (spu_extract (count_avail, 0) < number_of_tags
&& spu_extract (table_copy, 0) != 0);
index = spu_sub (index, count_avail);
/* is_valid is set to 0xFFFFFFFF if table_copy == 0, 0 otherwise. */
is_valid = spu_cmpeq (table_copy, 0);
index = spu_sel (index, is_valid, is_valid);
/* Now I need to actually mark the tags as used. */
table_copy = spu_sl (one, number_of_tags);
table_copy = spu_rl (table_copy, -number_of_tags - spu_extract (index, 0));
table_copy = spu_sel (table_copy, __mfc_tag_table, table_copy);
__mfc_tag_table = spu_sel (table_copy, __mfc_tag_table, is_valid);
return spu_extract (index, 0);
}
int main(int argc, char **argv) {
int i;
vector unsigned int all_ones = (vector unsigned int)
{0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF};
vector unsigned int all_zeroes = (vector unsigned int)
{0x00000000, 0x00000000, 0x00000000, 0x00000000};
/* These bits will form the selection mask */
unsigned short mask = 0x9;
/* Each bit in 0x9 forms a word in the mask */
vector unsigned int resultw =
spu_sel(all_zeroes, all_ones, spu_maskw(mask));
printf("resultw: ");
for (i=0; i<4; i++) {
printf("%08x", spu_extract(resultw, i));
}
/* Each bit in 0x09 forms a halfword in the mask */
vector unsigned short resulth =
spu_sel((vector unsigned short)all_zeroes,
(vector unsigned short)all_ones,
spu_maskh(mask));
printf("\nresulth: ");
for (i=0; i<8; i++) {
printf("%04x", spu_extract(resulth, i));
}
/* Each bit in 0x0009 forms a byte in the mask */
vector unsigned char resultb =
spu_sel((vector unsigned char)all_zeroes,
(vector unsigned char)all_ones,
spu_maskb(mask));
printf("\nresultb: ");
for (i=0; i<16; i++) {
printf("%02x", spu_extract(resultb, i));
}
printf("\n");
return 0;
}
void cp_buffer(int side){
int avail_out = num_free_in_buffer(OUT);
int avail_side = num_in_buffer(side);
int max = avail_out < avail_side ? avail_out : avail_side;
vector signed int *out_head;
if(mcb[am].local[OUT] < 255)
out_head = (vector signed int*) &md[ mcb[am].local[OUT] ].idx[ (mcb[am].id+1)&1 ][HEAD];
else
out_head = (vector signed int*) &md[am].idx[OUT][HEAD];
vector unsigned int cmp_v;
vector signed int from_size = spu_splats( mcb[am].buffer_size[side] );
vector signed int out_size = spu_splats( mcb[ mcb[am].local[OUT] ].buffer_size[ (mcb[am].id+1)&1 ] );
vector signed int ones = {1,1,1,1};
vector signed int zeros = {0,0,0,0};
int i;
for(i = 0; i < max; i++){
md[am].buffer[OUT][spu_extract( *out_head,0)] = md[am].buffer[side][spu_extract(md[am].idx[side][TAIL],0)];
// update idx
md[am].idx[side][TAIL] = spu_add(md[am].idx[side][TAIL], ones);
cmp_v = spu_cmpeq(md[am].idx[side][TAIL],from_size);
md[am].idx[side][TAIL] = spu_sel(md[am].idx[side][TAIL], zeros, cmp_v);
*out_head = spu_add(*out_head,ones);
cmp_v = spu_cmpeq(*out_head, out_size);
*out_head = spu_sel(*out_head,zeros,cmp_v);
}
update_tail(side);
md[am].consumed[side] += max;
if(mcb[am].local[OUT] < 255 && md[am].consumed[side] == mcb[am].data_size[side]){
md[am].depleted[side] = 1;
md[am].done = 1;
--num_active_mergers;
}
}
unsigned int
__mfc_multi_tag_release (unsigned int first_tag, unsigned int number_of_tags)
{
vector unsigned int table_copy, tmp, tmp1;
vector unsigned int one = (vector unsigned int)
{ 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF };
vector unsigned int is_invalid;
unsigned int last_tag;
vector unsigned int has_been_reserved;
last_tag = first_tag + number_of_tags;
table_copy = spu_sl (one, number_of_tags);
table_copy = spu_rl (table_copy, -last_tag);
table_copy = spu_xor (table_copy, -1);
/* Make sure the tags are in range and valid. */
tmp = spu_cmpgt (spu_promote(last_tag, 0), 32);
tmp1 = spu_cmpgt (spu_promote(number_of_tags, 0), 32);
is_invalid = spu_cmpgt (spu_promote(first_tag, 0), 31);
/* All bits are set to 1 if invalid, 0 if valid. */
is_invalid = spu_or (tmp, is_invalid);
is_invalid = spu_or (tmp1, is_invalid);
/* check whether these tags have been reserved */
tmp = spu_rlmask (one, (int)-number_of_tags);
tmp1 = spu_sl (__mfc_tag_table, first_tag);
has_been_reserved = spu_cmpgt(tmp1, tmp);
is_invalid = spu_or (has_been_reserved, is_invalid);
table_copy = spu_sel (__mfc_tag_table, table_copy, table_copy);
__mfc_tag_table = spu_sel (table_copy, __mfc_tag_table, is_invalid);
return spu_extract (is_invalid, 0);
}
unsigned int
__mfc_tag_reserve (void)
{
vector unsigned int mask = (vector unsigned int)
{ 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
vector unsigned int count_zeros, is_valid;
vector signed int count_neg;
count_zeros = spu_cntlz (__mfc_tag_table);
count_neg = spu_sub (0, (vector signed int) count_zeros);
mask = spu_rlmask (mask, (vector signed int) count_neg);
__mfc_tag_table = spu_andc (__mfc_tag_table, mask);
is_valid = spu_cmpeq (count_zeros, 32);
count_zeros = spu_sel (count_zeros, is_valid, is_valid);
return spu_extract (count_zeros, 0);
}
/* Scans the string pointed to by s for the character c and
* returns a pointer to the last occurance of c. If
* c is not found, then NULL is returned.
*/
char * strrchr(const char *s, int c)
{
int nskip;
vec_uchar16 *ptr, data, vc;
vec_uint4 cmp_c, cmp_0, cmp;
vec_uint4 res_ptr, res_cmp;
vec_uint4 mask, result;
vec_uint4 one = spu_splats(0xffffU);
/* Scan memory array a quadword at a time. Skip leading
* mis-aligned bytes.
*/
ptr = (vec_uchar16 *)s;
nskip = -((unsigned int)(ptr) & 15);
mask = spu_rlmask(one, nskip);
vc = spu_splats((unsigned char)(c));
data = *ptr++;
ptr = (vec_uchar16 *)((unsigned int)ptr & ~15);
cmp_c = spu_and(spu_gather(spu_cmpeq(data, vc)), mask);
cmp_0 = spu_and(spu_gather(spu_cmpeq(data, 0)), mask);
res_ptr = spu_splats(0U);
res_cmp = spu_splats(0U);
while (spu_extract(cmp_0, 0) == 0) {
cmp = spu_cmpeq(cmp_c, 0);
res_ptr = spu_sel(spu_promote((unsigned int)(ptr), 0), res_ptr, cmp);
res_cmp = spu_sel(cmp_c, res_cmp, cmp);
data = *ptr++;
cmp_c = spu_gather(spu_cmpeq(data, vc));
cmp_0 = spu_gather(spu_cmpeq(data, 0));
cmp = spu_cmpeq(cmp_c, 0);
}
/* Compute the location of the last character before termination
* character.
*
* First mask off compare results following the first termination character.
*/
mask = spu_sl(one, 31 - spu_extract(spu_cntlz(cmp_0), 0));
cmp_c = spu_and(cmp_c, mask);
/* Conditionally update res_ptr and res_cmd if a match was found in the last
* quadword.
*/
cmp = spu_cmpeq(cmp_c, 0);
res_ptr = spu_sel(spu_promote((unsigned int)(ptr), 0), res_ptr, cmp);
res_cmp = spu_sel(cmp_c, res_cmp, cmp);
/* Bit reserve res_cmp for locating last occurance.
*/
mask = spu_cmpeq(res_cmp, 0);
res_cmp = (vec_uint4)spu_maskb(spu_extract(res_cmp, 0));
res_cmp = spu_gather((vec_uchar16)spu_shuffle(res_cmp, res_cmp,
VEC_LITERAL(vec_uchar16,
15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0)));
/* Compute the location (ptr) of the last occurance of c. If no
* occurance was found (ie, element 0 of res_cmp == 0, then return
* NULL.
*/
result = spu_sub(spu_add(res_ptr, 15), spu_cntlz(res_cmp));
result = spu_andc(result, mask);
return ((char *)spu_extract(result, 0));
}
static inline
vec_uint4 vec_Maj(vec_uint4 x, vec_uint4 y, vec_uint4 z)
{
return spu_sel(x, y, spu_xor(x, z));
}
static inline
vec_uint4 vec_Ch(vec_uint4 x, vec_uint4 y, vec_uint4 z)
{
return spu_sel(z, y, x);
}
vector double
__divv2df3 (vector double a_in, vector double b_in)
{
/* Variables */
vec_int4 exp, exp_bias;
vec_uint4 no_underflow, overflow;
vec_float4 mant_bf, inv_bf;
vec_ullong2 exp_a, exp_b;
vec_ullong2 a_nan, a_zero, a_inf, a_denorm, a_denorm0;
vec_ullong2 b_nan, b_zero, b_inf, b_denorm, b_denorm0;
vec_ullong2 nan;
vec_uint4 a_exp, b_exp;
vec_ullong2 a_mant_0, b_mant_0;
vec_ullong2 a_exp_1s, b_exp_1s;
vec_ullong2 sign_exp_mask;
vec_double2 a, b;
vec_double2 mant_a, mant_b, inv_b, q0, q1, q2, mult;
/* Constants */
vec_uint4 exp_mask_u32 = spu_splats((unsigned int)0x7FF00000);
vec_uchar16 splat_hi = (vec_uchar16) {
0,1,2,3, 0,1,2,3, 8, 9,10,11, 8,9,10,11
};
vec_uchar16 swap_32 = (vec_uchar16) {
4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11
};
vec_ullong2 exp_mask = spu_splats(0x7FF0000000000000ULL);
vec_ullong2 sign_mask = spu_splats(0x8000000000000000ULL);
vec_float4 onef = spu_splats(1.0f);
vec_double2 one = spu_splats(1.0);
vec_double2 exp_53 = (vec_double2)spu_splats(0x0350000000000000ULL);
sign_exp_mask = spu_or(sign_mask, exp_mask);
/* Extract the floating point components from each of the operands including
* exponent and mantissa.
*/
a_exp = (vec_uint4)spu_and((vec_uint4)a_in, exp_mask_u32);
a_exp = spu_shuffle(a_exp, a_exp, splat_hi);
b_exp = (vec_uint4)spu_and((vec_uint4)b_in, exp_mask_u32);
b_exp = spu_shuffle(b_exp, b_exp, splat_hi);
a_mant_0 = (vec_ullong2)spu_cmpeq((vec_uint4)spu_andc((vec_ullong2)a_in, sign_exp_mask), 0);
a_mant_0 = spu_and(a_mant_0, spu_shuffle(a_mant_0, a_mant_0, swap_32));
b_mant_0 = (vec_ullong2)spu_cmpeq((vec_uint4)spu_andc((vec_ullong2)b_in, sign_exp_mask), 0);
b_mant_0 = spu_and(b_mant_0, spu_shuffle(b_mant_0, b_mant_0, swap_32));
a_exp_1s = (vec_ullong2)spu_cmpeq(a_exp, exp_mask_u32);
b_exp_1s = (vec_ullong2)spu_cmpeq(b_exp, exp_mask_u32);
/* Identify all possible special values that must be accommodated including:
* +-denorm, +-0, +-infinity, and NaNs.
*/
a_denorm0= (vec_ullong2)spu_cmpeq(a_exp, 0);
a_nan = spu_andc(a_exp_1s, a_mant_0);
a_zero = spu_and (a_denorm0, a_mant_0);
a_inf = spu_and (a_exp_1s, a_mant_0);
a_denorm = spu_andc(a_denorm0, a_zero);
b_denorm0= (vec_ullong2)spu_cmpeq(b_exp, 0);
b_nan = spu_andc(b_exp_1s, b_mant_0);
b_zero = spu_and (b_denorm0, b_mant_0);
b_inf = spu_and (b_exp_1s, b_mant_0);
b_denorm = spu_andc(b_denorm0, b_zero);
/* Scale denorm inputs to into normalized numbers by conditionally scaling the
* input parameters.
*/
a = spu_sub(spu_or(a_in, exp_53), spu_sel(exp_53, a_in, sign_mask));
a = spu_sel(a_in, a, a_denorm);
b = spu_sub(spu_or(b_in, exp_53), spu_sel(exp_53, b_in, sign_mask));
b = spu_sel(b_in, b, b_denorm);
/* Extract the divisor and dividend exponent and force parameters into the signed
* range [1.0,2.0) or [-1.0,2.0).
*/
exp_a = spu_and((vec_ullong2)a, exp_mask);
exp_b = spu_and((vec_ullong2)b, exp_mask);
mant_a = spu_sel(a, one, (vec_ullong2)exp_mask);
mant_b = spu_sel(b, one, (vec_ullong2)exp_mask);
/* Approximate the single reciprocal of b by using
* the single precision reciprocal estimate followed by one
* single precision iteration of Newton-Raphson.
*/
mant_bf = spu_roundtf(mant_b);
inv_bf = spu_re(mant_bf);
inv_bf = spu_madd(spu_nmsub(mant_bf, inv_bf, onef), inv_bf, inv_bf);
/* Perform 2 more Newton-Raphson iterations in double precision. The
* result (q1) is in the range (0.5, 2.0).
*/
inv_b = spu_extend(inv_bf);
inv_b = spu_madd(spu_nmsub(mant_b, inv_b, one), inv_b, inv_b);
q0 = spu_mul(mant_a, inv_b);
q1 = spu_madd(spu_nmsub(mant_b, q0, mant_a), inv_b, q0);
/* Determine the exponent correction factor that must be applied
* to q1 by taking into account the exponent of the normalized inputs
* and the scale factors that were applied to normalize them.
*/
exp = spu_rlmaska(spu_sub((vec_int4)exp_a, (vec_int4)exp_b), -20);
exp = spu_add(exp, (vec_int4)spu_add(spu_and((vec_int4)a_denorm, -0x34), spu_and((vec_int4)b_denorm, 0x34)));
/* Bias the quotient exponent depending on the sign of the exponent correction
* factor so that a single multiplier will ensure the entire double precision
* domain (including denorms) can be achieved.
*
* exp bias q1 adjust exp
* ===== ======== ==========
* positive 2^+65 -65
* negative 2^-64 +64
*/
exp_bias = spu_xor(spu_rlmaska(exp, -31), 64);
exp = spu_sub(exp, exp_bias);
q1 = spu_sel(q1, (vec_double2)spu_add((vec_int4)q1, spu_sl(exp_bias, 20)), exp_mask);
/* Compute a multiplier (mult) to applied to the quotient (q1) to produce the
* expected result. On overflow, clamp the multiplier to the maximum non-infinite
* number in case the rounding mode is not round-to-nearest.
*/
exp = spu_add(exp, 0x3FF);
no_underflow = spu_cmpgt(exp, 0);
overflow = spu_cmpgt(exp, 0x7FE);
exp = spu_and(spu_sl(exp, 20), (vec_int4)no_underflow);
exp = spu_and(exp, (vec_int4)exp_mask);
mult = spu_sel((vec_double2)exp, (vec_double2)(spu_add((vec_uint4)exp_mask, -1)), (vec_ullong2)overflow);
/* Handle special value conditions. These include:
*
* 1) IF either operand is a NaN OR both operands are 0 or INFINITY THEN a NaN
* results.
* 2) ELSE IF the dividend is an INFINITY OR the divisor is 0 THEN a INFINITY results.
* 3) ELSE IF the dividend is 0 OR the divisor is INFINITY THEN a 0 results.
*/
mult = spu_andc(mult, (vec_double2)spu_or(a_zero, b_inf));
mult = spu_sel(mult, (vec_double2)exp_mask, spu_or(a_inf, b_zero));
nan = spu_or(a_nan, b_nan);
nan = spu_or(nan, spu_and(a_zero, b_zero));
nan = spu_or(nan, spu_and(a_inf, b_inf));
mult = spu_or(mult, (vec_double2)nan);
/* Scale the final quotient */
q2 = spu_mul(q1, mult);
return (q2);
}
void draw_frame(uint64_t buf_ea) {
vec_uint4 buf[2*1920/4];
int row, col, i, tag = 0;
float step = 4.0f/spu.width*spu.zoom;
float xbeg = spu.xc - spu.width*step*0.5f;
vec_float4 vxbeg = spu_splats(xbeg)
+ spu_splats(step) * (vec_float4) {
0.f,1.f,2.f,3.f
};
vec_float4 xstep = spu_splats(step)*spu_splats(4.f);
vec_float4 vyp = spu_splats(spu.yc - spu.height*step*0.5f + step*spu.rank);
const vec_float4 vinc = spu_splats(spu.count * step);
const vec_float4 esc2 = spu_splats(BAILOUT*BAILOUT);
#if BAILBITS != 1
const vec_float4 esc21 = spu_splats(4.f/(BAILOUT*BAILOUT));
#endif
const vec_float4 two = spu_splats(2.f);
const vec_float4 zero = spu_splats(0.f);
const vec_float4 colsc = spu_splats(255.f);
const vec_float4 ccr = spu_splats(4.f*BAILOUT/(3.5f*3.141592654f));
const vec_float4 ccg = spu_splats(4.f*BAILOUT/(5.f*3.141592654f));
const vec_float4 ccb = spu_splats(4.f*BAILOUT/(9.f*3.141592654f));
vec_float4 x, y, x2, y2, m2, vxp;
vec_uint4 cmp, inc;
vec_uint4 vi;
vec_uint4 *p, *b;
vec_float4 co;
/* Process the full image. As there are 6 SPUs working in parallel, each with
* a different rank from 0 to 5, each SPU processes only the line numbers:
* rank, rank+6, rank+12, ...
* The program uses a SPU DMA programming technique known as "double buffering",
* where the previously generated line is transmitted to main memory while we
* compute the next one, hence the need for a local buffer containing two lines.
*/
for (row = spu.rank; row < spu.height; row += spu.count) {
/* Pixel buffer address (in local memory) of the next line to be drawn */
b = p = buf + ((1920/4)&-tag);
vxp = vxbeg; /* first four x coordinates */
/* Process a whole screen line by packets of 4 pixels */
for (col = spu.width/4; col > 0 ; col--) {
vi = spu_splats(0u);
x = vxp;
y = vyp;
i = 0;
cmp = spu_splats(-1u);
inc = spu_splats(1u);
m2 = zero;
/* This loop processes the Mandelbrot suite for the four complex numbers
* whose real part are the components of the x vector, and the imaginary
* part are in y (as we process the same line, all initial values of y
* are equal).
* We perform loop unrolling for SPU performance optimization reasons,
* hence the 4x replication of the same computation block.
*/
do {
x2 = x*x;
y2 = y*y;
m2 = spu_sel(m2, x2+y2, cmp);
cmp = spu_cmpgt(esc2, m2);
inc = spu_and(inc, cmp); /* increment the iteration count only if */
vi = vi + inc; /* we're still inside the bailout radius */
y = two*x*y + vyp;
x = x2-y2 + vxp;
x2 = x*x;
y2 = y*y;
m2 = spu_sel(m2, x2+y2, cmp);
cmp = spu_cmpgt(esc2, m2);
inc = spu_and(inc, cmp);
vi = vi + inc;
y = two*x*y + vyp;
x = x2-y2 + vxp;
x2 = x*x;
y2 = y*y;
m2 = spu_sel(m2, x2+y2, cmp);
cmp = spu_cmpgt(esc2, m2);
inc = spu_and(inc, cmp);
vi = vi + inc;
y = two*x*y + vyp;
x = x2-y2 + vxp;
x2 = x*x;
y2 = y*y;
m2 = spu_sel(m2, x2+y2, cmp);
cmp = spu_cmpgt(esc2, m2);
inc = spu_and(inc, cmp);
vi = vi + inc;
y = two*x*y + vyp;
x = x2-y2 + vxp;
i += 4;
}
/* Exit the loop only if the iteration limit of 128 has been reached,
* or all current four points are outside the bailout radius.
* The __builtin_expect(xxx, 1) construct hints the compiler that the xxx
* test has greater chance of being true (1), so a branch hinting
* instruction is inserted into the binary code to make the conditional
* branch faster in most cases (except the last one when we exit the
* loop). This results in performance increase.
*/
while (__builtin_expect((i < 128) &
(si_to_int((qword)spu_gather(cmp)) != 0), 1));
/* smooth coloring: compute the fractional part */
co = spu_convtf(vi, 0) + spu_splats(1.f);
co -= fast_logf(fast_logf(m2) * spu_splats(.5f));
#if BAILBITS != 1
co = spu_re(spu_rsqrte(co*esc21));
#endif
/* Compute the red, green an blue pixel components */
vec_uint4 cr = spu_convtu(mcos(co * ccr) * colsc, 0);
vec_uint4 cg = spu_convtu(mcos(co * ccg) * colsc, 0);
vec_uint4 cb = spu_convtu(mcos(co * ccb) * colsc, 0);
/* Put the 4 pixel values in the buffer */
*p++ = (spu_sl(cr, 16) | spu_sl(cg, 8) | cb) & ~-inc;
vxp += xstep;
}
/* double-buffered dma: initiate a dma transfer of last computed scanline
* then wait for completion of the second last transfer (previous computed
* line). This is done by changing the tag value.
*/
mfc_put(b, buf_ea+(spu.width*4)*row, spu.width*4, tag, 0, 0);
tag = 1 - tag;
wait_for_completion(tag);
vyp += vinc;
}
/* wait for completion of last sent image line */
wait_for_completion(1-tag);
}
void MinMaxBinFindBest3SIMD(minmaxbin_t *mmb, kdbuffer_t *result)
{
int i;
for(i=1; i < mmb->numbins; i++)
{
int j = mmb->numbins - i - 1;
vector float *min = (vector float *)mmb->minbins[i].b;
vector float *max = (vector float *)mmb->maxbins[j].b;
min[0] = spu_add(min[0], min[-1]);
max[0] = spu_add(max[0], max[1]);
}
vector float *vmax = (vector float*)result->baabb.max;
vector float *vmin = (vector float*)result->baabb.min;
vector float vwidth = spu_abs( spu_sub(*vmax, *vmin) );
vector float vnumbins = spu_splats(1/(float)mmb->numbins);
vector float vdelta = spu_mul(vwidth, vnumbins);
vector float vx = spu_add(*vmin, vdelta);
vector float vside = { vwidth[1] * vwidth[2], vwidth[0] * vwidth[2], vwidth[0] * vwidth[1], 0 };
vector float invarea = spu_splats( 1/(vwidth[0] * vside[0]));
vector float vctravers = spu_splats(2.0f);
vector float vbestcost = spu_splats(mmb->bestcost);
vector int vbesti = spu_splats(0);
vector float vbestx = vx;
for(i=0; i < mmb->numbins-1; i++)
{
vector float aleft, aright;
AreaLeftRight(*vmin, *vmax, vside, vx, &aleft, &aright);
vector float *vminbin = (vector float *)mmb->minbins[i].b;
vector float *vmaxbin = (vector float *)mmb->maxbins[i+1].b;
vector float cost = SAHCostSIMD(invarea, vctravers, *vminbin, aleft, *vmaxbin, aright);
vector unsigned int cmp = spu_cmpgt(cost, vbestcost);
vbestcost = spu_sel(cost, vbestcost, cmp);
vbesti = spu_sel(spu_splats(i), vbesti, cmp);
vbestx = spu_sel(vx, vbestx, cmp);
vx = spu_add(vx, vdelta);
}
int axis = 0;
float bestcost = vbestcost[axis];
if(vbestcost[1] < bestcost)
{
axis = 1;
bestcost = vbestcost[1];
}
if(vbestcost[2] < bestcost)
{
axis = 2;
bestcost = vbestcost[2];
}
int index = vbesti[axis];
result->plane = vbestx[axis];
result->axis = axis;
result->left_size = (int)mmb->minbins[ index ].b[axis];
result->right_size = (int)mmb->maxbins[ index+1 ].b[axis];
mmb->bestcost = vbestcost[axis];
}
void process_render_tasks(unsigned long eah_render_tasks, unsigned long eal_render_tasks)
{
const vec_uchar16 SHUFFLE_MERGE_BYTES = (vec_uchar16) { // merge lo bytes from unsigned shorts (array)
1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31
};
const vec_uchar16 SHUFFLE_GET_BUSY_WITH_ONES = (vec_uchar16) { // get busy flag with ones in unused bytes
0xc0, 0xc0, 2, 3, 0xc0,0xc0,0xc0,0xc0, 0xc0,0xc0,0xc0,0xc0
};
const vec_uchar16 ZERO_BYTES = (vec_uchar16) spu_splats(0);
char trianglebuffer[ 256 + TRIANGLE_MAX_SIZE ];
char sync_buffer[128+127];
void* aligned_sync_buffer = (void*) ( ((unsigned long)sync_buffer+127) & ~127 );
RenderableCacheLine* cache = (RenderableCacheLine*) aligned_sync_buffer;
unsigned long long cache_ea;
spu_mfcdma64(&cache_ea, eah_render_tasks, eal_render_tasks, sizeof(cache_ea), 0, MFC_GET_CMD);
mfc_write_tag_mask(1<<0);
mfc_read_tag_status_all();
while (cache_ea) {
// terminate immediately if possible
if (spu_stat_in_mbox())
return;
// read the cache line
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_GETLLAR_CMD);
spu_readch(MFC_RdAtomicStat);
unsigned int endTriangle = cache->endTriangle;
vec_ushort8 testTriangle = spu_splats((unsigned short) endTriangle);
// first look for short chunks
vec_uchar16 next = cache->chunkNext;
vec_uchar16 nextmask = spu_and(next, spu_splats((unsigned char)CHUNKNEXT_MASK));
// change next to word offset, note we don't care what the low bit shifted in is
vec_uchar16 firstshuf = (vec_uchar16) spu_sl( (vec_ushort8)nextmask, 1 );
vec_uchar16 trishufhi = spu_or ( firstshuf, spu_splats((unsigned char) 1));
vec_uchar16 trishuflo = spu_and( firstshuf, spu_splats((unsigned char) 254));
vec_ushort8 start0 = cache->chunkStart[0];
vec_ushort8 start1 = cache->chunkStart[1];
vec_ushort8 nstart0 = spu_shuffle( start0, start1, spu_shuffle( trishuflo, trishufhi, SHUF0 ) );
vec_ushort8 nstart1 = spu_shuffle( start0, start1, spu_shuffle( trishuflo, trishufhi, SHUF1 ) );
vec_ushort8 starteq0 = spu_cmpeq( nstart0, spu_splats((unsigned short)0) );
vec_ushort8 starteq1 = spu_cmpeq( nstart1, spu_splats((unsigned short)0) );
vec_ushort8 end0 = spu_sel( nstart0, spu_splats((unsigned short)4096), starteq0);
vec_ushort8 end1 = spu_sel( nstart1, spu_splats((unsigned short)4096), starteq1);
vec_ushort8 len0 = spu_sub( end0, start0);
vec_ushort8 len1 = spu_sub( end1, start1);
vec_ushort8 small0 = spu_cmpgt( spu_splats((unsigned short)17), len0);
vec_ushort8 small1 = spu_cmpgt( spu_splats((unsigned short)17), len1);
vec_uchar16 small = (vec_uchar16) spu_shuffle( small0, small1, MERGE );
vec_uint4 smallChunkGather = spu_gather(small);
// check to see if chunk is already at the last triangle
vec_uint4 doneChunkGather = spu_gather( (vec_uchar16) spu_shuffle(
(vec_uchar16) spu_cmpeq(testTriangle, cache->chunkTriangle[0]),
(vec_uchar16) spu_cmpeq(testTriangle, cache->chunkTriangle[1]),
SHUFFLE_MERGE_BYTES) );
// check if the chunk is free
vec_uint4 freeChunkGather = spu_gather(
spu_cmpeq( spu_splats( (unsigned char) CHUNKNEXT_FREE_BLOCK ), cache->chunkNext ) );
// check to see if the chunk is being processed
vec_uint4 busyChunkGather = spu_gather(
spu_cmpgt( cache->chunkNext, //spu_and(cache->chunkNext, CHUNKNEXT_MASK),
spu_splats( (unsigned char) (CHUNKNEXT_BUSY_BIT-1) ) ) );
// doneChunkGather, freeChunkGather, busyChunkGather - rightmost 16 bits of word 0
// note that if freeChunkGather is true then busyChunkGather must also be true
// done=false, free=false, busy=false -> can process
// free=false, busy=false -> can be merged
// decide which chunk to process
vec_uint4 mayProcessGather = spu_nor( doneChunkGather, busyChunkGather );
vec_uint4 mayProcessShortGather = spu_and( mayProcessGather, smallChunkGather );
vec_uint4 shortSelMask = spu_cmpeq( mayProcessShortGather, spu_splats(0U) );
vec_uint4 mayProcessSelection = spu_sel( mayProcessShortGather, mayProcessGather, shortSelMask );
/*
if (!spu_extract(shortSelMask, 0))
printf("taken short: may=%04x short=%04x mayshort=%04x mask=%04x sel=%04x\n",
spu_extract(mayProcessGather, 0) & 0xffff,
spu_extract(smallChunkGather, 0),
spu_extract(mayProcessShortGather, 0),
spu_extract(shortSelMask, 0) & 0xffff,
spu_extract(mayProcessSelection, 0) & 0xffff );
*/
vec_uint4 mayProcessBits = spu_sl( mayProcessSelection, 16);
unsigned int chunkToProcess = spu_extract( spu_cntlz( mayProcessBits ), 0);
unsigned int freeChunk = spu_extract( spu_cntlz( spu_sl( freeChunkGather, 16 ) ), 0);
// if there's nothing to process, try the next cache line in the rendering tasks list
if (!spu_extract(mayProcessBits, 0)) {
trynextcacheline:
cache_ea = cache->next;
// sleep();
continue;
}
unsigned int chunkStart = cache->chunkStartArray [chunkToProcess];
unsigned int chunkTriangle = cache->chunkTriangleArray[chunkToProcess];
unsigned int chunkNext = cache->chunkNextArray [chunkToProcess] & CHUNKNEXT_MASK;
unsigned int chunkEnd = (cache->chunkStartArray [chunkNext]-1) & (NUMBER_OF_TILES-1);
unsigned int chunkLength = 1 + chunkEnd-chunkStart;
// only need an extra block if the block is especially long
if (chunkLength <= NUMBER_OF_TILES_PER_CHUNK) {
freeChunk = 32;
}
// mark this block as busy
cache->chunkNextArray[chunkToProcess] |= CHUNKNEXT_BUSY_BIT;
// if there's at least one free chunk, claim it
if (freeChunk != 32) {
cache->chunkNextArray[freeChunk] = CHUNKNEXT_RESERVED;
cache->chunkTriangleArray[freeChunk] = chunkTriangle;
}
// write the cache line back
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_PUTLLC_CMD);
if (spu_readch(MFC_RdAtomicStat) & MFC_PUTLLC_STATUS)
continue;
#ifdef INFO
printf("[%d] Claimed chunk %d (%d-%d len %d) at tri %x end %x with free chunk %d\n", _SPUID,
chunkToProcess, chunkStart, chunkEnd, chunkLength, chunkTriangle, endTriangle,
freeChunk!=32 ? freeChunk : -1 );
// debug_render_tasks(cache);
#endif
Triangle* triangle;
int firstTile;
do {
// read the triangle data for the current triangle
unsigned int extra = chunkTriangle & 127;
unsigned long long trianglebuffer_ea = cache_ea + TRIANGLE_OFFSET_FROM_CACHE_LINE + (chunkTriangle & ~127);
triangle = (Triangle*) (trianglebuffer+extra);
unsigned int length = (extra + TRIANGLE_MAX_SIZE + 127) & ~127;
// ensure DMA slot available
do {} while (!spu_readchcnt(MFC_Cmd));
spu_mfcdma64(trianglebuffer, mfc_ea2h(trianglebuffer_ea), mfc_ea2l(trianglebuffer_ea),
length, 0, MFC_GET_CMD);
mfc_write_tag_mask(1<<0);
mfc_read_tag_status_all();
// get the triangle deltas
firstTile = findFirstTriangleTile(triangle, chunkStart, chunkEnd);
if (firstTile>=0)
break;
// no match, try next triangle
chunkTriangle = triangle->next_triangle;
} while (chunkTriangle != endTriangle);
// if we actually have something to process...
if (firstTile>=0) {
// the "normal" splitting will now become:
// chunkStart .. (firstTile-1) -> triangle->next_triangle
// firstTile .. (firstTile+NUM-1) -> chunkTriangle (BUSY)
// (firstTile+NUM) .. chunkEnd -> chunkTriangle (FREE)
int tailChunk;
int thisChunk;
int nextBlockStart;
int thisBlockStart;
int realBlockStart;
do {
retry:
// read the cache line
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_GETLLAR_CMD);
spu_readch(MFC_RdAtomicStat);
// calculate start of next block
nextBlockStart = firstTile + NUMBER_OF_TILES_PER_CHUNK;
if (nextBlockStart > chunkEnd)
nextBlockStart = chunkEnd+1;
// calculate start of block to mark as busy
thisBlockStart = nextBlockStart - NUMBER_OF_TILES_PER_CHUNK;
if (thisBlockStart < chunkStart)
thisBlockStart = chunkStart;
realBlockStart = thisBlockStart;
#ifdef INFO
printf("[%d] nextBlockStart=%d, realBlockStart=%d, thisBlockStart=%d, chunkStart=%d\n", _SPUID,
nextBlockStart, realBlockStart, thisBlockStart, chunkStart);
#endif
// allocate some more free chunks
vec_uint4 freeChunkGather2 = spu_sl(spu_gather(spu_cmpeq(
spu_splats((unsigned char)CHUNKNEXT_FREE_BLOCK), cache->chunkNext)), 16);
unsigned int freeChunk2 = spu_extract(spu_cntlz(freeChunkGather2), 0);
if (freeChunk == 32) {
// if we didn't have one before, try again
freeChunk = freeChunk2;
// and try to get the second one
freeChunkGather2 = spu_andc( freeChunkGather2, spu_promote(0x80000000>>freeChunk2, 0) );
freeChunk2 = spu_extract(spu_cntlz(freeChunkGather2), 0);
} else {
// speculatively clear the free chunk just in case we don't need it
cache->chunkNextArray[freeChunk] = CHUNKNEXT_FREE_BLOCK;
}
#ifdef INFO
printf("[%d] Free chunks %d and %d, cN=%d, nBS=%d, cE=%d, tBS=%d, cS=%d\n",
_SPUID, freeChunk, freeChunk2, chunkNext, nextBlockStart, chunkEnd, thisBlockStart, chunkStart );
#endif
// mark region after as available for processing if required
if (nextBlockStart < chunkEnd) {
if (freeChunk==32) {
// if no free chunk, relinquish entire block and write back
cache->chunkNextArray[chunkToProcess] = chunkNext;
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_PUTLLC_CMD);
// if writeback failed, we *might* have a free block, retry
if (spu_readch(MFC_RdAtomicStat) & MFC_PUTLLC_STATUS)
goto retry;
// otherwise give up and try the next cache line
goto trynextcacheline;
}
cache->chunkStartArray[freeChunk] = nextBlockStart;
cache->chunkNextArray[freeChunk] = chunkNext;
cache->chunkTriangleArray[freeChunk] = chunkTriangle;
cache->chunkNextArray[chunkToProcess] = freeChunk | CHUNKNEXT_BUSY_BIT;
tailChunk = freeChunk;
#ifdef INFO
printf("[%d] Insert tail, tailChunk=%d, chunkNext=%d, chunkToProcess=%d\n", _SPUID, tailChunk, chunkNext, chunkToProcess);
debug_render_tasks(cache);
#endif
} else {
// we're gonna use freeChunk2 for the "in front" block, as we've not
// used freeChunk, let's use it as it's more likely to have a free chunk
freeChunk2 = freeChunk;
tailChunk = chunkNext;
}
// mark region before as available if required and possible
thisChunk = chunkToProcess;
if (thisBlockStart > chunkStart) {
if (freeChunk2 != 32) {
// mark this region as busy
cache->chunkStartArray[freeChunk2]=thisBlockStart;
cache->chunkNextArray[freeChunk2]=tailChunk | CHUNKNEXT_BUSY_BIT;
cache->chunkTriangleArray[freeChunk2]=chunkTriangle;
// mark region before as available for processing
cache->chunkNextArray[chunkToProcess]=freeChunk2;
cache->chunkTriangleArray[chunkToProcess]=triangle->next_triangle;
thisChunk = freeChunk2;
#ifdef INFO
printf("[%d] Insert new head, tailChunk=%d, chunkNext=%d, thisChunk=%d\n", _SPUID, tailChunk, chunkNext, thisChunk);
debug_render_tasks(cache);
#endif
} else {
// need to keep whole block, update info and mark bust
cache->chunkTriangleArray[chunkToProcess]=chunkTriangle;
cache->chunkNextArray[chunkToProcess]=tailChunk | CHUNKNEXT_BUSY_BIT;
realBlockStart = chunkStart;
printf("[%d] Keep whole block, tailChunk=%d, chunkNext=%d, thisChunk=%d\n", _SPUID, tailChunk, chunkNext, thisChunk);
debug_render_tasks(cache);
#ifdef INFO
#endif
sleep();
}
}
// merge chunks
merge_cache_blocks(cache);
// write the cache line back
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_PUTLLC_CMD);
} while (spu_readch(MFC_RdAtomicStat) & MFC_PUTLLC_STATUS);
// finally after the write succeeded, update the variables
chunkNext = tailChunk;
chunkToProcess = thisChunk;
chunkStart = firstTile; //thisBlockStart;
chunkLength = nextBlockStart - firstTile;
chunkEnd = chunkStart + chunkLength - 1;
freeChunk = 32;
// now we can process the block up to endTriangle
initTileBuffers(thisBlockStart, chunkEnd);
int ok=0;
while (chunkTriangle != endTriangle) {
#ifdef INFO
printf("[%d] Processing chunk %d at %4d len %4d, triangle %04x first=%d tbs=%d\n",
_SPUID, chunkToProcess, chunkStart, chunkLength,
chunkTriangle, firstTile, thisBlockStart);
#endif
// and actually process that triangle on these chunks
processTriangleChunks(triangle, cache, thisBlockStart, chunkEnd, chunkTriangle, ok);
ok=1;
#ifdef PAUSE
sleep();
#endif
// and advance to the next-triangle
chunkTriangle = triangle->next_triangle;
// this should only ever happen if we're running really low on cache line slots
// basically, if we pick up a block with more than NUMBER_OF_TILES_PER_CHUNK and
// there's no slot to store the pre-NUMBER_OF_TILES_PER_CHUNK tiles.
// in this case, we process from thisBlockStart only (because we know that from
// chunkStart to there has no result) and then we only process one triangle
if (chunkStart != realBlockStart) {
/*
printf("[%d] chunkStart=%d != realBlockStart %d, chunkEnd=%d, "
"firstTile=%d chunk=%d\n",
_SPUID, chunkStart, realBlockStart, chunkEnd,
firstTile, chunkToProcess);
debug_render_tasks(cache);
*/
// abort the while loop
break;
}
// read the next triangle
unsigned int extra = chunkTriangle & 127;
unsigned long long trianglebuffer_ea = cache_ea + TRIANGLE_OFFSET_FROM_CACHE_LINE + (chunkTriangle & ~127);
triangle = (Triangle*) (trianglebuffer+extra);
unsigned int length = (extra + TRIANGLE_MAX_SIZE + 127) & ~127;
// ensure DMA slot available
do {} while (!spu_readchcnt(MFC_Cmd));
spu_mfcdma64(trianglebuffer, mfc_ea2h(trianglebuffer_ea),
mfc_ea2l(trianglebuffer_ea), length, 0, MFC_GET_CMD);
mfc_write_tag_mask(1<<0);
mfc_read_tag_status_all();
} // until chunkTriangle == endTriangle
// flush any output buffers
flushTileBuffers(thisBlockStart, chunkEnd);
} // firstTile>=0
static inline
vec_uint4 vec_Maj(vec_uint4 x, vec_uint4 y, vec_uint4 z)
{
return spu_sel(spu_and(y, z), spu_or(y, z), x);
}
void merge_buffers(){
vector unsigned int cmp_v, cmp_v2;
const vector signed int one_at_0 = {1,0,0,0};
const vector signed int one_at_1 = {0,1,0,0};
const vector signed int one_at_2 = {0,0,1,0};
const vector signed int ones = {1,1,1,1};
const vector signed int zeros = {0,0,0,0};
const vector unsigned char cmp_v_shuffle_mask = {31,31,31,31,
31,31,31,31,
31,31,31,31,
31,31,31,31};
vector unsigned char rev_mask;
const vector unsigned char rev_left = {12,13,14,15,
8,9,10,11,
4,5,6,7,
0,1,2,3};
const vector unsigned char rev_right = {28,29,30,31,
24,25,26,27,
20,21,22,23,
16,17,18,19};
vector signed int *out_head_idx;
if(mcb[am].local[OUT] < 255){
int parent_idx = mcb[am].local[OUT];
int side = (mcb[am].id+1)&1;
out_head_idx = (vector signed int*) &md[parent_idx].idx[side][HEAD];
} else {
out_head_idx = (vector signed int*) &md[am].idx[OUT][HEAD];
}
vector signed int *left_tail_idx = (vector signed int*) &md[am].idx[LEFT][TAIL];
vector signed int *right_tail_idx = (vector signed int*) &md[am].idx[RIGHT][TAIL];
vector signed int size_v = {mcb[am].buffer_size[LEFT], mcb[am].buffer_size[RIGHT], mcb[am].buffer_size[OUT], 0};
vector signed int avail_v = {num_in_buffer(LEFT), num_in_buffer(RIGHT), num_free_in_buffer(OUT), 1};
vector signed int avail_before = { spu_extract(avail_v, 0), spu_extract(avail_v, 1), 0, 0 };
vector unsigned int avail = spu_gather( spu_cmpgt(avail_v, zeros) ); // avail = 0x0F if all avail_v > zeros
vector signed int *left, *right, *out;
left = (vector signed int*) &md[am].buffer[LEFT][ spu_extract(*left_tail_idx,0) ];
right = (vector signed int*) &md[am].buffer[RIGHT][ spu_extract(*right_tail_idx,0) ];
out = (vector signed int*) &md[am].buffer[OUT][ spu_extract(*out_head_idx,0) ];
#ifdef TRACE_TIME
dec_val2 = spu_read_decrementer();
#endif
while(spu_extract(avail,0) == 0x0F){
// cmp left and right to determine who gets eaten
cmp_v = spu_cmpgt(*left,*right);
cmp_v = spu_shuffle(cmp_v, cmp_v, cmp_v_shuffle_mask);
// cmp_v = {FFFF,FFFF,FFFF,FFFF} if left[3] > right[3]
*out = spu_sel(*left,*right,cmp_v);
rev_mask = spu_sel(rev_right,rev_left,(vector unsigned char)cmp_v);
*left = spu_shuffle(*left,*right,rev_mask);
// data to be sorted is now in out and left, left in descending order
sort_vectors(out,left);
// update index of the used side
if( spu_extract(cmp_v,0) ){
// left[3] > right[3]
*right_tail_idx = spu_add(*right_tail_idx,ones);
avail_v = spu_sub(avail_v, one_at_1);
right++;
// modulus hack
cmp_v2 = spu_cmpeq(*right_tail_idx, size_v);
if( __builtin_expect( spu_extract(cmp_v2,0) ,0) ){
*right_tail_idx = zeros;
right = (vector signed int*) &md[am].buffer[RIGHT][0];
}
} else {
*right = *left;
*left_tail_idx = spu_add(*left_tail_idx,ones);
avail_v = spu_sub(avail_v, one_at_0);
left++;
// modulus hack
cmp_v2 = spu_cmpeq(*left_tail_idx, size_v);
if( __builtin_expect( spu_extract(cmp_v2,0) ,0) ){
*left_tail_idx = zeros;
left = (vector signed int*) &md[am].buffer[LEFT][0];
}
}
// update out head idx
*out_head_idx = spu_add(*out_head_idx,ones);
avail_v = spu_sub(avail_v, one_at_2);
out++;
// modulus hack
cmp_v2 = spu_cmpeq(*out_head_idx, size_v);
if( __builtin_expect(spu_extract(cmp_v2,0),0) ){
out = (vector signed int*) &md[am].buffer[OUT][0];
*out_head_idx = zeros;
}
// is there data still available?
avail = spu_gather(spu_cmpgt(avail_v, zeros));
}
#ifdef TRACE_TIME
merge_loop_ticks += -(spu_read_decrementer() - dec_val2);
#endif
// how much got produced?
vector signed int consumed = spu_sub(avail_before, avail_v);
int consumed_left = spu_extract(consumed, 0);
int consumed_right = spu_extract(consumed, 1);
if(consumed_left)
update_tail(LEFT);
if(consumed_right)
update_tail(RIGHT);
md[am].consumed[LEFT] += consumed_left;
md[am].consumed[RIGHT] += consumed_right;
if(md[am].consumed[LEFT] == mcb[am].data_size[LEFT])
md[am].depleted[LEFT] = 1;
if(md[am].consumed[RIGHT] == mcb[am].data_size[RIGHT])
md[am].depleted[RIGHT] = 1;
if(mcb[am].local[OUT] < 255 && md[am].depleted[LEFT] && md[am].depleted[RIGHT]){
md[am].done = 1;
--num_active_mergers;
}
}
inline vector float spu_max(vector float a, vector float b)
{
return spu_sel( b, a, spu_cmpgt( a, b ) );
}
Triangle* getTriangleBuffer(Context* context)
{
// if we've already allocated a triangle buffer (and we're in the same context)
if (context == _currentTriangleContext && _currentTriangle)
return _currentTriangle;
// trash the default values
_currentTriangleContext = context;
_currentTriangle = NULL;
// read the current renderable cache line to ensure there is room for the triangle data
// in the cache line buffer; we do this by comparing against all 16 cache line blocks
// to make sure that extending the write pointer wouldn't clobber the data
unsigned long long cache_ea = context->renderableCacheLine;
if (cache_ea == 0)
return NULL;
char cachebuffer[128+127];
RenderableCacheLine* cache = (RenderableCacheLine*) ( ((unsigned int)cachebuffer+127) & ~127 );
// printf("GTB: reading to %x from %x:%x\n", cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea));
spu_mfcdma64(cache, mfc_ea2h(cache_ea), mfc_ea2l(cache_ea), 128, 0, MFC_GETLLAR_CMD);
spu_readch(MFC_RdAtomicStat);
// extendvalid = ( read<=write && test<end ) || ( read>write && test<read )
// extendvalid = ( read>write && read>test ) || ( read<=write && end>test )
// simplifies to extendvalid = selb(end, read, read>write) > test
// or extendvalid = selb(end>test, read>test, read>write)
// rewind = next >= end
// rewindvalid = read != 0
// valid = extendvalid && (!rewind || rewindvalid)
// = extendvalid && (!rewind || !rewindinvalid)
// = extendvalid && !(rewind && rewindinvalid)
// invalid = ! (extendvalid && !(rewind && rewindinvalid))
// = (!extendvalid || (rewind && rewindinvalid))
vec_ushort8 v_writeptr = spu_splats( cache->endTriangle );
vec_ushort8 v_readptr0 = cache->chunkTriangle[0];
vec_ushort8 v_readptr1 = cache->chunkTriangle[1];
vec_ushort8 v_testptr = spu_add(v_writeptr, TRIANGLE_MAX_SIZE);
vec_ushort8 v_nextptr = spu_add(v_writeptr, 2*TRIANGLE_MAX_SIZE);
vec_ushort8 v_endptr = spu_splats( (unsigned short)TRIANGLE_BUFFER_SIZE);
vec_ushort8 v_zero = spu_splats( (unsigned short) 0 );
vec_uchar16 v_merger = (vec_uchar16) { 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31 };
vec_ushort8 v_max0_test = spu_sel( v_endptr, v_readptr0, spu_cmpgt( v_readptr0, v_writeptr ) );
vec_ushort8 v_max1_test = spu_sel( v_endptr, v_readptr1, spu_cmpgt( v_readptr1, v_writeptr ) );
vec_ushort8 v_extend0_valid = spu_cmpgt( v_max0_test, v_testptr );
vec_ushort8 v_extend1_valid = spu_cmpgt( v_max1_test, v_testptr );
vec_ushort8 v_rewind0_invalid = spu_cmpeq( v_readptr0, v_zero );
vec_ushort8 v_rewind1_invalid = spu_cmpeq( v_readptr1, v_zero );
vec_ushort8 v_rewind8 = spu_cmpgt( v_nextptr, v_endptr );
vec_uchar16 v_extend_valid = (vec_uchar16) spu_shuffle( v_extend0_valid, v_extend1_valid, v_merger );
vec_uchar16 v_rewind_invalid = (vec_uchar16) spu_shuffle( v_rewind0_invalid, v_rewind1_invalid, v_merger );
vec_uchar16 v_rewind = (vec_uchar16) v_rewind8;
vec_uchar16 v_valid_rhs = spu_and( v_rewind_invalid, v_rewind );
vec_uchar16 v_invalid = spu_orc( v_valid_rhs, v_extend_valid );
// check to see if the chunk is being processed
vec_uint4 v_free = spu_gather(
spu_cmpeq( spu_splats( (unsigned char) CHUNKNEXT_FREE_BLOCK ), cache->chunkNext ) );
vec_uint4 v_invalid_bits = spu_andc( spu_gather( v_invalid ), (vec_uint4) v_free );
// if any of the bits are invalid, then no can do
if ( spu_extract(v_invalid_bits, 0) ) {
return NULL;
}
// fetch in the data before this triangle in the cache buffer
unsigned int offset = cache->endTriangle;
_currentTriangleBufferExtra = offset & 127;
unsigned long long trianglebuffer_ea = cache_ea + TRIANGLE_OFFSET_FROM_CACHE_LINE + (offset & ~127);
if (_currentTriangleBufferExtra) {
spu_mfcdma64(_currentTriangleBuffer, mfc_ea2h(trianglebuffer_ea), mfc_ea2l(trianglebuffer_ea), 128, 0, MFC_GET_CMD);
// ensure DMA did actually complete
mfc_write_tag_mask(1<<0);
mfc_read_tag_status_all();
}
// final bit of initialisation
_currentTriangle = (Triangle*) (_currentTriangleBuffer+_currentTriangleBufferExtra);
_currentTriangleOffset = offset;
_currentTriangleRewind = v_rewind8;
_currentTriangleCacheEndTriangleEAL = mfc_ea2l(cache_ea) + (((char*)&cache->endTriangle) - ((char*)cache));
_currentTriangleCacheEndTriangleEAH = mfc_ea2h(cache_ea);
_currentTriangleBufferEA = trianglebuffer_ea;
// printf("Allocated new triangle buffer: %x\n", offset);
// and return the buffer ready to go
return _currentTriangle;
}
inline vector float spu_min(vector float a, vector float b)
{
return spu_sel( a, b, spu_cmpgt( a, b ) );
}