void time_pwm_nv50(unsigned int cnum)
{
int i;
for (i = 0; i < 2; i++) {
uint32_t a = 0xe114 + i * 8;
uint32_t b = a + 4;
uint32_t save0 = nva_rd32(cnum, a);
uint32_t save1 = nva_rd32(cnum, b);
struct timeval start, end;
nva_wr32(cnum, a, 0x200000);
nva_wr32(cnum, b, 0x80080000);
while (nva_rd32(cnum, b) & 0x80000000);
nva_wr32(cnum, b, 0x80080000);
while (nva_rd32(cnum, b) & 0x80000000);
gettimeofday(&start, NULL);
nva_wr32(cnum, b, 0x80080000);
while (nva_rd32(cnum, b) & 0x80000000);
gettimeofday(&end, NULL);
uint64_t td = (time_diff_us(start, end));
printf("PWM %d: %dHz\n", i, (int)(1000000ll * 0x200000 / td));
nva_wr32(cnum, a, save0);
nva_wr32(cnum, b, 0x80000000 | save1);
}
}
void time_pgraph_dispatch_clock(unsigned int card)
{
struct timeval start, end;
ptime_t t_start, t_end;
u32 reg;
if (nva_cards[card]->chipset.card_type == 0x50)
reg = 0x4008f8;
else if (nva_cards[card]->chipset.card_type == 0xc0)
reg = 0x4040f4;
else {
printf("pgraph_dispatch_clock is only available on nv50+ chipsets\n");
return;
}
gettimeofday(&start, NULL);
t_start = nva_rd32(card, reg);
do
{
gettimeofday(&end, NULL);
} while (time_diff_us(start, end) < 1000000);
gettimeofday(&end, NULL);
t_end = nva_rd32(card, reg);
printf("PGRAPH.DISPATCH's clock: 1s = %llu cycles --> frequency = %f MHz\n",
(t_end - t_start), (t_end - t_start)/1000000.0);
}
ptime_t time_ptimer_clock(unsigned int card)
{
struct timeval start, end;
ptime_t t_start, t_end;
gettimeofday(&start, NULL);
t_start = get_time(card);
do
{
gettimeofday(&end, NULL);
} while (time_diff_us(start, end) < 1000000);
gettimeofday(&end, NULL);
t_end = get_time(card);
return t_end - t_start;
}
static ptime_t
pms_launch(int cnum, struct pms_ucode* pms, ptime_t *wall_time)
{
u32 pbus1098;
u32 pms_data, pms_kick;
ptime_t ptimer_start, ptimer_end;
struct timeval wall_start, wall_end;
int i;
if (nva_cards[cnum].chipset < 0x90) {
pms_data = 0x001400;
pms_kick = 0x00000003;
} else {
pms_data = 0x080000;
pms_kick = 0x00000001;
}
/* upload ucode */
pbus1098 = nva_mask(cnum, 0x001098, 0x00000008, 0x00000000);
nva_wr32(cnum, 0x001304, 0x00000000);
for (i = 0; i < pms->len / 4; i++)
nva_wr32(cnum, pms_data + (i * 4), pms->ptr.u32[i]);
nva_wr32(cnum, 0x001098, pbus1098 | 0x18);
/* and run it! */
gettimeofday(&wall_start, NULL);
ptimer_start = get_time(cnum);
nva_wr32(cnum, 0x00130c, pms_kick);
/* Wait for completion */
while (nva_rd32(cnum, 0x001308) & 0x100);
ptimer_end = get_time(cnum);
gettimeofday(&wall_end, NULL);
if (wall_time)
*wall_time = time_diff_us(wall_start, wall_end);
return ptimer_end - ptimer_start - (get_time(cnum) - get_time(cnum));
}
/*------------------------------------------------------------
* 時間から作ってもいいかどうかチェックする。
* 引数
* cindex int インデックス
* 返り値
* BOOL 作っていい場合 TRUE(1)
* BOOL 作っていけない場合 FALSE(0)
------------------------------------------------------------*/
BOOL NPC_createCheckGenerateFromTime( int cindex )
{
struct timeval old;
if( !NPC_CHECKCREATEINDEX(cindex) )
return FALSE;
if( NPC_create[cindex].workdata[NPC_CREATEWORKNEVERMAKE] )
return FALSE;
if( NPC_create[cindex].intdata[NPC_CREATEBORNNUM]
<= NPC_create[cindex].workdata[NPC_CREATEWORKENEMYNUM] )
return FALSE;
if( NPC_create[cindex].intdata[NPC_CREATETIME] < 0 )return FALSE;
old.tv_sec = NPC_create[cindex].workdata[NPC_CREATEWORKMAKESTARTSEC];
old.tv_usec= NPC_create[cindex].workdata[NPC_CREATEWORKMAKESTARTUSEC];
if( time_diff_us( NowTime , old ) <=
NPC_create[cindex].intdata[NPC_CREATETIME]*1000 )
return FALSE;
return TRUE;
}
int main(int argc, char** argv){
if(argc < 3){
std::cout << "usage:" << argv[0] << " src.png dst.png" << std::endl;
return 1;
}
int mode = 0; // for cuda
if(argc == 4){ // this is silly implementation
char* m = argv[3];
mode = m[0] - '0';
}
const char* input_file = argv[1];
const char* output_file = argv[2];
std::vector<unsigned char> in_image;
unsigned int width, height;
const unsigned input_error = lodepng::decode(in_image, width, height, input_file);
if(input_error){
std::cout << "decoder error " << input_error << ": " << lodepng_error_text(input_error) << std::endl;
return 1;
}
unsigned char* input_image = new unsigned char[in_image.size()];
unsigned char* output_image = new unsigned char[in_image.size()];
std::copy(in_image.begin(), in_image.end(), input_image);
// first CUDA call takes a whlie.
// For benchmark, I do not take account of first call.
const int N = 5;
long sum_of_n_minus_one = 0;
for(int i=0;i<N;i++){
struct timeval st;
struct timeval et;
gettimeofday(&st, NULL);
if(mode == 0){
rgb_invert_in_cuda(output_image, input_image, width, height);
}else if(mode == 1){
rgb_invert_in_cpu(output_image, input_image, width, height);
}else if(mode == 2){
rgb_invert_in_cuda_uchar_array(output_image, input_image, width, height);
}else{
std::cout << "ERROR: unknown mode: " << mode << std::endl;
return 1;
}
gettimeofday(&et, NULL);
long us = time_diff_us(st, et);
if(i != 0){
sum_of_n_minus_one += us;
}
}
printf("%lu\n",sum_of_n_minus_one/(N-1));
std::vector<unsigned char> out_image(output_image, output_image+in_image.size());
unsigned output_error = lodepng::encode(output_file, out_image, width, height);
if(output_error){
std::cout << "encoder error " << output_error << ": "<< lodepng_error_text(output_error) << std::endl;
return 1;
}
return 0;
}
