// system interrupt handler
extern void sys_handler(void) {
unsigned st_status = AT91C_BASE_ST->ST_SR & AT91C_BASE_ST->ST_IMR;
unsigned rtc_status = AT91C_BASE_RTC->RTC_SR & AT91C_BASE_RTC->RTC_IMR;
unsigned dbgu_status = AT91C_BASE_DBGU->DBGU_CSR & AT91C_BASE_DBGU->DBGU_IMR;
static unsigned char counter = '0';
if (dbgu_status & AT91C_US_RXRDY) {
// disable RXRDY interrupt in DBGU
AT91C_BASE_DBGU->DBGU_IDR |= AT91C_US_RXRDY;
// disable rtt interrupt flag
AT91C_BASE_ST->ST_IDR = AT91C_ST_RTTINC;
if (wait_status == 1) {
wait_status = 0;
}
else {
transfer_size = dbgu_xmod_recv((void *)LINUX_BASE_ADDRESS);
run_kernel();
}
}
// handler of rtt - rttinc
if (st_status & AT91C_ST_RTTINC) {
AT91C_BASE_PIOB->PIO_ODSR ^= AT91C_PIO_PB27;
if (wait_status == 1) {
put_char(counter);
counter++;
if (counter == '6')
run_kernel();
put_char(' ');
}
else
put_char('C');
}
}
bool
check(misc::runner const & i_runner, host::generic_program i_program)
{
chrono::steady_clock::time_point tp = chrono::steady_clock::now();
host::buffer<pfm::int_> bufWrite(i_runner.m_context, item_count);
typedef host::buffer<pfm::int_>::const_iterator iterator;
i_runner.m_queue(
run_kernel(i_program,
fill_index(bufWrite),
item_count));
i_runner.m_queue(
run_kernel(i_program,
twice(bufWrite),
item_count));
auto future =
i_runner.m_queue(
bufWrite.with_range(
[](iterator i_begin, iterator i_end) {
return std::accumulate(i_begin, i_end, 0);
}));
std::future_status result = future.wait_until(tp + chrono::seconds(5));
assert(result == std::future_status::ready);
assert(future.get() == arith(2, item_count));
return true;
}
END_TEST
START_TEST (test_builtins)
{
uint32_t rs = run_kernel(builtins_source, NormalKind);
const char *errstr = 0;
switch (rs)
{
case 1:
errstr = "float2 cos(float2) doesn't behave correctly";
break;
case 2:
errstr = "float cos(float) doesn't behave correctly";
break;
case 3:
errstr = "float copysign(float) doesn't behave correctly";
break;
case 4:
errstr = "float2 copysign(float2) doesn't behave correctly";
break;
case 5:
errstr = "exp2() doesn't behave correctly";
break;
default:
errstr = default_error(rs);
}
fail_if(
errstr != 0,
errstr
);
}
END_TEST
START_TEST (test_image)
{
uint32_t rs = run_kernel(image_source, ImageKind);
const char *errstr = 0;
switch (rs)
{
case 1:
errstr = "Image1 must have width of 4";
break;
case 2:
errstr = "Image1 must have width of 4";
break;
case 3:
errstr = "Image2 must have type SIGNED_FLOAT16";
break;
case 4:
errstr = "Image2 must have channel order RGBA";
break;
case 5:
errstr = "The value read from the image is not good";
break;
default:
errstr = default_error(rs);
}
fail_if(
errstr != 0,
errstr
);
}
void process_cfm_by_gpu(unsigned char *pDataDst, int nDstWidth, int nDstHeight, short *pSrcData, int nSrcWidth, int nSrcHeight) {
int i,j;
int m,n;
#ifdef OPENCL_MU1
LOGD("MU1 ---------------------- input start");
set_input_i_to_kernel();
set_input_o_to_kernel();
LOGD("MU1 ---------------------- run kernel");
run_kernel();
LOGD("MU1 ---------------------- get output");
get_output_from_kernel();
LOGD("MU1 ---------------------- end");
#else
LOGD("MU1 ---------------------- start C");
for(i=0; i<512; i++) {
for(j=0; j<1024; j++) {
table_o[i][j] = (table_i[i][j]+3)*(table_q[i][j]+3)*(table_i[i][j]+2)*(table_q[i][j]+2)*(table_i[i][j]+1)*(table_q[i][j]+1)*(table_i[i][j])*(table_q[i][j])*(table_i[i][j]+3)*(table_q[i][j]+3)*(table_i[i][j]+2)*(table_q[i][j]+2)*(table_i[i][j]+1)*(table_q[i][j]+1)*(table_i[i][j])*(table_q[i][j]);
table_o[i][j] +=sin((100.0)/(table_i[i][j]+table_q[i][j]))*1000;
table_o[i][j] += sqrt(table_i[i][j]) + sqrt(table_q[i][j]) + sqrt(table_i[i][j]+table_q[i][j]);
}
}
LOGD("MU1 ---------------------- end");
#endif
}
END_TEST
START_TEST (test_barrier)
{
uint32_t rs = run_kernel(barrier_source, BarrierKind);
fail_if(
rs != 0x40,
default_error(rs)
);
}
bool
check(misc::runner const & i_runner, host::generic_program i_program)
{
chrono::steady_clock::time_point tp = chrono::steady_clock::now();
host::buffer<pfm::int_> bufWrite(i_runner.m_context, item_count);
typedef host::buffer<pfm::int_>::const_iterator iterator;
// kernel内で使用できる事の確認
i_runner.m_queue(
run_kernel(
i_program,
fill_index(bufWrite),
item_count));
auto future =
i_runner.m_queue(
bufWrite.with_range(
[](iterator i_begin, iterator i_end){
return std::accumulate(i_begin, i_end, 0);
}));
std::future_status result =
future.wait_until(tp + chrono::seconds(5));
assert(result == std::future_status::ready);
assert(future.get() == arith(2, item_count));
// ホストから呼べない事の確認
try {
int const a = 1; //gcc-4.7.2 twice(1)と書くと内部エラー
i_runner.m_queue(
host::run_kernel(
i_program,
twice(a),
1)
);
assert(false);
}
catch (cl::Error err) {
assert(err.err() == CL_INVALID_KERNEL_NAME);
}
return true;
}
static int selftest (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param)
{
hashconfig_t *hashconfig = hashcat_ctx->hashconfig;
hashes_t *hashes = hashcat_ctx->hashes;
status_ctx_t *status_ctx = hashcat_ctx->status_ctx;
user_options_extra_t *user_options_extra = hashcat_ctx->user_options_extra;
cl_int CL_err;
int CL_rc;
if (hashconfig->st_hash == NULL) return 0;
// init : replace hashes with selftest hash
device_param->kernel_params[15] = &device_param->d_st_digests_buf;
device_param->kernel_params[17] = &device_param->d_st_salts_buf;
device_param->kernel_params[18] = &device_param->d_st_esalts_buf;
device_param->kernel_params_buf32[31] = 1;
device_param->kernel_params_buf32[32] = 0;
// password : move the known password into a fake buffer
u32 highest_pw_len = 0;
if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
{
if (user_options_extra->attack_kern == ATTACK_KERN_STRAIGHT)
{
device_param->kernel_params_buf32[30] = 1;
pw_t pw; memset (&pw, 0, sizeof (pw));
char *pw_ptr = (char *) &pw.i;
const size_t pw_len = strlen (hashconfig->st_pass);
memcpy (pw_ptr, hashconfig->st_pass, pw_len);
pw.pw_len = (u32) pw_len;
if (hashconfig->opts_type & OPTS_TYPE_PT_UPPER)
{
uppercase ((u8 *) pw_ptr, pw.pw_len);
}
CL_err = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
}
else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
{
device_param->kernel_params_buf32[30] = 1;
device_param->kernel_params_buf32[33] = COMBINATOR_MODE_BASE_LEFT;
pw_t pw; memset (&pw, 0, sizeof (pw));
char *pw_ptr = (char *) &pw.i;
const size_t pw_len = strlen (hashconfig->st_pass);
memcpy (pw_ptr, hashconfig->st_pass, pw_len - 1);
pw.pw_len = (u32) pw_len - 1;
if (hashconfig->opts_type & OPTS_TYPE_PT_UPPER)
{
uppercase ((u8 *) pw_ptr, pw.pw_len);
}
pw_t comb; memset (&comb, 0, sizeof (comb));
char *comb_ptr = (char *) &comb.i;
memcpy (comb_ptr, hashconfig->st_pass + pw_len - 1, 1);
comb.pw_len = 1;
if (hashconfig->opts_type & OPTS_TYPE_PT_UPPER)
{
uppercase ((u8 *) comb_ptr, comb.pw_len);
}
if (hashconfig->opts_type & OPTS_TYPE_PT_ADD01)
{
comb_ptr[comb.pw_len] = 0x01;
}
if (hashconfig->opts_type & OPTS_TYPE_PT_ADD80)
{
comb_ptr[comb.pw_len] = 0x80;
}
CL_err = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_combs_c, CL_TRUE, 0, 1 * sizeof (pw_t), &comb, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
CL_err = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
}
else if (user_options_extra->attack_kern == ATTACK_KERN_BF)
{
device_param->kernel_params_buf32[30] = 1;
if (hashconfig->opts_type & OPTS_TYPE_PT_BITSLICE)
{
pw_t pw; memset (&pw, 0, sizeof (pw));
char *pw_ptr = (char *) &pw.i;
const size_t pw_len = strlen (hashconfig->st_pass);
memcpy (pw_ptr, hashconfig->st_pass, pw_len);
if (hashconfig->opts_type & OPTS_TYPE_PT_UPPER)
{
uppercase ((u8 *) pw_ptr, pw_len);
}
pw.pw_len = (u32) pw_len;
CL_err = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
}
else
{
bf_t bf; memset (&bf, 0, sizeof (bf));
char *bf_ptr = (char *) &bf.i;
memcpy (bf_ptr, hashconfig->st_pass, 1);
if (hashconfig->opts_type & OPTS_TYPE_PT_UTF16LE)
{
memset (bf_ptr, 0, 4);
for (int i = 0, j = 0; i < 1; i += 1, j += 2)
{
bf_ptr[j + 0] = hashconfig->st_pass[i];
bf_ptr[j + 1] = 0;
}
}
else if (hashconfig->opts_type & OPTS_TYPE_PT_UTF16BE)
{
memset (bf_ptr, 0, 4);
for (int i = 0, j = 0; i < 1; i += 1, j += 2)
{
bf_ptr[j + 0] = 0;
bf_ptr[j + 1] = hashconfig->st_pass[i];
}
}
if (hashconfig->opts_type & OPTS_TYPE_PT_UPPER)
{
uppercase ((u8 *) bf_ptr, 4);
}
if (hashconfig->opts_type & OPTS_TYPE_PT_GENERATE_BE)
{
bf.i = byte_swap_32 (bf.i);
}
CL_err = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_bfs_c, CL_TRUE, 0, 1 * sizeof (bf_t), &bf, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
pw_t pw; memset (&pw, 0, sizeof (pw));
char *pw_ptr = (char *) &pw.i;
const size_t pw_len = strlen (hashconfig->st_pass);
memcpy (pw_ptr + 1, hashconfig->st_pass + 1, pw_len - 1);
size_t new_pass_len = pw_len;
if (hashconfig->opts_type & OPTS_TYPE_PT_UTF16LE)
{
memset (pw_ptr, 0, pw_len);
for (size_t i = 1, j = 2; i < new_pass_len; i += 1, j += 2)
{
pw_ptr[j + 0] = hashconfig->st_pass[i];
pw_ptr[j + 1] = 0;
}
new_pass_len *= 2;
}
else if (hashconfig->opts_type & OPTS_TYPE_PT_UTF16BE)
{
memset (pw_ptr, 0, pw_len);
for (size_t i = 1, j = 2; i < new_pass_len; i += 1, j += 2)
{
pw_ptr[j + 0] = 0;
pw_ptr[j + 1] = hashconfig->st_pass[i];
}
new_pass_len *= 2;
}
if (hashconfig->opts_type & OPTS_TYPE_PT_UPPER)
{
uppercase ((u8 *) pw_ptr, new_pass_len);
}
if (hashconfig->opti_type & OPTI_TYPE_SINGLE_HASH)
{
if (hashconfig->opti_type & OPTI_TYPE_APPENDED_SALT)
{
memcpy (pw_ptr + new_pass_len, (char *) hashes->st_salts_buf[0].salt_buf, 64 - new_pass_len);
new_pass_len += hashes->st_salts_buf[0].salt_len;
}
}
pw.pw_len = (u32) new_pass_len;
if (hashconfig->opts_type & OPTS_TYPE_PT_ADD01)
{
pw_ptr[new_pass_len] = 0x01;
}
if (hashconfig->opts_type & OPTS_TYPE_PT_ADD80)
{
pw_ptr[new_pass_len] = 0x80;
}
if (hashconfig->opts_type & OPTS_TYPE_PT_ADDBITS14)
{
pw.i[14] = (u32) new_pass_len * 8;
pw.i[15] = 0;
}
if (hashconfig->opts_type & OPTS_TYPE_PT_ADDBITS15)
{
pw.i[14] = 0;
pw.i[15] = (u32) new_pass_len * 8;
}
if (hashconfig->opts_type & OPTS_TYPE_PT_GENERATE_BE)
{
for (int i = 0; i < 14; i++) pw.i[i] = byte_swap_32 (pw.i[i]);
}
CL_err = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
highest_pw_len = pw.pw_len;
}
}
}
else
{
pw_t pw; memset (&pw, 0, sizeof (pw));
char *pw_ptr = (char *) &pw.i;
const size_t pw_len = strlen (hashconfig->st_pass);
memcpy (pw_ptr, hashconfig->st_pass, pw_len);
pw.pw_len = (u32) pw_len;
CL_err = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, 1 * sizeof (pw_t), &pw, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
}
// main : run the kernel
if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
{
if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
{
if (highest_pw_len < 16)
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_1, 1, false, 0);
if (CL_rc == -1) return -1;
}
else if (highest_pw_len < 32)
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_2, 1, false, 0);
if (CL_rc == -1) return -1;
}
else
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_3, 1, false, 0);
if (CL_rc == -1) return -1;
}
}
else
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_4, 1, false, 0);
if (CL_rc == -1) return -1;
}
}
else
{
// missing handling hooks
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_1, 1, false, 0);
if (CL_rc == -1) return -1;
if (hashconfig->opts_type & OPTS_TYPE_HOOK12)
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_12, 1, false, 0);
if (CL_rc == -1) return -1;
CL_rc = hc_clEnqueueReadBuffer (hashcat_ctx, device_param->command_queue, device_param->d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL);
if (CL_rc == -1) return -1;
// do something with data
CL_rc = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL);
if (CL_rc == -1) return -1;
}
const u32 salt_pos = 0;
salt_t *salt_buf = &hashes->st_salts_buf[salt_pos];
const u32 kernel_loops_fixed = hashconfig_get_kernel_loops (hashcat_ctx);
const u32 loop_step = (kernel_loops_fixed) ? kernel_loops_fixed : 1;
const u32 iter = salt_buf->salt_iter;
for (u32 loop_pos = 0; loop_pos < iter; loop_pos += loop_step)
{
u32 loop_left = iter - loop_pos;
loop_left = MIN (loop_left, loop_step);
device_param->kernel_params_buf32[28] = loop_pos;
device_param->kernel_params_buf32[29] = loop_left;
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_2, 1, false, 0);
if (CL_rc == -1) return -1;
}
if (hashconfig->opts_type & OPTS_TYPE_HOOK23)
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_23, 1, false, 0);
if (CL_rc == -1) return -1;
CL_rc = hc_clEnqueueReadBuffer (hashcat_ctx, device_param->command_queue, device_param->d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL);
if (CL_rc == -1) return -1;
/*
* The following section depends on the hash mode
*/
switch (hashconfig->hash_mode)
{
// for 7z we only need device_param->hooks_buf, but other hooks could use any info from device_param. All of them should/must update hooks_buf
case 11600: seven_zip_hook_func (device_param, hashes->st_hook_salts_buf, 0, 1); break;
}
/*
* END of hash mode specific hook operations
*/
CL_rc = hc_clEnqueueWriteBuffer (hashcat_ctx, device_param->command_queue, device_param->d_hooks, CL_TRUE, 0, device_param->size_hooks, device_param->hooks_buf, 0, NULL, NULL);
if (CL_rc == -1) return -1;
}
if (hashconfig->opts_type & OPTS_TYPE_INIT2)
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_INIT2, 1, false, 0);
if (CL_rc == -1) return -1;
}
if (hashconfig->opts_type & OPTS_TYPE_LOOP2)
{
const u32 iter2 = salt_buf->salt_iter2;
for (u32 loop_pos = 0; loop_pos < iter2; loop_pos += loop_step)
{
u32 loop_left = iter2 - loop_pos;
loop_left = MIN (loop_left, loop_step);
device_param->kernel_params_buf32[28] = loop_pos;
device_param->kernel_params_buf32[29] = loop_left;
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_LOOP2, 1, false, 0);
if (CL_rc == -1) return -1;
}
}
if ((hashconfig->hash_mode == 2500) || (hashconfig->hash_mode == 2501))
{
device_param->kernel_params_buf32[28] = 0;
device_param->kernel_params_buf32[29] = 1;
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_AUX1, 1, false, 0);
if (CL_rc == -1) return -1;
}
else
{
CL_rc = run_kernel (hashcat_ctx, device_param, KERN_RUN_3, 1, false, 0);
if (CL_rc == -1) return -1;
}
}
// check : check if cracked
u32 num_cracked;
CL_err = hc_clEnqueueReadBuffer (hashcat_ctx, device_param->command_queue, device_param->d_result, CL_TRUE, 0, sizeof (u32), &num_cracked, 0, NULL, NULL);
if (CL_err != CL_SUCCESS) return -1;
// finish : cleanup and restore
device_param->kernel_params_buf32[27] = 0;
device_param->kernel_params_buf32[28] = 0;
device_param->kernel_params_buf32[29] = 0;
device_param->kernel_params_buf32[30] = 0;
device_param->kernel_params_buf32[31] = 0;
device_param->kernel_params_buf32[32] = 0;
device_param->kernel_params_buf32[33] = 0;
device_param->kernel_params_buf64[34] = 0;
device_param->kernel_params[15] = &device_param->d_digests_buf;
device_param->kernel_params[17] = &device_param->d_salt_bufs;
device_param->kernel_params[18] = &device_param->d_esalt_bufs;
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_pws_buf, device_param->size_pws); if (CL_rc == -1) return -1;
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_tmps, device_param->size_tmps); if (CL_rc == -1) return -1;
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_hooks, device_param->size_hooks); if (CL_rc == -1) return -1;
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_plain_bufs, device_param->size_plains); if (CL_rc == -1) return -1;
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_digests_shown, device_param->size_shown); if (CL_rc == -1) return -1;
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_result, device_param->size_results); if (CL_rc == -1) return -1;
if (user_options_extra->attack_kern == ATTACK_KERN_STRAIGHT)
{
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_rules_c, device_param->size_rules_c);
if (CL_rc == -1) return -1;
}
else if (user_options_extra->attack_kern == ATTACK_KERN_COMBI)
{
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_combs_c, device_param->size_combs);
if (CL_rc == -1) return -1;
}
else if (user_options_extra->attack_kern == ATTACK_KERN_BF)
{
CL_rc = run_kernel_bzero (hashcat_ctx, device_param, device_param->d_bfs_c, device_param->size_bfs);
if (CL_rc == -1) return -1;
}
// check return
if (num_cracked == 0)
{
hc_thread_mutex_lock (status_ctx->mux_display);
event_log_error (hashcat_ctx, "* Device #%u: ATTENTION! OpenCL kernel self-test failed.", device_param->device_id + 1);
event_log_warning (hashcat_ctx, "Your device driver installation is probably broken.");
event_log_warning (hashcat_ctx, "See also: https://hashcat.net/faq/wrongdriver");
event_log_warning (hashcat_ctx, NULL);
hc_thread_mutex_unlock (status_ctx->mux_display);
return -1;
}
return 0;
}
int clPeak::runComputeDP(cl::CommandQueue &queue, cl::Program &prog, device_info_t &devInfo)
{
float timed, gflops;
cl_uint workPerWI;
cl::NDRange globalSize, localSize;
cl_double A = 1.3f;
int iters = devInfo.computeIters;
if(!isComputeDP)
return 0;
if(!devInfo.doubleSupported)
{
cout << NEWLINE TAB TAB "No double precision support! Skipped" << endl;
return 0;
}
try
{
cl::Context ctx = queue.getInfo<CL_QUEUE_CONTEXT>();
uint globalWIs = (devInfo.numCUs) * (devInfo.computeWgsPerCU) * (devInfo.maxWGSize);
uint t = MIN((globalWIs * sizeof(cl_double)), devInfo.maxAllocSize);
t = roundToPowOf2(t);
globalWIs = t / sizeof(cl_double);
cl::Buffer outputBuf = cl::Buffer(ctx, CL_MEM_WRITE_ONLY, (globalWIs * sizeof(cl_double)));
globalSize = globalWIs;
localSize = devInfo.maxWGSize;
cl::Kernel kernel_v1(prog, "compute_dp_v1");
kernel_v1.setArg(0, outputBuf), kernel_v1.setArg(1, A);
cl::Kernel kernel_v2(prog, "compute_dp_v2");
kernel_v2.setArg(0, outputBuf), kernel_v2.setArg(1, A);
cl::Kernel kernel_v4(prog, "compute_dp_v4");
kernel_v4.setArg(0, outputBuf), kernel_v4.setArg(1, A);
cl::Kernel kernel_v8(prog, "compute_dp_v8");
kernel_v8.setArg(0, outputBuf), kernel_v8.setArg(1, A);
cl::Kernel kernel_v16(prog, "compute_dp_v16");
kernel_v16.setArg(0, outputBuf), kernel_v16.setArg(1, A);
cout << NEWLINE TAB TAB "Double-precision compute (GFLOPS)" << endl;
cout << setprecision(2) << fixed;
///////////////////////////////////////////////////////////////////////////
// Vector width 1
cout << TAB TAB TAB "double : "; cout.flush();
workPerWI = 4096; // Indicates flops executed per work-item
timed = run_kernel(queue, kernel_v1, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 2
cout << TAB TAB TAB "double2 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v2, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 4
cout << TAB TAB TAB "double4 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v4, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 8
cout << TAB TAB TAB "double8 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v8, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 16
cout << TAB TAB TAB "double16 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v16, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
}
catch(cl::Error error)
{
cerr << error.what() << "(" << error.err() << ")" << endl;
cerr << TAB TAB TAB "Tests skipped" << endl;
return -1;
}
return 0;
}
// main function
extern void main(void) {
unsigned char flag, numb_pages;
unsigned i, write_size, read_size,
temp_six_vector, six_vector, errors, write_offset;
unsigned boot_args[AT45_PAGE_SIZE / 4];
// calculate temp six vector
numb_pages = 0;
i = AT45_PAGE_NUMB;
while(i >>= 1)
numb_pages++;
temp_six_vector = (numb_pages << 13) + (AT45_PAGE_SIZE << 17);
// setup SYS interrupt
aic_configure_irq(AT91C_ID_SYS, AT91C_AIC_PRIOR_LOWEST,
AT91C_AIC_SRCTYPE_INT_LEVEL_SENSITIVE, aic_asm_sys_handler);
// enable SYS interrupt
aic_enable_irq(AT91C_ID_SYS);
// setup rtt - 1Hz clock
AT91C_BASE_ST->ST_RTMR = 0x4000;
upoint_r = pt_mem_area, upoint_w = (pt_mem_area + AT45DB642D_SIZE);
put_string("Init AT45DB642D and get device information\n");
if (!at45_init())
put_string("Device inited and ready\n");
else
put_string("Error!\n");
put_string("Press any key to load boot menu\n: ");
// setup rtt interrupt flag
AT91C_BASE_ST->ST_IER = AT91C_ST_RTTINC;
// enable RXRDY interrupt in DBGU
AT91C_BASE_DBGU->DBGU_IER |= AT91C_US_RXRDY;
wait_status = 1;
get_char();
while (flag != 'q') {
put_string("\nload (l), write(w), run kernel(r), quit(q), erase(e): ");
flag = get_char();
switch(flag) {
// loading data to sdram
case 'l':
put_string("Please trasfer the boot file:\n");
transfer_size = 0;
// setup rtt interrupt flag
AT91C_BASE_ST->ST_IER = AT91C_ST_RTTINC;
// enable RXRDY interrupt in DBGU
AT91C_BASE_DBGU->DBGU_IER |= AT91C_US_RXRDY;
while(!transfer_size);
delay(100000);
if (transfer_size > 0) {
put_string("Transfer complete\n");
util_printf("Byte's sended: %x\n", transfer_size);
}
break;
// writing bytes from data flash
case 'w':
if (transfer_size == 0) {
put_string("Please transfer begin, write end\n");
break;
}
else if (transfer_size > (AT45DB642D_SIZE)) {
put_string("Trasfer is larger than flash size\n");
break;
}
else {
if ((unsigned)transfer_size % AT45_PAGE_SIZE)
write_size = ((unsigned)transfer_size / AT45_PAGE_SIZE + 1) * AT45_PAGE_SIZE;
else
write_size = transfer_size;
put_string("Write boot(b) or linux kernel(n): ");
flag = get_char();
util_printf("%c\n", flag);
if (flag == 'b') {
write_offset = BOOT_OFFSET;
put_string("\nModification of Arm Interrupt Vector #6\n");
six_vector = (write_size / 512) + 1 + temp_six_vector;
util_printf("Six vector is 0x%x\n", six_vector);
upoint_w[5] = six_vector;
}
else {
write_offset = LINUX_OFFSET;
put_string("Writing args\n");
boot_args[0] = write_size;
if (!at45_write(BOOT_2_ARGS_OFFSET, boot_args, AT45_PAGE_SIZE))
put_string("Write success\n");
else
put_string("Error!\n");
}
util_printf("Write 0x%x bytes\n", write_size);
if (!at45_write(write_offset, upoint_w, write_size))
put_string("Write success\n");
else
put_string("Error!\n");
if (!at45_read(write_offset, upoint_r, write_size)) {
put_string("Read success\nStart verification\n");
six_vector = upoint_r[5];
if (write_offset == BOOT_OFFSET) {
if ((six_vector & 0xfffff000) - temp_six_vector) {
util_printf("Six vector is damage, current 0x%x, original 0x%x\n",
six_vector, temp_six_vector);
break;
}
else {
put_string("Six vector is correct\nStart code verification\n");
}
}
errors = 0;
for (i = 0; i < write_size / 4; i++) {
if (upoint_r[i] != upoint_w[i]) {
errors++;
util_printf("Addr - %x, write - %x, read - %x\n", i, upoint_w[i], upoint_r[i]);
}
}
put_string("Stop code verification\n");
if (errors)
put_string("Verification failed!\n");
else
put_string("Verification success!\n");
}
}
break;
// erase first page
case 'e':
if (!at45_read(0x0, upoint_r, 0x20)) {
six_vector = upoint_r[5];
read_size = (six_vector & 0xff) * 512;
for (i = 0; i <= read_size; i += AT45_PAGE_SIZE) {
if (!at45_page_erase(i))
put_string("Erase success\n");
else
put_string("Error!\n");
}
}
break;
// run 2boot code
case 'r':
run_kernel();
break;
// exit of loop
case 'q':
put_string("\nQuit & Reset\n");
AT91C_BASE_ST->ST_WDMR = 256 | AT91C_ST_RSTEN;
AT91C_BASE_ST->ST_CR = AT91C_ST_WDRST;
break;
// undef
default:
put_string("\nUndefined command\n");
break;
}
}
// Infinity loop
while (1) {
// Disable pck for idle cpu mode
AT91C_BASE_PMC->PMC_SCDR = AT91C_PMC_PCK;
}
}
static void run_kernel_loop()
{
int i;
for(i = 0; i < 2000; ++i)
run_kernel();
}
int clPeak::runComputeInteger(cl::CommandQueue &queue, cl::Program &prog, device_info_t &devInfo)
{
float timed, gflops;
cl_uint workPerWI;
cl::NDRange globalSize, localSize;
cl_int A = 4;
uint iters = devInfo.computeIters;
if(!isComputeInt)
return 0;
try
{
log->print(NEWLINE TAB TAB "Integer compute (GIOPS)" NEWLINE);
log->xmlOpenTag("integer_compute");
log->xmlAppendAttribs("unit", "gflops");
cl::Context ctx = queue.getInfo<CL_QUEUE_CONTEXT>();
uint64_t globalWIs = (devInfo.numCUs) * (devInfo.computeWgsPerCU) * (devInfo.maxWGSize);
uint64_t t = MIN((globalWIs * sizeof(cl_int)), devInfo.maxAllocSize) / sizeof(cl_int);
globalWIs = roundToMultipleOf(t, devInfo.maxWGSize);
cl::Buffer outputBuf = cl::Buffer(ctx, CL_MEM_WRITE_ONLY, (globalWIs * sizeof(cl_int)));
globalSize = globalWIs;
localSize = devInfo.maxWGSize;
cl::Kernel kernel_v1(prog, "compute_integer_v1");
kernel_v1.setArg(0, outputBuf), kernel_v1.setArg(1, A);
cl::Kernel kernel_v2(prog, "compute_integer_v2");
kernel_v2.setArg(0, outputBuf), kernel_v2.setArg(1, A);
cl::Kernel kernel_v4(prog, "compute_integer_v4");
kernel_v4.setArg(0, outputBuf), kernel_v4.setArg(1, A);
cl::Kernel kernel_v8(prog, "compute_integer_v8");
kernel_v8.setArg(0, outputBuf), kernel_v8.setArg(1, A);
cl::Kernel kernel_v16(prog, "compute_integer_v16");
kernel_v16.setArg(0, outputBuf), kernel_v16.setArg(1, A);
///////////////////////////////////////////////////////////////////////////
// Vector width 1
log->print(TAB TAB TAB "int : ");
workPerWI = 2048; // Indicates integer operations executed per work-item
timed = run_kernel(queue, kernel_v1, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 2
log->print(TAB TAB TAB "int2 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v2, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int2", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 4
log->print(TAB TAB TAB "int4 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v4, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int4", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 8
log->print(TAB TAB TAB "int8 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v8, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int8", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 16
log->print(TAB TAB TAB "int16 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v16, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int16", gflops);
///////////////////////////////////////////////////////////////////////////
log->xmlCloseTag(); // integer_compute
}
catch(cl::Error &error)
{
stringstream ss;
ss << error.what() << " (" << error.err() << ")" NEWLINE
<< TAB TAB TAB "Tests skipped" NEWLINE;
log->print(ss.str());
return -1;
}
return 0;
}