int main(int argc, char* argv[]) {
using namespace asmjit;
Performance perf;
uint32_t kNumRepeats = 10;
uint32_t kNumIterations = 10000;
JitRuntime runtime;
X86Assembler a(&runtime);
X86Compiler c(&runtime);
uint32_t r, i;
// --------------------------------------------------------------------------
// [Bench - Opcode]
// --------------------------------------------------------------------------
perf.reset();
for (r = 0; r < kNumRepeats; r++) {
perf.start();
for (i = 0; i < kNumIterations; i++) {
asmgen::opcode(a);
void *p = a.make();
runtime.release(p);
a.reset();
}
perf.end();
}
printf("Opcode | Time: %-6u [ms] | Speed: %-9u [inst/s]\n",
perf.best, instPerMs(perf.best, kNumIterations, asmgen::kGenOpCodeInstCount));
// --------------------------------------------------------------------------
// [Bench - Blend]
// --------------------------------------------------------------------------
perf.reset();
for (r = 0; r < kNumRepeats; r++) {
perf.start();
for (i = 0; i < kNumIterations; i++) {
asmgen::blend(c);
void* p = c.make();
runtime.release(p);
c.reset();
}
perf.end();
}
printf("Blend | Time: %-6u [ms] | Speed: %-9u [inst/s]\n",
perf.best, instPerMs(perf.best, kNumIterations, asmgen::kGenBlendInstCount));
return 0;
}
void run() {
Performance p;
printf("Allocating buffers\n");
text_buf = alloc_shared_buffer<char> (LENGTH_OF_TEXT, &text_c);
pattern_buf= alloc_shared_buffer<char> (LENGTH_OF_PATTERN, &pattern_c);
result_buf= alloc_shared_buffer<char> (1, &result_c);
result_c[0] = 0;
init_string( text_c, pattern_c );
printf("Initializing kernels\n");
size_t task_dim = 1;
clKernelSet kernel_set (device, context, program);
// kernel_set.addKernel ("text_processor", 1, &task_dim, text_buf, LENGTH_OF_TEXT);
// kernel_set.addKernel ("word_processor", 1, &task_dim);
kernel_set.addKernel ("word_processor", 1, &task_dim, text_buf, LENGTH_OF_TEXT);
kernel_set.addKernel ("matching", 1, &task_dim, pattern_buf, LENGTH_OF_PATTERN, result_buf);
printf("Launching the kernel...\n");
p.start();
kernel_set.launch();
printf(" start waiting.... \n");
kernel_set.finish();
p.stop();
printf(" done Execution (OpenCL Channel) time = (%u,%u), result = %d\n", p.report_sec(), p.report_usec(),result_c[0]);
test_in_cpu( text_c, pattern_c );
return;
}
void test_in_cpu( char* text, char* pattern )
{
Performance p;
p.start();
bool match = false;
int count = 0;
for( int n = 0 ; n < LENGTH_OF_TEXT - 1; n ++ ) {
int m = 0;
if( n == 0 || text[n-1] == ' ' ) {
for( ; m < LENGTH_OF_PATTERN - 1; m ++ ) {
if( pattern[m] != text[n+m] )
break;
}
}
if( m == LENGTH_OF_PATTERN - 1 )
count ++;
}
p.stop();
printf(" done Execution (CPU) time = (%u,%u), result = %d\n", p.report_sec(), p.report_usec(),count);
}
static void benchX86(uint32_t arch, uint32_t callConv) {
using namespace asmjit;
Performance perf;
TestRuntime runtime(arch, callConv);
X86Assembler a(&runtime, arch);
X86Compiler c;
uint32_t r, i;
const char* archName = arch == kArchX86 ? "X86" : "X64";
// --------------------------------------------------------------------------
// [Bench - Opcode]
// --------------------------------------------------------------------------
size_t asmOutputSize = 0;
size_t cmpOutputSize = 0;
perf.reset();
for (r = 0; r < kNumRepeats; r++) {
asmOutputSize = 0;
perf.start();
for (i = 0; i < kNumIterations; i++) {
asmgen::opcode(a);
void *p = a.make();
runtime.release(p);
asmOutputSize += a.getCodeSize();
a.reset();
}
perf.end();
}
printf("%-12s (%s) | Time: %-6u [ms] | Speed: %7.3f [MB/s]\n",
"X86Assembler", archName, perf.best, mbps(perf.best, asmOutputSize));
// --------------------------------------------------------------------------
// [Bench - Blend]
// --------------------------------------------------------------------------
perf.reset();
for (r = 0; r < kNumRepeats; r++) {
cmpOutputSize = 0;
perf.start();
for (i = 0; i < kNumIterations; i++) {
c.attach(&a);
asmgen::blend(c);
c.finalize();
void* p = a.make();
runtime.release(p);
cmpOutputSize += a.getCodeSize();
a.reset();
}
perf.end();
}
printf("%-12s (%s) | Time: %-6u [ms] | Speed: %7.3f [MB/s]\n",
"X86Compiler", archName, perf.best, mbps(perf.best, cmpOutputSize));
}