void SeqToolBarView::CacheDrawingInfo()
{
mPrefW = 0;
mPrefH = 0;
FreeCachedInfo();
toolref_vec toolRefs;
FillVec(toolRefs);
for (uint32 k = 0; k < toolRefs.size(); k++) {
// READ TOOL BLOCK
const AmTool* tool = toolRefs[k].ReadLock();
if (tool) {
BRect b = tool->IconBounds();
mPrefW += b.Width() + 1 + mSpace;
if (b.Height() > mPrefH) mPrefH = b.Height();
mCachedData.push_back( _SeqCachedTool(tool) );
}
toolRefs[k].ReadUnlock(tool);
// END READ TOOL BLOCK
}
/* Oops, I'm empty, prepare myself to just draw a string.
*/
if (mCachedData.size() < 1) {
mPrefW = 4 + StringWidth(EMPTY_STR) + 4;
mPrefH = 4 + arp_get_font_height(this) + 4;
return;
}
/* Extra space at the end to make it easier to drop in tools.
*/
mPrefW += LAST_BUTTON_PAD;
}
void BenchmarkLoop(CuContext* context, searchEngine_t engine, int count,
int numIterations, double* mgpuBenchmarks, double* thrustBenchmarks) {
////////////////////////////////////////////////////////////////////////////
// Prepare the device assets.
printf("ARRAY = %d elements.\n", count);
// Fill the array with sorted values of random repetition.
std::vector<T> values;
FillVec(values, count);
// Get a list of randomized keys for lower_bound search.
std::vector<T> keys(MaxQuerySize);
std::tr1::uniform_int<T> r(0, values.back());
for(int i(0); i < MaxQuerySize; ++i)
keys[i] = r(mt19937);
// Move both arrays to device memory.
DeviceMemPtr valuesDevice, keysDevice;
CUresult result = context->MemAlloc(values, &valuesDevice);
result = context->MemAlloc(keys, &keysDevice);
// Allocate result space.
DeviceMemPtr indicesDevice;
result = context->MemAlloc<uint>(MaxQuerySize, &indicesDevice);
if(CUDA_SUCCESS != result) {
printf("Error allocating.\n");
exit(0);
}
// Build the b-tree for MGPU Search.
searchType_t type = (8 == sizeof(T)) ? SEARCH_TYPE_INT64 :
SEARCH_TYPE_INT32;
int treeSize = searchTreeSize(count, type);
DeviceMemPtr btreeDevice;
result = context->ByteAlloc(treeSize, &btreeDevice);
searchStatus_t status = searchBuildTree(engine, count, type,
valuesDevice->Handle(), btreeDevice->Handle());
////////////////////////////////////////////////////////////////////////////
for(int q(0); q < NumQuerySizes; ++q) {
int querySize = QuerySizes[q][0];
int iterations = numIterations * QuerySizes[q][1];
mgpuBenchmarks[q] = 0;
thrustBenchmarks[q] = 0;
printf("\t(%d, %d):\n", count, querySize);
for(int t(0); t < NumTests; ++t) {
// Test MGPU
double throughput = MgpuBenchmark(engine, count, valuesDevice,
type, btreeDevice, iterations, querySize, keysDevice,
indicesDevice, &values[0], &keys[0]);
printf("\t\t%10.3f M/s\t\t", throughput / 1.0e6);
mgpuBenchmarks[q] = std::max(mgpuBenchmarks[q], throughput);
// Test thrust
throughput = ThrustBenchmark(count, valuesDevice, iterations,
querySize, keysDevice, indicesDevice, &values[0], &keys[0]);
printf("\t\t%10.3f M/s\n", throughput / 1.0e6);
thrustBenchmarks[q] = std::max(thrustBenchmarks[q], throughput);
}
}
printf("\n");
}
