Closed. This question is not reproducible or was caused by typos. It is not currently accepting answers.
This question was caused by a typo or a problem that can no longer be reproduced. While similar questions may be on-topic here, this one was resolved in a way less likely to help future readers.
Closed 6 years ago.
Improve this question
When I call curand, I always get the same number in a thread. They are different for each thread however. What am I doing wrong in the next code?
#define MAXTHREADS 2
#define NBBLOCKS 2
__global__ void testRand ( curandState * state, int nb ){
int id = threadIdx.x + blockIdx.x * blockDim.x;
int value;
for (int i=0;i<nb;i++){
curandState localState = state[id];
value = curand(&localState);
printf("Id %i, value %i\n",id,value);
}
}
__global__ void setup_kernel ( curandState * state, unsigned long seed )
{
int id = threadIdx.x + blockIdx.x * blockDim.x;
curand_init ( seed, id , 0, &state[id] );
}
/**
* Image comes in in horizontal lines
*/
void findOptimum() {
const dim3 blockSize(MAXTHREADS);
const dim3 gridSize(NBBLOCKS);
curandState* devStates;
cudaMalloc ( &devStates,MAXTHREADS*NBBLOCKS*sizeof( curandState ) );
time_t t;
time(&t);
setup_kernel <<< gridSize, blockSize >>> ( devStates, (unsigned long) t );
int nb = 4;
testRand <<< gridSize, blockSize >>> ( devStates,nb);
testRand <<< gridSize, blockSize >>> ( devStates,nb);
cudaFree(devStates);
}
It outputs:
Id 0, value -1075808309
Id 1, value -1660353324
Id 2, value 1282291714
Id 3, value -1892750252
Id 0, value -1075808309
Id 1, value -1660353324
Id 2, value 1282291714
Id 3, value -1892750252
...
This repeats a few times more.
As talonmies pointed out, I didn't modify the global state.
Adding state[id] = localState after te line with curand(localState) fixed the problem.
Related
I have a testing code that needs to update keys inside a device_vector of a class. Therefore, how do I divide portions of the work to especific threads?
Example of the code without the division:
__global__ void UpdateKeys(Request* vector, int size, int seed, int qt_threads){
curandState_t state;
curand_init(seed, threadIdx.x, 0, &state);
int id = blockIdx.x * blockDim.x + threadIdx.x;
if(id < size){
vector[i].key_ = (curand(&state % 100) / 100;
}
}
That vector is passed as a thrust::device_vector.
Examples of what I want:
1000 keys and 2000 threads: use only 1000 and give a key to each one.
1000 keys and 1000 threads: use it all.
1 key and 100 threads: use 1 thread.
500 keys and 250 threads: each thread take care of 2.
240 keys and 80 threads: each thread take care of 3.
If you modify your basic kernel structure like this:
__global__ void UpdateKeys(Request* vector, int size, int seed, int qt_threads){
curandState_t state;
curand_init(seed, threadIdx.x, 0, &state);
int id = blockIdx.x * blockDim.x + threadIdx.x;
int gid = blockDim.x * gridDim.x;
for(; id < size; id += gid){
vector[id].key_ = (curand(&state) % 100) / 100;
}
}
then it should be possible for any legal one dimensional block size (and number of one dimensional blocks) to process as many or as few inputs as you choose to provide via the size parameter. If you run more threads than keys, some threads will do nothing. If you run less threads than keys, some threads will process multiple keys.
I am trying to speed up a single program by using prefetches. The purpose of my program is just for test. Here is what it does:
It uses two int buffers of the same size
It reads one-by-one all the values of the first buffer
It reads the value at the index in the second buffer
It sums all the values taken from the second buffer
It does all the previous steps for bigger and bigger
At the end, I print the number of voluntary and involuntary CPU
In the very first time, values in the first buffers contains the values of its index (cf. function createIndexBuffer in the code just below) .
It will be more clear in the code of my program:
#include <stdio.h>
#include <stdlib.h>
#include <limits.h>
#include <sys/time.h>
#define BUFFER_SIZE ((unsigned long) 4096 * 100000)
unsigned int randomUint()
{
int value = rand() % UINT_MAX;
return value;
}
unsigned int * createValueBuffer()
{
unsigned int * valueBuffer = (unsigned int *) malloc(BUFFER_SIZE * sizeof(unsigned int));
for (unsigned long i = 0 ; i < BUFFER_SIZE ; i++)
{
valueBuffer[i] = randomUint();
}
return (valueBuffer);
}
unsigned int * createIndexBuffer()
{
unsigned int * indexBuffer = (unsigned int *) malloc(BUFFER_SIZE * sizeof(unsigned int));
for (unsigned long i = 0 ; i < BUFFER_SIZE ; i++)
{
indexBuffer[i] = i;
}
return (indexBuffer);
}
unsigned long long computeSum(unsigned int * indexBuffer, unsigned int * valueBuffer)
{
unsigned long long sum = 0;
for (unsigned int i = 0 ; i < BUFFER_SIZE ; i++)
{
unsigned int index = indexBuffer[i];
sum += valueBuffer[index];
}
return (sum);
}
unsigned int computeTimeInMicroSeconds()
{
unsigned int * valueBuffer = createValueBuffer();
unsigned int * indexBuffer = createIndexBuffer();
struct timeval startTime, endTime;
gettimeofday(&startTime, NULL);
unsigned long long sum = computeSum(indexBuffer, valueBuffer);
gettimeofday(&endTime, NULL);
printf("Sum = %llu\n", sum);
free(indexBuffer);
free(valueBuffer);
return ((endTime.tv_sec - startTime.tv_sec) * 1000 * 1000) + (endTime.tv_usec - startTime.tv_usec);
}
int main()
{
printf("sizeof buffers = %ldMb\n", BUFFER_SIZE * sizeof(unsigned int) / (1024 * 1024));
unsigned int timeInMicroSeconds = computeTimeInMicroSeconds();
printf("Time: %u micro-seconds = %.3f seconds\n", timeInMicroSeconds, (double) timeInMicroSeconds / (1000 * 1000));
}
If I launch it, I get the following output:
$ gcc TestPrefetch.c -O3 -o TestPrefetch && ./TestPrefetch
sizeof buffers = 1562Mb
Sum = 439813150288855829
Time: 201172 micro-seconds = 0.201 seconds
Quick and fast!!!
According to my knowledge (I may be wrong), one of the reason for having such a fast program is that, as I access my two buffers sequentially, data can be prefetched in the CPU cache.
We can make it more complex in order that data is (almost) prefeched in CPU cache. For example, we can just change the createIndexBuffer function in:
unsigned int * createIndexBuffer()
{
unsigned int * indexBuffer = (unsigned int *) malloc(BUFFER_SIZE * sizeof(unsigned int));
for (unsigned long i = 0 ; i < BUFFER_SIZE ; i++)
{
indexBuffer[i] = rand() % BUFFER_SIZE;
}
return (indexBuffer);
}
Let's try the program once again:
$ gcc TestPrefetch.c -O3 -o TestPrefetch && ./TestPrefetch
sizeof buffers = 1562Mb
Sum = 439835307963131237
Time: 3730387 micro-seconds = 3.730 seconds
More than 18 times slower!!!
We now arrive to my problem. Given the new createIndexBuffer function, I would like to speed up computeSum function using prefetch
unsigned long long computeSum(unsigned int * indexBuffer, unsigned int * valueBuffer)
{
unsigned long long sum = 0;
for (unsigned int i = 0 ; i < BUFFER_SIZE ; i++)
{
__builtin_prefetch((char *) &indexBuffer[i + 1], 0, 0);
unsigned int index = indexBuffer[i];
sum += valueBuffer[index];
}
return (sum);
}
of course I also have to change my createIndexBuffer in order it allocates a buffer having one more element
I relaunch my program: not better! As prefetch may be slower than one "for" loop iteration, I may prefetch not one element before but two elements before
__builtin_prefetch((char *) &indexBuffer[i + 2], 0, 0);
not better! two loops iterations? not better? Three? **I tried it until 50 (!!!) but I cannot enhance the performance of my function computeSum.
Can I would like help to understand why
Thank you very much for your help
I believe that above code is automatically optimized by CPU without any further space for manual optimization.
1. Main problem is that indexBuffer is sequentially accessed. Hardware prefetcher senses it and prefetches further values automatically, without need to call prefetch manually. So, during iteration #i, values indexBuffer[i+1], indexBuffer[i+2],... are already in cache. (By the way, there is no need to add artificial element to the end of array: memory access errors are silently ignored by prefetch instructions).
What you really need to do is to prefetch valueBuffer instead:
__builtin_prefetch((char *) &valueBuffer[indexBuffer[i + 1]], 0, 0);
2. But adding above line of code won't help either in such simple scenario. Cost of accessing memory is hundreds of cycles, while add instruction is ~1 cycle. Your code already spends 99% of time in memory accesses. Adding manual prefetch will make it this one cycle faster and no better.
Manual prefetch would really work well if your math were much more heavy (try it), like using an expression with large number of non-optimized out divisions (20-30 cycles each) or calling some math function (log, sin).
3. But even this doesn't guarantee to help. Dependency between loop iterations is very weak, it is only via sum variable. This allows CPU to execute instructions speculatively: it may start fetching valueBuffer[i+1] concurrently while still executing math for valueBuffer[i].
Prefetch fetches normally a full cache line. This is typically 64 bytes. So the random example fetches always 64 bytes for a 4 byte int. 16 times the data you actually need which fits very well with the slow down by a factor of 18. So the code is simply limited by memory throughput and not latency.
Sorry. What I gave you was not the correct version of my code. The correct version is, what you said:
__builtin_prefetch((char *) &valueBuffer[indexBuffer[i + prefetchStep]], 0, 0);
However, even with the right version, it is unfortunately not better
Then I adapted my program to try your suggestion using the sin function.
My adapted program is the following one:
#include <stdio.h>
#include <stdlib.h>
#include <limits.h>
#include <sys/time.h>
#include <math.h>
#define BUFFER_SIZE ((unsigned long) 4096 * 50000)
unsigned int randomUint()
{
int value = rand() % UINT_MAX;
return value;
}
unsigned int * createValueBuffer()
{
unsigned int * valueBuffer = (unsigned int *) malloc(BUFFER_SIZE * sizeof(unsigned int));
for (unsigned long i = 0 ; i < BUFFER_SIZE ; i++)
{
valueBuffer[i] = randomUint();
}
return (valueBuffer);
}
unsigned int * createIndexBuffer(unsigned short prefetchStep)
{
unsigned int * indexBuffer = (unsigned int *) malloc((BUFFER_SIZE + prefetchStep) * sizeof(unsigned int));
for (unsigned long i = 0 ; i < BUFFER_SIZE ; i++)
{
indexBuffer[i] = rand() % BUFFER_SIZE;
}
return (indexBuffer);
}
double computeSum(unsigned int * indexBuffer, unsigned int * valueBuffer, unsigned short prefetchStep)
{
double sum = 0;
for (unsigned int i = 0 ; i < BUFFER_SIZE ; i++)
{
__builtin_prefetch((char *) &valueBuffer[indexBuffer[i + prefetchStep]], 0, 0);
unsigned int index = indexBuffer[i];
sum += sin(valueBuffer[index]);
}
return (sum);
}
unsigned int computeTimeInMicroSeconds(unsigned short prefetchStep)
{
unsigned int * valueBuffer = createValueBuffer();
unsigned int * indexBuffer = createIndexBuffer(prefetchStep);
struct timeval startTime, endTime;
gettimeofday(&startTime, NULL);
double sum = computeSum(indexBuffer, valueBuffer, prefetchStep);
gettimeofday(&endTime, NULL);
printf("prefetchStep = %d, Sum = %f - ", prefetchStep, sum);
free(indexBuffer);
free(valueBuffer);
return ((endTime.tv_sec - startTime.tv_sec) * 1000 * 1000) + (endTime.tv_usec - startTime.tv_usec);
}
int main()
{
printf("sizeof buffers = %ldMb\n", BUFFER_SIZE * sizeof(unsigned int) / (1024 * 1024));
for (unsigned short prefetchStep = 0 ; prefetchStep < 250 ; prefetchStep++)
{
unsigned int timeInMicroSeconds = computeTimeInMicroSeconds(prefetchStep);
printf("Time: %u micro-seconds = %.3f seconds\n", timeInMicroSeconds, (double) timeInMicroSeconds / (1000 * 1000));
}
}
The output is:
$ gcc TestPrefetch.c -O3 -o TestPrefetch -lm && taskset -c 7 ./TestPrefetch
sizeof buffers = 781Mb
prefetchStep = 0, Sum = -1107.523504 - Time: 20895326 micro-seconds = 20.895 seconds
prefetchStep = 1, Sum = 13456.262424 - Time: 12706720 micro-seconds = 12.707 seconds
prefetchStep = 2, Sum = -20179.289469 - Time: 12136174 micro-seconds = 12.136 seconds
prefetchStep = 3, Sum = 12068.302534 - Time: 11233803 micro-seconds = 11.234 seconds
prefetchStep = 4, Sum = 21071.238160 - Time: 10855348 micro-seconds = 10.855 seconds
prefetchStep = 5, Sum = -22648.280105 - Time: 10517861 micro-seconds = 10.518 seconds
prefetchStep = 6, Sum = 22665.381676 - Time: 9205809 micro-seconds = 9.206 seconds
prefetchStep = 7, Sum = 2461.741268 - Time: 11391088 micro-seconds = 11.391 seconds
...
So here, it works better! Honestly, I was almost sure that it will not be better because the math function cost is higher compared to the memory access.
If anyone could give me more information about why it is better now, I would appreciate it
Thank you very much
Summary:
Any ideas about how to further improve upon the basic scatter operation in CUDA? Especially if one knows it will only be used to compact a larger array into a smaller one? or why the below methods of vectorizing memory ops and shared memory didn't work? I feel like there may be something fundamental I am missing and any help would be appreciated.
EDIT 03/09/15: So I found this Parallel For All Blog post "Optimized Filtering with Warp-Aggregated Atomics". I had assumed atomics would be intrinsically slower for this purpose, however I was wrong - especially since I don't think I care about maintaining element order in the array during my simulation. I'll have to think about it some more and then implement it to see what happens!
EDIT 01/04/16: I realized I never wrote about my results. Unfortunately in that Parallel for All Blog post they compared the global atomic method for compact to the Thrust prefix-sum compact method, which is actually quite slow. CUB's Device::IF is much faster than Thrust's - as is the prefix-sum version I wrote using CUB's Device::Scan + custom code. The warp-aggregrate global atomic method is still faster by about 5-10%, but nowhere near the 3-4x faster I had been hoping for based on the results in the blog. I'm still using the prefix-sum method as while maintaining element order is not necessary, I prefer the consistency of the prefix-sum results and the advantage from the atomics is not very big. I still try various methods to improve compact, but so far only marginal improvements (2%) at best for dramatically increased code complexity.
Details:
I am writing a simulation in CUDA where I compact out elements I am no longer interested in simulating every 40-60 time steps. From profiling it seems that the scatter op takes up the most amount of time when compacting - more so than the filter kernel or the prefix sum. Right now I use a pretty basic scatter function:
__global__ void scatter_arrays(float * new_freq, const float * const freq, const int * const flag, const int * const scan_Index, const int freq_Index){
int myID = blockIdx.x*blockDim.x + threadIdx.x;
for(int id = myID; id < freq_Index; id+= blockDim.x*gridDim.x){
if(flag[id]){
new_freq[scan_Index[id]] = freq[id];
}
}
}
freq_Index is the number of elements in the old array. The flag array is the result from the filter. Scan_ID is the result from the prefix sum on the flag array.
Attempts I've made to improve it are to read the flagged frequencies into shared memory first and then write from shared memory to global memory - the idea being that the writes to global memory would be more coalesced amongst the warps (e.g. instead of thread 0 writing to position 0 and thread 128 writing to position 1, thread 0 would write to 0 and thread 1 would write to 1). I also tried vectorizing the reads and the writes - instead of reading and writing floats/ints I read/wrote float4/int4 from the global arrays when possible, so four numbers at a time. This I thought might speed up the scatter by having fewer memory ops transferring larger amounts of memory. The "kitchen sink" code with both vectorized memory loads/stores and shared memory is below:
const int compact_threads = 256;
__global__ void scatter_arrays2(float * new_freq, const float * const freq, const int * const flag, const int * const scan_Index, const int freq_Index){
int gID = blockIdx.x*blockDim.x + threadIdx.x; //global ID
int tID = threadIdx.x; //thread ID within block
__shared__ float row[4*compact_threads];
__shared__ int start_index[1];
__shared__ int end_index[1];
float4 myResult;
int st_index;
int4 myFlag;
int4 index;
for(int id = gID; id < freq_Index/4; id+= blockDim.x*gridDim.x){
if(tID == 0){
index = reinterpret_cast<const int4*>(scan_Index)[id];
myFlag = reinterpret_cast<const int4*>(flag)[id];
start_index[0] = index.x;
st_index = index.x;
myResult = reinterpret_cast<const float4*>(freq)[id];
if(myFlag.x){ row[0] = myResult.x; }
if(myFlag.y){ row[index.y-st_index] = myResult.y; }
if(myFlag.z){ row[index.z-st_index] = myResult.z; }
if(myFlag.w){ row[index.w-st_index] = myResult.w; }
}
__syncthreads();
if(tID > 0){
myFlag = reinterpret_cast<const int4*>(flag)[id];
st_index = start_index[0];
index = reinterpret_cast<const int4*>(scan_Index)[id];
myResult = reinterpret_cast<const float4*>(freq)[id];
if(myFlag.x){ row[index.x-st_index] = myResult.x; }
if(myFlag.y){ row[index.y-st_index] = myResult.y; }
if(myFlag.z){ row[index.z-st_index] = myResult.z; }
if(myFlag.w){ row[index.w-st_index] = myResult.w; }
if(tID == blockDim.x -1 || gID == mutations_Index/4 - 1){ end_index[0] = index.w + myFlag.w; }
}
__syncthreads();
int count = end_index[0] - st_index;
int rem = st_index & 0x3; //equivalent to modulo 4
int offset = 0;
if(rem){ offset = 4 - rem; }
if(tID < offset && tID < count){
new_mutations_freq[population*new_array_Length+st_index+tID] = row[tID];
}
int tempID = 4*tID+offset;
if((tempID+3) < count){
reinterpret_cast<float4*>(new_freq)[tID] = make_float4(row[tempID],row[tempID+1],row[tempID+2],row[tempID+3]);
}
tempID = tID + offset + (count-offset)/4*4;
if(tempID < count){ new_freq[st_index+tempID] = row[tempID]; }
}
int id = gID + freq_Index/4 * 4;
if(id < freq_Index){
if(flag[id]){
new_freq[scan_Index[id]] = freq[id];
}
}
}
Obviously it gets a bit more complicated. :) While the above kernel seems stable when there are hundreds of thousands of elements in the array, I've noticed a race condition when the array numbers in the tens of millions. I'm still trying to track the bug down.
But regardless, neither method (shared memory or vectorization) together or alone improved performance. I was especially surprised by the lack of benefit from vectorizing the memory ops. It had helped in other functions I had written, though now I am wondering if maybe it helped because it increased Instruction-Level-Parallelism in the calculation steps of those other functions rather than the fewer memory ops.
I found the algorithm mentioned in this poster (similar algorithm also discussed in this paper) works pretty well, especially for compacting large arrays. It uses less memory to do it and is slightly faster than my previous method (5-10%). I put in a few tweaks to the poster's algorithm: 1) eliminating the final warp shuffle reduction in phase 1, can simply sum the elements as they are calculated, 2) giving the function the ability to work over more than just arrays sized as a multiple of 1024 + adding grid-strided loops, and 3) allowing each thread to load their registers simultaneously in phase 3 instead of one at a time. I also use CUB instead of Thrust for Inclusive sum for faster scans. There may be more tweaks I can make, but for now this is good.
//kernel phase 1
int myID = blockIdx.x*blockDim.x + threadIdx.x;
//padded_length is nearest multiple of 1024 > true_length
for(int id = myID; id < (padded_length >> 5); id+= blockDim.x*gridDim.x){
int lnID = threadIdx.x % warp_size;
int warpID = id >> 5;
unsigned int mask;
unsigned int cnt=0;//;//
for(int j = 0; j < 32; j++){
int index = (warpID<<10)+(j<<5)+lnID;
bool pred;
if(index > true_length) pred = false;
else pred = predicate(input[index]);
mask = __ballot(pred);
if(lnID == 0) {
flag[(warpID<<5)+j] = mask;
cnt += __popc(mask);
}
}
if(lnID == 0) counter[warpID] = cnt; //store sum
}
//kernel phase 2 -> CUB Inclusive sum transforms counter array to scan_Index array
//kernel phase 3
int myID = blockIdx.x*blockDim.x + threadIdx.x;
for(int id = myID; id < (padded_length >> 5); id+= blockDim.x*gridDim.x){
int lnID = threadIdx.x % warp_size;
int warpID = id >> 5;
unsigned int predmask;
unsigned int cnt;
predmask = flag[(warpID<<5)+lnID];
cnt = __popc(predmask);
//parallel prefix sum
#pragma unroll
for(int offset = 1; offset < 32; offset<<=1){
unsigned int n = __shfl_up(cnt, offset);
if(lnID >= offset) cnt += n;
}
unsigned int global_index = 0;
if(warpID > 0) global_index = scan_Index[warpID - 1];
for(int i = 0; i < 32; i++){
unsigned int mask = __shfl(predmask, i); //broadcast from thread i
unsigned int sub_group_index = 0;
if(i > 0) sub_group_index = __shfl(cnt, i-1);
if(mask & (1 << lnID)){
compacted_array[global_index + sub_group_index + __popc(mask & ((1 << lnID) - 1))] = input[(warpID<<10)+(i<<5)+lnID];
}
}
}
}
EDIT: There is a newer article by a subset of the poster authors where they examine a faster variation of compact than what is written above. However, their new version is not order preserving, so not useful for myself and I haven't implemented it to test it out. That said, if your project doesn't rely on object order, their newer compact version can probably speed up your algorithm.
The original problem was launching more threads that it is possible like this:
someKernel<<<1 , 1025>>> ( ... );
and not detecting the error, as I did not know how to detect kernel call errors. This is explained well in talonmies answer in this question:
What is the canonical way to check for errors using the CUDA runtime API?
Instead of modifying the code I presented I wrote my own for conciseness:
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <stdio.h>
#define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__); }
inline void gpuAssert(cudaError_t cudaError, char *file, int line, bool abort=true)
{
if (cudaError != cudaSuccess)
{
fprintf(stderr,"GPUassert: %s %s %d\n", cudaGetErrorString(cudaError), file, line);
}
}
__global__ void addKernel(const int *dev_a, const int *dev_b, int *dev_c)
{
int i = threadIdx.x;
if ( i < 5 )
dev_c[i] = dev_a[i] + dev_b[i];
}
int main()
{
const int arraySize = 5;
const int a[arraySize] = { 1, 2, 3, 4, 5 };
const int b[arraySize] = { 10, 20, 30, 40, 50 };
int c[arraySize] = { 0 };
int *dev_a(nullptr), *dev_b(nullptr), *dev_c(nullptr);
gpuErrchk( cudaMalloc((void**)&dev_a, arraySize * sizeof(int)) );
gpuErrchk( cudaMalloc((void**)&dev_b, arraySize * sizeof(int)) );
gpuErrchk( cudaMalloc((void**)&dev_c, arraySize * sizeof(int)) );
gpuErrchk( cudaMemcpy(dev_a, a, arraySize * sizeof(int), cudaMemcpyHostToDevice) );
gpuErrchk( cudaMemcpy(dev_b, b, arraySize * sizeof(int), cudaMemcpyHostToDevice) );
const int testMax1D = 1025;
dim3 testMax2D ( 32, 33 );
addKernel<<<1, testMax2D>>> ( dev_a , dev_b, dev_c );
gpuErrchk( cudaPeekAtLastError() );
gpuErrchk( cudaDeviceSynchronize() );
gpuErrchk( cudaMemcpy( c, dev_c, arraySize * sizeof(int), cudaMemcpyDeviceToHost) );
printf("{1,2,3,4,5} + {10,20,30,40,50} = {%d,%d,%d,%d,%d}\n",
c[0], c[1], c[2], c[3], c[4]);
cudaFree(dev_a);
cudaFree(dev_b);
cudaFree(dev_c);
return 0;
}
I now get correct error reports. Thank you for your patience.
I don't understand this call in the gpuAssert function, so I ommited it:
if (abort) exit(code);
Is exit a custom written function or something I missed?
There are two classes of errors that can occur with kernel launches and they need to be checked for in separate steps, following a particular order.
The first class of errors is reported synchronously when a kernel call is made and prior to the kernel actually being launched on the device, i.e. these are "pre-launch" errors. These errors typically involve requesting more of a particular resource than is available (e.g. too much shared memory, too many threads). Check for these by calling cudaGetLastError() immediately after a kernel call.
The second class of errors are those that occur at some point in time after the kernel was launched on the device (e.g. memory access violation, timeout of watchdog timer). These are "post-launch" errors. The reason they are reported some time after a kernel call, is a natural consequence of kernel launches occuring asynchronously. They are reported at the next opportunity, which is usually the next synchronous API call. Check for these by calling cudaDeviceSynchronize() and examining its status return.
The posted code is missing a check for errors of the first class.
I am trying to use CURAND library to generate random numbers which are completely independent of each other from 0 to 100. Hence I am giving time as seed to each thread and specifying the "id = threadIdx.x + blockDim.x * blockIdx.x" as sequence and offset .
Then after getting the random number as float, I multiply it by 100 and take its integer value.
Now, the problem I am facing is that its getting the same random number for the thread [0,0] and [0,1], no matter how many times I run the code which is 11. I am unable to understand what am I doing wrong. Please help.
I am pasting my code below:
#include <stdlib.h>
#include <stdio.h>
#include <math.h>
#include<curand_kernel.h>
#include "util/cuPrintf.cu"
#include<time.h>
#define NE WA*HA //Total number of random numbers
#define WA 2 // Matrix A width
#define HA 2 // Matrix A height
#define SAMPLE 100 //Sample number
#define BLOCK_SIZE 2 //Block size
__global__ void setup_kernel ( curandState * state, unsigned long seed )
{
int id = threadIdx.x + blockIdx.x + blockDim.x;
curand_init ( seed, id , id, &state[id] );
}
__global__ void generate( curandState* globalState, float* randomMatrix )
{
int ind = threadIdx.x + blockIdx.x * blockDim.x;
if(ind < NE){
curandState localState = globalState[ind];
float stopId = curand_uniform(&localState) * SAMPLE;
cuPrintf("Float random value is : %f",stopId);
int stop = stopId ;
cuPrintf("Random number %d\n",stop);
for(int i = 0; i < SAMPLE; i++){
if(i == stop){
float random = curand_normal( &localState );
cuPrintf("Random Value %f\t",random);
randomMatrix[ind] = random;
break;
}
}
globalState[ind] = localState;
}
}
/////////////////////////////////////////////////////////
// Program main
/////////////////////////////////////////////////////////
int main(int argc, char** argv)
{
// 1. allocate host memory for matrix A
unsigned int size_A = WA * HA;
unsigned int mem_size_A = sizeof(float) * size_A;
float* h_A = (float* ) malloc(mem_size_A);
time_t t;
// 2. allocate device memory
float* d_A;
cudaMalloc((void**) &d_A, mem_size_A);
// 3. create random states
curandState* devStates;
cudaMalloc ( &devStates, size_A*sizeof( curandState ) );
// 4. setup seeds
int n_blocks = size_A/BLOCK_SIZE;
time(&t);
printf("\nTime is : %u\n",(unsigned long) t);
setup_kernel <<< n_blocks, BLOCK_SIZE >>> ( devStates, (unsigned long) t );
// 4. generate random numbers
cudaPrintfInit();
generate <<< n_blocks, BLOCK_SIZE >>> ( devStates,d_A );
cudaPrintfDisplay(stdout, true);
cudaPrintfEnd();
// 5. copy result from device to host
cudaMemcpy(h_A, d_A, mem_size_A, cudaMemcpyDeviceToHost);
// 6. print out the results
printf("\n\nMatrix A (Results)\n");
for(int i = 0; i < size_A; i++)
{
printf("%f ", h_A[i]);
if(((i + 1) % WA) == 0)
printf("\n");
}
printf("\n");
// 7. clean up memory
free(h_A);
cudaFree(d_A);
}
Output that I get is :
Time is : 1347857063
[0, 0]: Float random value is : 11.675105[0, 0]: Random number 11
[0, 0]: Random Value 0.358356 [0, 1]: Float random value is : 11.675105[0, 1]: Random number 11
[0, 1]: Random Value 0.358356 [1, 0]: Float random value is : 63.840496[1, 0]: Random number 63
[1, 0]: Random Value 0.696459 [1, 1]: Float random value is : 44.712799[1, 1]: Random number 44
[1, 1]: Random Value 0.735049
There are a few things wrong here, I'm addressing the first ones here to get you started:
General points
Please check the return values of all CUDA API calls, see here for more info.
Please run cuda-memcheck to check for obvious things like out-of-bounds accesses.
Specific points
When allocating space for the RNG state, you should have space for one state per thread (not one per matrix element as you have now).
Your thread ID calculation in setup_kernel() is wrong, should be threadIdx.x + blockIdx.x * blockDim.x (* instead of +).
You use the thread ID as the sequence number as well as the offset, you should just set the offset to zero as described in the cuRAND manual:
For the highest quality parallel pseudorandom number generation, each
experiment should be assigned a unique seed. Within an experiment,
each thread of computation should be assigned a unique sequence
number.
Finally you're running two threads per block, that's incredibly inefficient. Check out the CUDA C Programming Guide, in the "maximize utilization" section for more information, but you should be looking to launch a multiple of 32 threads per block (e.g. 128, 256) and a large number of blocks (e.g. tens of thousands). If you're problem is small then consider running multiple problems at once (either batched in a single kernel launch or as kernels in different streams to get concurrent execution).