How to use CUDA programming to calculate and process the correct number

[vokoyo]

Bandwidth test - test memory bandwidth.

Especially important for PCIE capability. Different MB has different PCIE capability.

The CUDA adaptor performance is depend on the capability of PCIE. It could be the performance bottleneck.

On the following programming drills, the number of clock cycles necessary for computation and utilised memory bandwidth have to be computing.

(1) parallelization in the programs - using 256 threads

(2) improving the memory access modes

(3) testing the parallelization by using 512/1024

(4) utilizing BLOCKS in the computation

(5) utilizing shared memory

(6) improving the computation performance by using a Treesum algorithm

(7) resolving the memory band conflict issue, encountered in applying Treesum algorithm with the shared memory

[vokoyo]

My own coding sample however have some errors -

#include <stdio.h>
#include <stdlib.h>
#include <cuda_runtime.h>
#include <time.h>
#include <math.h>

int main()
{
float *a, *b, *c, *d;
int n = 1000;

if (!InitCUDA())
return 0;
a = (float*)malloc(sizeof(float)* n * n);
b = (float*)malloc(sizeof(float)* n * n);
c = (float*)malloc(sizeof(float)* n * n);
d = (float*)malloc(sizeof(float)* n * n);

srand(0);
matgen(a, n, n);
matgen(b, n, n);

clock_t time = matmultCUDA(a, n, b, n, c, n, n);
matmult(a, n, b, n, d, n, n);
compare_mat(c, n, d, n, n);

double sec = (double)time / CLOCKS_PER_SEC;

printf("Time used: %.2f (%.2lf GFLOPS)\n", sec, 2.0 * n * n * n / (sec * 1E9));

return 0;
}
void matgen(float* a, int lda, int n)
{
int i, j;
for (i = 0; i < n; i++)
{
for (j = 0; j < n; j++)
{
a[i* lda + j] = (float)rand() / RAND_MAX + (float)rand() / (RAND_MAX * RAND_MAX);
}
}
}

void matmult(const float* a, int lda, const float* b, int ldb, float* c, int ldc, int n)
{
int i, j, k;

for (i = 0; i< n; i++)
{
for (j = 0; j < n; j++)
{
double t = 0;
for (k = 0; k < n; k++) {
t += a[i* lda + k] * b[k * ldb + j];
}
c[i* ldc + j] = t;
}
}

void compare_mat(const float* a, int lda, const float* b, int ldb, int n)
{
float max_err = 0;
float average_err = 0; inti, j;
for (i = 0; i< n; i++)
{
for (j = 0; j < n; j++)
{
if (b[i* ldb + j] != 0)
{
float err = fabs((a[i* lda + j] - b[i* ldb + j]) / b[i* ldb + j]);
if (max_err< err) max_err = err;
average_err += err;
}
}
}
printf("Max error: %g Average error:%g\n", max_err, average_err / (n * n));
}

#define NUM_THREADS 256

clock_t matmultCUDA(const float* a, int lda, const float* b, int ldb, float* c, int ldc, int n)
{
float *ac, *bc, *cc;
clock_tstart, end;

start = clock();
cudaMalloc((void**)&ac, sizeof(float)* n * n);
cudaMalloc((void**)&bc, sizeof(float)* n * n);
cudaMalloc((void**)&cc, sizeof(float)* n * n);

cudaMemcpy2D(ac, sizeof(float)* n, a, sizeof(float)* lda, sizeof(float)* n, n, cudaMemcpyHostToDevice);
cudaMemcpy2D(bc, sizeof(float)* n, b, sizeof(float)* ldb, sizeof(float)* n, n, cudaMemcpyHostToDevice);

intblocks = (n + NUM_THREADS - 1) / NUM_THREADS;
matMultCUDA<<<blocks * n, NUM_THREADS >>>(ac, n, bc, n, cc, n, n);

cudaMemcpy2D(c, sizeof(float)* ldc, cc, sizeof(float)* n, sizeof(float)* n, n, cudaMemcpyDeviceToHost);

cudaFree(ac);
cudaFree(bc);
cudaFree(cc);

end = clock();
return end - start;
}

__global__ static void matMultCUDA(const float* a, size_t lda, const float* b, size_t ldb, float* c, size_t ldc, int n)
{
__shared__ float
matA[BLOCK_SIZE][BLOCK_SIZE];
__shared__ float matB[BLOCK_SIZE][BLOCK_SIZE];
constinttidc = threadIdx.x;
constinttidr = threadIdx.y;
constintbidc = blockIdx.x* BLOCK_SIZE;
constintbidr = blockIdx.y* BLOCK_SIZE;
int i, j;

float results = 0;
float comp = 0;

for (j = 0; j < n; j += BLOCK_SIZE)
{
matA[tidr][tidc] = a[(tidr + bidr) * lda + tidc + j];
matB[tidr][tidc] = b[(tidr + j) * ldb + tidc + bidc];

__syncthreads();

for (i = 0; i< BLOCK_SIZE; i++)
{
float t; comp -= matA[tidr][i] * matB[i][tidc];
t = results - comp; comp = (t - results) + comp; results = t;
}

__syncthreads();
}

c[(tidr + bidr) * ldc + tidc + bidc] = results;
}

[vokoyo]

I get some errors for my coding sample as below -



#include <stdio.h>
#include <stdlib.h>
#include <cuda_runtime.h>
#include <time.h>
#include <math.h>

#define DATA_SIZE 1048576
#define BLOCK_NUM 32
#define THREAD_NUM 256

int data[DATA_SIZE];

bool InitCUDA()
{
int count;

cudaGetDeviceCount(&count);
if (count == 0)
{
fprintf(stderr, "There is no device.\n");
return false;
}

int i;
for (i = 0; i < count; i++)
{
cudaDeviceProp prop;
if (cudaGetDeviceProperties(&prop, i) == cudaSuccess)
{
if (prop.major >= 1)
{
cudaGetDeviceProperties(&prop, i);
printf("Device Name: %s\n", prop.name);
printf("Total global mem: %1u bytes\n", prop.totalGlobalMem);
printf("Max threads per block: %d\n", prop.maxThreadsPerBlock);
printf("Clock rate: %.2f GHz\n", prop.clockRate*1e-6f);
printf("\n");
break;
}
}

cudaSetDevice(i);
}
return true;
}

__global__ static void sumOfSquares(int *num, int* result, clock_t* time)
{
const int tid = threadIdx.x;
const int size = DATA_SIZE / THREAD_NUM;
int sum = 0;
int i;
clock_t start;
if (tid == 0) start = clock();
for (i = tid * size; i < (tid + 1) * size; i++)
{
sum += num[i] * num[i];
}

result[tid] = sum;
if (tid == 0) *time = clock() - start;
}

int* gpudata, *result;
clock_t* time;
cudaMalloc((void**)&gpudata, sizeof(int)* DATA_SIZE);
cudaMalloc((void**)&result, sizeof(int)* THREAD_NUM);
cudaMalloc((void**)&time, sizeof(clock_t));
cudaMemcpy(gpudata, data, sizeof(int)* DATA_SIZE, cudaMemcpyHostToDevice);

sumOfSquares << <1, THREAD_NUM, 0 >> >(gpudata, result, time);

int sum[THREAD_NUM];
clock_t time_used;
cudaMemcpy(&sum, result, sizeof(int)* THREAD_NUM, cudaMemcpyDeviceToHost);
cudaMemcpy(&time_used, time, sizeof(clock_t), cudaMemcpyDeviceToHost);
cudaFree(gpudata);
cudaFree(result);
cudaFree(time);

int final_sum = 0;
for (int i = 0; i < THREAD_NUM; i++)
{
final_sum += sum[i];
}

printf("sum: %d time: %d\n", final_sum, time_used);

final_sum = 0;
for (int i = 0; i< DATA_SIZE; i++)
{
sum += data[i] * data[i];
}
printf("sum (CPU): %d\n", final_sum);

__global__ static void sumOfSquares(int *num, int* result, clock_t* time)
{
const int tid = threadIdx.x;
int sum = 0;
int i;
clock_t start;
if (tid == 0) start = clock();
for (i = tid; i< DATA_SIZE; i += THREAD_NUM)
{
sum += num[i] * num[i];
}

result[tid] = sum;
if (tid == 0) *time = clock() - start;
}

__global__ static void sumOfSquares(int *num, int* result, clock_t* time)
{
const int tid = threadIdx.x;
const int bid = blockIdx.x;
int sum = 0;
int i;
if (tid == 0) time[bid] = clock();
for (i = bid * THREAD_NUM + tid; i< DATA_SIZE; i += BLOCK_NUM * THREAD_NUM)
{
sum += num[i] * num[i];
}

result[bid * THREAD_NUM + tid] = sum;
if (tid == 0) time[bid + BLOCK_NUM] = clock();
}

int* gpudata, *result;
clock_t* time;
cudaMalloc((void**)&gpudata, sizeof(int)* DATA_SIZE);
cudaMalloc((void**)&result, sizeof(int)* THREAD_NUM * BLOCK_NUM);
cudaMalloc((void**)&time, sizeof(clock_t)* BLOCK_NUM * 2);
cudaMemcpy(gpudata, data, sizeof(int)* DATA_SIZE, cudaMemcpyHostToDevice);

sumOfSquares << <BLOCK_NUM, THREAD_NUM, 0 >> >(gpudata, result, time);

int sum[THREAD_NUM * BLOCK_NUM];
clock_t time_used[BLOCK_NUM * 2];
cudaMemcpy(&sum, result, sizeof(int)* THREAD_NUM * BLOCK_NUM, cudaMemcpyDeviceToHost);
cudaMemcpy(&time_used, time, sizeof(clock_t)* BLOCK_NUM * 2, cudaMemcpyDeviceToHost);
cudaFree(gpudata);
cudaFree(result);
cudaFree(time);

intfinal_sum = 0;
for (inti = 0; i< THREAD_NUM * BLOCK_NUM; i++)
{
final_sum += sum[i];
}

clock_t min_start, max_end;
min_start = time_used[0];
max_end = time_used[BLOCK_NUM];
for (inti = 1; i< BLOCK_NUM; i++)
{
if (min_start > time_used[i])min_start = time_used[i];
if (max_end < time_used[i + BLOCK_NUM])max_end = time_used[i + BLOCK_NUM];
}

printf("sum: %d time: %d\n", final_sum, max_end - min_start);

__global__ static void sumOfSquares(int *num, int* result, clock_t* time)
{
extern __shared__ int shared[];
const int tid = threadIdx.x;
const int bid = blockIdx.x;

int i;
if (tid == 0) time[bid] = clock();
shared[tid] = 0;

for (i = bid * THREAD_NUM + tid; i< DATA_SIZE; i += BLOCK_NUM * THREAD_NUM)
{
shared[tid] += num[i] * num[i];
}

__syncthreads();
if (tid == 0)
{
for (i = 1; i< THREAD_NUM; i++)
{
shared[0] += shared[i];
}
result[bid] = shared[0];
}

if (tid == 0) time[bid + BLOCK_NUM] = clock();
}

int* gpudata, *result;
clock_t* time;
cudaMalloc((void**)&gpudata, sizeof(int)* DATA_SIZE);
cudaMalloc((void**)&result, sizeof(int)* BLOCK_NUM);
cudaMalloc((void**)&time, sizeof(clock_t)* BLOCK_NUM * 2);
cudaMemcpy(gpudata, data, sizeof(int)* DATA_SIZE, cudaMemcpyHostToDevice);

sumOfSquares << <BLOCK_NUM, THREAD_NUM, THREAD_NUM * sizeof(int) >> >(gpudata, result, time);

int sum[BLOCK_NUM];
clock_t time_used[BLOCK_NUM * 2];
cudaMemcpy(&sum, result, sizeof(int)* BLOCK_NUM, cudaMemcpyDeviceToHost);
cudaMemcpy(&time_used, time, sizeof(clock_t)* BLOCK_NUM * 2, cudaMemcpyDeviceToHost);
cudaFree(gpudata);
cudaFree(result);
cudaFree(time);

int final_sum = 0;
for (int i = 0; i< BLOCK_NUM; i++)
{
final_sum += sum[i];
}

__global__ static void sumOfSquares(int*num, int* result, clock_t* time)
{
extern __shared__ int shared[];
const int tid = threadIdx.x;
const int bid = blockIdx.x;

int i;
into ffset = 1, mask = 1;
if (tid == 0) time[bid] = clock();
shared[tid] = 0;
for (i = bid * THREAD_NUM + tid; i< DATA_SIZE; i += BLOCK_NUM * THREAD_NUM)
{
shared[tid] += num[i] * num[i];
}

__syncthreads();
while (offset < THREAD_NUM)
{
if ((tid& mask) == 0)
{
shared[tid] += shared[tid + offset];
}
offset += offset;
mask = offset + mask;
__syncthreads();
}

if (tid == 0)
{
result[bid] = shared[0];
time[bid + BLOCK_NUM] = clock();
}
}

offset = THREAD_NUM / 2;
while (offset > 0)
{
if (tid< offset)
{
shared[tid] += shared[tid + offset];
}
offset >>= 1;
__syncthreads();
}

if (tid < 128)
{
shared[tid] += shared[tid + 128];
}
__syncthreads();

if (tid < 64)
{
shared[tid] += shared[tid + 64];
}
__syncthreads();

if (tid < 32)
{
shared[tid] += shared[tid + 32];
}
__syncthreads();

if (tid < 16)
{
shared[tid] += shared[tid + 16];
}
__syncthreads();

if (tid < 8)
{
shared[tid] += shared[tid + 8];
}
__syncthreads();

if (tid < 4)
{
shared[tid] += shared[tid + 4];
}
__syncthreads();

if (tid < 2)
{
shared[tid] += shared[tid + 2];
}
__syncthreads();

if (tid < 1)
{
shared[tid] += shared[tid + 1];
}
__syncthreads();