/g/ - Technology

OpenCL

Duck

loc-IN

51q9xuub4j7.png

Just learning it right now, will add whatever I make here. Wrote some matrix multiplication code today. [code] #include <CL/cl.h> #include <stdio.h> #include <stdlib.h> int main(int argc, char** argv){ cl_platform_id platform; cl_device_id device; clGetPlatformIDs(1, &platform, NULL); clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL); // 2. Create context and command queue cl_context context = clCreateContext(NULL, 1, &device, NULL, NULL, NULL); cl_command_queue queue = clCreateCommandQueue(context, device, 0, NULL); FILE* f = fopen("test_cl.cl", "rb"); fseek(f, 0, SEEK_END); size_t size = ftell(f); rewind(f); char* source = malloc(size+1); fread(source, 1, size, f); source[size] = '\0'; fclose(f); cl_program program = clCreateProgramWithSource(context, 1, (const char**)&source, &size, NULL); clBuildProgram(program, 1, &device, NULL, NULL, NULL); cl_kernel kernel = clCreateKernel(program, "test_kernel", NULL); size_t m=3; size_t n=2; size_t a=2; size_t b=4; float *A = (float*)malloc(sizeof(float) * m * n); float *B = (float*)malloc(sizeof(float) * a * b); float *C = (float*)malloc(sizeof(float) * m * b); for (int i = 0; i < m*n; i++) { A[i] = (float)i; } for (int i = 0; i < a*b; i++) { B[i] = 1024 - (float)i; } cl_mem d_A = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float)*(m*n), A, NULL); cl_mem d_B = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(float)*(a*b), B, NULL); cl_mem d_C = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(float)*(m*b), NULL, NULL); clSetKernelArg(kernel, 0, sizeof(cl_mem), &d_A); clSetKernelArg(kernel, 1, sizeof(cl_mem), &d_B); clSetKernelArg(kernel, 2, sizeof(cl_mem), &d_C); clSetKernelArg(kernel, 3, sizeof(unsigned int), &m); clSetKernelArg(kernel, 4, sizeof(unsigned int), &b); clSetKernelArg(kernel, 5, sizeof(unsigned int), &a); size_t globalSize[2] = {m, b}; clEnqueueNDRangeKernel(queue, kernel, 2, NULL, globalSize, NULL, 0, NULL, NULL); clFinish(queue); clEnqueueReadBuffer(queue, d_C, CL_TRUE, 0, m*b * sizeof(float), C, 0, NULL, NULL); int k=0; int i = 0; int z =0; while(k<m){ printf("%f\t", C[z]); z=z+1; i++; if(i==(b)){k=k+1; i=0; printf("\n");} } clReleaseMemObject(d_A); clReleaseMemObject(d_B); clReleaseMemObject(d_C); clReleaseProgram(program); clReleaseKernel(kernel); clReleaseCommandQueue(queue); clReleaseContext(context); free(A); free(B); free(C); } [/code]

Duck

loc-IN

forgot to attach kernel [code] __kernel void test_kernel(__global const float *a, __global const float *b, __global float *C, int HEIGHT, int WIDTH, int K) { int row = get_global_id(0); int col = get_global_id(1); if(row < HEIGHT && col < WIDTH){ float sum = 0.0; for(int k = 0; k < K; k++){ sum = sum + a[row * K + k] * b[k * WIDTH + col]; } C[row * WIDTH + col] = sum; } } [/code]

Anonymous

loc-IN

>>2158 love you anon

Duck

loc-IN

>>2159 Love you too saar

Duck

loc-IN

Downloaded some openCL book, none of their example codes work. Bruh

Anonymous

loc-IN

what are you writing opencl for? platform? each vendor has their own big fat libraries implementing the spec. those libs contains tons of test cases and example programs.

Anonymous

loc-IN

contain*

Duck

loc-IN

>>2162 Well I was planning to pair it with openGL, doing some simulations etc. The amd device isn't rocm supported so openCL is basically the only option

Anonymous

loc-IN

>>2164 hmm, your compiler sources are the best documentation btw

Anonymous

loc-IN

cant you sort of do a static analysis of the binary produced by the compiler to reconstruct the program logic?

Anonymous

loc-IN

join #indiachan on irc.libera.chat

Duck

loc-IN

>>2187 Nahi karni bhai

Anonymous

loc-IN

>>2157 What's the use case of opencl?

Duck

loc-IN

>>2189 Writing gpu kernels but you don't have to write separate code for amd and nvidia. Device can be selected at runtime

Anonymous

loc-IN

>>2190 Sounds really cool yaar. You are into low level programming?

Anonymous

loc-IN

>>2188 np bhai, keep your shitter and pisscord

Anonymous

loc-IN

>>2191 low-iq programming it's called, saar

Duck

loc-IN

I am fucking stupid, I have unified memory why am I copying the memory again like it's a discrete gpu

Duck

loc-IN

Solved some more CF today. Played laser tag today

Anonymous

loc-IN

>>2273 do you play CTFs?

Duck

loc-IN

>>2275 Not much I solved some problems during lockdowns like that mr robot challenge, and 4-5 more. Seem boring to me tbh

Anonymous

loc-IN

>>2276 maybe it's just that you're retarded

Duck

loc-IN

>>2277 Probably troo>>2277

Duck

loc-IN

Going home today. Sick of hearing some different southie language every morning

Duck

loc-IN

Will have to learn threejs, can't handle the fact that someone can made that run in a browser and I can't

Duck

loc-IN

New found appreciation for home, will go to buy a laptop tomorrow. Hope some pajeet laptop salesman doesn't convince/scam my dad into buying some junk

Duck

loc-IN

I want to kill myself but well

Duck

loc-IN

What are you supposed to do when you solve a leetcode question but it isn't the best possible solution? Solve all of them again?

Anonymous

loc-IN

>>2409 look how can you optimize it, or just look the optimal solution and submit it.

Duck

loc-IN

>I need to do better >I need to kill myself Fucking sucks that I can't bring myself to do either of the two. Feel like I am destined for mediocrity