| 1 |
|
|---|
| 2 |
|
|---|
| 3 | #include "cl.cvl"
|
|---|
| 4 | #include <stdio.h>
|
|---|
| 5 | #include <stdlib.h>
|
|---|
| 6 | #include <string.h>
|
|---|
| 7 | #include <civlc.h>
|
|---|
| 8 | #include "multiplekernel.cvl"
|
|---|
| 9 |
|
|---|
| 10 | int numArgs = 2;
|
|---|
| 11 | int numKernels = 4;
|
|---|
| 12 | /*
|
|---|
| 13 | This version of the kernel has a flag for each element in the param array
|
|---|
| 14 | If the mallocflag is set to 1, it means that it will be freed by clReleaseKernel
|
|---|
| 15 | */
|
|---|
| 16 | typedef struct
|
|---|
| 17 | {
|
|---|
| 18 | void * param[numArgs];
|
|---|
| 19 | int mallocflag[numArgs];
|
|---|
| 20 | }args;
|
|---|
| 21 | /*
|
|---|
| 22 | clCreateKernel in this iteration turns each flag to 0 to start off with, otherwise it gets a symbolic expression
|
|---|
| 23 | */
|
|---|
| 24 | cl_kernel clCreateKernel(args * argument, char * function)
|
|---|
| 25 | {
|
|---|
| 26 | cl_kernel kernel;
|
|---|
| 27 | kernel.arguments = argument;
|
|---|
| 28 | kernel.method = function;
|
|---|
| 29 |
|
|---|
| 30 | for(int j = 0; j < numArgs; j++)
|
|---|
| 31 | {
|
|---|
| 32 | ((args*)kernel.arguments)->mallocflag[j] = 0;
|
|---|
| 33 | }
|
|---|
| 34 |
|
|---|
| 35 | return kernel;
|
|---|
| 36 | }
|
|---|
| 37 | /*
|
|---|
| 38 | Checks the flag to make sure that it was a local variable that must be freed every step of the way
|
|---|
| 39 | */
|
|---|
| 40 | void clReleaseKernel(cl_kernel kernel)
|
|---|
| 41 | {
|
|---|
| 42 | for (int i = 0; i < numArgs; i++)
|
|---|
| 43 | {
|
|---|
| 44 | //printf("I am argument %d with value %d\n", i, ((args*)kernel.arguments)->mallocflag[i]);
|
|---|
| 45 | if (((args*)kernel.arguments)->mallocflag[i] == 1)
|
|---|
| 46 | {
|
|---|
| 47 | //printf("and I pass the flag check\n");
|
|---|
| 48 | free(((args*)kernel.arguments)->param[i]);
|
|---|
| 49 | }
|
|---|
| 50 | }
|
|---|
| 51 | }
|
|---|
| 52 | //Workfunc and clEnqueueNDRangeKernel are the same.
|
|---|
| 53 | void workfunc(size_t local, size_t global, cl_kernel param)
|
|---|
| 54 | {
|
|---|
| 55 | $proc procs[local];
|
|---|
| 56 | char * reduceChar = "add";
|
|---|
| 57 | for(int i = local * param.workgroup; i < local * param.workgroup + local; i++)
|
|---|
| 58 | {
|
|---|
| 59 | int n = *(int*)(((args*)param.arguments)->param[0]);
|
|---|
| 60 | param.local_id = i % local;
|
|---|
| 61 | param.global_id = i;
|
|---|
| 62 | //printf("My workgroup id is %d, my global id is %d, my local id is %d, and my method is %s\n", param.workgroup, param.global_id, param.local_id, param.method);
|
|---|
| 63 |
|
|---|
| 64 |
|
|---|
| 65 | //if(strcmp(param.method, reduceChar) == 0)
|
|---|
| 66 | //{
|
|---|
| 67 | procs[param.local_id] = $spawn add(param.workgroup, param.global_id, param.local_id, n, ((args*)param.arguments)->param[1]);
|
|---|
| 68 | //}
|
|---|
| 69 | }
|
|---|
| 70 | for(int j = 0; j < local; j++)
|
|---|
| 71 | {
|
|---|
| 72 | $wait(procs[j]);
|
|---|
| 73 | }
|
|---|
| 74 | }
|
|---|
| 75 |
|
|---|
| 76 | int clEnqueueNDRangeKernel(cl_kernel kernel, size_t global, size_t local)
|
|---|
| 77 | {
|
|---|
| 78 |
|
|---|
| 79 | $assert(global % local == 0);
|
|---|
| 80 | int numworkgroups = global/local;
|
|---|
| 81 | cl_kernel param[numworkgroups];
|
|---|
| 82 | $proc procs[numworkgroups];
|
|---|
| 83 | //consider $parfor
|
|---|
| 84 |
|
|---|
| 85 | /*
|
|---|
| 86 | $domain(1) dom = {0 .. numworkgroups - 1};
|
|---|
| 87 |
|
|---|
| 88 | $for(int i: dom)
|
|---|
| 89 | {
|
|---|
| 90 | param[i] = kernel;
|
|---|
| 91 | param[i].workgroup = i;
|
|---|
| 92 | }
|
|---|
| 93 | $parfor(int i: dom)
|
|---|
| 94 | {
|
|---|
| 95 | workfunc(local, global, param[i]);
|
|---|
| 96 | }
|
|---|
| 97 | */
|
|---|
| 98 |
|
|---|
| 99 | for(int i = 0; i < global/local; i++)
|
|---|
| 100 | {
|
|---|
| 101 | param[i] = kernel;
|
|---|
| 102 | param[i].workgroup = i;
|
|---|
| 103 | procs[i] = $spawn workfunc(local, global, param[i]);
|
|---|
| 104 | }
|
|---|
| 105 |
|
|---|
| 106 | //this part here is the new clFinish(commands);
|
|---|
| 107 | for(int i = 0; i < global/local; i++)
|
|---|
| 108 | {
|
|---|
| 109 | $wait(procs[i]);
|
|---|
| 110 | }
|
|---|
| 111 |
|
|---|
| 112 | return CL_SUCCESS;
|
|---|
| 113 |
|
|---|
| 114 | }
|
|---|
| 115 |
|
|---|
| 116 | int main(int argc, char** argv)
|
|---|
| 117 | {
|
|---|
| 118 | int global = numKernels;
|
|---|
| 119 | int local = 2;
|
|---|
| 120 |
|
|---|
| 121 | args * arguments;
|
|---|
| 122 | arguments = (args*)malloc(sizeof(args) * numKernels);
|
|---|
| 123 | //Must have enough arguments for each kernel
|
|---|
| 124 |
|
|---|
| 125 | int star = 1;
|
|---|
| 126 |
|
|---|
| 127 | int * one = (int*)malloc(sizeof(int));
|
|---|
| 128 | memcpy(one, &star, sizeof(int));
|
|---|
| 129 |
|
|---|
| 130 | //cl_kernel *kernels = (cl_kernel*)malloc(numKernels * sizeof(cl_kernel));
|
|---|
| 131 | cl_kernel kernels[numKernels];
|
|---|
| 132 |
|
|---|
| 133 | //printf("%d, %d\n", star, *one);
|
|---|
| 134 |
|
|---|
| 135 | //For loop to create multiple kernels, the programmer must put the loop himself
|
|---|
| 136 | for (int i = 0; i < numKernels; i++)
|
|---|
| 137 | {
|
|---|
| 138 | kernels[i] = clCreateKernel(arguments+i, "add");
|
|---|
| 139 |
|
|---|
| 140 |
|
|---|
| 141 | }
|
|---|
| 142 | //For each kernel, put in the variables
|
|---|
| 143 | for (int j = 0; j < numKernels; j++)
|
|---|
| 144 | {
|
|---|
| 145 |
|
|---|
| 146 | ((args*)kernels[j].arguments)->param[0] = (int*)malloc(sizeof(int));
|
|---|
| 147 | ((args*)kernels[j].arguments)->mallocflag[0] = 1;
|
|---|
| 148 | memcpy(((args*)kernels[j].arguments)->param[0], &star, sizeof(int));
|
|---|
| 149 |
|
|---|
| 150 | //param[0] is a __local variable, so it gets the flag put to 1
|
|---|
| 151 |
|
|---|
| 152 | //printf("I am kernel of index %d, with a value of %d\n", j, ((args*)kernels[j].arguments)->param[0]);
|
|---|
| 153 |
|
|---|
| 154 | ((args*)kernels[j].arguments)->param[1] = one;
|
|---|
| 155 | //This is a __global variable, so it does not get the flag changed in this iteration.
|
|---|
| 156 | //TODO: determine if global variables were freed by clReleaseKernel or the programmer.
|
|---|
| 157 | }
|
|---|
| 158 |
|
|---|
| 159 | //printf("assigned properly\n");
|
|---|
| 160 |
|
|---|
| 161 | for (int i = 0; i < numKernels; i++)
|
|---|
| 162 | {
|
|---|
| 163 | clEnqueueNDRangeKernel(kernels[i], global, local);
|
|---|
| 164 | }
|
|---|
| 165 | int ans = (int)((args*)kernels[0].arguments)->param[1];
|
|---|
| 166 | //memcpy(ans, &((args*)kernels[0].arguments)->param[1], sizeof(int));
|
|---|
| 167 | printf("There are %d kernels, and the final value is %d\n", numKernels, ans);
|
|---|
| 168 |
|
|---|
| 169 | for (int k = 0; k < numKernels; k++)
|
|---|
| 170 | {
|
|---|
| 171 | clReleaseKernel(kernels[k]);
|
|---|
| 172 | }
|
|---|
| 173 |
|
|---|
| 174 |
|
|---|
| 175 |
|
|---|
| 176 | free(one);
|
|---|
| 177 | free(arguments);
|
|---|
| 178 | return 0;
|
|---|
| 179 | }
|
|---|
| 180 |
|
|---|
| 181 |
|
|---|