Mercurial > hg > GearsTemplate

annotate src/parallel_execution/cuda.c @ 557:1eb2a22ec1e3

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
tweak
author anatofuz <anatofuz@cr.ie.u-ryukyu.ac.jp>
date Mon, 18 Nov 2019 21:22:34 +0900
parents d6983ce1015d
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
1 #include <stdio.h>
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
2 #include <sys/time.h>
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
3 #include <string.h>
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
4 #include <stdlib.h>
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
5
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
6 // includes, project
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
7 #include <driver_types.h>
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
8 #include <cuda_runtime.h>
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
9 #include <cuda.h>
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
10 #include "helper_cuda.h"
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
11 #include "pthread.h"
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
12
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
13 #include "context.h"
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
14
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
15 /*
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
16 struct Context {
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
17 int next;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
18 struct Worker* worker;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
19 struct TaskManager* taskManager;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
20 int codeNum;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
21 void (**code) (struct Context*);
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
22 void* heapStart;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
23 void* heap;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
24 long heapLimit;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
25 int dataNum;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
26 int idgCount; //number of waiting dataGear
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
27 int idg;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
28 int maxIdg;
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
29 int odg;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
30 int maxOdg;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
31 int workerId;
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
32 struct Context* task;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
33 struct Queue* tasks;
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
34 int num_exec;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
35 CUmodule module;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
36 CUfunction function;
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
37 union Data **data;
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
38
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
39 // multi dimension parameter
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
40 int iterate;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
41 struct Iterator* iterator;
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
42 };
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
43
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
44 struct CUDAWorker {
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
45 CUdevice device;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
46 CUcontext cuCtx;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
47 pthread_t thread;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
48 struct Context* context;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
49 int id;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
50 struct Queue* tasks;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
51 int runFlag;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
52 int next;
451
dcc42f3e7e97 Auto choice blockDim
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 436
diff changeset
53 int numStream;
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
54 CUstream *stream;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
55 } CUDAWorker;
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
56
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
57 struct LoopCounter {
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
58 int i;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
59 } LoopCounter;
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
60
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
61 struct Array {
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
62 int size;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
63 int index;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
64 int prefix;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
65 int* array;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
66 } Array;
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
67 */
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
68
451
dcc42f3e7e97 Auto choice blockDim
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 436
diff changeset
69 void cudaInit(struct CUDAWorker *cudaWorker,int phase, int deviceNum) {
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
70 // initialize and load kernel
451
dcc42f3e7e97 Auto choice blockDim
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 436
diff changeset
71 cudaWorker->numStream = 1; // number of stream
dcc42f3e7e97 Auto choice blockDim
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 436
diff changeset
72 // cudaWorker->stream = NEWN(cudaWorker->numStream, CUstream );
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
73 if (phase==0)
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
74 checkCudaErrors(cuInit(0));
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
75 if (phase==0)
451
dcc42f3e7e97 Auto choice blockDim
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 436
diff changeset
76 checkCudaErrors(cuDeviceGet(&cudaWorker->device, deviceNum));
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
77 if (phase==0)
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
78 checkCudaErrors(cuCtxCreate(&cudaWorker->cuCtx, CU_CTX_SCHED_SPIN, cudaWorker->device));
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
79 // if (cudaWorker->num_stream) {
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
80 // for (int i=0;i<cudaWorker->num_stream;i++)
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
81 // checkCudaErrors(cuStreamCreate(&cudaWorker->stream[i],0));
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
82 // }
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
83 printf("cuda Init: Done\n");
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
84 }
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
85
433
d920f3a3f037 Refactoring cuda.c
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 431
diff changeset
86 void cudaLoadFunction(struct Context* context, char* filename, char* function) {
431
b3359544adbb Edit cudaExec but not work
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 430
diff changeset
87 checkCudaErrors(cuModuleLoad(&context->module, filename));
b3359544adbb Edit cudaExec but not work
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 430
diff changeset
88 checkCudaErrors(cuModuleGetFunction(&context->function, context->module, function));
433
d920f3a3f037 Refactoring cuda.c
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 431
diff changeset
89 }
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
90
436
08a93fc2f0d3 Fix CudaExecutor but not work
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 433
diff changeset
91 void cudaShutdown(struct CUDAWorker *worker) {
410
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
92 // for (int i=0;i<worker->num_stream;i++)
85b0ddbf458e Fix CudaWorker
Tatsuki IHA <innparusu@cr.ie.u-ryukyu.ac.jp>
parents: 404
diff changeset
93 // checkCudaErrors(cuStreamDestroy(worker->stream[i]));
319
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
94 checkCudaErrors(cuCtxDestroy(worker->cuCtx));
a15511b1a6e0 separate cuda.c, and USE_CUDA_MAIN_THREAD flag
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents:
diff changeset
95 }