changeset 10392:32ee88f14239 libavcodec

Fix w32thread implementation to handle job count > thread_count.

--- a/w32thread.c	Tue Oct 13 10:30:10 2009 +0000
+++ b/w32thread.c	Tue Oct 13 13:03:12 2009 +0000
@@ -29,10 +29,13 @@
     AVCodecContext *avctx;
     HANDLE thread;
     HANDLE work_sem;
+    HANDLE job_sem;
     HANDLE done_sem;
     int (*func)(AVCodecContext *c, void *arg);
     void *arg;
-    int ret;
+    int argsize;
+    int *jobnr;
+    int *ret;
 }ThreadContext;
 
 
@@ -40,13 +43,19 @@
     ThreadContext *c= v;
 
     for(;;){
+        int ret, jobnr;
 //printf("thread_func %X enter wait\n", (int)v); fflush(stdout);
         WaitForSingleObject(c->work_sem, INFINITE);
+        WaitForSingleObject(c->job_sem, INFINITE);
+        jobnr = (*c->jobnr)++;
+        ReleaseSemaphore(c->job_sem, 1, 0);
 //printf("thread_func %X after wait (func=%X)\n", (int)v, (int)c->func); fflush(stdout);
         if(c->func)
-            c->ret= c->func(c->avctx, c->arg);
+            ret= c->func(c->avctx, (uint8_t *)c->arg + jobnr*c->argsize);
         else
             return 0;
+        if (c->ret)
+            c->ret[jobnr] = ret;
 //printf("thread_func %X signal complete\n", (int)v); fflush(stdout);
         ReleaseSemaphore(c->done_sem, 1, 0);
     }
@@ -65,12 +74,15 @@
     for(i=0; i<s->thread_count; i++){
 
         c[i].func= NULL;
-        ReleaseSemaphore(c[i].work_sem, 1, 0);
+    }
+    ReleaseSemaphore(c[0].work_sem, s->thread_count, 0);
+    for(i=0; i<s->thread_count; i++){
         WaitForSingleObject(c[i].thread, INFINITE);
-        if(c[i].work_sem) CloseHandle(c[i].work_sem);
-        if(c[i].done_sem) CloseHandle(c[i].done_sem);
         if(c[i].thread)   CloseHandle(c[i].thread);
     }
+    if(c[0].work_sem) CloseHandle(c[0].work_sem);
+    if(c[0].job_sem)  CloseHandle(c[0].job_sem);
+    if(c[0].done_sem) CloseHandle(c[0].done_sem);
 
     av_freep(&s->thread_opaque);
 }
@@ -78,25 +90,23 @@
 int avcodec_thread_execute(AVCodecContext *s, int (*func)(AVCodecContext *c2, void *arg2),void *arg, int *ret, int count, int size){
     ThreadContext *c= s->thread_opaque;
     int i;
+    int jobnr = 0;
 
     assert(s == c->avctx);
-    assert(count <= s->thread_count);
 
     /* note, we can be certain that this is not called with the same AVCodecContext by different threads at the same time */
 
-    for(i=0; i<count; i++){
-        c[i].arg= (char*)arg + i*size;
+    for(i=0; i<s->thread_count; i++){
+        c[i].arg= arg;
+        c[i].argsize= size;
         c[i].func= func;
-        c[i].ret= 12345;
-
-        ReleaseSemaphore(c[i].work_sem, 1, 0);
+        c[i].ret= ret;
+        c[i].jobnr = &jobnr;
     }
-    for(i=0; i<count; i++){
-        WaitForSingleObject(c[i].done_sem, INFINITE);
+    ReleaseSemaphore(c[0].work_sem, count, 0);
+    for(i=0; i<count; i++)
+        WaitForSingleObject(c[0].done_sem, INFINITE);
 
-        c[i].func= NULL;
-        if(ret) ret[i]= c[i].ret;
-    }
     return 0;
 }
 
@@ -110,15 +120,19 @@
     assert(!s->thread_opaque);
     c= av_mallocz(sizeof(ThreadContext)*thread_count);
     s->thread_opaque= c;
+    if(!(c[0].work_sem = CreateSemaphore(NULL, 0, INT_MAX, NULL)))
+        goto fail;
+    if(!(c[0].job_sem  = CreateSemaphore(NULL, 1, 1, NULL)))
+        goto fail;
+    if(!(c[0].done_sem = CreateSemaphore(NULL, 0, INT_MAX, NULL)))
+        goto fail;
 
     for(i=0; i<thread_count; i++){
 //printf("init semaphors %d\n", i); fflush(stdout);
         c[i].avctx= s;
-
-        if(!(c[i].work_sem = CreateSemaphore(NULL, 0, s->thread_count, NULL)))
-            goto fail;
-        if(!(c[i].done_sem = CreateSemaphore(NULL, 0, s->thread_count, NULL)))
-            goto fail;
+        c[i].work_sem = c[0].work_sem;
+        c[i].job_sem  = c[0].job_sem;
+        c[i].done_sem = c[0].done_sem;
 
 //printf("create thread %d\n", i); fflush(stdout);
         c[i].thread = (HANDLE)_beginthreadex(NULL, 0, thread_func, &c[i], 0, &threadid );