Changeset 97fed44 for benchmark

Timestamp:

Jan 25, 2022, 4:54:35 PM (4 years ago)

Author:

caparsons <caparson@…>

Branches:

ADT, ast-experimental, enum, forall-pointer-decay, master, pthread-emulation, qualifiedEnum, stuck-waitfor-destruct

Children:

6b2d444, a488783, f681823

Parents:

f57f6ea0 (diff), 4fcbf26 (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.

Message:

Merge branch 'master' of plg.uwaterloo.ca:software/cfa/cfa-cc

Location:

benchmark

Files:

• io/http/Makefile.am (modified) (1 diff)
• io/http/main.cfa (modified) (4 diffs)
• readyQ/churn.cfa (modified) (1 diff)
• readyQ/locality.cfa (modified) (2 diffs)
• readyQ/transfer.cfa (modified) (6 diffs)

benchmark/io/http/Makefile.am

@@ -21 +21 @@
 include $(top_srcdir)/tools/build/cfa.make
 
-AM_CFLAGS = -O2 -Wall -Wextra -I$(srcdir) -lrt -pthread # -Werror
+AM_CFLAGS = -O3 -Wall -Wextra -I$(srcdir) -lrt -pthread # -Werror
 AM_CFAFLAGS = -quiet -nodebug
 AM_LDFLAGS = -quiet -nodebug

benchmark/io/http/main.cfa

@@ -1 @@
-#define __USE_GNU
+#define _GNU_SOURCE
 
 #include <errno.h>
@@ -6 +6 @@
 #include <unistd.h>
 extern "C" {
+        #include <sched.h>
         #include <signal.h>
         #include <sys/socket.h>
@@ -67 +68 @@
         (this.self){ "Server Cluster", options.clopts.params };
 
+        cpu_set_t fullset;
+        CPU_ZERO(&fullset);
+        int ret = sched_getaffinity(getpid(), sizeof(fullset), &fullset);
+        if( ret != 0 ) abort | "sched_getaffinity failed with" | errno | strerror( errno );
+        int cnt = CPU_COUNT(&fullset);
+
         this.procs = alloc(options.clopts.nprocs);
         for(i; options.clopts.nprocs) {
                 (this.procs[i]){ "Benchmark Processor", this.self };
+
+                int c = 0;
+                int n = 1 + (i % cnt);
+                for(int j = 0; j < CPU_SETSIZE; j++) {
+                        if(CPU_ISSET(j, &fullset)) n--;
+                        if(n == 0) {
+                                c = j;
+                                break;
+                        }
+                }
+                cpu_set_t localset;
+                CPU_ZERO(&localset);
+                CPU_SET(c, &localset);
+                ret = pthread_setaffinity_np(this.procs[i].kernel_thread, sizeof(localset), &localset);
+                if( ret != 0 ) abort | "sched_getaffinity failed with" | ret | strerror( ret );
 
                 #if !defined(__CFA_NO_STATISTICS__)
@@ -146 +168 @@
         int waited = 0;
         for() {
-                ret = bind( server_fd, (struct sockaddr *)&address, sizeof(address) );
+                int sockfd = server_fd;
+                __CONST_SOCKADDR_ARG addr;
+                addr.__sockaddr__ = (struct sockaddr *)&address;
+                socklen_t addrlen = sizeof(address);
+                ret = bind( sockfd, addr, addrlen );
                 if(ret < 0) {
                         if(errno == EADDRINUSE) {

benchmark/readyQ/churn.cfa

@@ -21 +21 @@
         wait( sem );
         for() {
-                uint64_t r = thread_rand();
+                uint32_t r = prng();
                 bench_sem * next = __atomic_exchange_n(&spots[r % spot_cnt], &sem, __ATOMIC_SEQ_CST);
                 if(next) post( *next );

benchmark/readyQ/locality.cfa

@@ -128 +128 @@
 __attribute__((noinline)) void work(MyData & data, size_t cnt_, uint64_t & state) {
         for (cnt_) {
-                access(data, __xorshift64(state));
+                access(data, xorshift_13_7_17(state));
         }
 }
 
 void main(MyThread & this) {
-        uint64_t state = thread_rand();
+        uint64_t state = prng();
 
         // Wait for start
@@ -144 +144 @@
 
                 // Wait on a random spot
-                uint64_t idx = __xorshift64(state) % this.spots.len;
+                uint64_t idx = xorshift_13_7_17(state) % this.spots.len;
                 bool closed = put(*this.spots.ptr[idx], this, this.data, this.share);
 

benchmark/readyQ/transfer.cfa

@@ -1 +1 @@
 #include "rq_bench.hfa"
 #include <fstream.hfa>
+#include <locale.h>
 
 Duration default_preemption() {
@@ -8 +9 @@
 #define PRINT(...)
 
-__lehmer64_state_t lead_seed;
+__uint128_t lead_seed;
 volatile unsigned leader;
 volatile size_t lead_idx;
@@ -68 +69 @@
         waitgroup();
 
-        unsigned nleader = __lehmer64( lead_seed ) % nthreads;
+        unsigned nleader = lehmer64( lead_seed ) % nthreads;
         __atomic_store_n( &leader, nleader, __ATOMIC_SEQ_CST );
 
@@ -105 +106 @@
 // ==================================================
 int main(int argc, char * argv[]) {
-        __lehmer64_state_t lead_seed = getpid();
-        for(10) __lehmer64( lead_seed );
+        uint64_t lead_seed = getpid();
+        for(10) lehmer64( lead_seed );
         unsigned nprocs = 2;
 
@@ -126 +127 @@
 
         lead_idx = 0;
-        leader = __lehmer64( lead_seed ) % nthreads;
+        leader = lehmer64( lead_seed ) % nthreads;
 
         size_t rechecks = 0;
@@ -167 +168 @@
         }
 
+        setlocale( LC_NUMERIC, getenv( "LANG" ) );
         sout | "Duration (ms)           : " | ws(3, 3, unit(eng((end - start)`dms)));
         sout | "Number of processors    : " | nprocs;