Make realloc() go 100x faster on Linux/NetBSD

Cosmopolitan now supports mremap(), which is only supported on Linux and
NetBSD. First, it allows memory mappings to be relocated without copying
them; this can dramatically speed up data structures like std::vector if
the array size grows larger than 256kb. The mremap() system call is also
10x faster than munmap() when shrinking large memory mappings.

There's now two functions, getpagesize() and getgransize() which help to
write portable code that uses mmap(MAP_FIXED). Alternative sysconf() may
be called with our new _SC_GRANSIZE. The madvise() system call now has a
better wrapper with improved documentation.

tool/build/lib/elfwriter.c

@@ -163,7 +163,7 @@ struct ElfWriter *elfwriter_open(const char *path, int mode, int arch) {
   CHECK_NOTNULL((elf = calloc(1, sizeof(struct ElfWriter))));
   CHECK_NOTNULL((elf->path = strdup(path)));
   CHECK_NE(-1, (elf->fd = open(elf->path, O_CREAT | O_TRUNC | O_RDWR, mode)));
-  CHECK_NE(-1, ftruncate(elf->fd, (elf->mapsize = __granularity())));
+  CHECK_NE(-1, ftruncate(elf->fd, (elf->mapsize = getgransize())));
   CHECK_NE(MAP_FAILED, (elf->map = mmap((void *)(intptr_t)kFixedmapStart,
                                         elf->mapsize, PROT_READ | PROT_WRITE,
                                         MAP_SHARED | MAP_FIXED, elf->fd, 0)));
@@ -234,7 +234,7 @@ void *elfwriter_reserve(struct ElfWriter *elf, size_t size) {
     do {
       greed = greed + (greed >> 1);
     } while (need > greed);
-    greed = ROUNDUP(greed, __granularity());
+    greed = ROUNDUP(greed, getgransize());
     CHECK_NE(-1, ftruncate(elf->fd, greed));
     CHECK_NE(MAP_FAILED, mmap((char *)elf->map + elf->mapsize,
                               greed - elf->mapsize, PROT_READ | PROT_WRITE,

tool/lambda/lib/calloc.c

@@ -28,14 +28,14 @@ void *Calloc(size_t a, size_t b) {
   static size_t n;
   z = a * b;
   if (!p) {
-    n = __granularity();
-    p = mmap((void *)0x300000000000, __granularity(), PROT_READ | PROT_WRITE,
+    n = getgransize();
+    p = mmap((void *)0x300000000000, getgransize(), PROT_READ | PROT_WRITE,
              MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
   }
   if (i + z > n) {
-    mmap(p + i, __granularity(), PROT_READ | PROT_WRITE,
+    mmap(p + i, getgransize(), PROT_READ | PROT_WRITE,
          MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0);
-    n += __granularity();
+    n += getgransize();
   }
   r = p + i;
   i += z;

tool/net/redbean.c

@@ -5019,7 +5019,7 @@ static int LuaProgramTokenBucket(lua_State *L) {
       VERBOSEF("(token) please run the blackholed program; see our website!");
     }
   }
-  tokenbucket.b = _mapshared(ROUNDUP(1ul << cidr, __granularity()));
+  tokenbucket.b = _mapshared(ROUNDUP(1ul << cidr, getgransize()));
   memset(tokenbucket.b, 127, 1ul << cidr);
   tokenbucket.cidr = cidr;
   tokenbucket.reject = reject;
@@ -7339,7 +7339,7 @@ void RedBean(int argc, char *argv[]) {
   heartbeatinterval.tv_sec = 5;
   CHECK_GT(CLK_TCK, 0);
   CHECK_NE(MAP_FAILED,
-           (shared = mmap(NULL, ROUNDUP(sizeof(struct Shared), __granularity()),
+           (shared = mmap(NULL, ROUNDUP(sizeof(struct Shared), getgransize()),
                           PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS,
                           -1, 0)));
   if (daemonize) {

tool/plinko/lib/plinko.c

@@ -970,7 +970,7 @@ int Plinko(int argc, char *argv[]) {
   }
 
   if (mmap((void *)0x200000000000,
-           ROUNDUP((TERM + 1) * sizeof(g_mem[0]), __granularity()),
+           ROUNDUP((TERM + 1) * sizeof(g_mem[0]), getgransize()),
            PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1,
            0) == MAP_FAILED ||
       mmap((void *)(0x200000000000 +
@@ -979,7 +979,7 @@ int Plinko(int argc, char *argv[]) {
            PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1,
            0) == MAP_FAILED ||
       mmap((void *)0x400000000000,
-           ROUNDUP((TERM + 1) * sizeof(g_mem[0]), __granularity()),
+           ROUNDUP((TERM + 1) * sizeof(g_mem[0]), getgransize()),
            PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1,
            0) == MAP_FAILED ||
       mmap((void *)(0x400000000000 +

tool/viz/derasterize.c

@@ -452,7 +452,6 @@ static void PrintImage(unsigned yn, unsigned xn,
   size_t size;
   char *v, *vt;
   size = yn * (xn * (32 + (2 + (1 + 3) * 3) * 2 + 1 + 3)) * 1 + 5 + 1;
-  size = ROUNDUP(size, __granularity());
   CHECK_NOTNULL((vt = _mapanon(size)));
   v = RenderImage(vt, yn, xn, rgb);
   *v++ = '\r';
@@ -550,8 +549,8 @@ static void LoadFile(const char *path, size_t yn, size_t xn, void *rgb) {
   stbir_resize_uint8(data, gotx, goty, 0, rgb, xn * XS, yn * YS, 0, CN);
 #else
   CHECK_EQ(CN, 3);
-  data2size = ROUNDUP(sizeof(float) * goty * gotx * CN, __granularity());
-  data3size = ROUNDUP(sizeof(float) * yn * YS * xn * XS * CN, __granularity());
+  data2size = sizeof(float) * goty * gotx * CN;
+  data3size = sizeof(float) * yn * YS * xn * XS * CN;
   CHECK_NOTNULL((data2 = _mapanon(data2size)));
   CHECK_NOTNULL((data3 = _mapanon(data3size)));
   rgb2lin(goty * gotx * CN, data2, data);
@@ -625,7 +624,7 @@ int main(int argc, char *argv[]) {
   // FIXME: on the conversion stage should do 2Y because of halfblocks
   // printf( "filename >%s<\tx >%d<\ty >%d<\n\n", filename, x_, y_);
   size = y_ * YS * x_ * XS * CN;
-  CHECK_NOTNULL((rgb = _mapanon(ROUNDUP(size, __granularity()))));
+  CHECK_NOTNULL((rgb = _mapanon(size)));
   for (i = optind; i < argc; ++i) {
     if (!argv[i])
       continue;
@@ -636,7 +635,7 @@ int main(int argc, char *argv[]) {
     }
     PrintImage(y_, x_, rgb);
   }
-  munmap(rgb, ROUNDUP(size, __granularity()));
+  munmap(rgb, size);
   return 0;
 }
 

tool/viz/lib/sobel.c

@@ -37,8 +37,7 @@ forceinline void ConvolveGradient(unsigned yn, unsigned xn,
   size_t size;
   unsigned y, x, i, j, k;
   float py[4], px[4], (*tmp)[yn][xn][4];
-  tmp =
-      _mapanon((size = ROUNDUP(sizeof(float) * 4 * xn * yn, __granularity())));
+  tmp = _mapanon((size = ROUNDUP(sizeof(float) * 4 * xn * yn, getgransize())));
   for (y = 0; y < yn - KW + 1; ++y) {
     for (x = 0; x < xn - KW + 1; ++x) {
       for (k = 0; k < 4; ++k)

tool/viz/memzoom.c

@@ -281,8 +281,8 @@ static void SetupCanvas(void) {
     munmap(buffer, buffersize);
   }
   displaysize = ROUNDUP(ROUNDUP((tyn * txn) << zoom, 16), 1ul << zoom);
-  canvassize = ROUNDUP(displaysize, __granularity());
-  buffersize = ROUNDUP(tyn * txn * 16 + 4096, __granularity());
+  canvassize = ROUNDUP(displaysize, getgransize());
+  buffersize = ROUNDUP(tyn * txn * 16 + 4096, getgransize());
   canvas = Allocate(canvassize);
   buffer = Allocate(buffersize);
 }

tool/viz/printvideo.c

@@ -785,7 +785,7 @@ static void RasterIt(void) {
   static bool once;
   static void *buf;
   if (!once) {
-    buf = _mapanon(ROUNDUP(fb0_.size, __granularity()));
+    buf = _mapanon(ROUNDUP(fb0_.size, getgransize()));
     once = true;
   }
   WriteToFrameBuffer(fb0_.vscreen.yres_virtual, fb0_.vscreen.xres_virtual, buf,