Make stderr go faster

This change makes _IONBF (unbuffered) stdio handles go 20x faster for
certain kinds of formatting directives by being smarter about buffers

build/rules.mk

@@ -17,7 +17,6 @@
 MAKEFLAGS += --no-builtin-rules
 
 MAKE_ZIPCOPY = $(COMPILE) -AZIPCOPY -wT$@ $(ZIPCOPY) $< $@
-MAKE_PECHECK = $(COMPILE) -APECHECK -wT$@ $(PECHECK) $@
 ifneq ($(ARCH), aarch64)
 MAKE_OBJCOPY = $(COMPILE) -AOBJCOPY -T$@ $(OBJCOPY) -S -O binary $< $@ && $(MAKE_ZIPCOPY)
 else

libc/fmt/divmod10.internal.h

@@ -2,23 +2,23 @@
 #define COSMOPOLITAN_LIBC_FMT_DIVMOD10_H_
 #if !(__ASSEMBLER__ + __LINKER__ + 0)
 
-forceinline uint64_t DivMod10(uint64_t x, unsigned *r) {
+forceinline uint64_t __divmod10(uint64_t __x, unsigned *__r) {
 #if defined(__STRICT_ANSI__) || !defined(__GNUC__) || \
     (defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__))
-  *r = x % 10;
-  return x / 10;
+  *__r = __x % 10;
+  return __x / 10;
 #else
-  uint128_t dw;
-  unsigned long long hi, rm;
-  dw = x;
-  dw *= 0xcccccccccccccccdull;
-  hi = dw >> 64;
-  hi >>= 3;
-  rm = hi;
-  rm += rm << 2;
-  rm += rm;
-  *r = x - rm;
-  return hi;
+  uint128_t __dw;
+  unsigned long long __hi, __rm;
+  __dw = __x;
+  __dw *= 0xcccccccccccccccdull;
+  __hi = __dw >> 64;
+  __hi >>= 3;
+  __rm = __hi;
+  __rm += __rm << 2;
+  __rm += __rm;
+  *__r = __x - __rm;
+  return __hi;
 #endif
 }
 

libc/intrin/kprintf.greg.c

@@ -432,7 +432,7 @@ privileged static size_t kformat(char *b, size_t n, const char *fmt,
         FormatUnsigned:
           if (x && hash) sign = hash;
           for (i = j = 0;;) {
-            x = DivMod10(x, &rem);
+            x = __divmod10(x, &rem);
             z[i++ & 127] = '0' + rem;
             if (pdot ? i >= prec : !x) break;
             if (quot && ++j == 3) {

libc/stdio/fmt.c

@@ -244,15 +244,15 @@ static int __fmt_ntoa2(int out(const char *, void *, size_t), void *arg,
   unsigned len, count, digit;
   char buf[BUFFER_SIZE];
   len = 0;
-  /* we check for log2base != 3 because otherwise we'll print nothing for a
-   * value of 0 with precision 0 when # mandates that one be printed */
+  // we check for log2base!=3, since otherwise we'll print nothing for
+  // a value of 0 with precision 0 when # mandates that one be printed
   if (!value && log2base != 3) flags &= ~FLAGS_HASH;
   if (value || !(flags & FLAGS_PRECISION)) {
     count = 0;
     do {
       if (!log2base) {
         if (value <= UINT64_MAX) {
-          value = DivMod10(value, &digit);
+          value = __divmod10(value, &digit);
         } else {
           value = __udivmodti4(value, 10, &remainder);
           digit = remainder;

libc/stdio/vfprintf_unlocked.c

@@ -18,28 +18,35 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/calls.h"
 #include "libc/fmt/fmt.h"
-#include "libc/limits.h"
+#include "libc/stdckdint.h"
 #include "libc/stdio/stdio.h"
 #include "libc/sysv/errfuns.h"
 
+struct buf {
+  int n;
+  char p[512];
+};
+
 struct state {
   FILE *f;
   int n;
+  struct buf b;
 };
 
-static int vfprintfputchar(const char *s, struct state *t, size_t n) {
+static int __vfprintf_flbuf(const char *s, struct state *t, size_t n) {
   int rc;
   if (n) {
-    if (n == 1 && *s != '\n' && t->f->beg < t->f->size &&
-        t->f->bufmode != _IONBF) {
+    if (n == 1 && *s != '\n' && t->f->beg < t->f->size) {
       t->f->buf[t->f->beg++] = *s;
       t->n += n;
       rc = 0;
-    } else if (!fwrite_unlocked(s, 1, n, t->f)) {
-      rc = -1;
-    } else {
-      t->n += n;
+    } else if (fwrite_unlocked(s, 1, n, t->f)) {
       rc = 0;
+    } else {
+      rc = -1;
+    }
+    if (ckd_add(&t->n, t->n, n)) {
+      rc = eoverflow();
     }
   } else {
     rc = 0;
@@ -47,15 +54,49 @@ static int vfprintfputchar(const char *s, struct state *t, size_t n) {
   return rc;
 }
 
+static int __vfprintf_nbuf(const char *s, struct state *t, size_t n) {
+  size_t i;
+  for (i = 0; i < n; ++i) {
+    t->b.p[t->b.n++] = s[i];
+    if (t->b.n == sizeof(t->b.p)) {
+      if (!fwrite_unlocked(s, 1, t->b.n, t->f)) {
+        return -1;
+      }
+      t->b.n = 0;
+    }
+    if (ckd_add(&t->n, t->n, 1)) {
+      return eoverflow();
+    }
+  }
+  return 0;
+}
+
 /**
  * Formats and writes text to stream.
  * @see printf() for further documentation
  */
 int vfprintf_unlocked(FILE *f, const char *fmt, va_list va) {
   int rc;
-  struct state st[1] = {{f, 0}};
-  if ((rc = __fmt(vfprintfputchar, st, fmt, va)) != -1) {
-    rc = st->n;
+  struct state st;
+  int (*out)(const char *, struct state *, size_t);
+  if (f->bufmode != _IONBF) {
+    out = __vfprintf_flbuf;
+  } else {
+    out = __vfprintf_nbuf;
+  }
+  st.f = f;
+  st.n = 0;
+  st.b.n = 0;
+  if ((rc = __fmt(out, &st, fmt, va)) != -1) {
+    if (!st.b.n) {
+      rc = st.n;
+    } else if (fwrite_unlocked(st.b.p, 1, st.b.n, st.f)) {
+      if (ckd_add(&rc, st.n, st.b.n)) {
+        rc = eoverflow();
+      }
+    } else {
+      rc = -1;
+    }
   }
   return rc;
 }

tool/hello/hello.mk

@@ -1,9 +1,6 @@
 #-*-mode:makefile-gmake;indent-tabs-mode:t;tab-width:8;coding:utf-8-*-┐
 #───vi: set et ft=make ts=8 tw=8 fenc=utf-8 :vi───────────────────────┘
 
-# qemu-user execve() is broken so we need to build/bootstrap/ commands
-ifeq ($(ARCH), x86_64)
-
 PKGS += TOOL_HELLO
 
 TOOL_HELLO_FILES := $(wildcard tool/hello/*)
@@ -81,8 +78,3 @@ o/$(MODE)/tool/hello/hello-pe.com:				\
 	@$(COMPILE) -AELF2PE o/$(MODE)/tool/build/elf2pe.com -o $@ $<
 
 $(TOOL_HELLO_OBJS): tool/hello/hello.mk
-
-.PHONY: o/$(MODE)/tool/hello
-o/$(MODE)/tool/hello: $(TOOL_HELLO_BINS)
-
-endif