2020-06-15 14:18:57 +00:00
|
|
|
|
/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
|
2023-12-08 03:11:56 +00:00
|
|
|
|
│ vi: set et ft=c ts=2 sts=2 sw=2 fenc=utf-8 :vi │
|
2020-06-15 14:18:57 +00:00
|
|
|
|
╞══════════════════════════════════════════════════════════════════════════════╡
|
|
|
|
|
|
│ Copyright 2020 Justine Alexandra Roberts Tunney │
|
|
|
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
|
│ Permission to use, copy, modify, and/or distribute this software for │
|
|
|
|
|
|
│ any purpose with or without fee is hereby granted, provided that the │
|
|
|
|
|
|
│ above copyright notice and this permission notice appear in all copies. │
|
2020-06-15 14:18:57 +00:00
|
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
|
│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL │
|
|
|
|
|
|
│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED │
|
|
|
|
|
|
│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE │
|
|
|
|
|
|
│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL │
|
|
|
|
|
|
│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR │
|
|
|
|
|
|
│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER │
|
|
|
|
|
|
│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR │
|
|
|
|
|
|
│ PERFORMANCE OF THIS SOFTWARE. │
|
2020-06-15 14:18:57 +00:00
|
|
|
|
╚─────────────────────────────────────────────────────────────────────────────*/
|
|
|
|
|
|
#include "dsp/core/c11.h"
|
|
|
|
|
|
#include "dsp/core/c1331.h"
|
|
|
|
|
|
#include "dsp/core/c1331s.h"
|
|
|
|
|
|
#include "dsp/core/c161.h"
|
|
|
|
|
|
#include "dsp/core/core.h"
|
|
|
|
|
|
#include "dsp/core/half.h"
|
|
|
|
|
|
#include "dsp/core/illumination.h"
|
|
|
|
|
|
#include "dsp/core/q.h"
|
|
|
|
|
|
#include "dsp/scale/scale.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
#include "libc/assert.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/calls/calls.h"
|
|
|
|
|
|
#include "libc/calls/struct/sigset.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
#include "libc/calls/struct/timespec.h"
|
2023-07-30 01:44:15 +00:00
|
|
|
|
#include "libc/intrin/bsr.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/intrin/pmulhrsw.h"
|
|
|
|
|
|
#include "libc/log/check.h"
|
|
|
|
|
|
#include "libc/log/log.h"
|
2021-03-01 07:42:35 +00:00
|
|
|
|
#include "libc/macros.internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/math.h"
|
2024-01-08 18:07:35 +00:00
|
|
|
|
#include "libc/mem/gc.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/mem/mem.h"
|
2020-11-25 16:19:00 +00:00
|
|
|
|
#include "libc/nexgen32e/gc.internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
|
#include "libc/nexgen32e/nexgen32e.h"
|
|
|
|
|
|
#include "libc/nexgen32e/x86feature.h"
|
|
|
|
|
|
#include "libc/runtime/runtime.h"
|
|
|
|
|
|
#include "libc/str/str.h"
|
|
|
|
|
|
#include "libc/sysv/consts/sig.h"
|
|
|
|
|
|
#include "libc/sysv/errfuns.h"
|
|
|
|
|
|
#include "libc/time/time.h"
|
|
|
|
|
|
#include "libc/x/x.h"
|
|
|
|
|
|
#include "tool/viz/lib/graphic.h"
|
|
|
|
|
|
#include "tool/viz/lib/knobs.h"
|
|
|
|
|
|
#include "tool/viz/lib/ycbcr.h"
|
|
|
|
|
|
|
|
|
|
|
|
#define M 15
|
|
|
|
|
|
#define CLAMP(X) MIN(255, MAX(0, X))
|
|
|
|
|
|
|
|
|
|
|
|
const double kBt601Primaries[] = {.299, .587, .114};
|
|
|
|
|
|
const double kBt709Primaries[] = {871024 / 4096299., 8788810 / 12288897.,
|
|
|
|
|
|
887015 / 12288897.};
|
|
|
|
|
|
|
|
|
|
|
|
const double kSrgbToXyz[3][3] = {
|
|
|
|
|
|
{506752 / 1228815., 87881 / 245763., 12673 / 70218.},
|
|
|
|
|
|
{87098 / 409605., 175762 / 245763., 12673 / 175545.},
|
|
|
|
|
|
{7918 / 409605., 87881 / 737289., 1001167 / 1053270.},
|
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
long magikarp_latency_;
|
|
|
|
|
|
long gyarados_latency_;
|
|
|
|
|
|
long ycbcr2rgb_latency_;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
struct timespec magikarp_start_;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
|
|
|
|
|
|
struct YCbCr {
|
|
|
|
|
|
bool yonly;
|
|
|
|
|
|
int magnums[8][4];
|
|
|
|
|
|
int lighting[6][4];
|
|
|
|
|
|
unsigned char transfer[2][256];
|
|
|
|
|
|
struct YCbCrSamplingSolution {
|
|
|
|
|
|
long dyn, dxn;
|
|
|
|
|
|
long syn, sxn;
|
|
|
|
|
|
double ry, rx;
|
|
|
|
|
|
double oy, ox;
|
|
|
|
|
|
double py, px;
|
|
|
|
|
|
struct SamplingSolution *cy, *cx;
|
|
|
|
|
|
} luma, chroma;
|
|
|
|
|
|
};
|
|
|
|
|
|
|
2023-07-30 01:44:15 +00:00
|
|
|
|
static unsigned long roundup2pow(unsigned long x) {
|
|
|
|
|
|
return x > 1 ? 2ul << _bsrl(x - 1) : x ? 1 : 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static unsigned long rounddown2pow(unsigned long x) {
|
|
|
|
|
|
return x ? 1ul << _bsrl(x) : 0;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2020-06-15 14:18:57 +00:00
|
|
|
|
/**
|
|
|
|
|
|
* Computes magnums for Y′CbCr decoding.
|
|
|
|
|
|
*
|
|
|
|
|
|
* @param swing should be 219 for TV, or 255 for JPEG
|
|
|
|
|
|
* @param M is integer coefficient bits
|
|
|
|
|
|
*/
|
|
|
|
|
|
void YCbCrComputeCoefficients(int swing, double gamma,
|
|
|
|
|
|
const double primaries[3],
|
|
|
|
|
|
const double illuminant[3], int out_magnums[8][4],
|
|
|
|
|
|
int out_lighting[6][4],
|
|
|
|
|
|
unsigned char out_transfer[256]) {
|
|
|
|
|
|
int i, j;
|
|
|
|
|
|
double x;
|
|
|
|
|
|
double f1[6][3];
|
|
|
|
|
|
long longs[6][6];
|
2023-07-30 01:44:15 +00:00
|
|
|
|
long bitlimit = roundup2pow(swing);
|
|
|
|
|
|
long wordoffset = rounddown2pow((bitlimit - swing) / 2);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
long chromaswing = swing + 2 * (bitlimit / 2. - swing / 2. - wordoffset);
|
|
|
|
|
|
long lumamin = wordoffset;
|
|
|
|
|
|
long lumamax = wordoffset + swing;
|
|
|
|
|
|
long diffmax = wordoffset + chromaswing - bitlimit / 2;
|
|
|
|
|
|
long diffmin = -diffmax;
|
|
|
|
|
|
double rEb = 1 - primaries[2] + primaries[0] + primaries[1];
|
|
|
|
|
|
double rEgEb = 1 / primaries[1] * primaries[2] * rEb;
|
|
|
|
|
|
double rEr = 1 - primaries[0] + primaries[1] + primaries[2];
|
|
|
|
|
|
double rEgEr = 1 / primaries[1] * primaries[0] * rEr;
|
|
|
|
|
|
double unswing = 1. / swing * bitlimit;
|
|
|
|
|
|
double digital = 1. / swing * chromaswing;
|
|
|
|
|
|
double reals[6][6] = {
|
|
|
|
|
|
{rEr / digital},
|
|
|
|
|
|
{-rEgEb / digital, -rEgEr / digital},
|
|
|
|
|
|
{rEb / digital},
|
|
|
|
|
|
{0, 0, unswing},
|
|
|
|
|
|
};
|
|
|
|
|
|
for (i = 0; i < 4; ++i) {
|
|
|
|
|
|
GetIntegerCoefficients(longs[i], reals[i], M, diffmin, diffmax);
|
|
|
|
|
|
}
|
|
|
|
|
|
for (i = 0; i < 4; ++i) {
|
|
|
|
|
|
for (j = 0; j < 4; ++j) {
|
|
|
|
|
|
out_magnums[i][j] = longs[i][j];
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
out_magnums[3][0] = wordoffset;
|
|
|
|
|
|
out_magnums[3][1] = bitlimit / 2;
|
|
|
|
|
|
GetChromaticAdaptationMatrix(f1, kIlluminantD65, illuminant);
|
|
|
|
|
|
for (i = 0; i < 3; ++i) {
|
|
|
|
|
|
for (j = 0; j < 3; ++j) {
|
|
|
|
|
|
reals[i][j] = f1[i][j];
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
for (i = 0; i < 6; ++i) {
|
|
|
|
|
|
GetIntegerCoefficients(longs[i], reals[i], M, diffmin * 2, lumamax * 2);
|
|
|
|
|
|
}
|
|
|
|
|
|
for (i = 0; i < 6; ++i) {
|
|
|
|
|
|
for (j = 0; j < 3; ++j) {
|
|
|
|
|
|
out_lighting[i][j] = longs[i][j];
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
for (i = 0; i < 256; ++i) {
|
|
|
|
|
|
x = i;
|
|
|
|
|
|
x /= 255;
|
|
|
|
|
|
x = tv2pcgamma(x, gamma);
|
|
|
|
|
|
x *= 255;
|
|
|
|
|
|
out_transfer[i] = CLAMP(x);
|
|
|
|
|
|
}
|
|
|
|
|
|
memset(out_transfer, out_transfer[lumamin], lumamin);
|
|
|
|
|
|
memset(out_transfer + lumamax + 1, out_transfer[lumamax], bitlimit - lumamax);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void YCbCrInit(struct YCbCr **ycbcr, bool yonly, int swing, double gamma,
|
|
|
|
|
|
const double gamut[3], const double illuminant[3]) {
|
2023-05-10 06:35:10 +00:00
|
|
|
|
int i;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (!*ycbcr) *ycbcr = xcalloc(1, sizeof(struct YCbCr));
|
|
|
|
|
|
(*ycbcr)->yonly = yonly;
|
2022-04-24 16:59:22 +00:00
|
|
|
|
bzero((*ycbcr)->magnums, sizeof((*ycbcr)->magnums));
|
|
|
|
|
|
bzero((*ycbcr)->lighting, sizeof((*ycbcr)->lighting));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
YCbCrComputeCoefficients(swing, gamma, gamut, illuminant, (*ycbcr)->magnums,
|
|
|
|
|
|
(*ycbcr)->lighting, (*ycbcr)->transfer[0]);
|
2023-05-10 06:35:10 +00:00
|
|
|
|
for (i = 0; i < 256; ++i) {
|
|
|
|
|
|
(*ycbcr)->transfer[1][i] = i;
|
|
|
|
|
|
}
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void YCbCrFree(struct YCbCr **ycbcr) {
|
|
|
|
|
|
if (*ycbcr) {
|
|
|
|
|
|
FreeSamplingSolution((*ycbcr)->luma.cy), (*ycbcr)->luma.cy = NULL;
|
|
|
|
|
|
FreeSamplingSolution((*ycbcr)->luma.cx), (*ycbcr)->luma.cx = NULL;
|
|
|
|
|
|
FreeSamplingSolution((*ycbcr)->chroma.cy), (*ycbcr)->chroma.cy = NULL;
|
|
|
|
|
|
FreeSamplingSolution((*ycbcr)->chroma.cx), (*ycbcr)->chroma.cx = NULL;
|
|
|
|
|
|
free(*ycbcr), *ycbcr = NULL;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void *YCbCrReallocPlane(long ys, long xs, const unsigned char p[ys][xs],
|
|
|
|
|
|
long yn, long xn) {
|
|
|
|
|
|
long y;
|
|
|
|
|
|
unsigned char(*res)[yn][xn];
|
|
|
|
|
|
res = xmemalign(32, yn * xn);
|
|
|
|
|
|
for (y = 0; y < yn; ++y) {
|
|
|
|
|
|
memcpy((*res)[y], p[y], xn);
|
|
|
|
|
|
}
|
|
|
|
|
|
return res;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void YCbCrComputeSamplingSolution(struct YCbCrSamplingSolution *scale, long dyn,
|
|
|
|
|
|
long dxn, long syn, long sxn, double ry,
|
|
|
|
|
|
double rx, double oy, double ox, double py,
|
|
|
|
|
|
double px) {
|
|
|
|
|
|
if (scale->dyn != dyn || scale->dxn != dxn || scale->syn != syn ||
|
|
|
|
|
|
scale->sxn != sxn || fabs(scale->ry - ry) > .001 ||
|
|
|
|
|
|
fabs(scale->rx - rx) > .001 || fabs(scale->oy - oy) > .001 ||
|
|
|
|
|
|
fabs(scale->ox - ox) > .001 || fabs(scale->py - py) > .001 ||
|
|
|
|
|
|
fabs(scale->px - px) > .001) {
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("recomputing sampling solution");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
FreeSamplingSolution(scale->cy), scale->cy = NULL;
|
|
|
|
|
|
FreeSamplingSolution(scale->cx), scale->cx = NULL;
|
|
|
|
|
|
scale->cy = ComputeSamplingSolution(dyn, syn, ry, oy, py);
|
|
|
|
|
|
scale->cx = ComputeSamplingSolution(dxn, sxn, rx, ox, px);
|
|
|
|
|
|
scale->dyn = dyn, scale->dxn = dxn;
|
|
|
|
|
|
scale->syn = syn, scale->sxn = sxn;
|
|
|
|
|
|
scale->ry = ry, scale->rx = rx;
|
|
|
|
|
|
scale->oy = oy, scale->ox = ox;
|
|
|
|
|
|
scale->py = py, scale->px = px;
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void Y2Rgb(long yn, long xn, unsigned char RGB[restrict 3][yn][xn], long yys,
|
|
|
|
|
|
long yxs, const unsigned char Y[restrict yys][yxs],
|
|
|
|
|
|
const int K[8][4], const unsigned char T[256]) {
|
|
|
|
|
|
long i, j;
|
|
|
|
|
|
for (i = 0; i < yn; ++i) {
|
|
|
|
|
|
for (j = 0; j < xn; ++j) {
|
|
|
|
|
|
RGB[0][i][j] = T[Y[i][j]];
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
memcpy(RGB[1], RGB[0], yn * xn);
|
|
|
|
|
|
memcpy(RGB[2], RGB[0], yn * xn);
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
|
* Converts Y′CbCr samples to RGB.
|
|
|
|
|
|
*/
|
|
|
|
|
|
void YCbCr2Rgb(long yn, long xn, unsigned char RGB[restrict 3][yn][xn],
|
|
|
|
|
|
long yys, long yxs, const unsigned char Y[restrict yys][yxs],
|
|
|
|
|
|
long cys, long cxs, const unsigned char Cb[restrict cys][cxs],
|
|
|
|
|
|
const unsigned char Cr[restrict cys][cxs], const int K[8][4],
|
|
|
|
|
|
const int L[6][4], const unsigned char T[256]) {
|
|
|
|
|
|
long i, j;
|
2023-09-02 03:49:13 +00:00
|
|
|
|
short y, u, v, r, g, b;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
for (i = 0; i < yn; ++i) {
|
|
|
|
|
|
for (j = 0; j < xn; ++j) {
|
2020-08-25 11:23:25 +00:00
|
|
|
|
y = T[Y[i][j]];
|
2020-06-15 14:18:57 +00:00
|
|
|
|
u = Cb[i][j] - K[3][1];
|
|
|
|
|
|
v = Cr[i][j] - K[3][1];
|
|
|
|
|
|
r = y + QRS(M, v * K[0][0]);
|
|
|
|
|
|
g = y + QRS(M, u * K[1][0] + v * K[1][1]);
|
|
|
|
|
|
b = y + QRS(M, u * K[2][0]);
|
|
|
|
|
|
r = QRS(M, (MIN(235, MAX(16, r)) - K[3][0]) * K[3][2]);
|
|
|
|
|
|
g = QRS(M, (MIN(235, MAX(16, g)) - K[3][0]) * K[3][2]);
|
|
|
|
|
|
b = QRS(M, (MIN(235, MAX(16, b)) - K[3][0]) * K[3][2]);
|
|
|
|
|
|
RGB[0][i][j] = CLAMP(QRS(M, r * L[0][0] + g * L[0][1] + b * L[0][2]));
|
|
|
|
|
|
RGB[1][i][j] = CLAMP(QRS(M, r * L[1][0] + g * L[1][1] + b * L[1][2]));
|
|
|
|
|
|
RGB[2][i][j] = CLAMP(QRS(M, r * L[2][0] + g * L[2][1] + b * L[2][2]));
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void YCbCrConvert(struct YCbCr *me, long yn, long xn,
|
|
|
|
|
|
unsigned char RGB[restrict 3][yn][xn], long yys, long yxs,
|
|
|
|
|
|
const unsigned char Y[restrict yys][yxs], long cys, long cxs,
|
|
|
|
|
|
unsigned char Cb[restrict cys][cxs],
|
|
|
|
|
|
unsigned char Cr[restrict cys][cxs]) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
struct timespec ts = timespec_real();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
if (!me->yonly) {
|
|
|
|
|
|
YCbCr2Rgb(yn, xn, RGB, yys, yxs, Y, cys, cxs, Cb, Cr, me->magnums,
|
|
|
|
|
|
me->lighting, me->transfer[pf10_]);
|
|
|
|
|
|
} else {
|
|
|
|
|
|
Y2Rgb(yn, xn, RGB, yys, yxs, Y, me->magnums, me->transfer[pf10_]);
|
|
|
|
|
|
}
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
ycbcr2rgb_latency_ = timespec_tomicros(timespec_sub(timespec_real(), ts));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void YCbCr2RgbScaler(struct YCbCr *me, long dyn, long dxn,
|
|
|
|
|
|
unsigned char RGB[restrict 3][dyn][dxn], long yys,
|
|
|
|
|
|
long yxs, unsigned char Y[restrict yys][yxs], long cys,
|
|
|
|
|
|
long cxs, unsigned char Cb[restrict cys][cxs],
|
|
|
|
|
|
unsigned char Cr[restrict cys][cxs], long yyn, long yxn,
|
|
|
|
|
|
long cyn, long cxn, double syn, double sxn, double pry,
|
|
|
|
|
|
double prx) {
|
2023-09-02 03:49:13 +00:00
|
|
|
|
long scyn, scxn;
|
|
|
|
|
|
double yry, yrx, cry, crx, yoy, yox, coy, cox;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
scyn = syn * cyn / yyn;
|
|
|
|
|
|
scxn = sxn * cxn / yxn;
|
|
|
|
|
|
if (HALF(yxn) > dxn && HALF(scxn) > dxn) {
|
|
|
|
|
|
YCbCr2RgbScaler(me, dyn, dxn, RGB, yys, yxs,
|
|
|
|
|
|
Magikarp2xX(yys, yxs, Y, syn, sxn), cys, cxs,
|
|
|
|
|
|
Magkern2xX(cys, cxs, Cb, scyn, scxn),
|
|
|
|
|
|
Magkern2xX(cys, cxs, Cr, scyn, scxn), yyn, HALF(yxn), cyn,
|
|
|
|
|
|
HALF(cxn), syn, sxn / 2, pry, prx);
|
|
|
|
|
|
} else if (HALF(yyn) > dyn && HALF(scyn) > dyn) {
|
|
|
|
|
|
YCbCr2RgbScaler(me, dyn, dxn, RGB, yys, yxs,
|
|
|
|
|
|
Magikarp2xY(yys, yxs, Y, syn, sxn), cys, cxs,
|
|
|
|
|
|
Magkern2xY(cys, cxs, Cb, scyn, scxn),
|
|
|
|
|
|
Magkern2xY(cys, cxs, Cr, scyn, scxn), HALF(yyn), yxn,
|
|
|
|
|
|
HALF(cyn), scxn, syn / 2, sxn, pry, prx);
|
|
|
|
|
|
} else {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
struct timespec ts = timespec_real();
|
|
|
|
|
|
magikarp_latency_ = timespec_tomicros(timespec_sub(ts, magikarp_start_));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
yry = syn / dyn;
|
|
|
|
|
|
yrx = sxn / dxn;
|
|
|
|
|
|
cry = syn * cyn / yyn / dyn;
|
|
|
|
|
|
crx = sxn * cxn / yxn / dxn;
|
|
|
|
|
|
yoy = syn / scyn / 2 - pry * .5;
|
|
|
|
|
|
yox = sxn / scxn / 2 - prx * .5;
|
|
|
|
|
|
coy = syn / scyn / 2 - pry * .5;
|
|
|
|
|
|
cox = sxn / scxn / 2 - prx * .5;
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("gyarados pry=%.3f prx=%.3f syn=%.3f sxn=%.3f dyn=%ld dxn=%ld "
|
|
|
|
|
|
"yyn=%ld "
|
|
|
|
|
|
"yxn=%ld cyn=%ld cxn=%ld yry=%.3f yrx=%.3f cry=%.3f crx=%.3f "
|
|
|
|
|
|
"yoy=%.3f "
|
|
|
|
|
|
"yox=%.3f coy=%.3f cox=%.3f",
|
|
|
|
|
|
pry, prx, syn, sxn, dyn, dxn, yyn, yxn, cyn, cxn, yry, yrx, cry, crx,
|
|
|
|
|
|
yoy, yox, coy, cox);
|
2020-06-15 14:18:57 +00:00
|
|
|
|
YCbCrComputeSamplingSolution(&me->luma, dyn, dxn, syn, sxn, yry, yrx, yoy,
|
|
|
|
|
|
yox, pry, prx);
|
|
|
|
|
|
YCbCrComputeSamplingSolution(&me->chroma, dyn, dxn, scyn, scxn, cry, crx,
|
|
|
|
|
|
coy, cox, pry, prx);
|
|
|
|
|
|
if (pf8_) sharpen(1, yys, yxs, (void *)Y, yyn, yxn);
|
|
|
|
|
|
if (pf9_) unsharp(1, yys, yxs, (void *)Y, yyn, yxn);
|
|
|
|
|
|
GyaradosUint8(yys, yxs, Y, yys, yxs, Y, dyn, dxn, syn, sxn, 0, 255,
|
|
|
|
|
|
me->luma.cy, me->luma.cx, true);
|
|
|
|
|
|
GyaradosUint8(cys, cxs, Cb, cys, cxs, Cb, dyn, dxn, scyn, scxn, 0, 255,
|
|
|
|
|
|
me->chroma.cy, me->chroma.cx, false);
|
|
|
|
|
|
GyaradosUint8(cys, cxs, Cr, cys, cxs, Cr, dyn, dxn, scyn, scxn, 0, 255,
|
|
|
|
|
|
me->chroma.cy, me->chroma.cx, false);
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
gyarados_latency_ = timespec_tomicros(timespec_sub(timespec_real(), ts));
|
2020-06-15 14:18:57 +00:00
|
|
|
|
YCbCrConvert(me, dyn, dxn, RGB, yys, yxs, Y, cys, cxs, Cb, Cr);
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("done");
|
2020-06-15 14:18:57 +00:00
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
|
* Converts Y′CbCr frame for PC display.
|
|
|
|
|
|
*
|
|
|
|
|
|
* "[The] experiments of Professor J. D. Forbes, which I
|
|
|
|
|
|
* witnessed… [established] that blue and yellow do not
|
|
|
|
|
|
* make green but a pinkish tint, when neither prevails
|
|
|
|
|
|
* in the combination [and the] result of mixing yellow
|
|
|
|
|
|
* and blue was, I believe, not previously known.
|
|
|
|
|
|
* — James Clerk Maxwell
|
|
|
|
|
|
*
|
|
|
|
|
|
* This function converts TV to PC graphics. We do that by
|
|
|
|
|
|
*
|
|
|
|
|
|
* 1. decimating w/ facebook magikarp photoshop cubic sharpen
|
|
|
|
|
|
* 2. upsampling color difference planes, to be as big as luma plane
|
|
|
|
|
|
* 3. converting color format
|
|
|
|
|
|
* 4. expanding dynamic range
|
|
|
|
|
|
* 5. transferring gamma from TV to PC convention
|
|
|
|
|
|
* 6. resampling again to exact requested display / pixel geometry
|
|
|
|
|
|
*
|
|
|
|
|
|
* @param dyn/dxn is display height/width after scaling/conversion
|
|
|
|
|
|
* @param RGB points to memory for packed de-interlaced RGB output
|
|
|
|
|
|
* @param Y′ ∈ [16,235] is the luminance plane a gamma-corrected RGB
|
|
|
|
|
|
* weighted sum; a.k.a. black/white legacy component part of the
|
|
|
|
|
|
* TV signal; which may be used independently of the chrominance
|
|
|
|
|
|
* planes; and decodes to the range [0,1]
|
|
|
|
|
|
* @param Cb/Cr ∈ [16,240] is blue/red chrominance difference planes
|
|
|
|
|
|
* which (if sampled at a different rate) will get stretched out
|
|
|
|
|
|
* over the luma plane appropriately
|
|
|
|
|
|
* @param yys/yxs dimensions luma sample array
|
|
|
|
|
|
* @param cys/cxs dimensions chroma sample arrays
|
|
|
|
|
|
* @param yyn/yxn is number of samples in luma signal
|
|
|
|
|
|
* @param cyn/cxn is number of samples in each chroma signal
|
|
|
|
|
|
* @param syn/sxn is size of source signal
|
|
|
|
|
|
* @param pry/prx is pixel aspect ratio, e.g. 1,1
|
|
|
|
|
|
* @return RGB
|
|
|
|
|
|
*/
|
|
|
|
|
|
void *YCbCr2RgbScale(long dyn, long dxn,
|
|
|
|
|
|
unsigned char RGB[restrict 3][dyn][dxn], long yys,
|
|
|
|
|
|
long yxs, unsigned char Y[restrict yys][yxs], long cys,
|
|
|
|
|
|
long cxs, unsigned char Cb[restrict cys][cxs],
|
|
|
|
|
|
unsigned char Cr[restrict cys][cxs], long yyn, long yxn,
|
|
|
|
|
|
long cyn, long cxn, double syn, double sxn, double pry,
|
|
|
|
|
|
double prx, struct YCbCr **ycbcr) {
|
|
|
|
|
|
long minyys, minyxs, mincys, mincxs;
|
|
|
|
|
|
CHECK_LE(yyn, yys);
|
|
|
|
|
|
CHECK_LE(yxn, yxs);
|
|
|
|
|
|
CHECK_LE(cyn, cys);
|
|
|
|
|
|
CHECK_LE(cxn, cxs);
|
2021-09-04 04:14:26 +00:00
|
|
|
|
INFOF("magikarp2x");
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
|
magikarp_start_ = timespec_real();
|
2020-06-15 14:18:57 +00:00
|
|
|
|
minyys = MAX(ceil(syn), MAX(yyn, ceil(dyn * pry)));
|
|
|
|
|
|
minyxs = MAX(ceil(sxn), MAX(yxn, ceil(dxn * prx)));
|
|
|
|
|
|
mincys = MAX(cyn, ceil(dyn * pry));
|
|
|
|
|
|
mincxs = MAX(cxn, ceil(dxn * prx));
|
|
|
|
|
|
YCbCr2RgbScaler(*ycbcr, dyn, dxn, RGB, MAX(yys, minyys), MAX(yxs, minyxs),
|
|
|
|
|
|
(yys >= minyys && yxs >= minyxs
|
|
|
|
|
|
? Y
|
|
|
|
|
|
: gc(YCbCrReallocPlane(yys, yxs, Y, minyys, minyxs))),
|
|
|
|
|
|
MAX(cys, mincys), MAX(cxs, mincxs),
|
|
|
|
|
|
(cys >= mincys && cxs >= mincxs
|
|
|
|
|
|
? Cb
|
|
|
|
|
|
: gc(YCbCrReallocPlane(cys, cxs, Cb, mincys, mincxs))),
|
|
|
|
|
|
(cys >= mincys && cxs >= mincxs
|
|
|
|
|
|
? Cr
|
|
|
|
|
|
: gc(YCbCrReallocPlane(cys, cxs, Cr, mincys, mincxs))),
|
|
|
|
|
|
yyn, yxn, cyn, cxn, syn, sxn, pry, prx);
|
|
|
|
|
|
return RGB;
|
|
|
|
|
|
}
|
