Get enough of libc added for stb_image, maybe

This commit is contained in:
Ben Visness 2023-06-23 20:38:19 -05:00 committed by Martin Fouilleul
parent 571e8a6f8e
commit 47ea91ef66
21 changed files with 551 additions and 9 deletions

View File

@ -1,5 +1,7 @@
#!/bin/bash #!/bin/bash
set -eo pipefail
target="$1" target="$1"
if [ -z $target ] ; then if [ -z $target ] ; then

23
cstdlib/include/assert.h Normal file
View File

@ -0,0 +1,23 @@
#include <features.h>
#undef assert
#ifdef NDEBUG
#define assert(x) (void)0
#else
#define assert(x) ((void)((x) || (__assert_fail(#x, __FILE__, __LINE__, __func__),0)))
#endif
#if __STDC_VERSION__ >= 201112L && !defined(__cplusplus)
#define static_assert _Static_assert
#endif
#ifdef __cplusplus
extern "C" {
#endif
_Noreturn void __assert_fail (const char *, const char *, int, const char *);
#ifdef __cplusplus
}
#endif

View File

10
cstdlib/include/math.h Normal file
View File

@ -0,0 +1,10 @@
// NOTE(orca): not doing anything fancy for float_t and double_t
typedef float float_t;
typedef double double_t;
#define NAN __builtin_nanf("")
#define INFINITY __builtin_inff()
double fabs(double);
double pow(double, double);

5
cstdlib/include/stdio.h Normal file
View File

@ -0,0 +1,5 @@
struct _IO_FILE { char __x; };
typedef struct _IO_FILE FILE;
// TODO(orca)
// int fprintf(FILE *__restrict, const char *__restrict, ...);

3
cstdlib/include/stdlib.h Normal file
View File

@ -0,0 +1,3 @@
_Noreturn void abort (void);
int abs (int);

0
cstdlib/include/string.h Normal file
View File

View File

@ -0,0 +1,6 @@
#include "libm.h"
double __math_invalid(double x)
{
return (x - x) / (x - x);
}

View File

@ -0,0 +1,6 @@
#include "libm.h"
double __math_oflow(uint32_t sign)
{
return __math_xflow(sign, 0x1p769);
}

View File

@ -0,0 +1,6 @@
#include "libm.h"
double __math_uflow(uint32_t sign)
{
return __math_xflow(sign, 0x1p-767);
}

View File

@ -0,0 +1,8 @@
#include "libm.h"
double __math_xflow(uint32_t sign, double y)
{
// NOTE(orca): no fp barriers
// return eval_as_double(fp_barrier(sign ? -y : y) * y);
return eval_as_double((sign ? -y : y) * y);
}

6
cstdlib/src/abort.c Normal file
View File

@ -0,0 +1,6 @@
// NOTE(orca): This is a clang intrinsic. I hope it generates a wasm unreachable. I have not verified this.
// TODO(orca): Verify this.
_Noreturn void abort(void)
{
__builtin_unreachable();
}

6
cstdlib/src/abs.c Normal file
View File

@ -0,0 +1,6 @@
#include <stdlib.h>
int abs(int a)
{
return a>0 ? a : -a;
}

9
cstdlib/src/assert.c Normal file
View File

@ -0,0 +1,9 @@
#include <stdio.h>
#include <stdlib.h>
_Noreturn void __assert_fail(const char *expr, const char *file, int line, const char *func)
{
// TODO(orca)
// fprintf(stderr, "Assertion failed: %s (%s: %s: %d)\n", expr, file, func, line);
abort();
}

26
cstdlib/src/exp_data.h Normal file
View File

@ -0,0 +1,26 @@
/*
* Copyright (c) 2018, Arm Limited.
* SPDX-License-Identifier: MIT
*/
#ifndef _EXP_DATA_H
#define _EXP_DATA_H
#include <features.h>
#include <stdint.h>
#define EXP_TABLE_BITS 7
#define EXP_POLY_ORDER 5
#define EXP_USE_TOINT_NARROW 0
#define EXP2_POLY_ORDER 5
extern const struct exp_data {
double invln2N;
double shift;
double negln2hiN;
double negln2loN;
double poly[4]; /* Last four coefficients. */
double exp2_shift;
double exp2_poly[EXP2_POLY_ORDER];
uint64_t tab[2*(1 << EXP_TABLE_BITS)];
} __exp_data;
#endif

9
cstdlib/src/fabs.c Normal file
View File

@ -0,0 +1,9 @@
#include <math.h>
#include <stdint.h>
double fabs(double x)
{
union {double f; uint64_t i;} u = {x};
u.i &= -1ULL/2;
return u.f;
}

41
cstdlib/src/libm.h Normal file
View File

@ -0,0 +1,41 @@
#include <stdint.h>
#define WANT_ROUNDING 1
#if WANT_SNAN
#error SNaN is unsupported
#else
#define issignalingf_inline(x) 0
#define issignaling_inline(x) 0
#endif
/* Helps static branch prediction so hot path can be better optimized. */
#ifdef __GNUC__
#define predict_true(x) __builtin_expect(!!(x), 1)
#define predict_false(x) __builtin_expect(x, 0)
#else
#define predict_true(x) (x)
#define predict_false(x) (x)
#endif
static inline float eval_as_float(float x)
{
float y = x;
return y;
}
static inline double eval_as_double(double x)
{
double y = x;
return y;
}
#define asuint(f) ((union{float _f; uint32_t _i;}){f})._i
#define asfloat(i) ((union{uint32_t _i; float _f;}){i})._f
#define asuint64(f) ((union{double _f; uint64_t _i;}){f})._i
#define asdouble(i) ((union{uint64_t _i; double _f;}){i})._f
double __math_xflow(uint32_t, double);
double __math_uflow(uint32_t);
double __math_oflow(uint32_t);
double __math_invalid(double);

346
cstdlib/src/pow.c Normal file
View File

@ -0,0 +1,346 @@
/*
* Double-precision x^y function.
*
* Copyright (c) 2018, Arm Limited.
* SPDX-License-Identifier: MIT
*/
#include <math.h>
#include <stdint.h>
#include "libm.h"
#include "exp_data.h"
#include "pow_data.h"
/*
Worst-case error: 0.54 ULP (~= ulperr_exp + 1024*Ln2*relerr_log*2^53)
relerr_log: 1.3 * 2^-68 (Relative error of log, 1.5 * 2^-68 without fma)
ulperr_exp: 0.509 ULP (ULP error of exp, 0.511 ULP without fma)
*/
#define T __pow_log_data.tab
#define A __pow_log_data.poly
#define Ln2hi __pow_log_data.ln2hi
#define Ln2lo __pow_log_data.ln2lo
#define N (1 << POW_LOG_TABLE_BITS)
#define OFF 0x3fe6955500000000
/* Top 12 bits of a double (sign and exponent bits). */
static inline uint32_t top12(double x)
{
return asuint64(x) >> 52;
}
/* Compute y+TAIL = log(x) where the rounded result is y and TAIL has about
additional 15 bits precision. IX is the bit representation of x, but
normalized in the subnormal range using the sign bit for the exponent. */
static inline double_t log_inline(uint64_t ix, double_t *tail)
{
/* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
double_t z, r, y, invc, logc, logctail, kd, hi, t1, t2, lo, lo1, lo2, p;
uint64_t iz, tmp;
int k, i;
/* x = 2^k z; where z is in range [OFF,2*OFF) and exact.
The range is split into N subintervals.
The ith subinterval contains z and c is near its center. */
tmp = ix - OFF;
i = (tmp >> (52 - POW_LOG_TABLE_BITS)) % N;
k = (int64_t)tmp >> 52; /* arithmetic shift */
iz = ix - (tmp & 0xfffULL << 52);
z = asdouble(iz);
kd = (double_t)k;
/* log(x) = k*Ln2 + log(c) + log1p(z/c-1). */
invc = T[i].invc;
logc = T[i].logc;
logctail = T[i].logctail;
/* Note: 1/c is j/N or j/N/2 where j is an integer in [N,2N) and
|z/c - 1| < 1/N, so r = z/c - 1 is exactly representible. */
#if __FP_FAST_FMA
r = __builtin_fma(z, invc, -1.0);
#else
/* Split z such that rhi, rlo and rhi*rhi are exact and |rlo| <= |r|. */
double_t zhi = asdouble((iz + (1ULL << 31)) & (-1ULL << 32));
double_t zlo = z - zhi;
double_t rhi = zhi * invc - 1.0;
double_t rlo = zlo * invc;
r = rhi + rlo;
#endif
/* k*Ln2 + log(c) + r. */
t1 = kd * Ln2hi + logc;
t2 = t1 + r;
lo1 = kd * Ln2lo + logctail;
lo2 = t1 - t2 + r;
/* Evaluation is optimized assuming superscalar pipelined execution. */
double_t ar, ar2, ar3, lo3, lo4;
ar = A[0] * r; /* A[0] = -0.5. */
ar2 = r * ar;
ar3 = r * ar2;
/* k*Ln2 + log(c) + r + A[0]*r*r. */
#if __FP_FAST_FMA
hi = t2 + ar2;
lo3 = __builtin_fma(ar, r, -ar2);
lo4 = t2 - hi + ar2;
#else
double_t arhi = A[0] * rhi;
double_t arhi2 = rhi * arhi;
hi = t2 + arhi2;
lo3 = rlo * (ar + arhi);
lo4 = t2 - hi + arhi2;
#endif
/* p = log1p(r) - r - A[0]*r*r. */
p = (ar3 * (A[1] + r * A[2] +
ar2 * (A[3] + r * A[4] + ar2 * (A[5] + r * A[6]))));
lo = lo1 + lo2 + lo3 + lo4 + p;
y = hi + lo;
*tail = hi - y + lo;
return y;
}
#undef N
#undef T
#define N (1 << EXP_TABLE_BITS)
#define InvLn2N __exp_data.invln2N
#define NegLn2hiN __exp_data.negln2hiN
#define NegLn2loN __exp_data.negln2loN
#define Shift __exp_data.shift
#define T __exp_data.tab
#define C2 __exp_data.poly[5 - EXP_POLY_ORDER]
#define C3 __exp_data.poly[6 - EXP_POLY_ORDER]
#define C4 __exp_data.poly[7 - EXP_POLY_ORDER]
#define C5 __exp_data.poly[8 - EXP_POLY_ORDER]
#define C6 __exp_data.poly[9 - EXP_POLY_ORDER]
/* Handle cases that may overflow or underflow when computing the result that
is scale*(1+TMP) without intermediate rounding. The bit representation of
scale is in SBITS, however it has a computed exponent that may have
overflown into the sign bit so that needs to be adjusted before using it as
a double. (int32_t)KI is the k used in the argument reduction and exponent
adjustment of scale, positive k here means the result may overflow and
negative k means the result may underflow. */
static inline double specialcase(double_t tmp, uint64_t sbits, uint64_t ki)
{
double_t scale, y;
if ((ki & 0x80000000) == 0) {
/* k > 0, the exponent of scale might have overflowed by <= 460. */
sbits -= 1009ull << 52;
scale = asdouble(sbits);
y = 0x1p1009 * (scale + scale * tmp);
return eval_as_double(y);
}
/* k < 0, need special care in the subnormal range. */
sbits += 1022ull << 52;
/* Note: sbits is signed scale. */
scale = asdouble(sbits);
y = scale + scale * tmp;
if (fabs(y) < 1.0) {
/* Round y to the right precision before scaling it into the subnormal
range to avoid double rounding that can cause 0.5+E/2 ulp error where
E is the worst-case ulp error outside the subnormal range. So this
is only useful if the goal is better than 1 ulp worst-case error. */
double_t hi, lo, one = 1.0;
if (y < 0.0)
one = -1.0;
lo = scale - y + scale * tmp;
hi = one + y;
lo = one - hi + y + lo;
y = eval_as_double(hi + lo) - one;
/* Fix the sign of 0. */
if (y == 0.0)
y = asdouble(sbits & 0x8000000000000000);
/* The underflow exception needs to be signaled explicitly. */
// NOTE(orca): removing special fp functions
// fp_force_eval(fp_barrier(0x1p-1022) * 0x1p-1022);
}
y = 0x1p-1022 * y;
return eval_as_double(y);
}
#define SIGN_BIAS (0x800 << EXP_TABLE_BITS)
/* Computes sign*exp(x+xtail) where |xtail| < 2^-8/N and |xtail| <= |x|.
The sign_bias argument is SIGN_BIAS or 0 and sets the sign to -1 or 1. */
static inline double exp_inline(double_t x, double_t xtail, uint32_t sign_bias)
{
uint32_t abstop;
uint64_t ki, idx, top, sbits;
/* double_t for better performance on targets with FLT_EVAL_METHOD==2. */
double_t kd, z, r, r2, scale, tail, tmp;
abstop = top12(x) & 0x7ff;
if (predict_false(abstop - top12(0x1p-54) >=
top12(512.0) - top12(0x1p-54))) {
if (abstop - top12(0x1p-54) >= 0x80000000) {
/* Avoid spurious underflow for tiny x. */
/* Note: 0 is common input. */
double_t one = WANT_ROUNDING ? 1.0 + x : 1.0;
return sign_bias ? -one : one;
}
if (abstop >= top12(1024.0)) {
/* Note: inf and nan are already handled. */
if (asuint64(x) >> 63)
return __math_uflow(sign_bias);
else
return __math_oflow(sign_bias);
}
/* Large x is special cased below. */
abstop = 0;
}
/* exp(x) = 2^(k/N) * exp(r), with exp(r) in [2^(-1/2N),2^(1/2N)]. */
/* x = ln2/N*k + r, with int k and r in [-ln2/2N, ln2/2N]. */
z = InvLn2N * x;
#if TOINT_INTRINSICS
kd = roundtoint(z);
ki = converttoint(z);
#elif EXP_USE_TOINT_NARROW
/* z - kd is in [-0.5-2^-16, 0.5] in all rounding modes. */
kd = eval_as_double(z + Shift);
ki = asuint64(kd) >> 16;
kd = (double_t)(int32_t)ki;
#else
/* z - kd is in [-1, 1] in non-nearest rounding modes. */
kd = eval_as_double(z + Shift);
ki = asuint64(kd);
kd -= Shift;
#endif
r = x + kd * NegLn2hiN + kd * NegLn2loN;
/* The code assumes 2^-200 < |xtail| < 2^-8/N. */
r += xtail;
/* 2^(k/N) ~= scale * (1 + tail). */
idx = 2 * (ki % N);
top = (ki + sign_bias) << (52 - EXP_TABLE_BITS);
tail = asdouble(T[idx]);
/* This is only a valid scale when -1023*N < k < 1024*N. */
sbits = T[idx + 1] + top;
/* exp(x) = 2^(k/N) * exp(r) ~= scale + scale * (tail + exp(r) - 1). */
/* Evaluation is optimized assuming superscalar pipelined execution. */
r2 = r * r;
/* Without fma the worst case error is 0.25/N ulp larger. */
/* Worst case error is less than 0.5+1.11/N+(abs poly error * 2^53) ulp. */
tmp = tail + r + r2 * (C2 + r * C3) + r2 * r2 * (C4 + r * C5);
if (predict_false(abstop == 0))
return specialcase(tmp, sbits, ki);
scale = asdouble(sbits);
/* Note: tmp == 0 or |tmp| > 2^-200 and scale > 2^-739, so there
is no spurious underflow here even without fma. */
return eval_as_double(scale + scale * tmp);
}
/* Returns 0 if not int, 1 if odd int, 2 if even int. The argument is
the bit representation of a non-zero finite floating-point value. */
static inline int checkint(uint64_t iy)
{
int e = iy >> 52 & 0x7ff;
if (e < 0x3ff)
return 0;
if (e > 0x3ff + 52)
return 2;
if (iy & ((1ULL << (0x3ff + 52 - e)) - 1))
return 0;
if (iy & (1ULL << (0x3ff + 52 - e)))
return 1;
return 2;
}
/* Returns 1 if input is the bit representation of 0, infinity or nan. */
static inline int zeroinfnan(uint64_t i)
{
return 2 * i - 1 >= 2 * asuint64(INFINITY) - 1;
}
double pow(double x, double y)
{
uint32_t sign_bias = 0;
uint64_t ix, iy;
uint32_t topx, topy;
ix = asuint64(x);
iy = asuint64(y);
topx = top12(x);
topy = top12(y);
if (predict_false(topx - 0x001 >= 0x7ff - 0x001 ||
(topy & 0x7ff) - 0x3be >= 0x43e - 0x3be)) {
/* Note: if |y| > 1075 * ln2 * 2^53 ~= 0x1.749p62 then pow(x,y) = inf/0
and if |y| < 2^-54 / 1075 ~= 0x1.e7b6p-65 then pow(x,y) = +-1. */
/* Special cases: (x < 0x1p-126 or inf or nan) or
(|y| < 0x1p-65 or |y| >= 0x1p63 or nan). */
if (predict_false(zeroinfnan(iy))) {
if (2 * iy == 0)
return issignaling_inline(x) ? x + y : 1.0;
if (ix == asuint64(1.0))
return issignaling_inline(y) ? x + y : 1.0;
if (2 * ix > 2 * asuint64(INFINITY) ||
2 * iy > 2 * asuint64(INFINITY))
return x + y;
if (2 * ix == 2 * asuint64(1.0))
return 1.0;
if ((2 * ix < 2 * asuint64(1.0)) == !(iy >> 63))
return 0.0; /* |x|<1 && y==inf or |x|>1 && y==-inf. */
return y * y;
}
if (predict_false(zeroinfnan(ix))) {
double_t x2 = x * x;
if (ix >> 63 && checkint(iy) == 1)
x2 = -x2;
/* Without the barrier some versions of clang hoist the 1/x2 and
thus division by zero exception can be signaled spuriously. */
// NOTE(orca): I hope my version of clang is not affected lol
// return iy >> 63 ? fp_barrier(1 / x2) : x2;
return iy >> 63 ? (1 / x2) : x2;
}
/* Here x and y are non-zero finite. */
if (ix >> 63) {
/* Finite x < 0. */
int yint = checkint(iy);
if (yint == 0)
return __math_invalid(x);
if (yint == 1)
sign_bias = SIGN_BIAS;
ix &= 0x7fffffffffffffff;
topx &= 0x7ff;
}
if ((topy & 0x7ff) - 0x3be >= 0x43e - 0x3be) {
/* Note: sign_bias == 0 here because y is not odd. */
if (ix == asuint64(1.0))
return 1.0;
if ((topy & 0x7ff) < 0x3be) {
/* |y| < 2^-65, x^y ~= 1 + y*log(x). */
if (WANT_ROUNDING)
return ix > asuint64(1.0) ? 1.0 + y :
1.0 - y;
else
return 1.0;
}
return (ix > asuint64(1.0)) == (topy < 0x800) ?
__math_oflow(0) :
__math_uflow(0);
}
if (topx == 0) {
/* Normalize subnormal x so exponent becomes negative. */
ix = asuint64(x * 0x1p52);
ix &= 0x7fffffffffffffff;
ix -= 52ULL << 52;
}
}
double_t lo;
double_t hi = log_inline(ix, &lo);
double_t ehi, elo;
#if __FP_FAST_FMA
ehi = y * hi;
elo = y * lo + __builtin_fma(y, hi, -ehi);
#else
double_t yhi = asdouble(iy & -1ULL << 27);
double_t ylo = y - yhi;
double_t lhi = asdouble(asuint64(hi) & -1ULL << 27);
double_t llo = hi - lhi + lo;
ehi = yhi * lhi;
elo = ylo * lhi + y * llo; /* |elo| < |ehi| * 2^-25. */
#endif
return exp_inline(ehi, elo, sign_bias);
}

22
cstdlib/src/pow_data.h Normal file
View File

@ -0,0 +1,22 @@
/*
* Copyright (c) 2018, Arm Limited.
* SPDX-License-Identifier: MIT
*/
#ifndef _POW_DATA_H
#define _POW_DATA_H
#include <features.h>
#define POW_LOG_TABLE_BITS 7
#define POW_LOG_POLY_ORDER 8
extern const struct pow_log_data {
double ln2hi;
double ln2lo;
double poly[POW_LOG_POLY_ORDER - 1]; /* First coefficient is 1. */
/* Note: the pad field is unused, but allows slightly faster indexing. */
struct {
double invc, pad, logc, logctail;
} tab[1 << POW_LOG_TABLE_BITS];
} __pow_log_data;
#endif

View File

@ -1,14 +1,26 @@
#!/bin/bash #!/bin/bash
set -euo pipefail
if [[ -x /usr/local/opt/llvm/bin/clang ]]; then
CLANG=/usr/local/opt/llvm/bin/clang
elif [[ -x /opt/homebrew/opt/llvm/bin/clang ]]; then
CLANG=/opt/homebrew/opt/llvm/bin/clang
else
echo "Could not find Homebrew clang; this script will probably not work."
CLANG=clang
fi
wasmFlags="--target=wasm32 \ wasmFlags="--target=wasm32 \
--no-standard-libraries \ --no-standard-libraries \
-fno-builtin \ -fno-builtin \
-Wl,--no-entry \ -Wl,--no-entry \
-Wl,--export-dynamic \ -Wl,--export-dynamic \
-g \ -g \
-O2 \
-D__ORCA__ \ -D__ORCA__ \
-I ../../src -I ../../sdk -I../../milepost/ext -I ../../milepost -I ../../milepost/src -I ../../milepost/src/util -I ../../milepost/src/platform -I../.." -isystem ../../cstdlib/include -I ../../sdk -I../../milepost/ext -I ../../milepost -I ../../milepost/src -I ../../milepost/src/util -I ../../milepost/src/platform -I../.."
/usr/local/opt/llvm/bin/clang $wasmFlags -o ./module.wasm ../../sdk/orca.c src/main.c $CLANG $wasmFlags -o ./module.wasm ../../cstdlib/src/*.c ../../sdk/orca.c src/main.c
python3 ../../scripts/mkapp.py --orca-dir ../.. --name Pong --icon icon.png --data-file data/ball.png module.wasm python3 ../../scripts/mkapp.py --orca-dir ../.. --name Pong --icon icon.png --data-file data/ball.png module.wasm

View File

@ -42,17 +42,18 @@ ORCA_EXPORT void OnInit(void)
surface = mg_surface_main(); surface = mg_surface_main();
canvas = mg_canvas_create(); canvas = mg_canvas_create();
//NOTE: file test //NOTE: file test
file_handle file = file_open(STR8("/ball.png"), FILE_ACCESS_READ, 0); file_handle file = file_open(STR8("/ball.png"), FILE_ACCESS_READ, 0);
if(file_last_error(file) != IO_OK) if(file_last_error(file) != IO_OK)
{ {
log_error("Couldn't open file ball.png\n"); log_error("Couldn't open file ball.png\n");
} }
u64 size = file_size(file); u64 size = file_size(file);
char* buffer = mem_arena_alloc(mem_scratch(), size); char* buffer = mem_arena_alloc(mem_scratch(), size);
file_read(file, size, buffer); file_read(file, size, buffer);
file_close(file); file_close(file);
image = mg_image_create_from_data(surface, str8_from_buffer(size, buffer), false);
file = file_open(STR8("/test.txt"), FILE_ACCESS_WRITE, FILE_OPEN_CREATE); file = file_open(STR8("/test.txt"), FILE_ACCESS_WRITE, FILE_OPEN_CREATE);
if(file_last_error(file) != IO_OK) if(file_last_error(file) != IO_OK)
@ -63,10 +64,6 @@ ORCA_EXPORT void OnInit(void)
file_write(file, test_string.len, test_string.ptr); file_write(file, test_string.len, test_string.ptr);
file_close(file); file_close(file);
/*NOTE: Do this when we can compile stb to wasm
image = mg_image_create_from_data(surface, str8_from_buffer(size, buffer), false);
*/
mem_arena_clear(mem_scratch()); mem_arena_clear(mem_scratch());
} }
@ -186,8 +183,7 @@ ORCA_EXPORT void OnFrameRefresh(void)
mg_set_color(paddleColor); mg_set_color(paddleColor);
mg_rectangle_fill(paddle.x, paddle.y, paddle.w, paddle.h); mg_rectangle_fill(paddle.x, paddle.y, paddle.w, paddle.h);
mg_set_color(ballColor); mg_image_draw(image, ball);
mg_circle_fill(ball.x+ball.w/2, ball.y + ball.w/2, ball.w/2.);
mg_matrix_pop(); mg_matrix_pop();