Fix wasm3 alignment errors and silence some warnings #135

Merged
MartinFouilleul merged 5 commits from fix_wasm3_align_and_warnings into main 2023-09-21 06:52:03 +00:00
5 changed files with 99 additions and 102 deletions

View File

@ -170,7 +170,7 @@ def build_platform_layer_lib_win(release):
def build_platform_layer_lib_mac(release):
sdk_dir = "/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk"
flags = ["-mmacos-version-min=10.15.4", "-maes"]
flags = ["-mmacos-version-min=10.15.4"]
cflags = ["-std=c11"]
debug_flags = ["-O3"] if release else ["-g", "-DOC_DEBUG", "-DOC_LOG_COMPILE_DEBUG"]
ldflags = [f"-L{sdk_dir}/usr/lib", f"-F{sdk_dir}/System/Library/Frameworks/"]
@ -280,6 +280,7 @@ def build_wasm3_lib_mac(release):
"-foptimize-sibling-calls",
"-Wno-extern-initializer",
"-Dd_m3VerboseErrorMessages",
"-mmacos-version-min=10.15.4"
]
for f in glob.iglob("src/ext/wasm3/source/*.c"):
@ -289,7 +290,7 @@ def build_wasm3_lib_mac(release):
"-o", f"build/obj/{name}",
f,
], check=True)
subprocess.run(["ar", "-rcs", "build/lib/libwasm3.a", *glob.glob("build/obj/*.o")], check=True)
subprocess.run(["libtool", "-static", "-o", "build/lib/libwasm3.a", "-no_warning_for_no_symbols", *glob.glob("build/obj/*.o")], check=True)
subprocess.run(["rm", "-rf", "build/obj"], check=True)
@ -348,9 +349,7 @@ def build_orca_mac(release):
debug_flags = ["-O2"] if release else ["-g", "-DOC_DEBUG -DOC_LOG_COMPILE_DEBUG"]
flags = [
*debug_flags,
"-mmacos-version-min=10.15.4",
"-maes",
]
"-mmacos-version-min=10.15.4"]
gen_all_bindings()

View File

@ -121,7 +121,7 @@ typedef int8_t i8;
# if defined (M3_COMPILER_MSVC)
# define vectorcall // For MSVC, better not to specify any call convention
# elif defined(__x86_64__)
# define vectorcall __attribute__((aligned(32)))
# define vectorcall
//# elif defined(__riscv) && (__riscv_xlen == 64)
//# define vectorcall __attribute__((aligned(16)))
# elif defined(__MINGW32__)

View File

@ -15,9 +15,13 @@
#include "mtl_renderer.h"
const int OC_MTL_INPUT_BUFFERS_COUNT = 3,
OC_MTL_TILE_SIZE = 16,
OC_MTL_MSAA_COUNT = 8;
enum
{
OC_MTL_INPUT_BUFFERS_COUNT = 3,
OC_MTL_TILE_SIZE = 16,
OC_MTL_MSAA_COUNT = 8,
OC_MTL_OFFSET_CURVE_SAMPLE_COUNT = 5,
};
typedef struct oc_mtl_canvas_backend
{
@ -511,8 +515,7 @@ void oc_mtl_render_stroke_quadratic(oc_mtl_canvas_backend* backend, oc_vec2* p)
}
else
{
const int CHECK_SAMPLE_COUNT = 5;
f32 checkSamples[CHECK_SAMPLE_COUNT] = { 1. / 6, 2. / 6, 3. / 6, 4. / 6, 5. / 6 };
f32 checkSamples[OC_MTL_OFFSET_CURVE_SAMPLE_COUNT] = { 1. / 6, 2. / 6, 3. / 6, 4. / 6, 5. / 6 };
f32 d2LowBound = oc_square(0.5 * width - tolerance);
f32 d2HighBound = oc_square(0.5 * width + tolerance);
@ -520,7 +523,7 @@ void oc_mtl_render_stroke_quadratic(oc_mtl_canvas_backend* backend, oc_vec2* p)
f32 maxOvershoot = 0;
f32 maxOvershootParameter = 0;
for(int i = 0; i < CHECK_SAMPLE_COUNT; i++)
for(int i = 0; i < OC_MTL_OFFSET_CURVE_SAMPLE_COUNT; i++)
{
f32 t = checkSamples[i];
@ -613,8 +616,7 @@ void oc_mtl_render_stroke_cubic(oc_mtl_canvas_backend* backend, oc_vec2* p)
}
else
{
const int CHECK_SAMPLE_COUNT = 5;
f32 checkSamples[CHECK_SAMPLE_COUNT] = { 1. / 6, 2. / 6, 3. / 6, 4. / 6, 5. / 6 };
f32 checkSamples[OC_MTL_OFFSET_CURVE_SAMPLE_COUNT] = { 1. / 6, 2. / 6, 3. / 6, 4. / 6, 5. / 6 };
f32 d2LowBound = oc_square(0.5 * width - tolerance);
f32 d2HighBound = oc_square(0.5 * width + tolerance);
@ -622,7 +624,7 @@ void oc_mtl_render_stroke_cubic(oc_mtl_canvas_backend* backend, oc_vec2* p)
f32 maxOvershoot = 0;
f32 maxOvershootParameter = 0;
for(int i = 0; i < CHECK_SAMPLE_COUNT; i++)
for(int i = 0; i < OC_MTL_OFFSET_CURVE_SAMPLE_COUNT; i++)
{
f32 t = checkSamples[i];

View File

@ -8,8 +8,84 @@
#include "hash.h"
#include "platform/platform.h"
#if OC_ARCH_X64
#include <immintrin.h>
//xxhash64, copy-pasted from https://github.com/demetri/scribbles/blob/master/hashing/hash_functions.c
// Thanks to Demetri Spanos
uint64_t xxh_64(const void* key, int len, uint64_t h)
{
// primes used in mul-rot updates
uint64_t p1 = 0x9e3779b185ebca87, p2 = 0xc2b2ae3d27d4eb4f,
p3 = 0x165667b19e3779f9, p4 = 0x85ebca77c2b2ae63, p5 = 0x27d4eb2f165667c5;
// inital 32-byte (4x8) wide hash state
uint64_t s[4] = { h + p1 + p2, h + p2, h, h - p1 };
// bulk work: process all 32 byte blocks
uint64_t* k32 = (uint64_t*)key;
for(int i = 0; i < (len / 32); i += 4)
{
uint64_t b[4] = { k32[i + 0], k32[i + 1], k32[i + 2], k32[i + 3] };
for(int j = 0; j < 4; j++)
b[j] = b[j] * p2 + s[j];
for(int j = 0; j < 4; j++)
s[j] = ((b[j] << 31) | (b[j] >> 33)) * p1;
}
// mix 32-byte state down to 8-byte state, initalize to value for short keys
uint64_t s64 = (s[2] + p5);
if(len > 32)
{
s64 = ((s[0] << 1) | (s[0] >> 63)) + ((s[1] << 7) | (s[1] >> 57)) + ((s[2] << 12) | (s[2] >> 52)) + ((s[3] << 18) | (s[3] >> 46));
for(int i = 0; i < 4; i++)
{
uint64_t ps = (((s[i] * p2) << 31) | ((s[i] * p2) >> 33)) * p1;
s64 = (s64 ^ ps) * p1 + p4;
}
}
s64 += len;
// up to 31 bytes remain, process 0-3 8 byte blocks
uint8_t* tail = (uint8_t*)(((char*)key) + (len / 32) * 32);
for(int i = 0; i < (len & 31) / 8; i++, tail += 8)
{
uint64_t b = (*((uint64_t*)tail)) * p2;
b = (((b << 31) | (b >> 33)) * p1) ^ s64;
s64 = ((b << 27) | (b >> 37)) * p1 + p4;
}
// up to 7 bytes remain, process 0-1 4 byte block
for(int i = 0; i < (len & 7) / 4; i++, tail += 4)
{
uint64_t b = s64 ^ (*(uint32_t*)tail) * p1;
s64 = ((b << 23) | (b >> 41)) * p2 + p3;
}
// up to 3 bytes remain, process 0-3 1 byte blocks
for(int i = 0; i < (len & 3); i++, tail++)
{
uint64_t b = s64 ^ (*tail) * p5;
s64 = ((b << 11) | (b >> 53)) * p1;
}
// finalization mix
s64 = (s64 ^ (s64 >> 33)) * p2;
s64 = (s64 ^ (s64 >> 29)) * p3;
return (s64 ^ (s64 >> 32));
}
u64 oc_hash_xx64_string_seed(oc_str8 string, u64 seed)
{
return (xxh_64(string.ptr, string.len, seed));
}
u64 oc_hash_xx64_string(oc_str8 string)
{
return (xxh_64(string.ptr, string.len, 0));
}
#if 0 //NOTE(martin): keep that here cause we could want to use them when aes is available, but we don't for now
#if OC_ARCH_X64
#include <immintrin.h>
u64 oc_hash_aes_u64(u64 x)
{
@ -127,79 +203,5 @@ u64 oc_hash_aes_string_seed(oc_str8 string, u64 seed)
u64 result = _mm_extract_epi64(hash, 0);
return (result);
}
#endif // OC_ARCH_X64
//xxhash64, copy-pasted from https://github.com/demetri/scribbles/blob/master/hashing/hash_functions.c
// Thanks to Demetri Spanos
uint64_t xxh_64(const void* key, int len, uint64_t h)
{
// primes used in mul-rot updates
uint64_t p1 = 0x9e3779b185ebca87, p2 = 0xc2b2ae3d27d4eb4f,
p3 = 0x165667b19e3779f9, p4 = 0x85ebca77c2b2ae63, p5 = 0x27d4eb2f165667c5;
// inital 32-byte (4x8) wide hash state
uint64_t s[4] = { h + p1 + p2, h + p2, h, h - p1 };
// bulk work: process all 32 byte blocks
uint64_t* k32 = (uint64_t*)key;
for(int i = 0; i < (len / 32); i += 4)
{
uint64_t b[4] = { k32[i + 0], k32[i + 1], k32[i + 2], k32[i + 3] };
for(int j = 0; j < 4; j++)
b[j] = b[j] * p2 + s[j];
for(int j = 0; j < 4; j++)
s[j] = ((b[j] << 31) | (b[j] >> 33)) * p1;
}
// mix 32-byte state down to 8-byte state, initalize to value for short keys
uint64_t s64 = (s[2] + p5);
if(len > 32)
{
s64 = ((s[0] << 1) | (s[0] >> 63)) + ((s[1] << 7) | (s[1] >> 57)) + ((s[2] << 12) | (s[2] >> 52)) + ((s[3] << 18) | (s[3] >> 46));
for(int i = 0; i < 4; i++)
{
uint64_t ps = (((s[i] * p2) << 31) | ((s[i] * p2) >> 33)) * p1;
s64 = (s64 ^ ps) * p1 + p4;
}
}
s64 += len;
// up to 31 bytes remain, process 0-3 8 byte blocks
uint8_t* tail = (uint8_t*)(((char*)key) + (len / 32) * 32);
for(int i = 0; i < (len & 31) / 8; i++, tail += 8)
{
uint64_t b = (*((uint64_t*)tail)) * p2;
b = (((b << 31) | (b >> 33)) * p1) ^ s64;
s64 = ((b << 27) | (b >> 37)) * p1 + p4;
}
// up to 7 bytes remain, process 0-1 4 byte block
for(int i = 0; i < (len & 7) / 4; i++, tail += 4)
{
uint64_t b = s64 ^ (*(uint32_t*)tail) * p1;
s64 = ((b << 23) | (b >> 41)) * p2 + p3;
}
// up to 3 bytes remain, process 0-3 1 byte blocks
for(int i = 0; i < (len & 3); i++, tail++)
{
uint64_t b = s64 ^ (*tail) * p5;
s64 = ((b << 11) | (b >> 53)) * p1;
}
// finalization mix
s64 = (s64 ^ (s64 >> 33)) * p2;
s64 = (s64 ^ (s64 >> 29)) * p3;
return (s64 ^ (s64 >> 32));
}
u64 oc_hash_xx64_string_seed(oc_str8 string, u64 seed)
{
return (xxh_64(string.ptr, string.len, seed));
}
u64 oc_hash_xx64_string(oc_str8 string)
{
return (xxh_64(string.ptr, string.len, 0));
}
#endif // OC_ARCH_X64
#endif // 0

View File

@ -12,17 +12,11 @@
#include "typedefs.h"
#ifdef __cplusplus
extern "C"
{
extern "C" {
#endif
ORCA_API u64 oc_hash_aes_u64(u64 x);
ORCA_API u64 oc_hash_aes_u64_x2(u64 x, u64 y);
ORCA_API u64 oc_hash_aes_string(oc_str8 string);
ORCA_API u64 oc_hash_aes_string_seed(oc_str8 string, u64 seed);
ORCA_API u64 oc_hash_xx64_string_seed(oc_str8 string, u64 seed);
ORCA_API u64 oc_hash_xx64_string(oc_str8 string);
ORCA_API u64 oc_hash_xx64_string_seed(oc_str8 string, u64 seed);
ORCA_API u64 oc_hash_xx64_string(oc_str8 string);
#ifdef __cplusplus
} // extern "C"