@shd101wyy/yo 0.0.2

package/vendor/mimalloc/src/prim/windows/readme.md

@@ -0,0 +1,17 @@
+## Primitives:
+
+- `prim.c` contains Windows primitives for OS allocation.
+
+## Event Tracing for Windows (ETW)
+
+- `etw.h` is generated from `etw.man` which contains the manifest for mimalloc events.
+  (100 is an allocation, 101 is for a free)
+
+- `etw-mimalloc.wprp` is a profile for the Windows Performance Recorder (WPR).
+  In an admin prompt, you can use:
+  ```
+  > wpr -start src\prim\windows\etw-mimalloc.wprp -filemode
+  > <my mimalloc program>
+  > wpr -stop test.etl
+  ``` 
+  and then open `test.etl` in the Windows Performance Analyzer (WPA).

package/vendor/mimalloc/src/random.c

@@ -0,0 +1,258 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2019-2021, Microsoft Research, Daan Leijen
+This is free software; you can redistribute it and/or modify it under the
+terms of the MIT license. A copy of the license can be found in the file
+"LICENSE" at the root of this distribution.
+-----------------------------------------------------------------------------*/
+#include "mimalloc.h"
+#include "mimalloc/internal.h"
+#include "mimalloc/prim.h"    // _mi_prim_random_buf
+#include <string.h>       // memset
+
+/* ----------------------------------------------------------------------------
+We use our own PRNG to keep predictable performance of random number generation
+and to avoid implementations that use a lock. We only use the OS provided
+random source to initialize the initial seeds. Since we do not need ultimate
+performance but we do rely on the security (for secret cookies in secure mode)
+we use a cryptographically secure generator (chacha20).
+-----------------------------------------------------------------------------*/
+
+#define MI_CHACHA_ROUNDS (20)   // perhaps use 12 for better performance?
+
+
+/* ----------------------------------------------------------------------------
+Chacha20 implementation as the original algorithm with a 64-bit nonce
+and counter: https://en.wikipedia.org/wiki/Salsa20
+The input matrix has sixteen 32-bit values:
+Position  0 to  3: constant key
+Position  4 to 11: the key
+Position 12 to 13: the counter.
+Position 14 to 15: the nonce.
+
+The implementation uses regular C code which compiles very well on modern compilers.
+(gcc x64 has no register spills, and clang 6+ uses SSE instructions)
+-----------------------------------------------------------------------------*/
+
+static inline uint32_t rotl(uint32_t x, uint32_t shift) {
+  return (x << shift) | (x >> (32 - shift));
+}
+
+static inline void qround(uint32_t x[16], size_t a, size_t b, size_t c, size_t d) {
+  x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 16);
+  x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 12);
+  x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 8);
+  x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 7);
+}
+
+static void chacha_block(mi_random_ctx_t* ctx)
+{
+  // scramble into `x`
+  uint32_t x[16];
+  for (size_t i = 0; i < 16; i++) {
+    x[i] = ctx->input[i];
+  }
+  for (size_t i = 0; i < MI_CHACHA_ROUNDS; i += 2) {
+    qround(x, 0, 4,  8, 12);
+    qround(x, 1, 5,  9, 13);
+    qround(x, 2, 6, 10, 14);
+    qround(x, 3, 7, 11, 15);
+    qround(x, 0, 5, 10, 15);
+    qround(x, 1, 6, 11, 12);
+    qround(x, 2, 7,  8, 13);
+    qround(x, 3, 4,  9, 14);
+  }
+
+  // add scrambled data to the initial state
+  for (size_t i = 0; i < 16; i++) {
+    ctx->output[i] = x[i] + ctx->input[i];
+  }
+  ctx->output_available = 16;
+
+  // increment the counter for the next round
+  ctx->input[12] += 1;
+  if (ctx->input[12] == 0) {
+    ctx->input[13] += 1;
+    if (ctx->input[13] == 0) {  // and keep increasing into the nonce
+      ctx->input[14] += 1;
+    }
+  }
+}
+
+static uint32_t chacha_next32(mi_random_ctx_t* ctx) {
+  if (ctx->output_available <= 0) {
+    chacha_block(ctx);
+    ctx->output_available = 16; // (assign again to suppress static analysis warning)
+  }
+  const uint32_t x = ctx->output[16 - ctx->output_available];
+  ctx->output[16 - ctx->output_available] = 0; // reset once the data is handed out
+  ctx->output_available--;
+  return x;
+}
+
+static inline uint32_t read32(const uint8_t* p, size_t idx32) {
+  const size_t i = 4*idx32;
+  return ((uint32_t)p[i+0] | (uint32_t)p[i+1] << 8 | (uint32_t)p[i+2] << 16 | (uint32_t)p[i+3] << 24);
+}
+
+static void chacha_init(mi_random_ctx_t* ctx, const uint8_t key[32], uint64_t nonce)
+{
+  // since we only use chacha for randomness (and not encryption) we
+  // do not _need_ to read 32-bit values as little endian but we do anyways
+  // just for being compatible :-)
+  memset(ctx, 0, sizeof(*ctx));
+  for (size_t i = 0; i < 4; i++) {
+    const uint8_t* sigma = (uint8_t*)"expand 32-byte k";
+    ctx->input[i] = read32(sigma,i);
+  }
+  for (size_t i = 0; i < 8; i++) {
+    ctx->input[i + 4] = read32(key,i);
+  }
+  ctx->input[12] = 0;
+  ctx->input[13] = 0;
+  ctx->input[14] = (uint32_t)nonce;
+  ctx->input[15] = (uint32_t)(nonce >> 32);
+}
+
+static void chacha_split(mi_random_ctx_t* ctx, uint64_t nonce, mi_random_ctx_t* ctx_new) {
+  memset(ctx_new, 0, sizeof(*ctx_new));
+  _mi_memcpy(ctx_new->input, ctx->input, sizeof(ctx_new->input));
+  ctx_new->input[12] = 0;
+  ctx_new->input[13] = 0;
+  ctx_new->input[14] = (uint32_t)nonce;
+  ctx_new->input[15] = (uint32_t)(nonce >> 32);
+  mi_assert_internal(ctx->input[14] != ctx_new->input[14] || ctx->input[15] != ctx_new->input[15]); // do not reuse nonces!
+  chacha_block(ctx_new);
+}
+
+
+/* ----------------------------------------------------------------------------
+Random interface
+-----------------------------------------------------------------------------*/
+
+#if MI_DEBUG>1
+static bool mi_random_is_initialized(mi_random_ctx_t* ctx) {
+  return (ctx != NULL && ctx->input[0] != 0);
+}
+#endif
+
+void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* ctx_new) {
+  mi_assert_internal(mi_random_is_initialized(ctx));
+  mi_assert_internal(ctx != ctx_new);
+  chacha_split(ctx, (uintptr_t)ctx_new /*nonce*/, ctx_new);
+}
+
+uintptr_t _mi_random_next(mi_random_ctx_t* ctx) {
+  mi_assert_internal(mi_random_is_initialized(ctx));
+  uintptr_t r;
+  do {
+    #if MI_INTPTR_SIZE <= 4
+    r = chacha_next32(ctx);
+    #elif MI_INTPTR_SIZE == 8
+    r = (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx));
+    #else
+    # error "define mi_random_next for this platform"
+    #endif
+  } while (r==0);
+  return r;
+}
+
+
+/* ----------------------------------------------------------------------------
+To initialize a fresh random context.
+If we cannot get good randomness, we fall back to weak randomness based on a timer and ASLR.
+-----------------------------------------------------------------------------*/
+
+uintptr_t _mi_os_random_weak(uintptr_t extra_seed) {
+  uintptr_t x = (uintptr_t)&_mi_os_random_weak ^ extra_seed; // ASLR makes the address random
+  x ^= _mi_prim_clock_now();  
+  // and do a few randomization steps
+  uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1;
+  for (uintptr_t i = 0; i < max || x==0; i++, x++) {
+    x = _mi_random_shuffle(x);
+  }
+  mi_assert_internal(x != 0);
+  return x;
+}
+
+static void mi_random_init_ex(mi_random_ctx_t* ctx, bool use_weak) {
+  uint8_t key[32];
+  if (use_weak || !_mi_prim_random_buf(key, sizeof(key))) {
+    // if we fail to get random data from the OS, we fall back to a
+    // weak random source based on the current time
+    #if !defined(__wasi__)
+    if (!use_weak) { _mi_warning_message("unable to use secure randomness\n"); }
+    #endif
+    uintptr_t x = _mi_os_random_weak(0);
+    for (size_t i = 0; i < 8; i++, x++) {  // key is eight 32-bit words.
+      x = _mi_random_shuffle(x);
+      ((uint32_t*)key)[i] = (uint32_t)x;
+    }
+    ctx->weak = true;
+  }
+  else {
+    ctx->weak = false;
+  }
+  chacha_init(ctx, key, (uintptr_t)ctx /*nonce*/ );
+}
+
+void _mi_random_init(mi_random_ctx_t* ctx) {
+  mi_random_init_ex(ctx, false);
+}
+
+void _mi_random_init_weak(mi_random_ctx_t * ctx) {
+  mi_random_init_ex(ctx, true);
+}
+
+void _mi_random_reinit_if_weak(mi_random_ctx_t * ctx) {
+  if (ctx->weak) {
+    _mi_random_init(ctx);
+  }
+}
+
+/* --------------------------------------------------------
+test vectors from <https://tools.ietf.org/html/rfc8439>
+----------------------------------------------------------- */
+/*
+static bool array_equals(uint32_t* x, uint32_t* y, size_t n) {
+  for (size_t i = 0; i < n; i++) {
+    if (x[i] != y[i]) return false;
+  }
+  return true;
+}
+static void chacha_test(void)
+{
+  uint32_t x[4] = { 0x11111111, 0x01020304, 0x9b8d6f43, 0x01234567 };
+  uint32_t x_out[4] = { 0xea2a92f4, 0xcb1cf8ce, 0x4581472e, 0x5881c4bb };
+  qround(x, 0, 1, 2, 3);
+  mi_assert_internal(array_equals(x, x_out, 4));
+
+  uint32_t y[16] = {
+       0x879531e0,  0xc5ecf37d,  0x516461b1,  0xc9a62f8a,
+       0x44c20ef3,  0x3390af7f,  0xd9fc690b,  0x2a5f714c,
+       0x53372767,  0xb00a5631,  0x974c541a,  0x359e9963,
+       0x5c971061,  0x3d631689,  0x2098d9d6,  0x91dbd320 };
+  uint32_t y_out[16] = {
+       0x879531e0,  0xc5ecf37d,  0xbdb886dc,  0xc9a62f8a,
+       0x44c20ef3,  0x3390af7f,  0xd9fc690b,  0xcfacafd2,
+       0xe46bea80,  0xb00a5631,  0x974c541a,  0x359e9963,
+       0x5c971061,  0xccc07c79,  0x2098d9d6,  0x91dbd320 };
+  qround(y, 2, 7, 8, 13);
+  mi_assert_internal(array_equals(y, y_out, 16));
+
+  mi_random_ctx_t r = {
+    { 0x61707865, 0x3320646e, 0x79622d32, 0x6b206574,
+      0x03020100, 0x07060504, 0x0b0a0908, 0x0f0e0d0c,
+      0x13121110, 0x17161514, 0x1b1a1918, 0x1f1e1d1c,
+      0x00000001, 0x09000000, 0x4a000000, 0x00000000 },
+    {0},
+    0
+  };
+  uint32_t r_out[16] = {
+       0xe4e7f110, 0x15593bd1, 0x1fdd0f50, 0xc47120a3,
+       0xc7f4d1c7, 0x0368c033, 0x9aaa2204, 0x4e6cd4c3,
+       0x466482d2, 0x09aa9f07, 0x05d7c214, 0xa2028bd9,
+       0xd19c12b5, 0xb94e16de, 0xe883d0cb, 0x4e3c50a2 };
+  chacha_block(&r);
+  mi_assert_internal(array_equals(r.output, r_out, 16));
+}
+*/

package/vendor/mimalloc/src/segment-map.c

@@ -0,0 +1,142 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2019-2023, Microsoft Research, Daan Leijen
+This is free software; you can redistribute it and/or modify it under the
+terms of the MIT license. A copy of the license can be found in the file
+"LICENSE" at the root of this distribution.
+-----------------------------------------------------------------------------*/
+
+/* -----------------------------------------------------------
+  The following functions are to reliably find the segment or
+  block that encompasses any pointer p (or NULL if it is not
+  in any of our segments).
+  We maintain a bitmap of all memory with 1 bit per MI_SEGMENT_SIZE (64MiB)
+  set to 1 if it contains the segment meta data.
+----------------------------------------------------------- */
+#include "mimalloc.h"
+#include "mimalloc/internal.h"
+#include "mimalloc/atomic.h"
+
+// Reduce total address space to reduce .bss  (due to the `mi_segment_map`)
+#if (MI_INTPTR_SIZE > 4) && MI_TRACK_ASAN
+#define MI_SEGMENT_MAP_MAX_ADDRESS    (128*1024ULL*MI_GiB)  // 128 TiB  (see issue #881)
+#elif (MI_INTPTR_SIZE > 4)
+#define MI_SEGMENT_MAP_MAX_ADDRESS    (48*1024ULL*MI_GiB)   // 48 TiB
+#else
+#define MI_SEGMENT_MAP_MAX_ADDRESS    (UINT32_MAX)
+#endif
+
+#define MI_SEGMENT_MAP_PART_SIZE      (MI_INTPTR_SIZE*MI_KiB - 128)      // 128 > sizeof(mi_memid_t) ! 
+#define MI_SEGMENT_MAP_PART_BITS      (8*MI_SEGMENT_MAP_PART_SIZE)
+#define MI_SEGMENT_MAP_PART_ENTRIES   (MI_SEGMENT_MAP_PART_SIZE / MI_INTPTR_SIZE)
+#define MI_SEGMENT_MAP_PART_BIT_SPAN  (MI_SEGMENT_ALIGN)                 // memory area covered by 1 bit
+
+#if (MI_SEGMENT_MAP_PART_BITS < (MI_SEGMENT_MAP_MAX_ADDRESS / MI_SEGMENT_MAP_PART_BIT_SPAN)) // prevent overflow on 32-bit (issue #1017)
+#define MI_SEGMENT_MAP_PART_SPAN      (MI_SEGMENT_MAP_PART_BITS * MI_SEGMENT_MAP_PART_BIT_SPAN)
+#else
+#define MI_SEGMENT_MAP_PART_SPAN      MI_SEGMENT_MAP_MAX_ADDRESS
+#endif
+
+#define MI_SEGMENT_MAP_MAX_PARTS      ((MI_SEGMENT_MAP_MAX_ADDRESS / MI_SEGMENT_MAP_PART_SPAN) + 1)
+
+// A part of the segment map.
+typedef struct mi_segmap_part_s {
+  mi_memid_t memid;
+  _Atomic(uintptr_t) map[MI_SEGMENT_MAP_PART_ENTRIES];
+} mi_segmap_part_t;
+
+// Allocate parts on-demand to reduce .bss footprint
+static _Atomic(mi_segmap_part_t*) mi_segment_map[MI_SEGMENT_MAP_MAX_PARTS]; // = { NULL, .. }
+
+static mi_segmap_part_t* mi_segment_map_index_of(const mi_segment_t* segment, bool create_on_demand, size_t* idx, size_t* bitidx) {
+  // note: segment can be invalid or NULL.
+  mi_assert_internal(_mi_ptr_segment(segment + 1) == segment); // is it aligned on MI_SEGMENT_SIZE?
+  *idx = 0;
+  *bitidx = 0;  
+  if ((uintptr_t)segment >= MI_SEGMENT_MAP_MAX_ADDRESS) return NULL;
+  const uintptr_t segindex = ((uintptr_t)segment) / MI_SEGMENT_MAP_PART_SPAN;
+  if (segindex >= MI_SEGMENT_MAP_MAX_PARTS) return NULL;
+  mi_segmap_part_t* part = mi_atomic_load_ptr_relaxed(mi_segmap_part_t, &mi_segment_map[segindex]);
+
+  // allocate on demand to reduce .bss footprint
+  if mi_unlikely(part == NULL) {
+    if (!create_on_demand) return NULL;
+    mi_memid_t memid;
+    part = (mi_segmap_part_t*)_mi_os_zalloc(sizeof(mi_segmap_part_t), &memid);
+    if (part == NULL) return NULL;
+    part->memid = memid;
+    mi_segmap_part_t* expected = NULL;
+    if (!mi_atomic_cas_ptr_strong_release(mi_segmap_part_t, &mi_segment_map[segindex], &expected, part)) {
+      _mi_os_free(part, sizeof(mi_segmap_part_t), memid);
+      part = expected;
+      if (part == NULL) return NULL;
+    }
+  }
+  mi_assert(part != NULL);
+  const uintptr_t offset = ((uintptr_t)segment) % MI_SEGMENT_MAP_PART_SPAN;
+  const uintptr_t bitofs = offset / MI_SEGMENT_MAP_PART_BIT_SPAN;
+  *idx = bitofs / MI_INTPTR_BITS;
+  *bitidx = bitofs % MI_INTPTR_BITS;
+  return part;
+}
+
+void _mi_segment_map_allocated_at(const mi_segment_t* segment) {
+  if (segment->memid.memkind == MI_MEM_ARENA) return; // we lookup segments first in the arena's and don't need the segment map
+  size_t index;
+  size_t bitidx;
+  mi_segmap_part_t* part = mi_segment_map_index_of(segment, true /* alloc map if needed */, &index, &bitidx);
+  if (part == NULL) return; // outside our address range..
+  uintptr_t mask = mi_atomic_load_relaxed(&part->map[index]);
+  uintptr_t newmask;
+  do {
+    newmask = (mask | ((uintptr_t)1 << bitidx));
+  } while (!mi_atomic_cas_weak_release(&part->map[index], &mask, newmask));
+}
+
+void _mi_segment_map_freed_at(const mi_segment_t* segment) {
+  if (segment->memid.memkind == MI_MEM_ARENA) return;
+  size_t index;
+  size_t bitidx;
+  mi_segmap_part_t* part = mi_segment_map_index_of(segment, false /* don't alloc if not present */, &index, &bitidx);
+  if (part == NULL) return; // outside our address range..
+  uintptr_t mask = mi_atomic_load_relaxed(&part->map[index]);
+  uintptr_t newmask;
+  do {
+    newmask = (mask & ~((uintptr_t)1 << bitidx));
+  } while (!mi_atomic_cas_weak_release(&part->map[index], &mask, newmask));
+}
+
+// Determine the segment belonging to a pointer or NULL if it is not in a valid segment.
+static mi_segment_t* _mi_segment_of(const void* p) {
+  if (p == NULL) return NULL;
+  mi_segment_t* segment = _mi_ptr_segment(p);  // segment can be NULL  
+  size_t index;
+  size_t bitidx;
+  mi_segmap_part_t* part = mi_segment_map_index_of(segment, false /* dont alloc if not present */, &index, &bitidx);
+  if (part == NULL) return NULL;  
+  const uintptr_t mask = mi_atomic_load_relaxed(&part->map[index]);
+  if mi_likely((mask & ((uintptr_t)1 << bitidx)) != 0) {
+    bool cookie_ok = (_mi_ptr_cookie(segment) == segment->cookie);
+    mi_assert_internal(cookie_ok); MI_UNUSED(cookie_ok);
+    return segment; // yes, allocated by us
+  }
+  return NULL;
+}
+
+// Is this a valid pointer in our heap?
+static bool mi_is_valid_pointer(const void* p) {
+  // first check if it is in an arena, then check if it is OS allocated
+  return (_mi_arena_contains(p) || _mi_segment_of(p) != NULL);
+}
+
+mi_decl_nodiscard mi_decl_export bool mi_is_in_heap_region(const void* p) mi_attr_noexcept {
+  return mi_is_valid_pointer(p);
+}
+
+void _mi_segment_map_unsafe_destroy(void) {
+  for (size_t i = 0; i < MI_SEGMENT_MAP_MAX_PARTS; i++) {
+    mi_segmap_part_t* part = mi_atomic_exchange_ptr_relaxed(mi_segmap_part_t, &mi_segment_map[i], NULL);
+    if (part != NULL) {
+      _mi_os_free(part, sizeof(mi_segmap_part_t), part->memid);
+    }
+  }
+}