internal static parse_string_helper find_bs_bits_and_quote_bits(uint8_t *src, uint8_t *dst)
{
if (Avx2.IsSupported)
{
// this can read up to 31 bytes beyond the buffer size, but we require
// SIMDJSON_PADDING of padding
var v = Avx.LoadVector256(src);
// store to dest unconditionally - we can overwrite the bits we don't like
// later
Avx.Store((dst), v);
var quote_mask = Avx2.CompareEqual(v, Vector256.Create((uint8_t)'"'));
return(new parse_string_helper
{
bs_bits = (uint32_t)Avx2.MoveMask(Avx2.CompareEqual(v, Vector256.Create((uint8_t)'\\'))), // bs_bits
quote_bits = (uint32_t)Avx2.MoveMask(quote_mask) // quote_bits
});
}
else // SSE42
{
// this can read up to 31 bytes beyond the buffer size, but we require
// SIMDJSON_PADDING of padding
var v = Sse2.LoadVector128((src));
// store to dest unconditionally - we can overwrite the bits we don't like
// later
Sse2.Store((dst), v);
var quote_mask = Sse2.CompareEqual(v, Vector128.Create((uint8_t)'"'));
return(new parse_string_helper
{
bs_bits = (uint32_t)Sse2.MoveMask(Sse2.CompareEqual(v,
Vector128.Create((uint8_t)'\\'))), // bs_bits
quote_bits = (uint32_t)Sse2.MoveMask(quote_mask) // quote_bits
});
}
}
public static size_t codepoint_to_utf8(uint32_t cp, uint8_t *c)
{
if (cp <= 0x7F)
{
c[0] = (uint8_t)cp;
return(1); // ascii
}
else if (cp <= 0x7FF)
{
c[0] = (uint8_t)((cp >> 6) + 192);
c[1] = (uint8_t)((cp & 63) + 128);
return(2); // universal plane
// Surrogates are treated elsewhere...
//} //else if (0xd800 <= cp && cp <= 0xdfff) {
// return 0; // surrogates // could put assert here
}
else if (cp <= 0xFFFF)
{
c[0] = (uint8_t)((cp >> 12) + 224);
c[1] = (uint8_t)(((cp >> 6) & 63) + 128);
c[2] = (uint8_t)((cp & 63) + 128);
return(3);
}
else if (cp <= 0x10FFFF)
{ // if you know you have a valid code point, this is not needed
c[0] = (uint8_t)((cp >> 18) + 240);
c[1] = (uint8_t)(((cp >> 12) & 63) + 128);
c[2] = (uint8_t)(((cp >> 6) & 63) + 128);
c[3] = (uint8_t)((cp & 63) + 128);
return(4);
}
// will return 0 when the code point was too large.
return(0); // bad r
}
internal static JsonParseError JsonParse(uint8_t *jsonData, size_t length, ParsedJson pj, bool reallocIfNeeded = true)
{
if (pj.bytecapacity < length)
{
return(JsonParseError.Capacity);
}
bool reallocated = false;
if (reallocIfNeeded)
{
// realloc is needed if the end of the memory crosses a page
if ((size_t)(jsonData + length - 1) % (size_t)pagesize < SIMDJSON_PADDING)
{
uint8_t *tmpbuf = jsonData;
jsonData = (uint8_t *)allocate_padded_buffer(length);
if (jsonData == null)
{
return(JsonParseError.Memalloc);
}
memcpy(jsonData, tmpbuf, length);
reallocated = true;
}
}
var result = JsonParseError.Success;
if (stage1_find_marks.find_structural_bits(jsonData, length, pj))
{
result = stage2_build_tape.unified_machine(jsonData, length, pj);
}
if (reallocated)
aligned_free(jsonData); }
public static uint32_t hex_to_u32_nocheck(uint8_t *src)
{
uint8_t v1 = src[0];
uint8_t v2 = src[1];
uint8_t v3 = src[2];
uint8_t v4 = src[3];
return((uint32_t)(digittoval[v1] << 12 | digittoval[v2] << 8 | digittoval[v3] << 4 | digittoval[v4]));
}
internal static uint32_t hex_to_u32_nocheck(uint8_t *src) // strictly speaking, static inline is a C-ism
{
uint32_t v1 = digittoval32[630 + src[0]];
uint32_t v2 = digittoval32[420 + src[1]];
uint32_t v3 = digittoval32[210 + src[2]];
uint32_t v4 = digittoval32[0 + src[3]];
return(v1 | v2 | v3 | v4);
}
public static bool parse_string(uint8_t* buf, size_t len, ParsedJson pj, uint32_t depth, uint32_t offset)
{
if (Avx2.IsSupported)
return parse_string_avx2(buf, len, pj, depth, offset);
//if (Sse41.IsSupported)
// return parse_string_sse41(buf, len, pj, depth, offset);
ThrowHelper.ThrowPNSE();
return false;
}
// if needed, allocate memory so that the object is able to process JSON
// documents having up to len bytes and maxdepth "depth"
public bool AllocateCapacity(size_t len, size_t maxdepth = DEFAULTMAXDEPTH)
{
if ((maxdepth == 0) || (len == 0))
{
return(false);
}
if (len > SIMDJSON_MAXSIZE_BYTES)
{
return(false);
}
if ((len <= bytecapacity) && (depthcapacity < maxdepth))
{
return(true);
}
Deallocate();
isvalid = false;
bytecapacity = 0; // will only set it to len after allocations are a success
n_structural_indexes = 0;
uint32_t max_structures = (uint32_t)(ROUNDUP_N(len, 64) + 2 + 7);
structural_indexes = allocate <uint32_t>(max_structures);
// a pathological input like "[[[[..." would generate len tape elements, so need a capacity of len + 1
size_t localtapecapacity = ROUNDUP_N(len + 1, 64);
// a document with only zero-length strings... could have len/3 string
// and we would need len/3 * 5 bytes on the string buffer
size_t localstringcapacity = ROUNDUP_N(5 * len / 3 + 32, 64);
string_buf = allocate <uint8_t>(localstringcapacity);
tape = allocate <uint64_t>(localtapecapacity);
containing_scope_offset = allocate <uint32_t>(maxdepth);
ret_address = allocate <char1>(maxdepth);
if ((string_buf == null) || (tape == null) ||
(containing_scope_offset == null) || (ret_address == null) || (structural_indexes == null))
{
delete(ret_address);
delete(containing_scope_offset);
delete(tape);
delete(string_buf);
delete(structural_indexes);
return(false);
}
/*
* // We do not need to initialize this content for parsing, though we could
* // need to initialize it for safety.
* memset(string_buf, 0 , localstringcapacity);
* memset(structural_indexes, 0, max_structures * sizeof(uint32_t));
* memset(tape, 0, localtapecapacity * sizeof(uint64_t));
*/
bytecapacity = len;
depthcapacity = maxdepth;
tapecapacity = localtapecapacity;
stringcapacity = localstringcapacity;
return(true);
}
public static uint32_t hex_to_u32_nocheck(uint8_t *src)
{
// all these will sign-extend the chars looked up, placing 1-bits into the high 28 bits of every
// invalid value. After the shifts, this will *still* result in the outcome that the high 16 bits of any
// value with any invalid char will be all 1's. We check for this in the caller.
uint8_t v1 = (uint8_t)digittoval[src[0]];
uint8_t v2 = (uint8_t)digittoval[src[1]];
uint8_t v3 = (uint8_t)digittoval[src[2]];
uint8_t v4 = (uint8_t)digittoval[src[3]];
return((uint32_t)(v1 << 12 | v2 << 8 | v3 << 4 | v4));
}
internal static bool is_valid_null_atom(uint8_t *loc)
{
const uint64_t nv = 2314885532098524526; //* (uint64_t*)"null ";
const uint64_t mask4 = 0x00000000ffffffff;
uint32_t error = 0;
uint64_t locval; // we want to avoid unaligned 64-bit loads (undefined in C/C++)
memcpy(&locval, loc, sizeof(uint64_t));
error = (uint32_t)((locval & mask4) ^ nv);
error |= is_not_structural_or_whitespace(loc[4]);
return(error == 0);
}
internal static bool is_valid_false_atom(uint8_t *loc)
{
const uint64_t fv = 2314885828568703334; //* (uint64_t*)"false ";
const uint64_t mask5 = 0x000000ffffffffff;
uint32_t error = 0;
uint64_t locval; // we want to avoid unaligned 64-bit loads (undefined in C/C++)
memcpy(&locval, loc, sizeof(uint64_t));
error = (uint32_t)((locval & mask5) ^ fv);
error |= is_not_structural_or_whitespace(loc[5]);
return(error == 0);
}
internal static bool is_valid_true_atom(uint8_t *loc)
{
uint64_t tv = 2314885531981673076; //* (uint64_t*)"true ";
uint64_t mask4 = 0x00000000ffffffff;
uint32_t error = 0;
uint64_t locval; // we want to avoid unaligned 64-bit loads (undefined in C/C++)
memcpy(&locval, loc, sizeof(uint64_t));
error = (uint32_t)((locval & mask4) ^ tv);
error |= is_not_structural_or_whitespace(loc[4]);
return(error == 0);
}
internal static bool is_valid_null_atom(uint8_t *loc)
{
uint64_t nv = 2314885532098524526; //* (uint64_t*)"null ";
uint64_t mask4 = 0x00000000ffffffff;
uint32_t error = 0;
uint64_t locval; // we want to avoid unaligned 64-bit loads (undefined in C/C++)
// this can read up to 7 bytes beyond the buffer size, but we require
// SIMDJSON_PADDING of padding
memcpy(&locval, loc, sizeof(uint64_t));
error = (uint32_t)((locval & mask4) ^ nv);
error |= is_not_structural_or_whitespace(loc[4]);
return(error == 0);
}
public static ParsedJson build_parsed_json(uint8_t *buf, size_t len, bool reallocifneeded = true)
{
ParsedJson pj = new ParsedJson();
bool ok = pj.allocateCapacity(len);
if (ok)
{
ok = json_parse(buf, len, &pj, reallocifneeded);
}
else
{
throw new InvalidOperationException("failure during memory allocation");
}
return(pj);
}
public static bool json_parse(uint8_t *buf, size_t len, ParsedJson *pj, bool reallocifneeded = true)
{
if (pj->bytecapacity < len)
{
Debug.WriteLine("Your ParsedJson cannot support documents that big: " + len);
return(false);
}
bool reallocated = false;
if (reallocifneeded)
{
// realloc is needed if the end of the memory crosses a page
long pagesize = System.Environment.SystemPageSize;
if (((size_t)(buf + len - 1) % (size_t)pagesize) < SIMDJSON_PADDING)
{
uint8_t *tmpbuf = buf;
buf = (uint8_t *)Utils.allocate_padded_buffer(len);
if (buf == null)
{
return(false);
}
memcpy((void *)buf, tmpbuf, len);
reallocated = true;
}
}
bool isok = stage1_find_marks.find_structural_bits(buf, len, pj);
if (isok)
{
isok = stage2_build_tape.unified_machine(buf, len, pj);
}
else
{
if (reallocated)
{
Utils.free((void *)buf);
}
return(false);
}
if (reallocated)
{
Utils.free((void *)buf);
}
return(isok);
}
internal static simd_input fill_input(uint8_t *ptr)
{
if (Avx2.IsSupported)
{
simd_input @in = new simd_input();
@in.lo = Avx.LoadVector256((ptr + 0));
@in.hi = Avx.LoadVector256((ptr + 32));
return(@in);
}
else
{
simd_input @in = new simd_input();
@in.v0 = Sse2.LoadVector128((ptr + 0));
@in.v1 = Sse2.LoadVector128((ptr + 16));
@in.v2 = Sse2.LoadVector128((ptr + 32));
@in.v3 = Sse2.LoadVector128((ptr + 48));
return(@in);
}
}
// if needed, allocate memory so that the object is able to process JSON
// documents having up to len butes and maxdepth "depth"
public bool AllocateCapacity(size_t len, size_t maxdepth = DEFAULTMAXDEPTH)
{
if ((maxdepth == 0) || (len == 0))
{
Debug.WriteLine("capacities must be non-zero ");
return(false);
}
if (len > 0)
{
if ((len <= bytecapacity) && (depthcapacity < maxdepth))
{
return(true);
}
Deallocate();
}
isvalid = false;
bytecapacity = 0; // will only set it to len after allocations are a success
n_structural_indexes = 0;
uint32_t max_structures = (uint32_t)ROUNDUP_N(len, 64) + 2 + 7;
structural_indexes = Utils.allocate <uint32_t>(max_structures);
size_t localtapecapacity = ROUNDUP_N(len, 64);
size_t localstringcapacity = ROUNDUP_N(len, 64);
string_buf = Utils.allocate <uint8_t>(localstringcapacity);
tape = Utils.allocate <uint64_t>(localtapecapacity);
containing_scope_offset = Utils.allocate <uint32_t>(maxdepth);
ret_address = Utils.allocate <bytechar>(maxdepth);
if ((string_buf == null) || (tape == null) ||
(containing_scope_offset == null) || (ret_address == null) || (structural_indexes == null))
{
Deallocate();
return(false);
}
bytecapacity = len;
depthcapacity = maxdepth;
tapecapacity = localtapecapacity;
stringcapacity = localstringcapacity;
return(true);
}
internal static bool JsonParse(uint8_t *jsonData, size_t length, ParsedJson pj, bool reallocIfNeeded = true)
{
if (pj.bytecapacity < length)
{
throw new InvalidOperationException("Your ParsedJson cannot support documents that big: " + length);
}
bool reallocated = false;
if (reallocIfNeeded)
{
// realloc is needed if the end of the memory crosses a page
if ((size_t)(jsonData + length - 1) % (size_t)pagesize < SIMDJSON_PADDING)
{
uint8_t *tmpbuf = jsonData;
jsonData = (uint8_t *)allocate_padded_buffer(length);
if (jsonData == null)
{
return(false);
}
memcpy(jsonData, tmpbuf, length);
reallocated = true;
}
}
bool isok = stage1_find_marks.find_structural_bits(jsonData, length, pj);
if (isok)
{
isok = stage2_build_tape.unified_machine(jsonData, length, pj);
}
else
{
if (reallocated)
{
free(jsonData);
}
return(false);
}
if (reallocated)
free(jsonData); }
internal static bool is_valid_false_atom(uint8_t *loc)
{
// We have to use an integer constant because the space in the cast
// below would lead to values illegally being qualified
// uint64_t fv = *reinterpret_cast<const uint64_t *>("false ");
// using this constant (that is the same false) but nulls out the
// unused bits solves that
uint64_t fv = 0x00000065736c6166; // takes into account endianness
uint64_t mask5 = 0x000000ffffffffff;
// we can't use the 32 bit value for checking for errors otherwise
// the last character of false (it being 5 byte long!) would be
// ignored
uint64_t error = 0;
uint64_t locval; // we want to avoid unaligned 64-bit loads (undefined in C/C++)
// this can read up to 7 bytes beyond the buffer size, but we require
// SIMDJSON_PADDING of padding
memcpy(&locval, loc, sizeof(uint64_t));
error = (locval & mask5) ^ fv;
error |= is_not_structural_or_whitespace(loc[5]);
return(error == 0);
}
private void Deallocate()
{
isvalid = false;
bytecapacity = 0;
depthcapacity = 0;
tapecapacity = 0;
stringcapacity = 0;
if (ret_address != null)
{
delete(ret_address);
ret_address = null;
}
if (containing_scope_offset != null)
{
delete(containing_scope_offset);
containing_scope_offset = null;
}
if (tape != null)
{
delete(tape);
tape = null;
}
if (string_buf != null)
{
delete(string_buf);
string_buf = null;
}
if (structural_indexes != null)
{
delete(structural_indexes);
structural_indexes = null;
}
}
/*************************************
*
* Decryption helpers
*
*************************************/
#if false
void galaxian_state::decode_mooncrst(int length, uint8_t *dest)
{
uint8_t *rom = memregion("maincpu")->base();
int offs;
for (offs = 0; offs < length; offs++)
{
uint8_t data = rom[offs];
uint8_t res = data;
if (BIT(data, 1))
{
res ^= 0x40;
}
if (BIT(data, 5))
{
res ^= 0x04;
}
if ((offs & 1) == 0)
{
res = BITSWAP8(res, 7, 2, 5, 4, 3, 6, 1, 0);
}
dest[offs] = res;
}
}
public static extern void RenderStatic(InputHeader *inputHeader, InputDataArrays *inputData, int32_t visualizeLightCount, uint8_t *framebuffer_r, uint8_t *framebuffer_g, uint8_t *framebuffer_b) /*x75*/;
// take input from buf and remove useless whitespace, input and output can be
// the same, result is null terminated, return the string length (minus the null termination)
public static size_t Minify(uint8_t *buf, size_t len, uint8_t * @out)
{
if (!Avx2.IsSupported)
{
throw new NotSupportedException("AVX2 is required form SimdJson");
}
//C#: load const vectors once (there is no `const _m256` in C#)
Vector256 <byte> lut_cntrl = s_lut_cntrl;
Vector256 <byte> low_nibble_mask = s_low_nibble_mask;
Vector256 <byte> high_nibble_mask = s_high_nibble_mask;
fixed(byte *mask128_epi8 = s_mask128_epi8)
{
// Useful constant masks
const uint64_t even_bits = 0x5555555555555555UL;
const uint64_t odd_bits = ~even_bits;
uint8_t * initout = @out;
uint64_t prev_iter_ends_odd_backslash =
0UL; // either 0 or 1, but a 64-bit value
uint64_t prev_iter_inside_quote = 0UL; // either all zeros or all ones
size_t idx = 0;
if (len >= 64)
{
size_t avxlen = len - 63;
for (; idx < avxlen; idx += 64)
{
Vector256 <byte> input_lo = Avx.LoadVector256((buf + idx + 0));
Vector256 <byte> input_hi = Avx.LoadVector256((buf + idx + 32));
uint64_t bs_bits = cmp_mask_against_input_mini(input_lo, input_hi,
Vector256.Create((byte)'\\'));
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
bool iter_ends_odd_backslash = add_overflow(
bs_bits, odd_starts, &odd_carries);
odd_carries |= prev_iter_ends_odd_backslash;
prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1UL : 0x0UL;
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
uint64_t quote_bits = cmp_mask_against_input_mini(input_lo, input_hi,
Vector256.Create((byte)'"'));
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = Sse2.X64.ConvertToUInt64(Pclmulqdq.CarrylessMultiply(
Vector128.Create(quote_bits, 0UL).AsUInt64(), Vector128.Create((byte)0xFF).AsUInt64(), 0));
quote_mask ^= prev_iter_inside_quote;
prev_iter_inside_quote =
(uint64_t)((int64_t)quote_mask >>
63); // might be undefined behavior, should be fully defined in C++20, ok according to John Regher from Utah University
Vector256 <byte> whitespace_shufti_mask = Vector256.Create((byte)0x18);
Vector256 <byte> v_lo = Avx2.And(
Avx2.Shuffle(low_nibble_mask, input_lo),
Avx2.Shuffle(high_nibble_mask,
Avx2.And(Avx2.ShiftRightLogical(input_lo.AsUInt32(), 4).AsByte(),
Vector256.Create((byte)0x7f))));
Vector256 <byte> v_hi = Avx2.And(
Avx2.Shuffle(low_nibble_mask, input_hi),
Avx2.Shuffle(high_nibble_mask,
Avx2.And(Avx2.ShiftRightLogical(input_hi.AsUInt32(), 4).AsByte(),
Vector256.Create((byte)0x7f))));
Vector256 <byte> tmp_ws_lo = Avx2.CompareEqual(
Avx2.And(v_lo, whitespace_shufti_mask), Vector256.Create((byte)0));
Vector256 <byte> tmp_ws_hi = Avx2.CompareEqual(
Avx2.And(v_hi, whitespace_shufti_mask), Vector256.Create((byte)0));
uint64_t ws_res_0 = (uint32_t)Avx2.MoveMask(tmp_ws_lo);
uint64_t ws_res_1 = (uint64_t)Avx2.MoveMask(tmp_ws_hi);
uint64_t whitespace = ~(ws_res_0 | (ws_res_1 << 32));
whitespace &= ~quote_mask;
int mask1 = (int)(whitespace & 0xFFFF);
int mask2 = (int)((whitespace >> 16) & 0xFFFF);
int mask3 = (int)((whitespace >> 32) & 0xFFFF);
int mask4 = (int)((whitespace >> 48) & 0xFFFF);
int pop1 = hamming((~whitespace) & 0xFFFF);
int pop2 = hamming((~whitespace) & (ulong)(0xFFFFFFFF));
int pop3 = hamming((~whitespace) & (ulong)(0xFFFFFFFFFFFF));
int pop4 = hamming((~whitespace));
var vmask1 =
_mm256_loadu2_m128i((ulong *)mask128_epi8 + (mask2 & 0x7FFF) * 2,
(ulong *)mask128_epi8 + (mask1 & 0x7FFF) * 2);
var vmask2 =
_mm256_loadu2_m128i((ulong *)mask128_epi8 + (mask4 & 0x7FFF) * 2,
(ulong *)mask128_epi8 + (mask3 & 0x7FFF) * 2);
var result1 = Avx2.Shuffle(input_lo, vmask1.AsByte());
var result2 = Avx2.Shuffle(input_hi, vmask2.AsByte());
_mm256_storeu2_m128i((@out + pop1), @out, result1);
_mm256_storeu2_m128i((@out + pop3), (@out + pop2),
result2);
@out += pop4;
}
}
// we finish off the job... copying and pasting the code is not ideal here,
// but it gets the job done.
if (idx < len)
{
uint8_t *buffer = stackalloc uint8_t[64];
memset(buffer, 0, 64);
memcpy(buffer, buf + idx, len - idx);
var input_lo = Avx.LoadVector256((buffer));
var input_hi = Avx.LoadVector256((buffer + 32));
uint64_t bs_bits =
cmp_mask_against_input_mini(input_lo, input_hi, Vector256.Create((byte)'\\'));
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
//bool iter_ends_odd_backslash =
add_overflow(bs_bits, odd_starts, &odd_carries);
odd_carries |= prev_iter_ends_odd_backslash;
//prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1ULL : 0x0ULL; // we never use it
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
uint64_t quote_bits =
cmp_mask_against_input_mini(input_lo, input_hi, Vector256.Create((byte)'"'));
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = Sse2.X64.ConvertToUInt64(Pclmulqdq.CarrylessMultiply(
Vector128.Create(quote_bits, 0UL), Vector128.Create((byte)0xFF).AsUInt64(), 0));
quote_mask ^= prev_iter_inside_quote;
// prev_iter_inside_quote = (uint64_t)((int64_t)quote_mask >> 63);// we don't need this anymore
Vector256 <byte> mask_20 = Vector256.Create((byte)0x20); // c==32
Vector256 <byte> mask_70 =
Vector256.Create((byte)0x70); // adding 0x70 does not check low 4-bits
// but moves any value >= 16 above 128
Vector256 <byte> tmp_ws_lo = Avx2.Or(
Avx2.CompareEqual(mask_20, input_lo),
Avx2.Shuffle(lut_cntrl, Avx2.AddSaturate(mask_70, input_lo)));
Vector256 <byte> tmp_ws_hi = Avx2.Or(
Avx2.CompareEqual(mask_20, input_hi),
Avx2.Shuffle(lut_cntrl, Avx2.AddSaturate(mask_70, input_hi)));
uint64_t ws_res_0 = (uint32_t)Avx2.MoveMask(tmp_ws_lo);
uint64_t ws_res_1 = (uint64_t)Avx2.MoveMask(tmp_ws_hi);
uint64_t whitespace = (ws_res_0 | (ws_res_1 << 32));
whitespace &= ~quote_mask;
if (len - idx < 64)
{
whitespace |= ((0xFFFFFFFFFFFFFFFF) << (int)(len - idx));
}
int mask1 = (int)(whitespace & 0xFFFF);
int mask2 = (int)((whitespace >> 16) & 0xFFFF);
int mask3 = (int)((whitespace >> 32) & 0xFFFF);
int mask4 = (int)((whitespace >> 48) & 0xFFFF);
int pop1 = hamming((~whitespace) & 0xFFFF);
int pop2 = hamming((~whitespace) & 0xFFFFFFFF);
int pop3 = hamming((~whitespace) & 0xFFFFFFFFFFFF);
int pop4 = hamming((~whitespace));
var vmask1 =
_mm256_loadu2_m128i((ulong *)mask128_epi8 + (mask2 & 0x7FFF) * 2,
(ulong *)mask128_epi8 + (mask1 & 0x7FFF) * 2);
var vmask2 =
_mm256_loadu2_m128i((ulong *)mask128_epi8 + (mask4 & 0x7FFF) * 2,
(ulong *)mask128_epi8 + (mask3 & 0x7FFF) * 2);
var result1 = Avx2.Shuffle(input_lo, vmask1.AsByte());
var result2 = Avx2.Shuffle(input_hi, vmask2.AsByte());
_mm256_storeu2_m128i((buffer + pop1), buffer,
result1);
_mm256_storeu2_m128i((buffer + pop3), (buffer + pop2),
result2);
memcpy(@out, buffer, (size_t)pop4);
@out += pop4;
}
*@out = (byte)'\0'; // NULL termination
return((size_t)@out - (size_t)initout);
}
}
// this should be called when parsing (right before writing the tapes)
public void Init()
{
current_string_buf_loc = string_buf;
current_loc = 0;
isvalid = false;
}
public static bool parse_number(uint8_t *buf, ParsedJson *pj, uint32_t offset, bool found_minus)
{
bytechar *p = (bytechar *)(buf + offset);
bool negative = false;
if (found_minus)
{
++p;
negative = true;
if (!is_integer(*p))
{
// a negative sign must be followed by an integer
return(false);
}
}
bytechar *startdigits = p;
int64_t i;
if (*p == '0')
{
// 0 cannot be followed by an integer
++p;
if (is_not_structural_or_whitespace_or_exponent_or_decimal((uint8_t)(*p)))
{
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
i = 0;
}
else
{
if (!(is_integer(*p)))
{
// must start with an integer
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
i = digit;
p++;
// the is_made_of_eight_digits_fast routine is unlikely to help here because
// we rarely see large integer parts like 123456789
while (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
i = 10 * i + digit; // might overflow
++p;
}
}
int64_t exponent = 0;
if ('.' == *p)
{
++p;
bytechar *firstafterperiod = p;
if (is_integer(*p))
{
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
++p;
i = i * 10 + digit;
}
else
{
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
#if SWAR_NUMBER_PARSING
// this helps if we have lots of decimals!
// this turns out to be frequent enough.
if (is_made_of_eight_digits_fast(p))
{
i = i * 100000000 + parse_eight_digits_unrolled(p);
p += 8;
// exponent -= 8;
}
#endif
while (is_integer(*p))
{
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
++p;
i = i * 10 + digit; // in rare cases, this will overflow, but that's ok because we have parse_highprecision_float later.
}
exponent = firstafterperiod - p;
}
int digitcount = (int)(p - startdigits - 1);
int64_t expnumber = 0; // exponential part
if (('e' == *p) || ('E' == *p))
{
++p;
bool negexp = false;
if ('-' == *p)
{
negexp = true;
++p;
}
else if ('+' == *p)
{
++p;
}
if (!is_integer(*p))
{
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
expnumber = digit;
p++;
while (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
expnumber = 10 * expnumber + digit;
++p;
}
if (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
expnumber = 10 * expnumber + digit;
++p;
}
if (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
expnumber = 10 * expnumber + digit;
++p;
}
if (is_integer(*p))
{
// we refuse to parse this
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
exponent += (negexp ? -expnumber : expnumber);
}
i = negative ? -i : i;
if ((exponent != 0) || (expnumber != 0))
{
if ((digitcount >= 19))
{
// this is uncommon!!!
// this is almost never going to get called!!!
// we start anew, going slowly!!!
return(parse_float(buf, pj, offset,
found_minus));
}
///////////
// We want 0.1e1 to be a float.
//////////
if (i == 0)
{
pj->write_tape_double(0.0);
#if JSON_TEST_NUMBERS // for unit testing
foundFloat(0.0, buf + offset);
#endif
}
else
{
if ((exponent > 308) || (exponent < -308))
{
// we refuse to parse this
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
double d = i;
d *= power_of_ten[308 + exponent];
// d = negative ? -d : d;
pj->write_tape_double(d);
#if JSON_TEST_NUMBERS // for unit testing
foundFloat(d, buf + offset);
#endif
}
}
else
{
if ((digitcount >= 18))
{
// this is uncommon!!!
return(parse_large_integer(buf, pj, offset,
found_minus));
}
pj->write_tape_s64(i);
#if JSON_TEST_NUMBERS // for unit testing
foundInteger(i, buf + offset);
#endif
}
return(is_structural_or_whitespace((uint8_t)(*p)) != 0);
}
// called by parse_number when we know that the output is an integer,
// but where there might be some integer overflow.
// we want to catch overflows!
// Do not call this function directly as it skips some of the checks from
// parse_number
//
// This function will almost never be called!!!
//
static bool parse_large_integer(uint8_t *buf, ParsedJson *pj, uint32_t offset, bool found_minus)
{
bytechar *p = (bytechar *)(buf + offset);
bool negative = false;
if (found_minus)
{
++p;
negative = true;
}
uint64_t i;
if (*p == '0')
{
// 0 cannot be followed by an integer
++p;
i = 0;
}
else
{
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
i = digit;
p++;
// the is_made_of_eight_digits_fast routine is unlikely to help here because
// we rarely see large integer parts like 123456789
while (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
if (mul_overflow(i, 10, &i))
{
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false); // overflow
}
if (add_overflow(i, digit, &i))
{
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false); // overflow
}
++p;
}
}
if (negative)
{
if (i > 0x8000000000000000)
{
// overflows!
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false); // overflow
}
}
else
{
if (i >= 0x8000000000000000)
{
// overflows!
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false); // overflow
}
}
int64_t signed_answer = negative ? -(int64_t)i : (int64_t)i;
pj->write_tape_s64(signed_answer);
#if JSON_TEST_NUMBERS // for unit testing
foundInteger(signed_answer, buf + offset);
#endif
return(is_structural_or_whitespace((byte)(*p)) != 0);
}
// called by parse_number when we know that the output is a float,
// but where there might be some integer overflow. The trick here is to
// parse using floats from the start.
// Do not call this function directly as it skips some of the checks from
// parse_number
//
// This function will almost never be called!!!
//
// Note: a redesign could avoid this function entirely.
//
private static bool parse_float(uint8_t *buf, ParsedJson *pj, uint32_t offset, bool found_minus)
{
bytechar *p = (bytechar *)(buf + offset);
bool negative = false;
if (found_minus)
{
++p;
negative = true;
}
double i;
if (*p == '0')
{
// 0 cannot be followed by an integer
++p;
i = 0;
}
else
{
unsigned_bytechar digit = (unsigned_bytechar)(*p - (bytechar)'0');
i = digit;
p++;
while (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
i = 10 * i + digit;
++p;
}
}
if ('.' == *p)
{
++p;
double fractionalweight = 1;
if (is_integer(*p))
{
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
++p;
fractionalweight *= 0.1;
i = i + digit * fractionalweight;
}
else
{
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
while (is_integer(*p))
{
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
++p;
fractionalweight *= 0.1;
i = i + digit * fractionalweight;
}
}
if (('e' == *p) || ('E' == *p))
{
++p;
bool negexp = false;
if ('-' == *p)
{
negexp = true;
++p;
}
else if ('+' == *p)
{
++p;
}
if (!is_integer(*p))
{
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
unsigned_bytechar digit = (unsigned_bytechar)(*p - '0');
int64_t expnumber = digit; // exponential part
p++;
if (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
expnumber = 10 * expnumber + digit;
++p;
}
if (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
expnumber = 10 * expnumber + digit;
++p;
}
if (is_integer(*p))
{
digit = (unsigned_bytechar)(*p - '0');
expnumber = 10 * expnumber + digit;
++p;
}
if (is_integer(*p))
{
// we refuse to parse this
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
int exponent = (int)(negexp ? -expnumber : expnumber);
if ((exponent > 308) || (exponent < -308))
{
// we refuse to parse this
#if JSON_TEST_NUMBERS // for unit testing
foundInvalidNumber(buf + offset);
#endif
return(false);
}
i *= power_of_ten[308 + exponent];
}
if (is_not_structural_or_whitespace((byte)*p) != 0)
{
return(false);
}
double d = negative ? -i : i;
pj->write_tape_double(d);
#if JSON_TEST_NUMBERS // for unit testing
foundFloat(d, buf + offset);
#endif
return(is_structural_or_whitespace((byte)(*p)) != 0);
}
/*
* find all devices in the piconet
*/
int sdp_general_inquiry(inquiry_info *ii, int dev_num, int duration, uint8_t *found);
internal static bool find_structural_bits(uint8_t* buf, size_t len, ParsedJson pj)
{
if (len > pj.bytecapacity)
{
Console.WriteLine("Your ParsedJson object only supports documents up to " + pj.bytecapacity +
" bytes but you are trying to process " + len + " bytes\n");
return false;
}
uint32_t* base_ptr = pj.structural_indexes;
uint32_t @base = 0;
#if SIMDJSON_UTF8VALIDATE // NOT TESTED YET!
var has_error = Vector256<byte>.Zero;
var previous = new avx_processed_utf_bytes();
previous.rawbytes = Vector256<byte>.Zero;
previous.high_nibbles = Vector256<byte>.Zero;
previous.carried_continuations = Vector256<byte>.Zero;
var highbit = Vector256.Create((byte)0x80);
#endif
const uint64_t even_bits = 0x5555555555555555UL;
const uint64_t odd_bits = ~even_bits;
// for now, just work in 64-byte chunks
// we have padded the input out to 64 byte multiple with the remainder being
// zeros
// persistent state across loop
uint64_t prev_iter_ends_odd_backslash = 0UL; // either 0 or 1, but a 64-bit value
uint64_t prev_iter_inside_quote = 0UL; // either all zeros or all ones
// effectively the very first char is considered to follow "whitespace" for the
// purposes of psuedo-structural character detection
uint64_t prev_iter_ends_pseudo_pred = 1UL;
size_t lenminus64 = len < 64 ? 0 : len - 64;
size_t idx = 0;
uint64_t structurals = 0;
// C#: assign static readonly fields to locals before the loop
Vector256<byte> low_nibble_mask = s_low_nibble_mask;
Vector256<byte> high_nibble_mask = s_high_nibble_mask;
Vector256<byte> utf8ValidVec = s_utf8ValidVec;
var structural_shufti_mask = Vector256.Create((byte)0x7);
var whitespace_shufti_mask = Vector256.Create((byte)0x18);
var slashVec = Vector256.Create((bytechar) '\\').AsByte();
var ffVec = Vector128.Create((byte) 0xFF).AsUInt64();
var doubleQuoteVec = Vector256.Create((byte)'"');
var zeroBVec = Vector256.Create((byte) 0);
var vec7f = Vector256.Create((byte) 0x7f);
for (; idx < lenminus64; idx += 64)
{
var input_lo = Avx.LoadVector256(buf + idx + 0);
var input_hi = Avx.LoadVector256(buf + idx + 32);
#if SIMDJSON_UTF8VALIDATE // NOT TESTED YET!
if ((Avx.TestZ(Avx2.Or(input_lo, input_hi), highbit)) == true)
{
// it is ascii, we just check continuation
has_error = Avx2.Or(
Avx2.CompareGreaterThan(previous.carried_continuations.AsSByte(), utf8ValidVec, has_error);
}
else
{
// it is not ascii so we have to do heavy work
previous = Utf8Validation.avxcheckUTF8Bytes(input_lo, ref previous, ref has_error);
previous = Utf8Validation.avxcheckUTF8Bytes(input_hi, ref previous, ref has_error);
}
#endif
////////////////////////////////////////////////////////////////////////////////////////////
// Step 1: detect odd sequences of backslashes
////////////////////////////////////////////////////////////////////////////////////////////
///
uint64_t bs_bits =
cmp_mask_against_input(input_lo, input_hi, slashVec);
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
// flip lowest if we have an odd-length run at the end of the prior
// iteration
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
// must record the carry-out of our odd-carries out of bit 63; this
// indicates whether the sense of any edge going to the next iteration
// should be flipped
bool iter_ends_odd_backslash =
add_overflow(bs_bits, odd_starts, &odd_carries);
odd_carries |=
prev_iter_ends_odd_backslash; // push in bit zero as a potential end
// if we had an odd-numbered run at the
// end of the previous iteration
prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1UL : 0x0UL;
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
////////////////////////////////////////////////////////////////////////////////////////////
// Step 2: detect insides of quote pairs
////////////////////////////////////////////////////////////////////////////////////////////
uint64_t quote_bits =
cmp_mask_against_input(input_lo, input_hi, doubleQuoteVec);
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = Sse2.X64.ConvertToUInt64(Pclmulqdq.CarrylessMultiply(
Vector128.Create(quote_bits, 0UL /*C# reversed*/), ffVec, 0));
uint32_t cnt = (uint32_t) hamming(structurals);
uint32_t next_base = @base + cnt;
while (structurals != 0)
{
base_ptr[@base + 0] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 1] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 2] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 3] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 4] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 5] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 6] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 7] = (uint32_t) idx - 64 + (uint32_t) trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
@base += 8;
}
@base = next_base;
quote_mask ^= prev_iter_inside_quote;
prev_iter_inside_quote =
(uint64_t) ((int64_t) quote_mask >>
63); // right shift of a signed value expected to be well-defined and standard compliant as of C++20, John Regher from Utah U. says this is fine code
var v_lo = Avx2.And(
Avx2.Shuffle(low_nibble_mask, input_lo),
Avx2.Shuffle(high_nibble_mask,
Avx2.And(Avx2.ShiftRightLogical(input_lo.AsUInt32(), 4).AsByte(),
vec7f)));
var v_hi = Avx2.And(
Avx2.Shuffle(low_nibble_mask, input_hi),
Avx2.Shuffle(high_nibble_mask,
Avx2.And(Avx2.ShiftRightLogical(input_hi.AsUInt32(), 4).AsByte(),
vec7f)));
var tmp_lo = Avx2.CompareEqual(
Avx2.And(v_lo, structural_shufti_mask), zeroBVec);
var tmp_hi = Avx2.CompareEqual(
Avx2.And(v_hi, structural_shufti_mask), zeroBVec);
uint64_t structural_res_0 = (uint32_t) Avx2.MoveMask(tmp_lo);
uint64_t structural_res_1 = (uint64_t) Avx2.MoveMask(tmp_hi);
structurals = ~(structural_res_0 | (structural_res_1 << 32));
var tmp_ws_lo = Avx2.CompareEqual(
Avx2.And(v_lo, whitespace_shufti_mask), zeroBVec);
var tmp_ws_hi = Avx2.CompareEqual(
Avx2.And(v_hi, whitespace_shufti_mask), zeroBVec);
uint64_t ws_res_0 = (uint32_t) Avx2.MoveMask(tmp_ws_lo);
uint64_t ws_res_1 = (uint64_t) Avx2.MoveMask(tmp_ws_hi);
uint64_t whitespace = ~(ws_res_0 | (ws_res_1 << 32));
// mask off anything inside quotes
structurals &= ~quote_mask;
// add the real quote bits back into our bitmask as well, so we can
// quickly traverse the strings we've spent all this trouble gathering
structurals |= quote_bits;
// Now, establish "pseudo-structural characters". These are non-whitespace
// characters that are (a) outside quotes and (b) have a predecessor that's
// either whitespace or a structural character. This means that subsequent
// passes will get a chance to encounter the first character of every string
// of non-whitespace and, if we're parsing an atom like true/false/null or a
// number we can stop at the first whitespace or structural character
// following it.
// a qualified predecessor is something that can happen 1 position before an
// psuedo-structural character
uint64_t pseudo_pred = structurals | whitespace;
uint64_t shifted_pseudo_pred = (pseudo_pred << 1) | prev_iter_ends_pseudo_pred;
prev_iter_ends_pseudo_pred = pseudo_pred >> 63;
uint64_t pseudo_structurals =
shifted_pseudo_pred & (~whitespace) & (~quote_mask);
structurals |= pseudo_structurals;
// now, we've used our close quotes all we need to. So let's switch them off
// they will be off in the quote mask and on in quote bits.
structurals &= ~(quote_bits & ~quote_mask);
//Console.WriteLine($"Iter: {idx}, satur: {structurals}");
//*(uint64_t *)(pj.structurals + idx / 8) = structurals;
}
////////////////
/// we use a giant copy-paste which is ugly.
/// but otherwise the string needs to be properly padded or else we
/// risk invalidating the UTF-8 checks.
////////////
if (idx < len)
{
uint8_t* tmpbuf = stackalloc uint8_t[64];
memset(tmpbuf, 0x20, 64);
memcpy(tmpbuf, buf + idx, len - idx);
Vector256<byte> input_lo = Avx.LoadVector256(tmpbuf + 0);
Vector256<byte> input_hi = Avx.LoadVector256(tmpbuf + 32);
#if SIMDJSON_UTF8VALIDATE // NOT TESTED YET!
var highbit = Vector256.Create((byte)0x80);
if ((Avx.TestZ(Avx2.Or(input_lo, input_hi), highbit)) == true)
{
// it is ascii, we just check continuation
has_error = Avx2.Or(
Avx2.CompareGreaterThan(previous.carried_continuations.AsSByte(),
utf8ValidVec).AsByte(), has_error);
}
else
{
// it is not ascii so we have to do heavy work
previous = Utf8Validation.avxcheckUTF8Bytes(input_lo, ref previous, ref has_error);
previous = Utf8Validation.avxcheckUTF8Bytes(input_hi, ref previous, ref has_error);
}
#endif
////////////////////////////////////////////////////////////////////////////////////////////
// Step 1: detect odd sequences of backslashes
////////////////////////////////////////////////////////////////////////////////////////////
uint64_t bs_bits =
cmp_mask_against_input(input_lo, input_hi, slashVec);
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
// flip lowest if we have an odd-length run at the end of the prior
// iteration
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
// must record the carry-out of our odd-carries out of bit 63; this
// indicates whether the sense of any edge going to the next iteration
// should be flipped
//bool iter_ends_odd_backslash =
add_overflow(bs_bits, odd_starts, &odd_carries);
odd_carries |=
prev_iter_ends_odd_backslash; // push in bit zero as a potential end
// if we had an odd-numbered run at the
// end of the previous iteration
//prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1ULL : 0x0ULL;
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
////////////////////////////////////////////////////////////////////////////////////////////
// Step 2: detect insides of quote pairs
////////////////////////////////////////////////////////////////////////////////////////////
uint64_t quote_bits =
cmp_mask_against_input(input_lo, input_hi, doubleQuoteVec);
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = (uint64_t)Sse2.X64.ConvertToInt64(Pclmulqdq.CarrylessMultiply(
Vector128.Create(quote_bits, 0UL /*C# reversed*/), ffVec, 0).AsInt64());
quote_mask ^= prev_iter_inside_quote;
//BUG? https://github.com/dotnet/coreclr/issues/22813
//quote_mask = 60;
//prev_iter_inside_quote = (uint64_t)((int64_t)quote_mask >> 63); // right shift of a signed value expected to be well-defined and standard compliant as of C++20
uint32_t cnt = (uint32_t)hamming(structurals);
uint32_t next_base = @base + cnt;
while (structurals != 0)
{
base_ptr[@base + 0] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 1] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 2] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 3] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 4] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 5] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 6] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 7] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
@base += 8;
}
@base = next_base;
// How do we build up a user traversable data structure
// first, do a 'shufti' to detect structural JSON characters
// they are { 0x7b } 0x7d : 0x3a [ 0x5b ] 0x5d , 0x2c
// these go into the first 3 buckets of the comparison (1/2/4)
// we are also interested in the four whitespace characters
// space 0x20, linefeed 0x0a, horizontal tab 0x09 and carriage return 0x0d
// these go into the next 2 buckets of the comparison (8/16)
var v_lo = Avx2.And(
Avx2.Shuffle(low_nibble_mask, input_lo),
Avx2.Shuffle(high_nibble_mask,
Avx2.And(Avx2.ShiftRightLogical(input_lo.AsUInt32(), 4).AsByte(),
vec7f)));
var v_hi = Avx2.And(
Avx2.Shuffle(low_nibble_mask, input_hi),
Avx2.Shuffle(high_nibble_mask,
Avx2.And(Avx2.ShiftRightLogical(input_hi.AsUInt32(), 4).AsByte(),
vec7f)));
var tmp_lo = Avx2.CompareEqual(
Avx2.And(v_lo, structural_shufti_mask), zeroBVec);
var tmp_hi = Avx2.CompareEqual(
Avx2.And(v_hi, structural_shufti_mask), zeroBVec);
uint64_t structural_res_0 = (uint32_t)Avx2.MoveMask(tmp_lo);
uint64_t structural_res_1 = (uint64_t)Avx2.MoveMask(tmp_hi);
structurals = ~(structural_res_0 | (structural_res_1 << 32));
// this additional mask and transfer is non-trivially expensive,
// unfortunately
var tmp_ws_lo = Avx2.CompareEqual(
Avx2.And(v_lo, whitespace_shufti_mask), zeroBVec);
var tmp_ws_hi = Avx2.CompareEqual(
Avx2.And(v_hi, whitespace_shufti_mask), zeroBVec);
uint64_t ws_res_0 = (uint32_t)Avx2.MoveMask(tmp_ws_lo);
uint64_t ws_res_1 = (uint64_t)Avx2.MoveMask(tmp_ws_hi);
uint64_t whitespace = ~(ws_res_0 | (ws_res_1 << 32));
// mask off anything inside quotes
structurals &= ~quote_mask;
// add the real quote bits back into our bitmask as well, so we can
// quickly traverse the strings we've spent all this trouble gathering
structurals |= quote_bits;
// Now, establish "pseudo-structural characters". These are non-whitespace
// characters that are (a) outside quotes and (b) have a predecessor that's
// either whitespace or a structural character. This means that subsequent
// passes will get a chance to encounter the first character of every string
// of non-whitespace and, if we're parsing an atom like true/false/null or a
// number we can stop at the first whitespace or structural character
// following it.
// a qualified predecessor is something that can happen 1 position before an
// psuedo-structural character
uint64_t pseudo_pred = structurals | whitespace;
uint64_t shifted_pseudo_pred = (pseudo_pred << 1) | prev_iter_ends_pseudo_pred;
prev_iter_ends_pseudo_pred = pseudo_pred >> 63;
uint64_t pseudo_structurals =
shifted_pseudo_pred & (~whitespace) & (~quote_mask);
structurals |= pseudo_structurals;
// now, we've used our close quotes all we need to. So let's switch them off
// they will be off in the quote mask and on in quote bits.
structurals &= ~(quote_bits & ~quote_mask);
//*(uint64_t *)(pj.structurals + idx / 8) = structurals;
idx += 64;
}
uint32_t cnt2 = (uint32_t)hamming(structurals);
uint32_t next_base2 = @base + cnt2;
while (structurals != 0)
{
base_ptr[@base + 0] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 1] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 2] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 3] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 4] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 5] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 6] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
base_ptr[@base + 7] = (uint32_t)idx - 64 + (uint32_t)trailingzeroes(structurals);
structurals = structurals & (structurals - 1);
@base += 8;
}
@base = next_base2;
pj.n_structural_indexes = @base;
if (base_ptr[pj.n_structural_indexes - 1] > len)
{
throw new InvalidOperationException("Internal bug");
}
if (len != base_ptr[pj.n_structural_indexes - 1])
{
// the string might not be NULL terminated, but we add a virtual NULL ending character.
base_ptr[pj.n_structural_indexes++] = (uint32_t)len;
}
base_ptr[pj.n_structural_indexes] = 0; // make it safe to dereference one beyond this array
#if SIMDJSON_UTF8VALIDATE // NOT TESTED YET!
return Avx.TestZ(has_error, has_error);
#else
return true;
#endif
}
public static bool parse_string(uint8_t *buf, size_t len, ParsedJson *pj, uint32_t depth, uint32_t offset)
{
#if SIMDJSON_SKIPSTRINGPARSING // for performance analysis, it is sometimes useful to skip parsing
pj->write_tape(0, '"'); // don't bother with the string parsing at all
return(true); // always succeeds
#else
uint8_t *src = &buf[offset + 1]; // we know that buf at offset is a "
uint8_t *dst = pj->current_string_buf_loc;
#if JSON_TEST_STRINGS // for unit testing
uint8_t *const start_of_string = dst;
#endif
var slashVec = Vector256.Create((byte)'\\');
var quoteVec = Vector256.Create((byte)'"');
while (true)
{
Vector256 <byte> v = Avx2.LoadVector256((src));
uint32_t bs_bits =
(uint32_t)Avx2.MoveMask(Avx2.CompareEqual(v, slashVec));
uint32_t quote_bits =
(uint32_t)Avx2.MoveMask(Avx2.CompareEqual(v, quoteVec));
// All Unicode characters may be placed within the
// quotation marks, except for the characters that MUST be escaped:
// quotation mark, reverse solidus, and the control characters (U+0000
//through U+001F).
// https://tools.ietf.org/html/rfc8259
#if CHECKUNESCAPED
var unitsep = Vector256.Create((byte)0x1F);
var unescaped_vec =
Avx2.CompareEqual(Avx2.Max(unitsep, v), unitsep); // could do it with saturated subtraction
#endif // CHECKUNESCAPED
uint32_t quote_dist = (uint32_t)trailingzeroes(quote_bits);
uint32_t bs_dist = (uint32_t)trailingzeroes(bs_bits);
// store to dest unconditionally - we can overwrite the bits we don't like
// later
Avx.Store((dst), v);
if (quote_dist < bs_dist)
{
// we encountered quotes first. Move dst to point to quotes and exit
dst[quote_dist] = 0; // null terminate and get out
pj->write_tape((size_t)pj->current_string_buf_loc - (size_t)pj->string_buf, (uint8_t)'"');
pj->current_string_buf_loc = dst + quote_dist + 1; // the +1 is due to the 0 value
#if CHECKUNESCAPED
// check that there is no unescaped char before the quote
uint32_t unescaped_bits = (uint32_t)Avx2.MoveMask(unescaped_vec);
bool is_ok = ((quote_bits - 1) & (~quote_bits) & unescaped_bits) == 0;
#if JSON_TEST_STRINGS // for unit testing
if (is_ok)
{
foundString(buf + offset, start_of_string, pj->current_string_buf_loc - 1);
}
else
{
foundBadString(buf + offset);
}
#endif // JSON_TEST_STRINGS
return(is_ok);
#else //CHECKUNESCAPED
#if JSON_TEST_STRINGS // for unit testing
foundString(buf + offset, start_of_string, pj->current_string_buf_loc - 1);
#endif // JSON_TEST_STRINGS
return(true);
#endif //CHECKUNESCAPED
}
else if (quote_dist > bs_dist)
{
uint8_t escape_char = src[bs_dist + 1];
#if CHECKUNESCAPED
// we are going to need the unescaped_bits to check for unescaped chars
uint32_t unescaped_bits = (uint32_t)Avx2.MoveMask(unescaped_vec);
if (((bs_bits - 1) & (~bs_bits) & unescaped_bits) != 0)
{
#if JSON_TEST_STRINGS // for unit testing
foundBadString(buf + offset);
#endif // JSON_TEST_STRINGS
return(false);
}
#endif //CHECKUNESCAPED
// we encountered backslash first. Handle backslash
if (escape_char == 'u')
{
// move src/dst up to the start; they will be further adjusted
// within the unicode codepoint handling code.
src += bs_dist;
dst += bs_dist;
if (!handle_unicode_codepoint(&src, &dst))
{
#if JSON_TEST_STRINGS // for unit testing
foundBadString(buf + offset);
#endif // JSON_TEST_STRINGS
return(false);
}
}
else
{
// simple 1:1 conversion. Will eat bs_dist+2 characters in input and
// write bs_dist+1 characters to output
// note this may reach beyond the part of the buffer we've actually
// seen. I think this is ok
uint8_t escape_result = escape_map[escape_char];
if (escape_result == 0)
{
#if JSON_TEST_STRINGS // for unit testing
foundBadString(buf + offset);
#endif // JSON_TEST_STRINGS
return(false); // bogus escape value is an error
}
dst[bs_dist] = escape_result;
src += bs_dist + 2;
dst += bs_dist + 1;
}
}
else
{
// they are the same. Since they can't co-occur, it means we encountered
// neither.
src += 32;
dst += 32;
#if CHECKUNESCAPED
// check for unescaped chars
if (Avx.TestZ(unescaped_vec, unescaped_vec) != true)
{
#if JSON_TEST_STRINGS // for unit testing
foundBadString(buf + offset);
#endif // JSON_TEST_STRINGS
return(false);
}
#endif // CHECKUNESCAPED
}
}
// can't be reached
return(true);
#endif // SIMDJSON_SKIPSTRINGPARSING
}
public static extern void ShadeTile(int32_t tileStartX, int32_t tileEndX, int32_t tileStartY, int32_t tileEndY, int32_t gBufferWidth, int32_t gBufferHeight, InputDataArrays *inputData, float cameraProj_11, float cameraProj_22, float cameraProj_33, float cameraProj_43, int32_t *tileLightIndices, int32_t tileNumLights, bool visualizeLightCount, uint8_t *framebuffer_r, uint8_t *framebuffer_g, uint8_t *framebuffer_b) /*x77*/;