From c9354d88530fea3485fe2e802eadcbc161759691 Mon Sep 17 00:00:00 2001 From: Taiki Date: Sat, 26 Jan 2019 15:04:04 +0100 Subject: [PATCH] Extend the capabilities of convert_ruby_to_v8 to convert more types, and to work better when the encoding of strings isn't UTF-8. --- ext/mini_racer_extension/extconf.rb | 5 + .../mini_racer_extension.cc | 208 +++++--- ext/mini_racer_extension/simdutf8check.h | 463 ++++++++++++++++++ test/mini_racer_test.rb | 62 ++- 4 files changed, 672 insertions(+), 66 deletions(-) create mode 100644 ext/mini_racer_extension/simdutf8check.h diff --git a/ext/mini_racer_extension/extconf.rb b/ext/mini_racer_extension/extconf.rb index 48196387..beee3729 100644 --- a/ext/mini_racer_extension/extconf.rb +++ b/ext/mini_racer_extension/extconf.rb @@ -11,6 +11,11 @@ $CPPFLAGS += " -fPIC" unless $CPPFLAGS.split.include? "-rdynamic" or IS_DARWIN $CPPFLAGS += " -std=c++0x" $CPPFLAGS += " -fpermissive" +if enable_config('avx2') + $CPPFLAGS += " -mavx2" +else + $CPPFLAGS += " -mssse3" +end $CPPFLAGS += " -Wno-reserved-user-defined-literal" if IS_DARWIN diff --git a/ext/mini_racer_extension/mini_racer_extension.cc b/ext/mini_racer_extension/mini_racer_extension.cc index 836f3d3f..bc71ec6c 100644 --- a/ext/mini_racer_extension/mini_racer_extension.cc +++ b/ext/mini_racer_extension/mini_racer_extension.cc @@ -9,6 +9,7 @@ #include #include #include +#include "simdutf8check.h" using namespace v8; @@ -453,7 +454,84 @@ static VALUE convert_v8_to_ruby(Isolate* isolate, Local::New(isolate, value)); } -static Local convert_ruby_to_v8(Isolate* isolate, VALUE value) { +static VALUE encode_as_utf8(VALUE string) +{ + return rb_funcall(string, rb_intern("encode"), 1, rb_str_new2("UTF-8")); +} + +#ifdef __AVX2__ +static bool (*best_utf8_validate_func(void))(const char *, size_t) +{ + __builtin_cpu_init(); + if (__builtin_cpu_supports("avx2")) { + return validate_utf8_fast_avx; + } else { + return validate_utf8_fast; + } +} +#endif + +static inline Local convert_ruby_str_to_v8( + HandleScope& scope, Isolate *isolate, VALUE value) +{ + static const rb_encoding *utf8_enc = rb_utf8_encoding(); + static const rb_encoding *ascii8bit_enc = rb_ascii8bit_encoding(); + static const rb_encoding *usascii_enc = rb_usascii_encoding(); + static const rb_encoding *latin1_enc = rb_enc_find("ISO-8859-1"); + assert(latin1_enc != nullptr); +#ifndef __AVX2__ +# define validate_utf8 validate_utf8_fast +#else + static const (*validate_utf8)(const char *, size_t) = + best_utf8_validate_func(); +#endif + + rb_encoding *enc = rb_enc_get(value); + char *str = RSTRING_PTR(value); + long len = RSTRING_LEN(value); + if (len < 0 || len > INT_MAX) { + return Null(isolate); + } + bool is_valid_utf8 = enc == utf8_enc && + validate_utf8(str, static_cast(len)); + + MaybeLocal v8str; + int int_len = static_cast(len); + if (is_valid_utf8) { +convert_from_utf8: + v8str = String::NewFromUtf8( + isolate, str, NewStringType::kNormal, int_len); + } else if (enc == utf8_enc || enc == ascii8bit_enc || + enc == usascii_enc || enc == latin1_enc || + rb_funcall(value, rb_intern("valid_encoding?"), 0) == Qfalse) { +treat_as_latin1: + // if ASCII, it could be that the string is invalid + // ignore that possibility (effectively treat it as latin1) + v8str = String::NewFromOneByte( + isolate, reinterpret_cast(str), + NewStringType::kNormal, int_len); + } else { + int state; + VALUE result = rb_protect(encode_as_utf8, value, &state); + + //Ran into an exception! + if (state) { + rb_set_errinfo(Qnil); + goto treat_as_latin1; + } else if (rb_enc_get(result) != utf8_enc) { + // conversion did not result in UTF-8. Odd! + goto treat_as_latin1; + } else { + str = RSTRING_PTR(result); + int_len = RSTRING_LEN(result); + goto convert_from_utf8; + } + } + return v8str.ToLocalChecked(); +} + +static Local convert_ruby_to_v8(Isolate* isolate, VALUE value) +{ EscapableHandleScope scope(isolate); Local array; @@ -466,67 +544,85 @@ static Local convert_ruby_to_v8(Isolate* isolate, VALUE value) { VALUE klass; switch (TYPE(value)) { - case T_FIXNUM: - fixnum = NUM2LONG(value); - if (fixnum > INT_MAX) + case T_FIXNUM: { - return scope.Escape(Number::New(isolate, (double)fixnum)); + fixnum = NUM2LONG(value); + if (fixnum > INT_MAX) + { + return scope.Escape(Number::New(isolate, (double)fixnum)); + } + return scope.Escape(Integer::New(isolate, (int)fixnum)); } - return scope.Escape(Integer::New(isolate, (int)fixnum)); - case T_FLOAT: - return scope.Escape(Number::New(isolate, NUM2DBL(value))); - case T_STRING: - return scope.Escape(String::NewFromUtf8(isolate, RSTRING_PTR(value), NewStringType::kNormal, (int)RSTRING_LEN(value)).ToLocalChecked()); - case T_NIL: - return scope.Escape(Null(isolate)); - case T_TRUE: - return scope.Escape(True(isolate)); - case T_FALSE: - return scope.Escape(False(isolate)); - case T_ARRAY: - length = RARRAY_LEN(value); - array = Array::New(isolate, (int)length); - for(i=0; iSet(i, convert_ruby_to_v8(isolate, rb_ary_entry(value, i))); - } - return scope.Escape(array); - case T_HASH: - object = Object::New(isolate); - hash_as_array = rb_funcall(value, rb_intern("to_a"), 0); - length = RARRAY_LEN(hash_as_array); - for(i=0; iSet(convert_ruby_to_v8(isolate, rb_ary_entry(pair, 0)), - convert_ruby_to_v8(isolate, rb_ary_entry(pair, 1))); - } - return scope.Escape(object); - case T_SYMBOL: - value = rb_funcall(value, rb_intern("to_s"), 0); - return scope.Escape(String::NewFromUtf8(isolate, RSTRING_PTR(value), NewStringType::kNormal, (int)RSTRING_LEN(value)).ToLocalChecked()); - case T_DATA: - klass = rb_funcall(value, rb_intern("class"), 0); - if (klass == rb_cTime || klass == rb_cDateTime) + case T_FLOAT: + return scope.Escape(Number::New(isolate, NUM2DBL(value))); + case T_STRING: + return scope.Escape(convert_ruby_str_to_v8(scope, isolate, value)); + case T_NIL: + return scope.Escape(Null(isolate)); + case T_TRUE: + return scope.Escape(True(isolate)); + case T_FALSE: + return scope.Escape(False(isolate)); + case T_ARRAY: + { + length = RARRAY_LEN(value); + array = Array::New(isolate, (int)length); + for(i=0; iSet(i, convert_ruby_to_v8(isolate, rb_ary_entry(value, i))); + } + return scope.Escape(array); + } + case T_HASH: + { + object = Object::New(isolate); + hash_as_array = rb_funcall(value, rb_intern("to_a"), 0); + length = RARRAY_LEN(hash_as_array); + for(i=0; iSet(convert_ruby_to_v8(isolate, rb_ary_entry(pair, 0)), + convert_ruby_to_v8(isolate, rb_ary_entry(pair, 1))); + } + return scope.Escape(object); + } + case T_SYMBOL: { - if (klass == rb_cDateTime) + value = rb_funcall(value, rb_intern("to_s"), 0); + return scope.Escape(convert_ruby_str_to_v8(scope, isolate, value)); + } + case T_DATA: + { + klass = rb_funcall(value, rb_intern("class"), 0); + if (klass == rb_cTime || klass == rb_cDateTime) { - value = rb_funcall(value, rb_intern("to_time"), 0); + if (klass == rb_cDateTime) + { + value = rb_funcall(value, rb_intern("to_time"), 0); + } + value = rb_funcall(value, rb_intern("to_f"), 0); + return scope.Escape(Date::New(isolate, NUM2DBL(value) * 1000)); + } + // break intentionally missing + } + case T_OBJECT: + case T_CLASS: + case T_ICLASS: + case T_MODULE: + case T_REGEXP: + case T_MATCH: + case T_STRUCT: + case T_BIGNUM: + case T_FILE: + case T_UNDEF: + case T_NODE: + default: + { + if (rb_respond_to(value, rb_intern("to_s"))) { + // TODO: if this throws we're screwed + value = rb_funcall(value, rb_intern("to_s"), 0); + return scope.Escape(convert_ruby_str_to_v8(scope, isolate, value)); } - value = rb_funcall(value, rb_intern("to_f"), 0); - return scope.Escape(Date::New(isolate, NUM2DBL(value) * 1000)); + return scope.Escape(String::NewFromUtf8(isolate, "Undefined Conversion")); } - case T_OBJECT: - case T_CLASS: - case T_ICLASS: - case T_MODULE: - case T_REGEXP: - case T_MATCH: - case T_STRUCT: - case T_BIGNUM: - case T_FILE: - case T_UNDEF: - case T_NODE: - default: - return scope.Escape(String::NewFromUtf8(isolate, "Undefined Conversion")); } } diff --git a/ext/mini_racer_extension/simdutf8check.h b/ext/mini_racer_extension/simdutf8check.h new file mode 100644 index 00000000..9954b2ec --- /dev/null +++ b/ext/mini_racer_extension/simdutf8check.h @@ -0,0 +1,463 @@ + +#ifndef SIMDUTF8CHECK_H +#define SIMDUTF8CHECK_H +#include +#include +#include +#include +#include +/* + * legal utf-8 byte sequence + * http://www.unicode.org/versions/Unicode6.0.0/ch03.pdf - page 94 + * + * Code Points 1st 2s 3s 4s + * U+0000..U+007F 00..7F + * U+0080..U+07FF C2..DF 80..BF + * U+0800..U+0FFF E0 A0..BF 80..BF + * U+1000..U+CFFF E1..EC 80..BF 80..BF + * U+D000..U+D7FF ED 80..9F 80..BF + * U+E000..U+FFFF EE..EF 80..BF 80..BF + * U+10000..U+3FFFF F0 90..BF 80..BF 80..BF + * U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF + * U+100000..U+10FFFF F4 80..8F 80..BF 80..BF + * + */ + +// all byte values must be no larger than 0xF4 +static inline void checkSmallerThan0xF4(__m128i current_bytes, + __m128i *has_error) { + // unsigned, saturates to 0 below max + *has_error = _mm_or_si128(*has_error, + _mm_subs_epu8(current_bytes, _mm_set1_epi8(0xF4))); +} + +static inline __m128i continuationLengths(__m128i high_nibbles) { + return _mm_shuffle_epi8( + _mm_setr_epi8(1, 1, 1, 1, 1, 1, 1, 1, // 0xxx (ASCII) + 0, 0, 0, 0, // 10xx (continuation) + 2, 2, // 110x + 3, // 1110 + 4), // 1111, next should be 0 (not checked here) + high_nibbles); +} + +static inline __m128i carryContinuations(__m128i initial_lengths, + __m128i previous_carries) { + + __m128i right1 = + _mm_subs_epu8(_mm_alignr_epi8(initial_lengths, previous_carries, 16 - 1), + _mm_set1_epi8(1)); + __m128i sum = _mm_add_epi8(initial_lengths, right1); + + __m128i right2 = _mm_subs_epu8(_mm_alignr_epi8(sum, previous_carries, 16 - 2), + _mm_set1_epi8(2)); + return _mm_add_epi8(sum, right2); +} + +static inline void checkContinuations(__m128i initial_lengths, __m128i carries, + __m128i *has_error) { + + // overlap || underlap + // carry > length && length > 0 || !(carry > length) && !(length > 0) + // (carries > length) == (lengths > 0) + __m128i overunder = + _mm_cmpeq_epi8(_mm_cmpgt_epi8(carries, initial_lengths), + _mm_cmpgt_epi8(initial_lengths, _mm_setzero_si128())); + + *has_error = _mm_or_si128(*has_error, overunder); +} + +// when 0xED is found, next byte must be no larger than 0x9F +// when 0xF4 is found, next byte must be no larger than 0x8F +// next byte must be continuation, ie sign bit is set, so signed < is ok +static inline void checkFirstContinuationMax(__m128i current_bytes, + __m128i off1_current_bytes, + __m128i *has_error) { + __m128i maskED = _mm_cmpeq_epi8(off1_current_bytes, _mm_set1_epi8(0xED)); + __m128i maskF4 = _mm_cmpeq_epi8(off1_current_bytes, _mm_set1_epi8(0xF4)); + + __m128i badfollowED = + _mm_and_si128(_mm_cmpgt_epi8(current_bytes, _mm_set1_epi8(0x9F)), maskED); + __m128i badfollowF4 = + _mm_and_si128(_mm_cmpgt_epi8(current_bytes, _mm_set1_epi8(0x8F)), maskF4); + + *has_error = _mm_or_si128(*has_error, _mm_or_si128(badfollowED, badfollowF4)); +} + +// map off1_hibits => error condition +// hibits off1 cur +// C => < C2 && true +// E => < E1 && < A0 +// F => < F1 && < 90 +// else false && false +static inline void checkOverlong(__m128i current_bytes, + __m128i off1_current_bytes, __m128i hibits, + __m128i previous_hibits, __m128i *has_error) { + __m128i off1_hibits = _mm_alignr_epi8(hibits, previous_hibits, 16 - 1); + __m128i initial_mins = _mm_shuffle_epi8( + _mm_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, // 10xx => false + 0xC2, -128, // 110x + 0xE1, // 1110 + 0xF1), + off1_hibits); + + __m128i initial_under = _mm_cmpgt_epi8(initial_mins, off1_current_bytes); + + __m128i second_mins = _mm_shuffle_epi8( + _mm_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, // 10xx => false + 127, 127, // 110x => true + 0xA0, // 1110 + 0x90), + off1_hibits); + __m128i second_under = _mm_cmpgt_epi8(second_mins, current_bytes); + *has_error = + _mm_or_si128(*has_error, _mm_and_si128(initial_under, second_under)); +} + +struct processed_utf_bytes { + __m128i rawbytes; + __m128i high_nibbles; + __m128i carried_continuations; +}; + +static inline void count_nibbles(__m128i bytes, + struct processed_utf_bytes *answer) { + answer->rawbytes = bytes; + answer->high_nibbles = + _mm_and_si128(_mm_srli_epi16(bytes, 4), _mm_set1_epi8(0x0F)); +} + +// check whether the current bytes are valid UTF-8 +// at the end of the function, previous gets updated +static struct processed_utf_bytes +checkUTF8Bytes(__m128i current_bytes, struct processed_utf_bytes *previous, + __m128i *has_error) { + struct processed_utf_bytes pb; + count_nibbles(current_bytes, &pb); + + checkSmallerThan0xF4(current_bytes, has_error); + + __m128i initial_lengths = continuationLengths(pb.high_nibbles); + + pb.carried_continuations = + carryContinuations(initial_lengths, previous->carried_continuations); + + checkContinuations(initial_lengths, pb.carried_continuations, has_error); + + __m128i off1_current_bytes = + _mm_alignr_epi8(pb.rawbytes, previous->rawbytes, 16 - 1); + checkFirstContinuationMax(current_bytes, off1_current_bytes, has_error); + + checkOverlong(current_bytes, off1_current_bytes, pb.high_nibbles, + previous->high_nibbles, has_error); + return pb; +} + +// alternative to _mm_testz_si128(x, x) (which requires SSE 4.1) +static inline bool is_all_zeros(__m128i xmm) { + return _mm_movemask_epi8(_mm_cmpeq_epi8(xmm, _mm_setzero_si128())) == 0xFFFF; +} + +static bool validate_utf8_fast(const char *src, size_t len) { + size_t i = 0; + __m128i has_error = _mm_setzero_si128(); + struct processed_utf_bytes previous = {.rawbytes = _mm_setzero_si128(), + .high_nibbles = _mm_setzero_si128(), + .carried_continuations = + _mm_setzero_si128()}; + if (len >= 16) { + for (; i <= len - 16; i += 16) { + __m128i current_bytes = _mm_loadu_si128((const __m128i *)(src + i)); + previous = checkUTF8Bytes(current_bytes, &previous, &has_error); + } + } + + // last part + if (i < len) { + char buffer[16]; + memset(buffer, 0, 16); + memcpy(buffer, src + i, len - i); + __m128i current_bytes = _mm_loadu_si128((const __m128i *)(buffer)); + previous = checkUTF8Bytes(current_bytes, &previous, &has_error); + } else { + has_error = + _mm_or_si128(_mm_cmpgt_epi8(previous.carried_continuations, + _mm_setr_epi8(9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 1)), + has_error); + } + + return is_all_zeros(has_error); +} + +#ifdef __AVX2__ + +/*****************************/ +static inline __m256i push_last_byte_of_a_to_b(__m256i a, __m256i b) { + return _mm256_alignr_epi8(b, _mm256_permute2x128_si256(a, b, 0x21), 15); +} + +static inline __m256i push_last_2bytes_of_a_to_b(__m256i a, __m256i b) { + return _mm256_alignr_epi8(b, _mm256_permute2x128_si256(a, b, 0x21), 14); +} + +// all byte values must be no larger than 0xF4 +static inline void avxcheckSmallerThan0xF4(__m256i current_bytes, + __m256i *has_error) { + // unsigned, saturates to 0 below max + *has_error = _mm256_or_si256( + *has_error, _mm256_subs_epu8(current_bytes, _mm256_set1_epi8(0xF4))); +} + +static inline __m256i avxcontinuationLengths(__m256i high_nibbles) { + return _mm256_shuffle_epi8( + _mm256_setr_epi8(1, 1, 1, 1, 1, 1, 1, 1, // 0xxx (ASCII) + 0, 0, 0, 0, // 10xx (continuation) + 2, 2, // 110x + 3, // 1110 + 4, // 1111, next should be 0 (not checked here) + 1, 1, 1, 1, 1, 1, 1, 1, // 0xxx (ASCII) + 0, 0, 0, 0, // 10xx (continuation) + 2, 2, // 110x + 3, // 1110 + 4 // 1111, next should be 0 (not checked here) + ), + high_nibbles); +} + +static inline __m256i avxcarryContinuations(__m256i initial_lengths, + __m256i previous_carries) { + + __m256i right1 = _mm256_subs_epu8( + push_last_byte_of_a_to_b(previous_carries, initial_lengths), + _mm256_set1_epi8(1)); + __m256i sum = _mm256_add_epi8(initial_lengths, right1); + + __m256i right2 = _mm256_subs_epu8( + push_last_2bytes_of_a_to_b(previous_carries, sum), _mm256_set1_epi8(2)); + return _mm256_add_epi8(sum, right2); +} + +static inline void avxcheckContinuations(__m256i initial_lengths, + __m256i carries, __m256i *has_error) { + + // overlap || underlap + // carry > length && length > 0 || !(carry > length) && !(length > 0) + // (carries > length) == (lengths > 0) + __m256i overunder = _mm256_cmpeq_epi8( + _mm256_cmpgt_epi8(carries, initial_lengths), + _mm256_cmpgt_epi8(initial_lengths, _mm256_setzero_si256())); + + *has_error = _mm256_or_si256(*has_error, overunder); +} + +// when 0xED is found, next byte must be no larger than 0x9F +// when 0xF4 is found, next byte must be no larger than 0x8F +// next byte must be continuation, ie sign bit is set, so signed < is ok +static inline void avxcheckFirstContinuationMax(__m256i current_bytes, + __m256i off1_current_bytes, + __m256i *has_error) { + __m256i maskED = + _mm256_cmpeq_epi8(off1_current_bytes, _mm256_set1_epi8(0xED)); + __m256i maskF4 = + _mm256_cmpeq_epi8(off1_current_bytes, _mm256_set1_epi8(0xF4)); + + __m256i badfollowED = _mm256_and_si256( + _mm256_cmpgt_epi8(current_bytes, _mm256_set1_epi8(0x9F)), maskED); + __m256i badfollowF4 = _mm256_and_si256( + _mm256_cmpgt_epi8(current_bytes, _mm256_set1_epi8(0x8F)), maskF4); + + *has_error = + _mm256_or_si256(*has_error, _mm256_or_si256(badfollowED, badfollowF4)); +} + +// map off1_hibits => error condition +// hibits off1 cur +// C => < C2 && true +// E => < E1 && < A0 +// F => < F1 && < 90 +// else false && false +static inline void avxcheckOverlong(__m256i current_bytes, + __m256i off1_current_bytes, __m256i hibits, + __m256i previous_hibits, + __m256i *has_error) { + __m256i off1_hibits = push_last_byte_of_a_to_b(previous_hibits, hibits); + __m256i initial_mins = _mm256_shuffle_epi8( + _mm256_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, // 10xx => false + 0xC2, -128, // 110x + 0xE1, // 1110 + 0xF1, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, -128, // 10xx => false + 0xC2, -128, // 110x + 0xE1, // 1110 + 0xF1), + off1_hibits); + + __m256i initial_under = _mm256_cmpgt_epi8(initial_mins, off1_current_bytes); + + __m256i second_mins = _mm256_shuffle_epi8( + _mm256_setr_epi8(-128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, // 10xx => false + 127, 127, // 110x => true + 0xA0, // 1110 + 0x90, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, -128, // 10xx => false + 127, 127, // 110x => true + 0xA0, // 1110 + 0x90), + off1_hibits); + __m256i second_under = _mm256_cmpgt_epi8(second_mins, current_bytes); + *has_error = _mm256_or_si256(*has_error, + _mm256_and_si256(initial_under, second_under)); +} + +struct avx_processed_utf_bytes { + __m256i rawbytes; + __m256i high_nibbles; + __m256i carried_continuations; +}; + +static inline void avx_count_nibbles(__m256i bytes, + struct avx_processed_utf_bytes *answer) { + answer->rawbytes = bytes; + answer->high_nibbles = + _mm256_and_si256(_mm256_srli_epi16(bytes, 4), _mm256_set1_epi8(0x0F)); +} + +// check whether the current bytes are valid UTF-8 +// at the end of the function, previous gets updated +static struct avx_processed_utf_bytes +avxcheckUTF8Bytes(__m256i current_bytes, + struct avx_processed_utf_bytes *previous, + __m256i *has_error) { + struct avx_processed_utf_bytes pb; + avx_count_nibbles(current_bytes, &pb); + + avxcheckSmallerThan0xF4(current_bytes, has_error); + + __m256i initial_lengths = avxcontinuationLengths(pb.high_nibbles); + + pb.carried_continuations = + avxcarryContinuations(initial_lengths, previous->carried_continuations); + + avxcheckContinuations(initial_lengths, pb.carried_continuations, has_error); + + __m256i off1_current_bytes = + push_last_byte_of_a_to_b(previous->rawbytes, pb.rawbytes); + avxcheckFirstContinuationMax(current_bytes, off1_current_bytes, has_error); + + avxcheckOverlong(current_bytes, off1_current_bytes, pb.high_nibbles, + previous->high_nibbles, has_error); + return pb; +} + +// check whether the current bytes are valid UTF-8 +// at the end of the function, previous gets updated +static struct avx_processed_utf_bytes +avxcheckUTF8Bytes_asciipath(__m256i current_bytes, + struct avx_processed_utf_bytes *previous, + __m256i *has_error) { + if (_mm256_testz_si256(current_bytes, + _mm256_set1_epi8(0x80))) { // fast ascii path + *has_error = _mm256_or_si256( + _mm256_cmpgt_epi8(previous->carried_continuations, + _mm256_setr_epi8(9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 1)), + *has_error); + return *previous; + } + + struct avx_processed_utf_bytes pb; + avx_count_nibbles(current_bytes, &pb); + + avxcheckSmallerThan0xF4(current_bytes, has_error); + + __m256i initial_lengths = avxcontinuationLengths(pb.high_nibbles); + + pb.carried_continuations = + avxcarryContinuations(initial_lengths, previous->carried_continuations); + + avxcheckContinuations(initial_lengths, pb.carried_continuations, has_error); + + __m256i off1_current_bytes = + push_last_byte_of_a_to_b(previous->rawbytes, pb.rawbytes); + avxcheckFirstContinuationMax(current_bytes, off1_current_bytes, has_error); + + avxcheckOverlong(current_bytes, off1_current_bytes, pb.high_nibbles, + previous->high_nibbles, has_error); + return pb; +} + +static bool validate_utf8_fast_avx_asciipath(const char *src, size_t len) { + size_t i = 0; + __m256i has_error = _mm256_setzero_si256(); + struct avx_processed_utf_bytes previous = { + .rawbytes = _mm256_setzero_si256(), + .high_nibbles = _mm256_setzero_si256(), + .carried_continuations = _mm256_setzero_si256()}; + if (len >= 32) { + for (; i <= len - 32; i += 32) { + __m256i current_bytes = _mm256_loadu_si256((const __m256i *)(src + i)); + previous = + avxcheckUTF8Bytes_asciipath(current_bytes, &previous, &has_error); + } + } + + // last part + if (i < len) { + char buffer[32]; + memset(buffer, 0, 32); + memcpy(buffer, src + i, len - i); + __m256i current_bytes = _mm256_loadu_si256((const __m256i *)(buffer)); + previous = avxcheckUTF8Bytes(current_bytes, &previous, &has_error); + } else { + has_error = _mm256_or_si256( + _mm256_cmpgt_epi8(previous.carried_continuations, + _mm256_setr_epi8(9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 1)), + has_error); + } + + return _mm256_testz_si256(has_error, has_error); +} + +static bool validate_utf8_fast_avx(const char *src, size_t len) { + size_t i = 0; + __m256i has_error = _mm256_setzero_si256(); + struct avx_processed_utf_bytes previous = { + .rawbytes = _mm256_setzero_si256(), + .high_nibbles = _mm256_setzero_si256(), + .carried_continuations = _mm256_setzero_si256()}; + if (len >= 32) { + for (; i <= len - 32; i += 32) { + __m256i current_bytes = _mm256_loadu_si256((const __m256i *)(src + i)); + previous = avxcheckUTF8Bytes(current_bytes, &previous, &has_error); + } + } + + // last part + if (i < len) { + char buffer[32]; + memset(buffer, 0, 32); + memcpy(buffer, src + i, len - i); + __m256i current_bytes = _mm256_loadu_si256((const __m256i *)(buffer)); + previous = avxcheckUTF8Bytes(current_bytes, &previous, &has_error); + } else { + has_error = _mm256_or_si256( + _mm256_cmpgt_epi8(previous.carried_continuations, + _mm256_setr_epi8(9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 1)), + has_error); + } + + return _mm256_testz_si256(has_error, has_error); +} + +#endif // __AVX2__ +#endif diff --git a/test/mini_racer_test.rb b/test/mini_racer_test.rb index 06558108..fd1c3b91 100644 --- a/test/mini_racer_test.rb +++ b/test/mini_racer_test.rb @@ -274,20 +274,62 @@ def test_return_int_max assert_equal(test_num, context.eval("test()")) end - def test_return_unknown + def test_non_utf8_argument context = MiniRacer::Context.new - test_unknown = Date.new # hits T_DATA in convert_ruby_to_v8 - context.attach("test", proc{test_unknown}) - assert_equal("Undefined Conversion", context.eval("test()")) + context.eval('function test(arg) { return "saw " + arg; }') - # clean up and start up a new context - context = nil - GC.start + bad_utf8 = "\x80" + if bad_utf8.encoding != Encoding::UTF_8 + bad_utf8 = bad_utf8.force_encoding('UTF-8') + end + + assert_equal bad_utf8.encoding, Encoding::UTF_8 + assert_equal("saw \u0080", context.call('test', bad_utf8)) + assert_equal("saw \u0080", context.call('test', "\x80".force_encoding('ASCII-8BIT'))) + assert_equal("saw \u0080", context.call('test', "\x80".force_encoding('ISO-8859-1'))) + end + + def test_conversion_to_utf8 + context = MiniRacer::Context.new + context.eval('function test(arg) { return "saw " + arg; }') + + assert_equal("saw real string", context.call("test", "real string")) + + euro = "\x80".force_encoding(Encoding::WINDOWS_1252) # euro! + assert_equal("saw \u20AC", context.call('test', euro)) + end + + def test_failed_conversion_to_utf8 + context = MiniRacer::Context.new + context.eval('function test(arg) { return "saw " + arg; }') + + # euro, but ruby doesn't do the conversion + euro = "\x80".force_encoding(Encoding::WINDOWS_1258) + begin + euro.encode(Encoding::UTF_8) + skip 'Expected ruby not to know how to do this conversion' + rescue Encoding::ConverterNotFoundError + # expected + end + + # in which case we treat it as latin1 + assert_equal("saw \u0080", context.call('test', euro)) + end + def test_non_utf_argument_deep context = MiniRacer::Context.new - test_unknown = Date.new # hits T_DATA in convert_ruby_to_v8 - context.attach("test", proc{test_unknown}) - assert_equal("Undefined Conversion", context.eval("test()")) + context.eval('function test(arg) { return "saw " + Object.keys(arg[0]) + " => " + Object.values(arg[0]); }') + + assert_equal "saw \u0081 => \u0080", context.call("test", [{"\x81" => ["\x80".force_encoding('ASCII-8BIT')]}]) + end + + def test_io_object_arg + # to_s should be called on this + context = MiniRacer::Context.new + context.eval('function test(arg) { return "saw " + arg; }') + + output = context.call("test", IO.new(1)) + assert_match /\Asaw #