/*
- * Copyright Andrey Semashev 2007 - 2015.
+ * Copyright Andrey Semashev 2007 - 2021.
* Distributed under the Boost Software License, Version 1.0.
* (See accompanying file LICENSE_1_0.txt or copy at
* http://www.boost.org/LICENSE_1_0.txt)
stride = packs_per_stride * 16
};
+template< typename CharT >
+BOOST_FORCEINLINE void store_characters(__m128i mm_chars, CharT* buf)
+{
+ switch (sizeof(CharT))
+ {
+ case 1:
+ _mm_store_si128(reinterpret_cast< __m128i* >(buf), mm_chars);
+ break;
+
+ case 2:
+ {
+ __m128i mm_0 = _mm_setzero_si128();
+ _mm_store_si128(reinterpret_cast< __m128i* >(buf), _mm_unpacklo_epi8(mm_chars, mm_0));
+ _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 1, _mm_unpackhi_epi8(mm_chars, mm_0));
+ }
+ break;
+
+ case 4:
+ {
+ __m128i mm_0 = _mm_setzero_si128();
+ __m128i mm = _mm_unpacklo_epi8(mm_chars, mm_0);
+ _mm_store_si128(reinterpret_cast< __m128i* >(buf), _mm_unpacklo_epi16(mm, mm_0));
+ _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 1, _mm_unpackhi_epi16(mm, mm_0));
+ mm = _mm_unpackhi_epi8(mm_chars, mm_0);
+ _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 2, _mm_unpacklo_epi16(mm, mm_0));
+ _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 3, _mm_unpackhi_epi16(mm, mm_0));
+ }
+ break;
+ }
+}
+
union xmm_constant
{
uint8_t as_bytes[16];
static const xmm_constant mm_shuffle_pattern2 = {{ 0, 1, 0x80, 2, 3, 0x80, 4, 5, 0x80, 6, 7, 0x80, 8, 9, 0x80, 10 }};
static const xmm_constant mm_shuffle_pattern3 = {{ 5, 0x80, 6, 7, 0x80, 8, 9, 0x80, 10, 11, 0x80, 12, 13, 0x80, 14, 15 }};
+#if defined(BOOST_LOG_AUX_X86_64)
+
+// x86-64 architecture has more registers which we can utilize to pass constants
+#define BOOST_LOG_AUX_MM_CONSTANT_ARGS_DECL __m128i mm_15, __m128i mm_char_space,
+#define BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_15, mm_char_space,
+#define BOOST_LOG_AUX_MM_CONSTANTS \
+ const __m128i mm_15 = _mm_set1_epi32(0x0F0F0F0F);\
+ const __m128i mm_char_space = _mm_set1_epi32(0x20202020);
+
+#else
+
+// MSVC in 32-bit mode is not able to pass all constants to dump_pack, and is also not able to align them on the stack, so we have to fetch them from global constants
+static const xmm_constant mm_15 = {{ 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F }};
+static const xmm_constant mm_char_space = {{ 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 }};
+#define BOOST_LOG_AUX_MM_CONSTANT_ARGS_DECL
+#define BOOST_LOG_AUX_MM_CONSTANT_ARGS
+#define BOOST_LOG_AUX_MM_CONSTANTS
+
+#endif
+
+//! Dumps a pack of input data into a string of 8 bit ASCII characters
+static BOOST_FORCEINLINE void dump_pack
+(
+ BOOST_LOG_AUX_MM_CONSTANT_ARGS_DECL
+ __m128i mm_char_table, __m128i mm_input,
+ __m128i& mm_output1, __m128i& mm_output2, __m128i& mm_output3
+)
+{
+ // Split half-bytes
+ __m128i mm_input_hi = _mm_and_si128(_mm_srli_epi16(mm_input, 4), mm_15);
+ __m128i mm_input_lo = _mm_and_si128(mm_input, mm_15);
+
+ // Stringize each of the halves
+ mm_input_hi = _mm_shuffle_epi8(mm_char_table, mm_input_hi);
+ mm_input_lo = _mm_shuffle_epi8(mm_char_table, mm_input_lo);
+
+ // Join them back together
+ __m128i mm_1 = _mm_unpacklo_epi8(mm_input_hi, mm_input_lo);
+ __m128i mm_2 = _mm_unpackhi_epi8(mm_input_hi, mm_input_lo);
+
+ // Insert spaces between stringized bytes:
+ // |0123456789abcdef|0123456789abcdef|
+ // | 01 23 45 67 89 |ab cd ef 01 23 4|5 67 89 ab cd ef|
+ mm_output1 = _mm_shuffle_epi8(mm_1, mm_shuffle_pattern1.as_mm);
+ mm_output2 = _mm_shuffle_epi8(_mm_alignr_epi8(mm_2, mm_1, 10), mm_shuffle_pattern2.as_mm);
+ mm_output3 = _mm_shuffle_epi8(mm_2, mm_shuffle_pattern3.as_mm);
+
+ mm_output1 = _mm_max_epu8(mm_output1, mm_char_space);
+ mm_output2 = _mm_max_epu8(mm_output2, mm_char_space);
+ mm_output3 = _mm_max_epu8(mm_output3, mm_char_space);
+}
+
+template< typename CharT >
+BOOST_FORCEINLINE void dump_data_ssse3(const void* data, std::size_t size, std::basic_ostream< CharT >& strm)
+{
+ typedef CharT char_type;
+
+ char_type buf_storage[stride * 3u + 16u];
+ // Align the temporary buffer at 16 bytes
+ char_type* const buf = reinterpret_cast< char_type* >((uint8_t*)buf_storage + (16u - (((uintptr_t)(char_type*)buf_storage) & 15u)));
+ char_type* buf_begin = buf + 1u; // skip the first space of the first chunk
+ char_type* buf_end = buf + stride * 3u;
+
+ const char* const char_table = g_hex_char_table[(strm.flags() & std::ios_base::uppercase) != 0];
+ const __m128i mm_char_table =
+#if !defined(BOOST_NO_ALIGNMENT)
+ _mm_load_si128(reinterpret_cast< const __m128i* >(char_table));
+#else
+ _mm_lddqu_si128(reinterpret_cast< const __m128i* >(char_table));
+#endif
+
+ // First, check the input alignment
+ const uint8_t* p = static_cast< const uint8_t* >(data);
+ const std::size_t prealign_size = ((16u - ((uintptr_t)p & 15u)) & 15u);
+ if (BOOST_UNLIKELY(prealign_size > 0))
+ {
+ __m128i mm_input = _mm_lddqu_si128(reinterpret_cast< const __m128i* >(p));
+ BOOST_LOG_AUX_MM_CONSTANTS
+
+ __m128i mm_output1, mm_output2, mm_output3;
+ dump_pack(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_table, mm_input, mm_output1, mm_output2, mm_output3);
+
+ store_characters(mm_output1, buf);
+ store_characters(mm_output2, buf + 16u);
+ store_characters(mm_output3, buf + 32u);
+
+ strm.write(buf_begin, prealign_size * 3u - 1u);
+
+ buf_begin = buf;
+ size -= prealign_size;
+ p += prealign_size;
+ }
+
+ const std::size_t stride_count = size / stride;
+ std::size_t tail_size = size % stride;
+ for (std::size_t i = 0; i < stride_count; ++i)
+ {
+ char_type* b = buf;
+ BOOST_LOG_AUX_MM_CONSTANTS
+
+ for (unsigned int j = 0; j < packs_per_stride; ++j, b += 3u * 16u, p += 16u)
+ {
+ __m128i mm_input = _mm_load_si128(reinterpret_cast< const __m128i* >(p));
+ __m128i mm_output1, mm_output2, mm_output3;
+ dump_pack(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_table, mm_input, mm_output1, mm_output2, mm_output3);
+
+ store_characters(mm_output1, b);
+ store_characters(mm_output2, b + 16u);
+ store_characters(mm_output3, b + 32u);
+ }
+
+ strm.write(buf_begin, buf_end - buf_begin);
+ buf_begin = buf;
+ }
+
+ if (BOOST_UNLIKELY(tail_size > 0))
+ {
+ char_type* b = buf;
+ while (tail_size >= 16u)
+ {
+ __m128i mm_input = _mm_load_si128(reinterpret_cast< const __m128i* >(p));
+ BOOST_LOG_AUX_MM_CONSTANTS
+
+ __m128i mm_output1, mm_output2, mm_output3;
+ dump_pack(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_table, mm_input, mm_output1, mm_output2, mm_output3);
+
+ store_characters(mm_output1, b);
+ store_characters(mm_output2, b + 16u);
+ store_characters(mm_output3, b + 32u);
+
+ b += 3u * 16u;
+ p += 16u;
+ tail_size -= 16u;
+ }
+
+ for (unsigned int i = 0; i < tail_size; ++i, ++p, b += 3u)
+ {
+ uint32_t n = *p;
+ b[0] = static_cast< char_type >(' ');
+ b[1] = static_cast< char_type >(char_table[n >> 4]);
+ b[2] = static_cast< char_type >(char_table[n & 0x0F]);
+ }
+
+ strm.write(buf_begin, b - buf_begin);
+ }
+}
+
+#undef BOOST_LOG_AUX_MM_CONSTANT_ARGS_DECL
+#undef BOOST_LOG_AUX_MM_CONSTANT_ARGS
+#undef BOOST_LOG_AUX_MM_CONSTANTS
+
+
#if defined(BOOST_LOG_AUX_X86_64)
// x86-64 architecture has more registers which we can utilize to pass constants
#else
// MSVC in 32-bit mode is not able to pass all constants to dump_pack, and is also not able to align them on the stack, so we have to fetch them from global constants
-static const xmm_constant mm_15 = {{ 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F, 0x0F }};
static const xmm_constant mm_9 = {{ 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09, 0x09 }};
static const xmm_constant mm_char_0 = {{ 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30, 0x30 }};
-static const xmm_constant mm_char_space = {{ 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 }};
#define BOOST_LOG_AUX_MM_CONSTANT_ARGS_DECL
#define BOOST_LOG_AUX_MM_CONSTANT_ARGS
#define BOOST_LOG_AUX_MM_CONSTANTS
#endif
//! Dumps a pack of input data into a string of 8 bit ASCII characters
-static BOOST_FORCEINLINE void dump_pack
+static BOOST_FORCEINLINE void dump_pack_slow_pshufb
(
BOOST_LOG_AUX_MM_CONSTANT_ARGS_DECL
__m128i mm_char_10_to_a, __m128i mm_input,
}
template< typename CharT >
-BOOST_FORCEINLINE void store_characters(__m128i mm_chars, CharT* buf)
-{
- switch (sizeof(CharT))
- {
- case 1:
- _mm_store_si128(reinterpret_cast< __m128i* >(buf), mm_chars);
- break;
-
- case 2:
- {
- __m128i mm_0 = _mm_setzero_si128();
- _mm_store_si128(reinterpret_cast< __m128i* >(buf), _mm_unpacklo_epi8(mm_chars, mm_0));
- _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 1, _mm_unpackhi_epi8(mm_chars, mm_0));
- }
- break;
-
- case 4:
- {
- __m128i mm_0 = _mm_setzero_si128();
- __m128i mm = _mm_unpacklo_epi8(mm_chars, mm_0);
- _mm_store_si128(reinterpret_cast< __m128i* >(buf), _mm_unpacklo_epi16(mm, mm_0));
- _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 1, _mm_unpackhi_epi16(mm, mm_0));
- mm = _mm_unpackhi_epi8(mm_chars, mm_0);
- _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 2, _mm_unpacklo_epi16(mm, mm_0));
- _mm_store_si128(reinterpret_cast< __m128i* >(buf) + 3, _mm_unpackhi_epi16(mm, mm_0));
- }
- break;
- }
-}
-
-template< typename CharT >
-BOOST_FORCEINLINE void dump_data_ssse3(const void* data, std::size_t size, std::basic_ostream< CharT >& strm)
+BOOST_FORCEINLINE void dump_data_ssse3_slow_pshufb(const void* data, std::size_t size, std::basic_ostream< CharT >& strm)
{
typedef CharT char_type;
BOOST_LOG_AUX_MM_CONSTANTS
__m128i mm_output1, mm_output2, mm_output3;
- dump_pack(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_10_to_a, mm_input, mm_output1, mm_output2, mm_output3);
+ dump_pack_slow_pshufb(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_10_to_a, mm_input, mm_output1, mm_output2, mm_output3);
store_characters(mm_output1, buf);
store_characters(mm_output2, buf + 16u);
{
__m128i mm_input = _mm_load_si128(reinterpret_cast< const __m128i* >(p));
__m128i mm_output1, mm_output2, mm_output3;
- dump_pack(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_10_to_a, mm_input, mm_output1, mm_output2, mm_output3);
+ dump_pack_slow_pshufb(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_10_to_a, mm_input, mm_output1, mm_output2, mm_output3);
store_characters(mm_output1, b);
store_characters(mm_output2, b + 16u);
BOOST_LOG_AUX_MM_CONSTANTS
__m128i mm_output1, mm_output2, mm_output3;
- dump_pack(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_10_to_a, mm_input, mm_output1, mm_output2, mm_output3);
+ dump_pack_slow_pshufb(BOOST_LOG_AUX_MM_CONSTANT_ARGS mm_char_10_to_a, mm_input, mm_output1, mm_output2, mm_output3);
store_characters(mm_output1, b);
store_characters(mm_output2, b + 16u);
}
}
+#undef BOOST_LOG_AUX_MM_CONSTANT_ARGS_DECL
+#undef BOOST_LOG_AUX_MM_CONSTANT_ARGS
+#undef BOOST_LOG_AUX_MM_CONSTANTS
+
} // namespace
void dump_data_char_ssse3(const void* data, std::size_t size, std::basic_ostream< char >& strm)
}
#endif
+void dump_data_char_ssse3_slow_pshufb(const void* data, std::size_t size, std::basic_ostream< char >& strm)
+{
+ if (size >= 16)
+ {
+ dump_data_ssse3_slow_pshufb(data, size, strm);
+ }
+ else
+ {
+ dump_data_generic(data, size, strm);
+ }
+}
+
+void dump_data_wchar_ssse3_slow_pshufb(const void* data, std::size_t size, std::basic_ostream< wchar_t >& strm)
+{
+ if (size >= 16)
+ {
+ dump_data_ssse3_slow_pshufb(data, size, strm);
+ }
+ else
+ {
+ dump_data_generic(data, size, strm);
+ }
+}
+
+#if !defined(BOOST_NO_CXX11_CHAR16_T)
+void dump_data_char16_ssse3_slow_pshufb(const void* data, std::size_t size, std::basic_ostream< char16_t >& strm)
+{
+ if (size >= 16)
+ {
+ dump_data_ssse3_slow_pshufb(data, size, strm);
+ }
+ else
+ {
+ dump_data_generic(data, size, strm);
+ }
+}
+#endif
+
+#if !defined(BOOST_NO_CXX11_CHAR32_T)
+void dump_data_char32_ssse3_slow_pshufb(const void* data, std::size_t size, std::basic_ostream< char32_t >& strm)
+{
+ if (size >= 16)
+ {
+ dump_data_ssse3_slow_pshufb(data, size, strm);
+ }
+ else
+ {
+ dump_data_generic(data, size, strm);
+ }
+}
+#endif
+
} // namespace aux
BOOST_LOG_CLOSE_NAMESPACE // namespace log