summaryrefslogtreecommitdiff
path: root/vendor/httparse/src/simd/avx2.rs
diff options
context:
space:
mode:
Diffstat (limited to 'vendor/httparse/src/simd/avx2.rs')
-rw-r--r--vendor/httparse/src/simd/avx2.rs206
1 files changed, 206 insertions, 0 deletions
diff --git a/vendor/httparse/src/simd/avx2.rs b/vendor/httparse/src/simd/avx2.rs
new file mode 100644
index 00000000..2adf0a28
--- /dev/null
+++ b/vendor/httparse/src/simd/avx2.rs
@@ -0,0 +1,206 @@
+use crate::iter::Bytes;
+
+#[inline]
+#[target_feature(enable = "avx2")]
+pub unsafe fn match_uri_vectored(bytes: &mut Bytes) {
+ while bytes.as_ref().len() >= 32 {
+
+ let advance = match_url_char_32_avx(bytes.as_ref());
+
+ bytes.advance(advance);
+
+ if advance != 32 {
+ return;
+ }
+ }
+ // NOTE: use SWAR for <32B, more efficient than falling back to SSE4.2
+ super::swar::match_uri_vectored(bytes)
+}
+
+#[inline(always)]
+#[allow(non_snake_case, overflowing_literals)]
+#[allow(unused)]
+unsafe fn match_url_char_32_avx(buf: &[u8]) -> usize {
+ // NOTE: This check might be not necessary since this function is only used in
+ // `match_uri_vectored` where buffer overflow is taken care of.
+ debug_assert!(buf.len() >= 32);
+
+ #[cfg(target_arch = "x86")]
+ use core::arch::x86::*;
+ #[cfg(target_arch = "x86_64")]
+ use core::arch::x86_64::*;
+
+ // pointer to buffer
+ let ptr = buf.as_ptr();
+
+ // %x21-%x7e %x80-%xff
+ //
+ // Character ranges allowed by this function, can also be interpreted as:
+ // 33 =< (x != 127) =< 255
+ //
+ // Create a vector full of DEL (0x7f) characters.
+ let DEL: __m256i = _mm256_set1_epi8(0x7f);
+ // Create a vector full of exclamation mark (!) (0x21) characters.
+ // Used as lower threshold, characters in URLs cannot be smaller than this.
+ let LOW: __m256i = _mm256_set1_epi8(0x21);
+
+ // Load a chunk of 32 bytes from `ptr` as a vector.
+ // We can check 32 bytes in parallel at most with AVX2 since
+ // YMM registers can only have 256 bits most.
+ let dat = _mm256_lddqu_si256(ptr as *const _);
+
+ // unsigned comparison dat >= LOW
+ //
+ // `_mm256_max_epu8` creates a new vector by comparing vectors `dat` and `LOW`
+ // and picks the max. values from each for all indices.
+ // So if a byte in `dat` is <= 32, it'll be represented as 33
+ // which is the smallest valid character.
+ //
+ // Then, we compare the new vector with `dat` for equality.
+ //
+ // `_mm256_cmpeq_epi8` returns a new vector where;
+ // * matching bytes are set to 0xFF (all bits set),
+ // * nonmatching bytes are set to 0 (no bits set).
+ let low = _mm256_cmpeq_epi8(_mm256_max_epu8(dat, LOW), dat);
+ // Similar to what we did before, but now invalid characters are set to 0xFF.
+ let del = _mm256_cmpeq_epi8(dat, DEL);
+
+ // We glue the both comparisons via `_mm256_andnot_si256`.
+ //
+ // Since the representation of truthiness differ in these comparisons,
+ // we are in need of bitwise NOT to convert valid characters of `del`.
+ let bit = _mm256_andnot_si256(del, low);
+ // This creates a bitmask from the most significant bit of each byte.
+ // Simply, we're converting a vector value to scalar value here.
+ let res = _mm256_movemask_epi8(bit) as u32;
+
+ // Count trailing zeros to find the first encountered invalid character.
+ // Bitwise NOT is required once again to flip truthiness.
+ // TODO: use .trailing_ones() once MSRV >= 1.46
+ (!res).trailing_zeros() as usize
+}
+
+#[target_feature(enable = "avx2")]
+pub unsafe fn match_header_value_vectored(bytes: &mut Bytes) {
+ while bytes.as_ref().len() >= 32 {
+ let advance = match_header_value_char_32_avx(bytes.as_ref());
+ bytes.advance(advance);
+
+ if advance != 32 {
+ return;
+ }
+ }
+ // NOTE: use SWAR for <32B, more efficient than falling back to SSE4.2
+ super::swar::match_header_value_vectored(bytes)
+}
+
+#[inline(always)]
+#[allow(non_snake_case)]
+#[allow(unused)]
+unsafe fn match_header_value_char_32_avx(buf: &[u8]) -> usize {
+ debug_assert!(buf.len() >= 32);
+
+ #[cfg(target_arch = "x86")]
+ use core::arch::x86::*;
+ #[cfg(target_arch = "x86_64")]
+ use core::arch::x86_64::*;
+
+ let ptr = buf.as_ptr();
+
+ // %x09 %x20-%x7e %x80-%xff
+ // Create a vector full of horizontal tab (\t) (0x09) characters.
+ let TAB: __m256i = _mm256_set1_epi8(0x09);
+ // Create a vector full of DEL (0x7f) characters.
+ let DEL: __m256i = _mm256_set1_epi8(0x7f);
+ // Create a vector full of space (0x20) characters.
+ let LOW: __m256i = _mm256_set1_epi8(0x20);
+
+ // Load a chunk of 32 bytes from `ptr` as a vector.
+ let dat = _mm256_lddqu_si256(ptr as *const _);
+
+ // unsigned comparison dat >= LOW
+ //
+ // Same as what we do in `match_url_char_32_avx`.
+ // This time the lower threshold is set to space character though.
+ let low = _mm256_cmpeq_epi8(_mm256_max_epu8(dat, LOW), dat);
+ // Check if `dat` includes `TAB` characters.
+ let tab = _mm256_cmpeq_epi8(dat, TAB);
+ // Check if `dat` includes `DEL` characters.
+ let del = _mm256_cmpeq_epi8(dat, DEL);
+
+ // Combine all comparisons together, notice that we're also using OR
+ // to connect `low` and `tab` but flip bits of `del`.
+ //
+ // In the end, this is simply:
+ // ~del & (low | tab)
+ let bit = _mm256_andnot_si256(del, _mm256_or_si256(low, tab));
+ // This creates a bitmask from the most significant bit of each byte.
+ // Creates a scalar value from vector value.
+ let res = _mm256_movemask_epi8(bit) as u32;
+
+ // Count trailing zeros to find the first encountered invalid character.
+ // Bitwise NOT is required once again to flip truthiness.
+ // TODO: use .trailing_ones() once MSRV >= 1.46
+ (!res).trailing_zeros() as usize
+}
+
+#[test]
+fn avx2_code_matches_uri_chars_table() {
+ if !is_x86_feature_detected!("avx2") {
+ return;
+ }
+
+ #[allow(clippy::undocumented_unsafe_blocks)]
+ unsafe {
+ assert!(byte_is_allowed(b'_', match_uri_vectored));
+
+ for (b, allowed) in crate::URI_MAP.iter().cloned().enumerate() {
+ assert_eq!(
+ byte_is_allowed(b as u8, match_uri_vectored), allowed,
+ "byte_is_allowed({:?}) should be {:?}", b, allowed,
+ );
+ }
+ }
+}
+
+#[test]
+fn avx2_code_matches_header_value_chars_table() {
+ if !is_x86_feature_detected!("avx2") {
+ return;
+ }
+
+ #[allow(clippy::undocumented_unsafe_blocks)]
+ unsafe {
+ assert!(byte_is_allowed(b'_', match_header_value_vectored));
+
+ for (b, allowed) in crate::HEADER_VALUE_MAP.iter().cloned().enumerate() {
+ assert_eq!(
+ byte_is_allowed(b as u8, match_header_value_vectored), allowed,
+ "byte_is_allowed({:?}) should be {:?}", b, allowed,
+ );
+ }
+ }
+}
+
+#[cfg(test)]
+unsafe fn byte_is_allowed(byte: u8, f: unsafe fn(bytes: &mut Bytes<'_>)) -> bool {
+ let slice = [
+ b'_', b'_', b'_', b'_',
+ b'_', b'_', b'_', b'_',
+ b'_', b'_', b'_', b'_',
+ b'_', b'_', b'_', b'_',
+ b'_', b'_', b'_', b'_',
+ b'_', b'_', b'_', b'_',
+ b'_', b'_', byte, b'_',
+ b'_', b'_', b'_', b'_',
+ ];
+ let mut bytes = Bytes::new(&slice);
+
+ f(&mut bytes);
+
+ match bytes.pos() {
+ 32 => true,
+ 26 => false,
+ _ => unreachable!(),
+ }
+}