deflate-ruby 1.0.1 → 1.0.2

data/ext/deflate_ruby/libdeflate/arm/crc32_pmull_wide.h

@@ -0,0 +1,226 @@
+/*
+ * arm/crc32_pmull_wide.h - gzip CRC-32 with PMULL (extra-wide version)
+ *
+ * Copyright 2022 Eric Biggers
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/*
+ * This file is a "template" for instantiating PMULL-based crc32_arm functions.
+ * The "parameters" are:
+ *
+ * SUFFIX:
+ *	Name suffix to append to all instantiated functions.
+ * ATTRIBUTES:
+ *	Target function attributes to use.
+ * ENABLE_EOR3:
+ *	Use the eor3 instruction (from the sha3 extension).
+ *
+ * This is the extra-wide version; it uses an unusually large stride length of
+ * 12, and it assumes that crc32 instructions are available too.  It's intended
+ * for powerful CPUs that support both pmull and crc32 instructions, but where
+ * throughput of pmull and xor (given enough instructions issued in parallel) is
+ * significantly higher than that of crc32, thus making the crc32 instructions
+ * (counterintuitively) not actually the fastest way to compute the CRC-32.  The
+ * Apple M1 processor is an example of such a CPU.
+ */
+
+#include "crc32_pmull_helpers.h"
+
+static ATTRIBUTES u32
+ADD_SUFFIX(crc32_arm)(u32 crc, const u8 *p, size_t len)
+{
+	uint8x16_t v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11;
+
+	if (len < 3 * 192) {
+		static const u64 _aligned_attribute(16) mults[3][2] = {
+			{ CRC32_X543_MODG, CRC32_X479_MODG }, /* 4 vecs */
+			{ CRC32_X287_MODG, CRC32_X223_MODG }, /* 2 vecs */
+			{ CRC32_X159_MODG, CRC32_X95_MODG },  /* 1 vecs */
+		};
+		poly64x2_t multipliers_4, multipliers_2, multipliers_1;
+
+		if (len < 64)
+			goto tail;
+		multipliers_4 = load_multipliers(mults[0]);
+		multipliers_2 = load_multipliers(mults[1]);
+		multipliers_1 = load_multipliers(mults[2]);
+		/*
+		 * Short length; don't bother aligning the pointer, and fold
+		 * 64 bytes (4 vectors) at a time, at most.
+		 */
+		v0 = veorq_u8(vld1q_u8(p + 0), u32_to_bytevec(crc));
+		v1 = vld1q_u8(p + 16);
+		v2 = vld1q_u8(p + 32);
+		v3 = vld1q_u8(p + 48);
+		p += 64;
+		len -= 64;
+		while (len >= 64) {
+			v0 = fold_vec(v0, vld1q_u8(p + 0), multipliers_4);
+			v1 = fold_vec(v1, vld1q_u8(p + 16), multipliers_4);
+			v2 = fold_vec(v2, vld1q_u8(p + 32), multipliers_4);
+			v3 = fold_vec(v3, vld1q_u8(p + 48), multipliers_4);
+			p += 64;
+			len -= 64;
+		}
+		v0 = fold_vec(v0, v2, multipliers_2);
+		v1 = fold_vec(v1, v3, multipliers_2);
+		if (len >= 32) {
+			v0 = fold_vec(v0, vld1q_u8(p + 0), multipliers_2);
+			v1 = fold_vec(v1, vld1q_u8(p + 16), multipliers_2);
+			p += 32;
+			len -= 32;
+		}
+		v0 = fold_vec(v0, v1, multipliers_1);
+	} else {
+		static const u64 _aligned_attribute(16) mults[4][2] = {
+			{ CRC32_X1567_MODG, CRC32_X1503_MODG }, /* 12 vecs */
+			{ CRC32_X799_MODG, CRC32_X735_MODG },   /* 6 vecs */
+			{ CRC32_X415_MODG, CRC32_X351_MODG },   /* 3 vecs */
+			{ CRC32_X159_MODG, CRC32_X95_MODG },    /* 1 vecs */
+		};
+		const poly64x2_t multipliers_12 = load_multipliers(mults[0]);
+		const poly64x2_t multipliers_6 = load_multipliers(mults[1]);
+		const poly64x2_t multipliers_3 = load_multipliers(mults[2]);
+		const poly64x2_t multipliers_1 = load_multipliers(mults[3]);
+		const size_t align = -(uintptr_t)p & 15;
+		const uint8x16_t *vp;
+
+		/* Align p to the next 16-byte boundary. */
+		if (align) {
+			if (align & 1)
+				crc = __crc32b(crc, *p++);
+			if (align & 2) {
+				crc = __crc32h(crc, le16_bswap(*(u16 *)p));
+				p += 2;
+			}
+			if (align & 4) {
+				crc = __crc32w(crc, le32_bswap(*(u32 *)p));
+				p += 4;
+			}
+			if (align & 8) {
+				crc = __crc32d(crc, le64_bswap(*(u64 *)p));
+				p += 8;
+			}
+			len -= align;
+		}
+		vp = (const uint8x16_t *)p;
+		v0 = veorq_u8(*vp++, u32_to_bytevec(crc));
+		v1 = *vp++;
+		v2 = *vp++;
+		v3 = *vp++;
+		v4 = *vp++;
+		v5 = *vp++;
+		v6 = *vp++;
+		v7 = *vp++;
+		v8 = *vp++;
+		v9 = *vp++;
+		v10 = *vp++;
+		v11 = *vp++;
+		len -= 192;
+		/* Fold 192 bytes (12 vectors) at a time. */
+		do {
+			v0 = fold_vec(v0, *vp++, multipliers_12);
+			v1 = fold_vec(v1, *vp++, multipliers_12);
+			v2 = fold_vec(v2, *vp++, multipliers_12);
+			v3 = fold_vec(v3, *vp++, multipliers_12);
+			v4 = fold_vec(v4, *vp++, multipliers_12);
+			v5 = fold_vec(v5, *vp++, multipliers_12);
+			v6 = fold_vec(v6, *vp++, multipliers_12);
+			v7 = fold_vec(v7, *vp++, multipliers_12);
+			v8 = fold_vec(v8, *vp++, multipliers_12);
+			v9 = fold_vec(v9, *vp++, multipliers_12);
+			v10 = fold_vec(v10, *vp++, multipliers_12);
+			v11 = fold_vec(v11, *vp++, multipliers_12);
+			len -= 192;
+		} while (len >= 192);
+
+		/*
+		 * Fewer than 192 bytes left.  Fold v0-v11 down to just v0,
+		 * while processing up to 144 more bytes.
+		 */
+		v0 = fold_vec(v0, v6, multipliers_6);
+		v1 = fold_vec(v1, v7, multipliers_6);
+		v2 = fold_vec(v2, v8, multipliers_6);
+		v3 = fold_vec(v3, v9, multipliers_6);
+		v4 = fold_vec(v4, v10, multipliers_6);
+		v5 = fold_vec(v5, v11, multipliers_6);
+		if (len >= 96) {
+			v0 = fold_vec(v0, *vp++, multipliers_6);
+			v1 = fold_vec(v1, *vp++, multipliers_6);
+			v2 = fold_vec(v2, *vp++, multipliers_6);
+			v3 = fold_vec(v3, *vp++, multipliers_6);
+			v4 = fold_vec(v4, *vp++, multipliers_6);
+			v5 = fold_vec(v5, *vp++, multipliers_6);
+			len -= 96;
+		}
+		v0 = fold_vec(v0, v3, multipliers_3);
+		v1 = fold_vec(v1, v4, multipliers_3);
+		v2 = fold_vec(v2, v5, multipliers_3);
+		if (len >= 48) {
+			v0 = fold_vec(v0, *vp++, multipliers_3);
+			v1 = fold_vec(v1, *vp++, multipliers_3);
+			v2 = fold_vec(v2, *vp++, multipliers_3);
+			len -= 48;
+		}
+		v0 = fold_vec(v0, v1, multipliers_1);
+		v0 = fold_vec(v0, v2, multipliers_1);
+		p = (const u8 *)vp;
+	}
+	/* Reduce 128 to 32 bits using crc32 instructions. */
+	crc = __crc32d(0, vgetq_lane_u64(vreinterpretq_u64_u8(v0), 0));
+	crc = __crc32d(crc, vgetq_lane_u64(vreinterpretq_u64_u8(v0), 1));
+tail:
+	/* Finish up the remainder using crc32 instructions. */
+	if (len & 32) {
+		crc = __crc32d(crc, get_unaligned_le64(p + 0));
+		crc = __crc32d(crc, get_unaligned_le64(p + 8));
+		crc = __crc32d(crc, get_unaligned_le64(p + 16));
+		crc = __crc32d(crc, get_unaligned_le64(p + 24));
+		p += 32;
+	}
+	if (len & 16) {
+		crc = __crc32d(crc, get_unaligned_le64(p + 0));
+		crc = __crc32d(crc, get_unaligned_le64(p + 8));
+		p += 16;
+	}
+	if (len & 8) {
+		crc = __crc32d(crc, get_unaligned_le64(p));
+		p += 8;
+	}
+	if (len & 4) {
+		crc = __crc32w(crc, get_unaligned_le32(p));
+		p += 4;
+	}
+	if (len & 2) {
+		crc = __crc32h(crc, get_unaligned_le16(p));
+		p += 2;
+	}
+	if (len & 1)
+		crc = __crc32b(crc, *p);
+	return crc;
+}
+
+#undef SUFFIX
+#undef ATTRIBUTES
+#undef ENABLE_EOR3

data/ext/deflate_ruby/libdeflate/bt_matchfinder.h

@@ -0,0 +1,342 @@
+/*
+ * bt_matchfinder.h - Lempel-Ziv matchfinding with a hash table of binary trees
+ *
+ * Copyright 2016 Eric Biggers
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * ----------------------------------------------------------------------------
+ *
+ * This is a Binary Trees (bt) based matchfinder.
+ *
+ * The main data structure is a hash table where each hash bucket contains a
+ * binary tree of sequences whose first 4 bytes share the same hash code.  Each
+ * sequence is identified by its starting position in the input buffer.  Each
+ * binary tree is always sorted such that each left child represents a sequence
+ * lexicographically lesser than its parent and each right child represents a
+ * sequence lexicographically greater than its parent.
+ *
+ * The algorithm processes the input buffer sequentially.  At each byte
+ * position, the hash code of the first 4 bytes of the sequence beginning at
+ * that position (the sequence being matched against) is computed.  This
+ * identifies the hash bucket to use for that position.  Then, a new binary tree
+ * node is created to represent the current sequence.  Then, in a single tree
+ * traversal, the hash bucket's binary tree is searched for matches and is
+ * re-rooted at the new node.
+ *
+ * Compared to the simpler algorithm that uses linked lists instead of binary
+ * trees (see hc_matchfinder.h), the binary tree version gains more information
+ * at each node visitation.  Ideally, the binary tree version will examine only
+ * 'log(n)' nodes to find the same matches that the linked list version will
+ * find by examining 'n' nodes.  In addition, the binary tree version can
+ * examine fewer bytes at each node by taking advantage of the common prefixes
+ * that result from the sort order, whereas the linked list version may have to
+ * examine up to the full length of the match at each node.
+ *
+ * However, it is not always best to use the binary tree version.  It requires
+ * nearly twice as much memory as the linked list version, and it takes time to
+ * keep the binary trees sorted, even at positions where the compressor does not
+ * need matches.  Generally, when doing fast compression on small buffers,
+ * binary trees are the wrong approach.  They are best suited for thorough
+ * compression and/or large buffers.
+ *
+ * ----------------------------------------------------------------------------
+ */
+
+#ifndef LIB_BT_MATCHFINDER_H
+#define LIB_BT_MATCHFINDER_H
+
+#include "matchfinder_common.h"
+
+#define BT_MATCHFINDER_HASH3_ORDER 16
+#define BT_MATCHFINDER_HASH3_WAYS  2
+#define BT_MATCHFINDER_HASH4_ORDER 16
+
+#define BT_MATCHFINDER_TOTAL_HASH_SIZE		\
+	(((1UL << BT_MATCHFINDER_HASH3_ORDER) * BT_MATCHFINDER_HASH3_WAYS + \
+	  (1UL << BT_MATCHFINDER_HASH4_ORDER)) * sizeof(mf_pos_t))
+
+/* Representation of a match found by the bt_matchfinder  */
+struct lz_match {
+
+	/* The number of bytes matched.  */
+	u16 length;
+
+	/* The offset back from the current position that was matched.  */
+	u16 offset;
+};
+
+struct MATCHFINDER_ALIGNED bt_matchfinder {
+
+	/* The hash table for finding length 3 matches  */
+	mf_pos_t hash3_tab[1UL << BT_MATCHFINDER_HASH3_ORDER][BT_MATCHFINDER_HASH3_WAYS];
+
+	/* The hash table which contains the roots of the binary trees for
+	 * finding length 4+ matches  */
+	mf_pos_t hash4_tab[1UL << BT_MATCHFINDER_HASH4_ORDER];
+
+	/* The child node references for the binary trees.  The left and right
+	 * children of the node for the sequence with position 'pos' are
+	 * 'child_tab[pos * 2]' and 'child_tab[pos * 2 + 1]', respectively.  */
+	mf_pos_t child_tab[2UL * MATCHFINDER_WINDOW_SIZE];
+};
+
+/* Prepare the matchfinder for a new input buffer.  */
+static forceinline void
+bt_matchfinder_init(struct bt_matchfinder *mf)
+{
+	STATIC_ASSERT(BT_MATCHFINDER_TOTAL_HASH_SIZE %
+		      MATCHFINDER_SIZE_ALIGNMENT == 0);
+
+	matchfinder_init((mf_pos_t *)mf, BT_MATCHFINDER_TOTAL_HASH_SIZE);
+}
+
+static forceinline void
+bt_matchfinder_slide_window(struct bt_matchfinder *mf)
+{
+	STATIC_ASSERT(sizeof(*mf) % MATCHFINDER_SIZE_ALIGNMENT == 0);
+
+	matchfinder_rebase((mf_pos_t *)mf, sizeof(*mf));
+}
+
+static forceinline mf_pos_t *
+bt_left_child(struct bt_matchfinder *mf, s32 node)
+{
+	return &mf->child_tab[2 * (node & (MATCHFINDER_WINDOW_SIZE - 1)) + 0];
+}
+
+static forceinline mf_pos_t *
+bt_right_child(struct bt_matchfinder *mf, s32 node)
+{
+	return &mf->child_tab[2 * (node & (MATCHFINDER_WINDOW_SIZE - 1)) + 1];
+}
+
+/* The minimum permissible value of 'max_len' for bt_matchfinder_get_matches()
+ * and bt_matchfinder_skip_byte().  There must be sufficiently many bytes
+ * remaining to load a 32-bit integer from the *next* position.  */
+#define BT_MATCHFINDER_REQUIRED_NBYTES	5
+
+/* Advance the binary tree matchfinder by one byte, optionally recording
+ * matches.  @record_matches should be a compile-time constant.  */
+static forceinline struct lz_match *
+bt_matchfinder_advance_one_byte(struct bt_matchfinder * const mf,
+				const u8 * const in_base,
+				const ptrdiff_t cur_pos,
+				const u32 max_len,
+				const u32 nice_len,
+				const u32 max_search_depth,
+				u32 * const next_hashes,
+				struct lz_match *lz_matchptr,
+				const bool record_matches)
+{
+	const u8 *in_next = in_base + cur_pos;
+	u32 depth_remaining = max_search_depth;
+	const s32 cutoff = cur_pos - MATCHFINDER_WINDOW_SIZE;
+	u32 next_hashseq;
+	u32 hash3;
+	u32 hash4;
+	s32 cur_node;
+#if BT_MATCHFINDER_HASH3_WAYS >= 2
+	s32 cur_node_2;
+#endif
+	const u8 *matchptr;
+	mf_pos_t *pending_lt_ptr, *pending_gt_ptr;
+	u32 best_lt_len, best_gt_len;
+	u32 len;
+	u32 best_len = 3;
+
+	STATIC_ASSERT(BT_MATCHFINDER_HASH3_WAYS >= 1 &&
+		      BT_MATCHFINDER_HASH3_WAYS <= 2);
+
+	next_hashseq = get_unaligned_le32(in_next + 1);
+
+	hash3 = next_hashes[0];
+	hash4 = next_hashes[1];
+
+	next_hashes[0] = lz_hash(next_hashseq & 0xFFFFFF, BT_MATCHFINDER_HASH3_ORDER);
+	next_hashes[1] = lz_hash(next_hashseq, BT_MATCHFINDER_HASH4_ORDER);
+	prefetchw(&mf->hash3_tab[next_hashes[0]]);
+	prefetchw(&mf->hash4_tab[next_hashes[1]]);
+
+	cur_node = mf->hash3_tab[hash3][0];
+	mf->hash3_tab[hash3][0] = cur_pos;
+#if BT_MATCHFINDER_HASH3_WAYS >= 2
+	cur_node_2 = mf->hash3_tab[hash3][1];
+	mf->hash3_tab[hash3][1] = cur_node;
+#endif
+	if (record_matches && cur_node > cutoff) {
+		u32 seq3 = load_u24_unaligned(in_next);
+		if (seq3 == load_u24_unaligned(&in_base[cur_node])) {
+			lz_matchptr->length = 3;
+			lz_matchptr->offset = in_next - &in_base[cur_node];
+			lz_matchptr++;
+		}
+	#if BT_MATCHFINDER_HASH3_WAYS >= 2
+		else if (cur_node_2 > cutoff &&
+			seq3 == load_u24_unaligned(&in_base[cur_node_2]))
+		{
+			lz_matchptr->length = 3;
+			lz_matchptr->offset = in_next - &in_base[cur_node_2];
+			lz_matchptr++;
+		}
+	#endif
+	}
+
+	cur_node = mf->hash4_tab[hash4];
+	mf->hash4_tab[hash4] = cur_pos;
+
+	pending_lt_ptr = bt_left_child(mf, cur_pos);
+	pending_gt_ptr = bt_right_child(mf, cur_pos);
+
+	if (cur_node <= cutoff) {
+		*pending_lt_ptr = MATCHFINDER_INITVAL;
+		*pending_gt_ptr = MATCHFINDER_INITVAL;
+		return lz_matchptr;
+	}
+
+	best_lt_len = 0;
+	best_gt_len = 0;
+	len = 0;
+
+	for (;;) {
+		matchptr = &in_base[cur_node];
+
+		if (matchptr[len] == in_next[len]) {
+			len = lz_extend(in_next, matchptr, len + 1, max_len);
+			if (!record_matches || len > best_len) {
+				if (record_matches) {
+					best_len = len;
+					lz_matchptr->length = len;
+					lz_matchptr->offset = in_next - matchptr;
+					lz_matchptr++;
+				}
+				if (len >= nice_len) {
+					*pending_lt_ptr = *bt_left_child(mf, cur_node);
+					*pending_gt_ptr = *bt_right_child(mf, cur_node);
+					return lz_matchptr;
+				}
+			}
+		}
+
+		if (matchptr[len] < in_next[len]) {
+			*pending_lt_ptr = cur_node;
+			pending_lt_ptr = bt_right_child(mf, cur_node);
+			cur_node = *pending_lt_ptr;
+			best_lt_len = len;
+			if (best_gt_len < len)
+				len = best_gt_len;
+		} else {
+			*pending_gt_ptr = cur_node;
+			pending_gt_ptr = bt_left_child(mf, cur_node);
+			cur_node = *pending_gt_ptr;
+			best_gt_len = len;
+			if (best_lt_len < len)
+				len = best_lt_len;
+		}
+
+		if (cur_node <= cutoff || !--depth_remaining) {
+			*pending_lt_ptr = MATCHFINDER_INITVAL;
+			*pending_gt_ptr = MATCHFINDER_INITVAL;
+			return lz_matchptr;
+		}
+	}
+}
+
+/*
+ * Retrieve a list of matches with the current position.
+ *
+ * @mf
+ *	The matchfinder structure.
+ * @in_base
+ *	Pointer to the next byte in the input buffer to process _at the last
+ *	time bt_matchfinder_init() or bt_matchfinder_slide_window() was called_.
+ * @cur_pos
+ *	The current position in the input buffer relative to @in_base (the
+ *	position of the sequence being matched against).
+ * @max_len
+ *	The maximum permissible match length at this position.  Must be >=
+ *	BT_MATCHFINDER_REQUIRED_NBYTES.
+ * @nice_len
+ *	Stop searching if a match of at least this length is found.
+ *	Must be <= @max_len.
+ * @max_search_depth
+ *	Limit on the number of potential matches to consider.  Must be >= 1.
+ * @next_hashes
+ *	The precomputed hash codes for the sequence beginning at @in_next.
+ *	These will be used and then updated with the precomputed hashcodes for
+ *	the sequence beginning at @in_next + 1.
+ * @lz_matchptr
+ *	An array in which this function will record the matches.  The recorded
+ *	matches will be sorted by strictly increasing length and (non-strictly)
+ *	increasing offset.  The maximum number of matches that may be found is
+ *	'nice_len - 2'.
+ *
+ * The return value is a pointer to the next available slot in the @lz_matchptr
+ * array.  (If no matches were found, this will be the same as @lz_matchptr.)
+ */
+static forceinline struct lz_match *
+bt_matchfinder_get_matches(struct bt_matchfinder *mf,
+			   const u8 *in_base,
+			   ptrdiff_t cur_pos,
+			   u32 max_len,
+			   u32 nice_len,
+			   u32 max_search_depth,
+			   u32 next_hashes[2],
+			   struct lz_match *lz_matchptr)
+{
+	return bt_matchfinder_advance_one_byte(mf,
+					       in_base,
+					       cur_pos,
+					       max_len,
+					       nice_len,
+					       max_search_depth,
+					       next_hashes,
+					       lz_matchptr,
+					       true);
+}
+
+/*
+ * Advance the matchfinder, but don't record any matches.
+ *
+ * This is very similar to bt_matchfinder_get_matches() because both functions
+ * must do hashing and tree re-rooting.
+ */
+static forceinline void
+bt_matchfinder_skip_byte(struct bt_matchfinder *mf,
+			 const u8 *in_base,
+			 ptrdiff_t cur_pos,
+			 u32 nice_len,
+			 u32 max_search_depth,
+			 u32 next_hashes[2])
+{
+	bt_matchfinder_advance_one_byte(mf,
+					in_base,
+					cur_pos,
+					nice_len,
+					nice_len,
+					max_search_depth,
+					next_hashes,
+					NULL,
+					false);
+}
+
+#endif /* LIB_BT_MATCHFINDER_H */

data/ext/deflate_ruby/libdeflate/common_defs.h

@@ -67,7 +67,8 @@
 #undef ARCH_ARM32
 #undef ARCH_RISCV
 #ifdef _MSC_VER
-#  if defined(_M_X64)
+   /* Way too many things are broken in ARM64EC to pretend that it is x86_64. */
+#  if defined(_M_X64) && !defined(_M_ARM64EC)
 #    define ARCH_X86_64
 #  elif defined(_M_IX86)
 #    define ARCH_X86_32

data/ext/deflate_ruby/libdeflate/cpu_features_common.h

@@ -0,0 +1,93 @@
+/*
+ * cpu_features_common.h - code shared by all lib/$arch/cpu_features.c
+ *
+ * Copyright 2020 Eric Biggers
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef LIB_CPU_FEATURES_COMMON_H
+#define LIB_CPU_FEATURES_COMMON_H
+
+#if defined(TEST_SUPPORT__DO_NOT_USE) && !defined(FREESTANDING)
+   /* for strdup() and strtok_r() */
+#  undef _ANSI_SOURCE
+#  ifndef __APPLE__
+#    undef _GNU_SOURCE
+#    define _GNU_SOURCE
+#  endif
+#  include <stdio.h>
+#  include <stdlib.h>
+#  include <string.h>
+#endif
+
+#include "lib_common.h"
+
+struct cpu_feature {
+	u32 bit;
+	const char *name;
+};
+
+#if defined(TEST_SUPPORT__DO_NOT_USE) && !defined(FREESTANDING)
+/* Disable any features that are listed in $LIBDEFLATE_DISABLE_CPU_FEATURES. */
+static inline void
+disable_cpu_features_for_testing(u32 *features,
+				 const struct cpu_feature *feature_table,
+				 size_t feature_table_length)
+{
+	char *env_value, *strbuf, *p, *saveptr = NULL;
+	size_t i;
+
+	env_value = getenv("LIBDEFLATE_DISABLE_CPU_FEATURES");
+	if (!env_value)
+		return;
+	strbuf = strdup(env_value);
+	if (!strbuf)
+		abort();
+	p = strtok_r(strbuf, ",", &saveptr);
+	while (p) {
+		for (i = 0; i < feature_table_length; i++) {
+			if (strcmp(p, feature_table[i].name) == 0) {
+				*features &= ~feature_table[i].bit;
+				break;
+			}
+		}
+		if (i == feature_table_length) {
+			fprintf(stderr,
+				"unrecognized feature in LIBDEFLATE_DISABLE_CPU_FEATURES: \"%s\"\n",
+				p);
+			abort();
+		}
+		p = strtok_r(NULL, ",", &saveptr);
+	}
+	free(strbuf);
+}
+#else /* TEST_SUPPORT__DO_NOT_USE */
+static inline void
+disable_cpu_features_for_testing(u32 *features,
+				 const struct cpu_feature *feature_table,
+				 size_t feature_table_length)
+{
+}
+#endif /* !TEST_SUPPORT__DO_NOT_USE */
+
+#endif /* LIB_CPU_FEATURES_COMMON_H */