@yowasp/yosys 0.37.58-dev.635 → 0.37.67-dev.638

package/gen/resources-yosys.js

@@ -132,7 +132,7 @@ export const filesystem = {
                                 "cxxrtl_capi_vcd.cc": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2020  whitequark <whitequark@whitequark.org>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n// This file is a part of the CXXRTL C API. It should be used together with `cxxrtl/capi/cxxrtl_capi_vcd.h`.\n\n#include <cxxrtl/capi/cxxrtl_capi_vcd.h>\n#include <cxxrtl/cxxrtl_vcd.h>\n\nextern const cxxrtl::debug_items &cxxrtl_debug_items_from_handle(cxxrtl_handle handle);\n\nstruct _cxxrtl_vcd {\n\tcxxrtl::vcd_writer writer;\n\tbool flush = false;\n};\n\ncxxrtl_vcd cxxrtl_vcd_create() {\n\treturn new _cxxrtl_vcd;\n}\n\nvoid cxxrtl_vcd_destroy(cxxrtl_vcd vcd) {\n\tdelete vcd;\n}\n\nvoid cxxrtl_vcd_timescale(cxxrtl_vcd vcd, int number, const char *unit) {\n\tvcd->writer.timescale(number, unit);\n}\n\nvoid cxxrtl_vcd_add(cxxrtl_vcd vcd, const char *name, cxxrtl_object *object) {\n\t// Note the copy. We don't know whether `object` came from a design (in which case it is\n\t// an instance of `debug_item`), or from user code (in which case it is an instance of\n\t// `cxxrtl_object`), so casting the pointer wouldn't be safe.\n\tvcd->writer.add(name, cxxrtl::debug_item(*object));\n}\n\nvoid cxxrtl_vcd_add_from(cxxrtl_vcd vcd, cxxrtl_handle handle) {\n\tvcd->writer.add(cxxrtl_debug_items_from_handle(handle));\n}\n\nvoid cxxrtl_vcd_add_from_if(cxxrtl_vcd vcd, cxxrtl_handle handle, void *data,\n\t\t\t\t\t\t\t\t\t\t\t\t\t\tint (*filter)(void *data, const char *name,\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t              const cxxrtl_object *object)) {\n\tvcd->writer.add(cxxrtl_debug_items_from_handle(handle),\n\t\t[=](const std::string &name, const cxxrtl::debug_item &item) {\n\t\t\treturn filter(data, name.c_str(), static_cast<const cxxrtl_object*>(&item));\n\t\t});\n}\n\nvoid cxxrtl_vcd_add_from_without_memories(cxxrtl_vcd vcd, cxxrtl_handle handle) {\n\tvcd->writer.add_without_memories(cxxrtl_debug_items_from_handle(handle));\n}\n\nvoid cxxrtl_vcd_sample(cxxrtl_vcd vcd, uint64_t time) {\n\tif (vcd->flush) {\n\t\tvcd->writer.buffer.clear();\n\t\tvcd->flush = false;\n\t}\n\tvcd->writer.sample(time);\n}\n\nvoid cxxrtl_vcd_read(cxxrtl_vcd vcd, const char **data, size_t *size) {\n\tif (vcd->flush) {\n\t\tvcd->writer.buffer.clear();\n\t\tvcd->flush = false;\n\t}\n\t*data = vcd->writer.buffer.c_str();\n\t*size = vcd->writer.buffer.size();\n\tvcd->flush = true;\n}\n",
                                 "cxxrtl_capi_vcd.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2020  whitequark <whitequark@whitequark.org>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#ifndef CXXRTL_CAPI_VCD_H\n#define CXXRTL_CAPI_VCD_H\n\n// This file is a part of the CXXRTL C API. It should be used together with `cxxrtl_vcd_capi.cc`.\n//\n// The CXXRTL C API for VCD writing makes it possible to insert virtual probes into designs and\n// dump waveforms to Value Change Dump files.\n\n#include <stddef.h>\n#include <stdint.h>\n\n#include <cxxrtl/capi/cxxrtl_capi.h>\n\n#ifdef __cplusplus\nextern \"C\" {\n#endif\n\n// Opaque reference to a VCD writer.\ntypedef struct _cxxrtl_vcd *cxxrtl_vcd;\n\n// Create a VCD writer.\ncxxrtl_vcd cxxrtl_vcd_create();\n\n// Release all resources used by a VCD writer.\nvoid cxxrtl_vcd_destroy(cxxrtl_vcd vcd);\n\n// Set VCD timescale.\n//\n// The `number` must be 1, 10, or 100, and the `unit` must be one of `\"s\"`, `\"ms\"`, `\"us\"`, `\"ns\"`,\n// `\"ps\"`, or `\"fs\"`.\n//\n// Timescale can only be set before the first call to `cxxrtl_vcd_sample`.\nvoid cxxrtl_vcd_timescale(cxxrtl_vcd vcd, int number, const char *unit);\n\n// Schedule a specific CXXRTL object to be sampled.\n//\n// The `name` is a full hierarchical name as described for `cxxrtl_get`; it does not need to match\n// the original name of `object`, if any. The `object` must outlive the VCD writer, but there are\n// no other requirements; if desired, it can be provided by user code, rather than come from\n// a design.\n//\n// Objects can only be scheduled before the first call to `cxxrtl_vcd_sample`.\nvoid cxxrtl_vcd_add(cxxrtl_vcd vcd, const char *name, struct cxxrtl_object *object);\n\n// Schedule all CXXRTL objects in a simulation.\n//\n// The design `handle` must outlive the VCD writer.\n//\n// Objects can only be scheduled before the first call to `cxxrtl_vcd_sample`.\nvoid cxxrtl_vcd_add_from(cxxrtl_vcd vcd, cxxrtl_handle handle);\n\n// Schedule CXXRTL objects in a simulation that match a given predicate.\n//\n// For every object in the simulation, `filter` is called with the provided `data`, the full\n// hierarchical name of the object (see `cxxrtl_get` for details), and the object description.\n// The object will be sampled if the predicate returns a non-zero value.\n//\n// Objects can only be scheduled before the first call to `cxxrtl_vcd_sample`.\nvoid cxxrtl_vcd_add_from_if(cxxrtl_vcd vcd, cxxrtl_handle handle, void *data,\n                            int (*filter)(void *data, const char *name,\n                                          const struct cxxrtl_object *object));\n\n// Schedule all CXXRTL objects in a simulation except for memories.\n//\n// The design `handle` must outlive the VCD writer.\n//\n// Objects can only be scheduled before the first call to `cxxrtl_vcd_sample`.\nvoid cxxrtl_vcd_add_from_without_memories(cxxrtl_vcd vcd, cxxrtl_handle handle);\n\n// Sample all scheduled objects.\n//\n// First, `time` is written to the internal buffer. Second, the values of every signal changed since\n// the previous call to `cxxrtl_vcd_sample` (all values if this is the first call) are written to\n// the internal buffer. The contents of the buffer can be retrieved with `cxxrtl_vcd_read`.\nvoid cxxrtl_vcd_sample(cxxrtl_vcd vcd, uint64_t time);\n\n// Retrieve buffered VCD data.\n//\n// The pointer to the start of the next chunk of VCD data is assigned to `*data`, and the length\n// of that chunk is assigned to `*size`. The pointer to the data is valid until the next call to\n// `cxxrtl_vcd_sample` or `cxxrtl_vcd_read`. Once all of the buffered data has been retrieved,\n// this function will always return zero sized chunks.\nvoid cxxrtl_vcd_read(cxxrtl_vcd vcd, const char **data, size_t *size);\n\n#ifdef __cplusplus\n}\n#endif\n\n#endif\n",
                             },
-                            "cxxrtl.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2019-2020  whitequark <whitequark@whitequark.org>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n// This file is included by the designs generated with `write_cxxrtl`. It is not used in Yosys itself.\n//\n// The CXXRTL support library implements compile time specialized arbitrary width arithmetics, as well as provides\n// composite lvalues made out of bit slices and concatenations of lvalues. This allows the `write_cxxrtl` pass\n// to perform a straightforward translation of RTLIL structures to readable C++, relying on the C++ compiler\n// to unwrap the abstraction and generate efficient code.\n\n#ifndef CXXRTL_H\n#define CXXRTL_H\n\n#include <cstddef>\n#include <cstdint>\n#include <cassert>\n#include <limits>\n#include <type_traits>\n#include <tuple>\n#include <vector>\n#include <map>\n#include <algorithm>\n#include <memory>\n#include <functional>\n#include <sstream>\n\n// `cxxrtl::debug_item` has to inherit from `cxxrtl_object` to satisfy strict aliasing requirements.\n#include <cxxrtl/capi/cxxrtl_capi.h>\n\n#ifndef __has_attribute\n#\tdefine __has_attribute(x) 0\n#endif\n\n// CXXRTL essentially uses the C++ compiler as a hygienic macro engine that feeds an instruction selector.\n// It generates a lot of specialized template functions with relatively large bodies that, when inlined\n// into the caller and (for those with loops) unrolled, often expose many new optimization opportunities.\n// Because of this, most of the CXXRTL runtime must be always inlined for best performance.\n#if __has_attribute(always_inline)\n#define CXXRTL_ALWAYS_INLINE inline __attribute__((__always_inline__))\n#else\n#define CXXRTL_ALWAYS_INLINE inline\n#endif\n// Conversely, some functions in the generated code are extremely large yet very cold, with both of these\n// properties being extreme enough to confuse C++ compilers into spending pathological amounts of time\n// on a futile (the code becomes worse) attempt to optimize the least important parts of code.\n#if __has_attribute(optnone)\n#define CXXRTL_EXTREMELY_COLD __attribute__((__optnone__))\n#elif __has_attribute(optimize)\n#define CXXRTL_EXTREMELY_COLD __attribute__((__optimize__(0)))\n#else\n#define CXXRTL_EXTREMELY_COLD\n#endif\n\n// CXXRTL uses assert() to check for C++ contract violations (which may result in e.g. undefined behavior\n// of the simulation code itself), and CXXRTL_ASSERT to check for RTL contract violations (which may at\n// most result in undefined simulation results).\n//\n// Though by default, CXXRTL_ASSERT() expands to assert(), it may be overridden e.g. when integrating\n// the simulation into another process that should survive violating RTL contracts.\n#ifndef CXXRTL_ASSERT\n#ifndef CXXRTL_NDEBUG\n#define CXXRTL_ASSERT(x) assert(x)\n#else\n#define CXXRTL_ASSERT(x)\n#endif\n#endif\n\nnamespace cxxrtl {\n\n// All arbitrary-width values in CXXRTL are backed by arrays of unsigned integers called chunks. The chunk size\n// is the same regardless of the value width to simplify manipulating values via FFI interfaces, e.g. driving\n// and introspecting the simulation in Python.\n//\n// It is practical to use chunk sizes between 32 bits and platform register size because when arithmetics on\n// narrower integer types is legalized by the C++ compiler, it inserts code to clear the high bits of the register.\n// However, (a) most of our operations do not change those bits in the first place because of invariants that are\n// invisible to the compiler, (b) we often operate on non-power-of-2 values and have to clear the high bits anyway.\n// Therefore, using relatively wide chunks and clearing the high bits explicitly and only when we know they may be\n// clobbered results in simpler generated code.\ntypedef uint32_t chunk_t;\ntypedef uint64_t wide_chunk_t;\n\ntemplate<typename T>\nstruct chunk_traits {\n\tstatic_assert(std::is_integral<T>::value && std::is_unsigned<T>::value,\n\t              \"chunk type must be an unsigned integral type\");\n\tusing type = T;\n\tstatic constexpr size_t bits = std::numeric_limits<T>::digits;\n\tstatic constexpr T mask = std::numeric_limits<T>::max();\n};\n\ntemplate<class T>\nstruct expr_base;\n\ntemplate<size_t Bits>\nstruct value : public expr_base<value<Bits>> {\n\tstatic constexpr size_t bits = Bits;\n\n\tusing chunk = chunk_traits<chunk_t>;\n\tstatic constexpr chunk::type msb_mask = (Bits % chunk::bits == 0) ? chunk::mask\n\t\t: chunk::mask >> (chunk::bits - (Bits % chunk::bits));\n\n\tstatic constexpr size_t chunks = (Bits + chunk::bits - 1) / chunk::bits;\n\tchunk::type data[chunks] = {};\n\n\tvalue() = default;\n\ttemplate<typename... Init>\n\texplicit constexpr value(Init ...init) : data{init...} {}\n\n\tvalue(const value<Bits> &) = default;\n\tvalue<Bits> &operator=(const value<Bits> &) = default;\n\n\tvalue(value<Bits> &&) = default;\n\tvalue<Bits> &operator=(value<Bits> &&) = default;\n\n\t// A (no-op) helper that forces the cast to value<>.\n\tCXXRTL_ALWAYS_INLINE\n\tconst value<Bits> &val() const {\n\t\treturn *this;\n\t}\n\n\tstd::string str() const {\n\t\tstd::stringstream ss;\n\t\tss << *this;\n\t\treturn ss.str();\n\t}\n\n\t// Conversion operations.\n\t//\n\t// These functions ensure that a conversion is never out of range, and should be always used, if at all\n\t// possible, instead of direct manipulation of the `data` member. For very large types, .slice() and\n\t// .concat() can be used to split them into more manageable parts.\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tIntegerT get() const {\n\t\tstatic_assert(std::numeric_limits<IntegerT>::is_integer && !std::numeric_limits<IntegerT>::is_signed,\n\t\t              \"get<T>() requires T to be an unsigned integral type\");\n\t\tstatic_assert(std::numeric_limits<IntegerT>::digits >= Bits,\n\t\t              \"get<T>() requires T to be at least as wide as the value is\");\n\t\tIntegerT result = 0;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult |= IntegerT(data[n]) << (n * chunk::bits);\n\t\treturn result;\n\t}\n\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tvoid set(IntegerT other) {\n\t\tstatic_assert(std::numeric_limits<IntegerT>::is_integer && !std::numeric_limits<IntegerT>::is_signed,\n\t\t              \"set<T>() requires T to be an unsigned integral type\");\n\t\tstatic_assert(std::numeric_limits<IntegerT>::digits >= Bits,\n\t\t              \"set<T>() requires the value to be at least as wide as T is\");\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tdata[n] = (other >> (n * chunk::bits)) & chunk::mask;\n\t}\n\n\t// Operations with compile-time parameters.\n\t//\n\t// These operations are used to implement slicing, concatenation, and blitting.\n\t// The trunc, zext and sext operations add or remove most significant bits (i.e. on the left);\n\t// the rtrunc and rzext operations add or remove least significant bits (i.e. on the right).\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> trunc() const {\n\t\tstatic_assert(NewBits <= Bits, \"trunc() may not increase width\");\n\t\tvalue<NewBits> result;\n\t\tfor (size_t n = 0; n < result.chunks; n++)\n\t\t\tresult.data[n] = data[n];\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> zext() const {\n\t\tstatic_assert(NewBits >= Bits, \"zext() may not decrease width\");\n\t\tvalue<NewBits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n];\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> sext() const {\n\t\tstatic_assert(NewBits >= Bits, \"sext() may not decrease width\");\n\t\tvalue<NewBits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n];\n\t\tif (is_neg()) {\n\t\t\tresult.data[chunks - 1] |= ~msb_mask;\n\t\t\tfor (size_t n = chunks; n < result.chunks; n++)\n\t\t\t\tresult.data[n] = chunk::mask;\n\t\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\t}\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> rtrunc() const {\n\t\tstatic_assert(NewBits <= Bits, \"rtrunc() may not increase width\");\n\t\tvalue<NewBits> result;\n\t\tconstexpr size_t shift_chunks = (Bits - NewBits) / chunk::bits;\n\t\tconstexpr size_t shift_bits   = (Bits - NewBits) % chunk::bits;\n\t\tchunk::type carry = 0;\n\t\tif (shift_chunks + result.chunks < chunks) {\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[shift_chunks + result.chunks] << (chunk::bits - shift_bits);\n\t\t}\n\t\tfor (size_t n = result.chunks; n > 0; n--) {\n\t\t\tresult.data[n - 1] = carry | (data[shift_chunks + n - 1] >> shift_bits);\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[shift_chunks + n - 1] << (chunk::bits - shift_bits);\n\t\t}\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> rzext() const {\n\t\tstatic_assert(NewBits >= Bits, \"rzext() may not decrease width\");\n\t\tvalue<NewBits> result;\n\t\tconstexpr size_t shift_chunks = (NewBits - Bits) / chunk::bits;\n\t\tconstexpr size_t shift_bits   = (NewBits - Bits) % chunk::bits;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tresult.data[shift_chunks + n] = (data[n] << shift_bits) | carry;\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[n] >> (chunk::bits - shift_bits);\n\t\t}\n\t\tif (shift_chunks + chunks < result.chunks)\n\t\t\tresult.data[shift_chunks + chunks] = carry;\n\t\treturn result;\n\t}\n\n\t// Bit blit operation, i.e. a partial read-modify-write.\n\ttemplate<size_t Stop, size_t Start>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<Bits> blit(const value<Stop - Start + 1> &source) const {\n\t\tstatic_assert(Stop >= Start, \"blit() may not reverse bit order\");\n\t\tconstexpr chunk::type start_mask = ~(chunk::mask << (Start % chunk::bits));\n\t\tconstexpr chunk::type stop_mask = (Stop % chunk::bits + 1 == chunk::bits) ? 0\n\t\t\t: (chunk::mask << (Stop % chunk::bits + 1));\n\t\tvalue<Bits> masked = *this;\n\t\tif (Start / chunk::bits == Stop / chunk::bits) {\n\t\t\tmasked.data[Start / chunk::bits] &= stop_mask | start_mask;\n\t\t} else {\n\t\t\tmasked.data[Start / chunk::bits] &= start_mask;\n\t\t\tfor (size_t n = Start / chunk::bits + 1; n < Stop / chunk::bits; n++)\n\t\t\t\tmasked.data[n] = 0;\n\t\t\tmasked.data[Stop / chunk::bits] &= stop_mask;\n\t\t}\n\t\tvalue<Bits> shifted = source\n\t\t\t.template rzext<Stop + 1>()\n\t\t\t.template zext<Bits>();\n\t\treturn masked.bit_or(shifted);\n\t}\n\n\t// Helpers for selecting extending or truncating operation depending on whether the result is wider or narrower\n\t// than the operand. In C++17 these can be replaced with `if constexpr`.\n\ttemplate<size_t NewBits, typename = void>\n\tstruct zext_cast {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template zext<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits>\n\tstruct zext_cast<NewBits, typename std::enable_if<(NewBits < Bits)>::type> {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template trunc<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits, typename = void>\n\tstruct sext_cast {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template sext<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits>\n\tstruct sext_cast<NewBits, typename std::enable_if<(NewBits < Bits)>::type> {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template trunc<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> zcast() const {\n\t\treturn zext_cast<NewBits>()(*this);\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> scast() const {\n\t\treturn sext_cast<NewBits>()(*this);\n\t}\n\n\t// Bit replication is far more efficient than the equivalent concatenation.\n\ttemplate<size_t Count>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<Bits * Count> repeat() const {\n\t\tstatic_assert(Bits == 1, \"repeat() is implemented only for 1-bit values\");\n\t\treturn *this ? value<Bits * Count>().bit_not() : value<Bits * Count>();\n\t}\n\n\t// Operations with run-time parameters (offsets, amounts, etc).\n\t//\n\t// These operations are used for computations.\n\tbool bit(size_t offset) const {\n\t\treturn data[offset / chunk::bits] & (1 << (offset % chunk::bits));\n\t}\n\n\tvoid set_bit(size_t offset, bool value = true) {\n\t\tsize_t offset_chunks = offset / chunk::bits;\n\t\tsize_t offset_bits = offset % chunk::bits;\n\t\tdata[offset_chunks] &= ~(1 << offset_bits);\n\t\tdata[offset_chunks] |= value ? 1 << offset_bits : 0;\n\t}\n\n\texplicit operator bool() const {\n\t\treturn !is_zero();\n\t}\n\n\tbool is_zero() const {\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tif (data[n] != 0)\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\n\tbool is_neg() const {\n\t\treturn data[chunks - 1] & (1 << ((Bits - 1) % chunk::bits));\n\t}\n\n\tbool operator ==(const value<Bits> &other) const {\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tif (data[n] != other.data[n])\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\n\tbool operator !=(const value<Bits> &other) const {\n\t\treturn !(*this == other);\n\t}\n\n\tvalue<Bits> bit_not() const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = ~data[n];\n\t\tresult.data[chunks - 1] &= msb_mask;\n\t\treturn result;\n\t}\n\n\tvalue<Bits> bit_and(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n] & other.data[n];\n\t\treturn result;\n\t}\n\n\tvalue<Bits> bit_or(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n] | other.data[n];\n\t\treturn result;\n\t}\n\n\tvalue<Bits> bit_xor(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n] ^ other.data[n];\n\t\treturn result;\n\t}\n\n\tvalue<Bits> update(const value<Bits> &val, const value<Bits> &mask) const {\n\t\treturn bit_and(mask.bit_not()).bit_or(val.bit_and(mask));\n\t}\n\n\ttemplate<size_t AmountBits>\n\tvalue<Bits> shl(const value<AmountBits> &amount) const {\n\t\t// Ensure our early return is correct by prohibiting values larger than 4 Gbit.\n\t\tstatic_assert(Bits <= chunk::mask, \"shl() of unreasonably large values is not supported\");\n\t\t// Detect shifts definitely large than Bits early.\n\t\tfor (size_t n = 1; n < amount.chunks; n++)\n\t\t\tif (amount.data[n] != 0)\n\t\t\t\treturn {};\n\t\t// Past this point we can use the least significant chunk as the shift size.\n\t\tsize_t shift_chunks = amount.data[0] / chunk::bits;\n\t\tsize_t shift_bits   = amount.data[0] % chunk::bits;\n\t\tif (shift_chunks >= chunks)\n\t\t\treturn {};\n\t\tvalue<Bits> result;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks - shift_chunks; n++) {\n\t\t\tresult.data[shift_chunks + n] = (data[n] << shift_bits) | carry;\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[n] >> (chunk::bits - shift_bits);\n\t\t}\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\ttemplate<size_t AmountBits, bool Signed = false>\n\tvalue<Bits> shr(const value<AmountBits> &amount) const {\n\t\t// Ensure our early return is correct by prohibiting values larger than 4 Gbit.\n\t\tstatic_assert(Bits <= chunk::mask, \"shr() of unreasonably large values is not supported\");\n\t\t// Detect shifts definitely large than Bits early.\n\t\tfor (size_t n = 1; n < amount.chunks; n++)\n\t\t\tif (amount.data[n] != 0)\n\t\t\t\treturn (Signed && is_neg()) ? value<Bits>().bit_not() : value<Bits>();\n\t\t// Past this point we can use the least significant chunk as the shift size.\n\t\tsize_t shift_chunks = amount.data[0] / chunk::bits;\n\t\tsize_t shift_bits   = amount.data[0] % chunk::bits;\n\t\tif (shift_chunks >= chunks)\n\t\t\treturn (Signed && is_neg()) ? value<Bits>().bit_not() : value<Bits>();\n\t\tvalue<Bits> result;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks - shift_chunks; n++) {\n\t\t\tresult.data[chunks - shift_chunks - 1 - n] = carry | (data[chunks - 1 - n] >> shift_bits);\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[chunks - 1 - n] << (chunk::bits - shift_bits);\n\t\t}\n\t\tif (Signed && is_neg()) {\n\t\t\tsize_t top_chunk_idx  = amount.data[0] > Bits ? 0 : (Bits - amount.data[0]) / chunk::bits;\n\t\t\tsize_t top_chunk_bits = amount.data[0] > Bits ? 0 : (Bits - amount.data[0]) % chunk::bits;\n\t\t\tfor (size_t n = top_chunk_idx + 1; n < chunks; n++)\n\t\t\t\tresult.data[n] = chunk::mask;\n\t\t\tif (amount.data[0] != 0)\n\t\t\t\tresult.data[top_chunk_idx] |= chunk::mask << top_chunk_bits;\n\t\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\t}\n\t\treturn result;\n\t}\n\n\ttemplate<size_t AmountBits>\n\tvalue<Bits> sshr(const value<AmountBits> &amount) const {\n\t\treturn shr<AmountBits, /*Signed=*/true>(amount);\n\t}\n\n\ttemplate<size_t ResultBits, size_t SelBits>\n\tvalue<ResultBits> bmux(const value<SelBits> &sel) const {\n\t\tstatic_assert(ResultBits << SelBits == Bits, \"invalid sizes used in bmux()\");\n\t\tsize_t amount = sel.data[0] * ResultBits;\n\t\tsize_t shift_chunks = amount / chunk::bits;\n\t\tsize_t shift_bits   = amount % chunk::bits;\n\t\tvalue<ResultBits> result;\n\t\tchunk::type carry = 0;\n\t\tif (ResultBits % chunk::bits + shift_bits > chunk::bits)\n\t\t\tcarry = data[result.chunks + shift_chunks] << (chunk::bits - shift_bits);\n\t\tfor (size_t n = 0; n < result.chunks; n++) {\n\t\t\tresult.data[result.chunks - 1 - n] = carry | (data[result.chunks + shift_chunks - 1 - n] >> shift_bits);\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[result.chunks + shift_chunks - 1 - n] << (chunk::bits - shift_bits);\n\t\t}\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\ttemplate<size_t ResultBits, size_t SelBits>\n\tvalue<ResultBits> demux(const value<SelBits> &sel) const {\n\t\tstatic_assert(Bits << SelBits == ResultBits, \"invalid sizes used in demux()\");\n\t\tsize_t amount = sel.data[0] * Bits;\n\t\tsize_t shift_chunks = amount / chunk::bits;\n\t\tsize_t shift_bits   = amount % chunk::bits;\n\t\tvalue<ResultBits> result;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tresult.data[shift_chunks + n] = (data[n] << shift_bits) | carry;\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[n] >> (chunk::bits - shift_bits);\n\t\t}\n\t\tif (Bits % chunk::bits + shift_bits > chunk::bits)\n\t\t\tresult.data[shift_chunks + chunks] = carry;\n\t\treturn result;\n\t}\n\n\tsize_t ctpop() const {\n\t\tsize_t count = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\t// This loop implements the population count idiom as recognized by LLVM and GCC.\n\t\t\tfor (chunk::type x = data[n]; x != 0; count++)\n\t\t\t\tx = x & (x - 1);\n\t\t}\n\t\treturn count;\n\t}\n\n\tsize_t ctlz() const {\n\t\tsize_t count = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tchunk::type x = data[chunks - 1 - n];\n\t\t\t// First add to `count` as if the chunk is zero\n\t\t\tconstexpr size_t msb_chunk_bits = Bits % chunk::bits != 0 ? Bits % chunk::bits : chunk::bits;\n\t\t\tcount += (n == 0 ? msb_chunk_bits : chunk::bits);\n\t\t\t// If the chunk isn't zero, correct the `count` value and return\n\t\t\tif (x != 0) {\n\t\t\t\tfor (; x != 0; count--)\n\t\t\t\t\tx >>= 1;\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\t\treturn count;\n\t}\n\n\ttemplate<bool Invert, bool CarryIn>\n\tstd::pair<value<Bits>, bool /*CarryOut*/> alu(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tbool carry = CarryIn;\n\t\tfor (size_t n = 0; n < result.chunks; n++) {\n\t\t\tresult.data[n] = data[n] + (Invert ? ~other.data[n] : other.data[n]) + carry;\n\t\t\tif (result.chunks - 1 == n)\n\t\t\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\t\tcarry = (result.data[n] <  data[n]) ||\n\t\t\t        (result.data[n] == data[n] && carry);\n\t\t}\n\t\treturn {result, carry};\n\t}\n\n\tvalue<Bits> add(const value<Bits> &other) const {\n\t\treturn alu</*Invert=*/false, /*CarryIn=*/false>(other).first;\n\t}\n\n\tvalue<Bits> sub(const value<Bits> &other) const {\n\t\treturn alu</*Invert=*/true, /*CarryIn=*/true>(other).first;\n\t}\n\n\tvalue<Bits> neg() const {\n\t\treturn value<Bits> { 0u }.sub(*this);\n\t}\n\n\tbool ucmp(const value<Bits> &other) const {\n\t\tbool carry;\n\t\tstd::tie(std::ignore, carry) = alu</*Invert=*/true, /*CarryIn=*/true>(other);\n\t\treturn !carry; // a.ucmp(b) ≡ a u< b\n\t}\n\n\tbool scmp(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tbool carry;\n\t\tstd::tie(result, carry) = alu</*Invert=*/true, /*CarryIn=*/true>(other);\n\t\tbool overflow = (is_neg() == !other.is_neg()) && (is_neg() != result.is_neg());\n\t\treturn result.is_neg() ^ overflow; // a.scmp(b) ≡ a s< b\n\t}\n\n\ttemplate<size_t ResultBits>\n\tvalue<ResultBits> mul(const value<Bits> &other) const {\n\t\tvalue<ResultBits> result;\n\t\twide_chunk_t wide_result[result.chunks + 1] = {};\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tfor (size_t m = 0; m < chunks && n + m < result.chunks; m++) {\n\t\t\t\twide_result[n + m] += wide_chunk_t(data[n]) * wide_chunk_t(other.data[m]);\n\t\t\t\twide_result[n + m + 1] += wide_result[n + m] >> chunk::bits;\n\t\t\t\twide_result[n + m] &= chunk::mask;\n\t\t\t}\n\t\t}\n\t\tfor (size_t n = 0; n < result.chunks; n++) {\n\t\t\tresult.data[n] = wide_result[n];\n\t\t}\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\tstd::pair<value<Bits>, value<Bits>> udivmod(value<Bits> divisor) const {\n\t\tvalue<Bits> quotient;\n\t\tvalue<Bits> dividend = *this;\n\t\tif (dividend.ucmp(divisor))\n\t\t\treturn {/*quotient=*/value<Bits>{0u}, /*remainder=*/dividend};\n\t\tint64_t divisor_shift = divisor.ctlz() - dividend.ctlz();\n\t\tassert(divisor_shift >= 0);\n\t\tdivisor = divisor.shl(value<Bits>{(chunk::type) divisor_shift});\n\t\tfor (size_t step = 0; step <= divisor_shift; step++) {\n\t\t\tquotient = quotient.shl(value<Bits>{1u});\n\t\t\tif (!dividend.ucmp(divisor)) {\n\t\t\t\tdividend = dividend.sub(divisor);\n\t\t\t\tquotient.set_bit(0, true);\n\t\t\t}\n\t\t\tdivisor = divisor.shr(value<Bits>{1u});\n\t\t}\n\t\treturn {quotient, /*remainder=*/dividend};\n\t}\n\n\tstd::pair<value<Bits>, value<Bits>> sdivmod(const value<Bits> &other) const {\n\t\tvalue<Bits + 1> quotient;\n\t\tvalue<Bits + 1> remainder;\n\t\tvalue<Bits + 1> dividend = sext<Bits + 1>();\n\t\tvalue<Bits + 1> divisor = other.template sext<Bits + 1>();\n\t\tif (dividend.is_neg()) dividend = dividend.neg();\n\t\tif (divisor.is_neg()) divisor = divisor.neg();\n\t\tstd::tie(quotient, remainder) = dividend.udivmod(divisor);\n\t\tif (dividend.is_neg() != divisor.is_neg()) quotient = quotient.neg();\n\t\tif (dividend.is_neg()) remainder = remainder.neg();\n\t\treturn {quotient.template trunc<Bits>(), remainder.template trunc<Bits>()};\n\t}\n};\n\n// Expression template for a slice, usable as lvalue or rvalue, and composable with other expression templates here.\ntemplate<class T, size_t Stop, size_t Start>\nstruct slice_expr : public expr_base<slice_expr<T, Stop, Start>> {\n\tstatic_assert(Stop >= Start, \"slice_expr() may not reverse bit order\");\n\tstatic_assert(Start < T::bits && Stop < T::bits, \"slice_expr() must be within bounds\");\n\tstatic constexpr size_t bits = Stop - Start + 1;\n\n\tT &expr;\n\n\tslice_expr(T &expr) : expr(expr) {}\n\tslice_expr(const slice_expr<T, Stop, Start> &) = delete;\n\n\tCXXRTL_ALWAYS_INLINE\n\toperator value<bits>() const {\n\t\treturn static_cast<const value<T::bits> &>(expr)\n\t\t\t.template rtrunc<T::bits - Start>()\n\t\t\t.template trunc<bits>();\n\t}\n\n\tCXXRTL_ALWAYS_INLINE\n\tslice_expr<T, Stop, Start> &operator=(const value<bits> &rhs) {\n\t\t// Generic partial assignment implemented using a read-modify-write operation on the sliced expression.\n\t\texpr = static_cast<const value<T::bits> &>(expr)\n\t\t\t.template blit<Stop, Start>(rhs);\n\t\treturn *this;\n\t}\n\n\t// A helper that forces the cast to value<>, which allows deduction to work.\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<bits> val() const {\n\t\treturn static_cast<const value<bits> &>(*this);\n\t}\n};\n\n// Expression template for a concatenation, usable as lvalue or rvalue, and composable with other expression templates here.\ntemplate<class T, class U>\nstruct concat_expr : public expr_base<concat_expr<T, U>> {\n\tstatic constexpr size_t bits = T::bits + U::bits;\n\n\tT &ms_expr;\n\tU &ls_expr;\n\n\tconcat_expr(T &ms_expr, U &ls_expr) : ms_expr(ms_expr), ls_expr(ls_expr) {}\n\tconcat_expr(const concat_expr<T, U> &) = delete;\n\n\tCXXRTL_ALWAYS_INLINE\n\toperator value<bits>() const {\n\t\tvalue<bits> ms_shifted = static_cast<const value<T::bits> &>(ms_expr)\n\t\t\t.template rzext<bits>();\n\t\tvalue<bits> ls_extended = static_cast<const value<U::bits> &>(ls_expr)\n\t\t\t.template zext<bits>();\n\t\treturn ms_shifted.bit_or(ls_extended);\n\t}\n\n\tCXXRTL_ALWAYS_INLINE\n\tconcat_expr<T, U> &operator=(const value<bits> &rhs) {\n\t\tms_expr = rhs.template rtrunc<T::bits>();\n\t\tls_expr = rhs.template trunc<U::bits>();\n\t\treturn *this;\n\t}\n\n\t// A helper that forces the cast to value<>, which allows deduction to work.\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<bits> val() const {\n\t\treturn static_cast<const value<bits> &>(*this);\n\t}\n};\n\n// Base class for expression templates, providing helper methods for operations that are valid on both rvalues and lvalues.\n//\n// Note that expression objects (slices and concatenations) constructed in this way should NEVER be captured because\n// they refer to temporaries that will, in general, only live until the end of the statement. For example, both of\n// these snippets perform use-after-free:\n//\n//    const auto &a = val.slice<7,0>().slice<1>();\n//    value<1> b = a;\n//\n//    auto &&c = val.slice<7,0>().slice<1>();\n//    c = value<1>{1u};\n//\n// An easy way to write code using slices and concatenations safely is to follow two simple rules:\n//   * Never explicitly name any type except `value<W>` or `const value<W> &`.\n//   * Never use a `const auto &` or `auto &&` in any such expression.\n// Then, any code that compiles will be well-defined.\ntemplate<class T>\nstruct expr_base {\n\ttemplate<size_t Stop, size_t Start = Stop>\n\tCXXRTL_ALWAYS_INLINE\n\tslice_expr<const T, Stop, Start> slice() const {\n\t\treturn {*static_cast<const T *>(this)};\n\t}\n\n\ttemplate<size_t Stop, size_t Start = Stop>\n\tCXXRTL_ALWAYS_INLINE\n\tslice_expr<T, Stop, Start> slice() {\n\t\treturn {*static_cast<T *>(this)};\n\t}\n\n\ttemplate<class U>\n\tCXXRTL_ALWAYS_INLINE\n\tconcat_expr<const T, typename std::remove_reference<const U>::type> concat(const U &other) const {\n\t\treturn {*static_cast<const T *>(this), other};\n\t}\n\n\ttemplate<class U>\n\tCXXRTL_ALWAYS_INLINE\n\tconcat_expr<T, typename std::remove_reference<U>::type> concat(U &&other) {\n\t\treturn {*static_cast<T *>(this), other};\n\t}\n};\n\ntemplate<size_t Bits>\nstd::ostream &operator<<(std::ostream &os, const value<Bits> &val) {\n\tauto old_flags = os.flags(std::ios::right);\n\tauto old_width = os.width(0);\n\tauto old_fill  = os.fill('0');\n\tos << val.bits << '\\'' << std::hex;\n\tfor (size_t n = val.chunks - 1; n != (size_t)-1; n--) {\n\t\tif (n == val.chunks - 1 && Bits % value<Bits>::chunk::bits != 0)\n\t\t\tos.width((Bits % value<Bits>::chunk::bits + 3) / 4);\n\t\telse\n\t\t\tos.width((value<Bits>::chunk::bits + 3) / 4);\n\t\tos << val.data[n];\n\t}\n\tos.fill(old_fill);\n\tos.width(old_width);\n\tos.flags(old_flags);\n\treturn os;\n}\n\ntemplate<size_t Bits>\nstruct value_formatted {\n\tconst value<Bits> &val;\n\tbool character;\n\tbool justify_left;\n\tchar padding;\n\tint width;\n\tint base;\n\tbool signed_;\n\tbool plus;\n\n\tvalue_formatted(const value<Bits> &val, bool character, bool justify_left, char padding, int width, int base, bool signed_, bool plus) :\n\t\tval(val), character(character), justify_left(justify_left), padding(padding), width(width), base(base), signed_(signed_), plus(plus) {}\n\tvalue_formatted(const value_formatted<Bits> &) = delete;\n\tvalue_formatted<Bits> &operator=(const value_formatted<Bits> &rhs) = delete;\n};\n\ntemplate<size_t Bits>\nstd::ostream &operator<<(std::ostream &os, const value_formatted<Bits> &vf)\n{\n\tvalue<Bits> val = vf.val;\n\n\tstd::string buf;\n\n\t// We might want to replace some of these bit() calls with direct\n\t// chunk access if it turns out to be slow enough to matter.\n\n\tif (!vf.character) {\n\t\tsize_t width = Bits;\n\t\tif (vf.base != 10) {\n\t\t\twidth = 0;\n\t\t\tfor (size_t index = 0; index < Bits; index++)\n\t\t\t\tif (val.bit(index))\n\t\t\t\t\twidth = index + 1;\n\t\t}\n\n\t\tif (vf.base == 2) {\n\t\t\tfor (size_t i = width; i > 0; i--)\n\t\t\t\tbuf += (val.bit(i - 1) ? '1' : '0');\n\t\t} else if (vf.base == 8 || vf.base == 16) {\n\t\t\tsize_t step = (vf.base == 16) ? 4 : 3;\n\t\t\tfor (size_t index = 0; index < width; index += step) {\n\t\t\t\tuint8_t value = val.bit(index) | (val.bit(index + 1) << 1) | (val.bit(index + 2) << 2);\n\t\t\t\tif (step == 4)\n\t\t\t\t\tvalue |= val.bit(index + 3) << 3;\n\t\t\t\tbuf += \"0123456789abcdef\"[value];\n\t\t\t}\n\t\t\tstd::reverse(buf.begin(), buf.end());\n\t\t} else if (vf.base == 10) {\n\t\t\tbool negative = vf.signed_ && val.is_neg();\n\t\t\tif (negative)\n\t\t\t\tval = val.neg();\n\t\t\tif (val.is_zero())\n\t\t\t\tbuf += '0';\n\t\t\twhile (!val.is_zero()) {\n\t\t\t\tvalue<Bits> quotient, remainder;\n\t\t\t\tif (Bits >= 4)\n\t\t\t\t\tstd::tie(quotient, remainder) = val.udivmod(value<Bits>{10u});\n\t\t\t\telse\n\t\t\t\t\tstd::tie(quotient, remainder) = std::make_pair(value<Bits>{0u}, val);\n\t\t\t\tbuf += '0' + remainder.template trunc<(Bits > 4 ? 4 : Bits)>().val().template get<uint8_t>();\n\t\t\t\tval = quotient;\n\t\t\t}\n\t\t\tif (negative || vf.plus)\n\t\t\t\tbuf += negative ? '-' : '+';\n\t\t\tstd::reverse(buf.begin(), buf.end());\n\t\t} else assert(false);\n\t} else {\n\t\tbuf.reserve(Bits/8);\n\t\tfor (int i = 0; i < Bits; i += 8) {\n\t\t\tchar ch = 0;\n\t\t\tfor (int j = 0; j < 8 && i + j < int(Bits); j++)\n\t\t\t\tif (val.bit(i + j))\n\t\t\t\t\tch |= 1 << j;\n\t\t\tif (ch != 0)\n\t\t\t\tbuf.append({ch});\n\t\t}\n\t\tstd::reverse(buf.begin(), buf.end());\n\t}\n\n\tassert(vf.width == 0 || vf.padding != '\\0');\n\tif (!vf.justify_left && buf.size() < vf.width) {\n\t\tsize_t pad_width = vf.width - buf.size();\n\t\tif (vf.padding == '0' && (buf.front() == '+' || buf.front() == '-')) {\n\t\t\tos << buf.front();\n\t\t\tbuf.erase(0, 1);\n\t\t}\n\t\tos << std::string(pad_width, vf.padding);\n\t}\n\tos << buf;\n\tif (vf.justify_left && buf.size() < vf.width)\n\t\tos << std::string(vf.width - buf.size(), vf.padding);\n\n\treturn os;\n}\n\ntemplate<size_t Bits>\nstruct wire {\n\tstatic constexpr size_t bits = Bits;\n\n\tvalue<Bits> curr;\n\tvalue<Bits> next;\n\n\twire() = default;\n\texplicit constexpr wire(const value<Bits> &init) : curr(init), next(init) {}\n\ttemplate<typename... Init>\n\texplicit constexpr wire(Init ...init) : curr{init...}, next{init...} {}\n\n\t// Copying and copy-assigning values is natural. If, however, a value is replaced with a wire,\n\t// e.g. because a module is built with a different optimization level, then existing code could\n\t// unintentionally copy a wire instead, which would create a subtle but serious bug. To make sure\n\t// this doesn't happen, prohibit copying and copy-assigning wires.\n\twire(const wire<Bits> &) = delete;\n\twire<Bits> &operator=(const wire<Bits> &) = delete;\n\n\twire(wire<Bits> &&) = default;\n\twire<Bits> &operator=(wire<Bits> &&) = default;\n\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tIntegerT get() const {\n\t\treturn curr.template get<IntegerT>();\n\t}\n\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tvoid set(IntegerT other) {\n\t\tnext.template set<IntegerT>(other);\n\t}\n\n\tbool commit() {\n\t\tif (curr != next) {\n\t\t\tcurr = next;\n\t\t\treturn true;\n\t\t}\n\t\treturn false;\n\t}\n};\n\ntemplate<size_t Bits>\nstd::ostream &operator<<(std::ostream &os, const wire<Bits> &val) {\n\tos << val.curr;\n\treturn os;\n}\n\ntemplate<size_t Width>\nstruct memory {\n\tconst size_t depth;\n\tstd::unique_ptr<value<Width>[]> data;\n\n\texplicit memory(size_t depth) : depth(depth), data(new value<Width>[depth]) {}\n\n\tmemory(const memory<Width> &) = delete;\n\tmemory<Width> &operator=(const memory<Width> &) = delete;\n\n\tmemory(memory<Width> &&) = default;\n\tmemory<Width> &operator=(memory<Width> &&other) {\n\t\tassert(depth == other.depth);\n\t\tdata = std::move(other.data);\n\t\twrite_queue = std::move(other.write_queue);\n\t\treturn *this;\n\t}\n\n\t// An operator for direct memory reads. May be used at any time during the simulation.\n\tconst value<Width> &operator [](size_t index) const {\n\t\tassert(index < depth);\n\t\treturn data[index];\n\t}\n\n\t// An operator for direct memory writes. May only be used before the simulation is started. If used\n\t// after the simulation is started, the design may malfunction.\n\tvalue<Width> &operator [](size_t index) {\n\t\tassert(index < depth);\n\t\treturn data[index];\n\t}\n\n\t// A simple way to make a writable memory would be to use an array of wires instead of an array of values.\n\t// However, there are two significant downsides to this approach: first, it has large overhead (2× space\n\t// overhead, and O(depth) time overhead during commit); second, it does not simplify handling write port\n\t// priorities. Although in principle write ports could be ordered or conditionally enabled in generated\n\t// code based on their priorities and selected addresses, the feedback arc set problem is computationally\n\t// expensive, and the heuristic based algorithms are not easily modified to guarantee (rather than prefer)\n\t// a particular write port evaluation order.\n\t//\n\t// The approach used here instead is to queue writes into a buffer during the eval phase, then perform\n\t// the writes during the commit phase in the priority order. This approach has low overhead, with both space\n\t// and time proportional to the amount of write ports. Because virtually every memory in a practical design\n\t// has at most two write ports, linear search is used on every write, being the fastest and simplest approach.\n\tstruct write {\n\t\tsize_t index;\n\t\tvalue<Width> val;\n\t\tvalue<Width> mask;\n\t\tint priority;\n\t};\n\tstd::vector<write> write_queue;\n\n\tvoid update(size_t index, const value<Width> &val, const value<Width> &mask, int priority = 0) {\n\t\tassert(index < depth);\n\t\t// Queue up the write while keeping the queue sorted by priority.\n\t\twrite_queue.insert(\n\t\t\tstd::upper_bound(write_queue.begin(), write_queue.end(), priority,\n\t\t\t\t[](const int a, const write& b) { return a < b.priority; }),\n\t\t\twrite { index, val, mask, priority });\n\t}\n\n\tbool commit() {\n\t\tbool changed = false;\n\t\tfor (const write &entry : write_queue) {\n\t\t\tvalue<Width> elem = data[entry.index];\n\t\t\telem = elem.update(entry.val, entry.mask);\n\t\t\tchanged |= (data[entry.index] != elem);\n\t\t\tdata[entry.index] = elem;\n\t\t}\n\t\twrite_queue.clear();\n\t\treturn changed;\n\t}\n};\n\nstruct metadata {\n\tconst enum {\n\t\tMISSING = 0,\n\t\tUINT   \t= 1,\n\t\tSINT   \t= 2,\n\t\tSTRING \t= 3,\n\t\tDOUBLE \t= 4,\n\t} value_type;\n\n\t// In debug mode, using the wrong .as_*() function will assert.\n\t// In release mode, using the wrong .as_*() function will safely return a default value.\n\tconst uint64_t    uint_value = 0;\n\tconst int64_t     sint_value = 0;\n\tconst std::string string_value = \"\";\n\tconst double      double_value = 0.0;\n\n\tmetadata() : value_type(MISSING) {}\n\tmetadata(uint64_t value) : value_type(UINT), uint_value(value) {}\n\tmetadata(int64_t value) : value_type(SINT), sint_value(value) {}\n\tmetadata(const std::string &value) : value_type(STRING), string_value(value) {}\n\tmetadata(const char *value) : value_type(STRING), string_value(value) {}\n\tmetadata(double value) : value_type(DOUBLE), double_value(value) {}\n\n\tmetadata(const metadata &) = default;\n\tmetadata &operator=(const metadata &) = delete;\n\n\tuint64_t as_uint() const {\n\t\tassert(value_type == UINT);\n\t\treturn uint_value;\n\t}\n\n\tint64_t as_sint() const {\n\t\tassert(value_type == SINT);\n\t\treturn sint_value;\n\t}\n\n\tconst std::string &as_string() const {\n\t\tassert(value_type == STRING);\n\t\treturn string_value;\n\t}\n\n\tdouble as_double() const {\n\t\tassert(value_type == DOUBLE);\n\t\treturn double_value;\n\t}\n};\n\ntypedef std::map<std::string, metadata> metadata_map;\n\n// Tag class to disambiguate values/wires and their aliases.\nstruct debug_alias {};\n\n// Tag declaration to disambiguate values and debug outlines.\nusing debug_outline = ::_cxxrtl_outline;\n\n// This structure is intended for consumption via foreign function interfaces, like Python's ctypes.\n// Because of this it uses a C-style layout that is easy to parse rather than more idiomatic C++.\n//\n// To avoid violating strict aliasing rules, this structure has to be a subclass of the one used\n// in the C API, or it would not be possible to cast between the pointers to these.\n//\n// The `attrs` member cannot be owned by this structure because a `cxxrtl_object` can be created\n// from external C code.\nstruct debug_item : ::cxxrtl_object {\n\t// Object types.\n\tenum : uint32_t {\n\t\tVALUE   = CXXRTL_VALUE,\n\t\tWIRE    = CXXRTL_WIRE,\n\t\tMEMORY  = CXXRTL_MEMORY,\n\t\tALIAS   = CXXRTL_ALIAS,\n\t\tOUTLINE = CXXRTL_OUTLINE,\n\t};\n\n\t// Object flags.\n\tenum : uint32_t {\n\t\tINPUT  = CXXRTL_INPUT,\n\t\tOUTPUT = CXXRTL_OUTPUT,\n\t\tINOUT  = CXXRTL_INOUT,\n\t\tDRIVEN_SYNC = CXXRTL_DRIVEN_SYNC,\n\t\tDRIVEN_COMB = CXXRTL_DRIVEN_COMB,\n\t\tUNDRIVEN    = CXXRTL_UNDRIVEN,\n\t};\n\n\tdebug_item(const ::cxxrtl_object &object) : cxxrtl_object(object) {}\n\n\ttemplate<size_t Bits>\n\tdebug_item(value<Bits> &item, size_t lsb_offset = 0, uint32_t flags_ = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = VALUE;\n\t\tflags   = flags_;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = item.data;\n\t\tnext    = item.data;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(const value<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = VALUE;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.data);\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(wire<Bits> &item, size_t lsb_offset = 0, uint32_t flags_ = 0) {\n\t\tstatic_assert(sizeof(item.curr) == value<Bits>::chunks * sizeof(chunk_t) &&\n\t\t              sizeof(item.next) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"wire<Bits> is not compatible with C layout\");\n\t\ttype    = WIRE;\n\t\tflags   = flags_;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = item.curr.data;\n\t\tnext    = item.next.data;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Width>\n\tdebug_item(memory<Width> &item, size_t zero_offset = 0) {\n\t\tstatic_assert(sizeof(item.data[0]) == value<Width>::chunks * sizeof(chunk_t),\n\t\t              \"memory<Width> is not compatible with C layout\");\n\t\ttype    = MEMORY;\n\t\tflags   = 0;\n\t\twidth   = Width;\n\t\tlsb_at  = 0;\n\t\tdepth   = item.depth;\n\t\tzero_at = zero_offset;\n\t\tcurr    = item.data ? item.data[0].data : nullptr;\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(debug_alias, const value<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = ALIAS;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.data);\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(debug_alias, const wire<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item.curr) == value<Bits>::chunks * sizeof(chunk_t) &&\n\t\t              sizeof(item.next) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"wire<Bits> is not compatible with C layout\");\n\t\ttype    = ALIAS;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.curr.data);\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(debug_outline &group, const value<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = OUTLINE;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.data);\n\t\tnext    = nullptr;\n\t\toutline = &group;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits, class IntegerT>\n\tIntegerT get() const {\n\t\tassert(width == Bits && depth == 1);\n\t\tvalue<Bits> item;\n\t\tstd::copy(curr, curr + value<Bits>::chunks, item.data);\n\t\treturn item.template get<IntegerT>();\n\t}\n\n\ttemplate<size_t Bits, class IntegerT>\n\tvoid set(IntegerT other) const {\n\t\tassert(width == Bits && depth == 1);\n\t\tvalue<Bits> item;\n\t\titem.template set<IntegerT>(other);\n\t\tstd::copy(item.data, item.data + value<Bits>::chunks, next);\n\t}\n};\nstatic_assert(std::is_standard_layout<debug_item>::value, \"debug_item is not compatible with C layout\");\n\n} // namespace cxxrtl\n\ntypedef struct _cxxrtl_attr_set {\n\tcxxrtl::metadata_map map;\n} *cxxrtl_attr_set;\n\nnamespace cxxrtl {\n\n// Representation of an attribute set in the C++ interface.\nusing debug_attrs = ::_cxxrtl_attr_set;\n\nstruct debug_items {\n\tstd::map<std::string, std::vector<debug_item>> table;\n\tstd::map<std::string, std::unique_ptr<debug_attrs>> attrs_table;\n\n\tvoid add(const std::string &name, debug_item &&item, metadata_map &&item_attrs = {}) {\n\t\tstd::unique_ptr<debug_attrs> &attrs = attrs_table[name];\n\t\tif (attrs.get() == nullptr)\n\t\t\tattrs = std::unique_ptr<debug_attrs>(new debug_attrs);\n\t\tfor (auto attr : item_attrs)\n\t\t\tattrs->map.insert(attr);\n\t\titem.attrs = attrs.get();\n\t\tstd::vector<debug_item> &parts = table[name];\n\t\tparts.emplace_back(item);\n\t\tstd::sort(parts.begin(), parts.end(),\n\t\t\t[](const debug_item &a, const debug_item &b) {\n\t\t\t\treturn a.lsb_at < b.lsb_at;\n\t\t\t});\n\t}\n\n\tsize_t count(const std::string &name) const {\n\t\tif (table.count(name) == 0)\n\t\t\treturn 0;\n\t\treturn table.at(name).size();\n\t}\n\n\tconst std::vector<debug_item> &parts_at(const std::string &name) const {\n\t\treturn table.at(name);\n\t}\n\n\tconst debug_item &at(const std::string &name) const {\n\t\tconst std::vector<debug_item> &parts = table.at(name);\n\t\tassert(parts.size() == 1);\n\t\treturn parts.at(0);\n\t}\n\n\tconst debug_item &operator [](const std::string &name) const {\n\t\treturn at(name);\n\t}\n\n\tconst metadata_map &attrs(const std::string &name) const {\n\t\treturn attrs_table.at(name)->map;\n\t}\n};\n\n// Tag class to disambiguate the default constructor used by the toplevel module that calls reset(),\n// and the constructor of interior modules that should not call it.\nstruct interior {};\n\nstruct module {\n\tmodule() {}\n\tvirtual ~module() {}\n\n\t// Modules with black boxes cannot be copied. Although not all designs include black boxes,\n\t// delete the copy constructor and copy assignment operator to make sure that any downstream\n\t// code that manipulates modules doesn't accidentally depend on their availability.\n\tmodule(const module &) = delete;\n\tmodule &operator=(const module &) = delete;\n\n\tmodule(module &&) = default;\n\tmodule &operator=(module &&) = default;\n\n\tvirtual void reset() = 0;\n\n\tvirtual bool eval() = 0;\n\tvirtual bool commit() = 0;\n\n\tunsigned int steps = 0;\n\n\tsize_t step() {\n\t\t++steps;\n\t\tsize_t deltas = 0;\n\t\tbool converged = false;\n\t\tdo {\n\t\t\tconverged = eval();\n\t\t\tdeltas++;\n\t\t} while (commit() && !converged);\n\t\treturn deltas;\n\t}\n\n\tvirtual void debug_info(debug_items &items, std::string path = \"\") {\n\t\t(void)items, (void)path;\n\t}\n};\n\n} // namespace cxxrtl\n\n// Internal structures used to communicate with the implementation of the C interface.\n\ntypedef struct _cxxrtl_toplevel {\n\tstd::unique_ptr<cxxrtl::module> module;\n} *cxxrtl_toplevel;\n\ntypedef struct _cxxrtl_outline {\n\tstd::function<void()> eval;\n} *cxxrtl_outline;\n\n// Definitions of internal Yosys cells. Other than the functions in this namespace, CXXRTL is fully generic\n// and indepenent of Yosys implementation details.\n//\n// The `write_cxxrtl` pass translates internal cells (cells with names that start with `$`) to calls of these\n// functions. All of Yosys arithmetic and logical cells perform sign or zero extension on their operands,\n// whereas basic operations on arbitrary width values require operands to be of the same width. These functions\n// bridge the gap by performing the necessary casts. They are named similar to `cell_A[B]`, where A and B are `u`\n// if the corresponding operand is unsigned, and `s` if it is signed.\nnamespace cxxrtl_yosys {\n\nusing namespace cxxrtl;\n\n// std::max isn't constexpr until C++14 for no particular reason (it's an oversight), so we define our own.\ntemplate<class T>\nCXXRTL_ALWAYS_INLINE\nconstexpr T max(const T &a, const T &b) {\n\treturn a > b ? a : b;\n}\n\n// Logic operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> logic_not(const value<BitsA> &a) {\n\treturn value<BitsY> { a ? 0u : 1u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> logic_and(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn value<BitsY> { (bool(a) && bool(b)) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> logic_or(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn value<BitsY> { (bool(a) || bool(b)) ? 1u : 0u };\n}\n\n// Reduction operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_and(const value<BitsA> &a) {\n\treturn value<BitsY> { a.bit_not().is_zero() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_or(const value<BitsA> &a) {\n\treturn value<BitsY> { a ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_xor(const value<BitsA> &a) {\n\treturn value<BitsY> { (a.ctpop() % 2) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_xnor(const value<BitsA> &a) {\n\treturn value<BitsY> { (a.ctpop() % 2) ? 0u : 1u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_bool(const value<BitsA> &a) {\n\treturn value<BitsY> { a ? 1u : 0u };\n}\n\n// Bitwise operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> not_u(const value<BitsA> &a) {\n\treturn a.template zcast<BitsY>().bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> not_s(const value<BitsA> &a) {\n\treturn a.template scast<BitsY>().bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> and_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_and(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> and_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_and(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> or_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_or(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> or_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_or(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_xor(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_xor(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xnor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_xor(b.template zcast<BitsY>()).bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xnor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_xor(b.template scast<BitsY>()).bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shl_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shl_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshl_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshl_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shr_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.shr(b).template zcast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shr_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.shr(b).template scast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshr_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.shr(b).template zcast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshr_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.sshr(b).template scast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shr_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shr_su<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_us(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn b.is_neg() ? shl_uu<BitsY>(a, b.template sext<BitsB + 1>().neg()) : shr_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn b.is_neg() ? shl_su<BitsY>(a, b.template sext<BitsB + 1>().neg()) : shr_su<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_su<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_us(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_us<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_ss<BitsY>(a, b);\n}\n\n// Comparison operations\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eq_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template zext<BitsExt>() == b.template zext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eq_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template sext<BitsExt>() == b.template sext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ne_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template zext<BitsExt>() != b.template zext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ne_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template sext<BitsExt>() != b.template sext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eqx_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn eq_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eqx_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn eq_ss<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> nex_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn ne_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> nex_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn ne_ss<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> gt_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { b.template zext<BitsExt>().ucmp(a.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> gt_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { b.template sext<BitsExt>().scmp(a.template sext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ge_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !a.template zext<BitsExt>().ucmp(b.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ge_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !a.template sext<BitsExt>().scmp(b.template sext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> lt_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { a.template zext<BitsExt>().ucmp(b.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> lt_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { a.template sext<BitsExt>().scmp(b.template sext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> le_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !b.template zext<BitsExt>().ucmp(a.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> le_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !b.template sext<BitsExt>().scmp(a.template sext<BitsExt>()) ? 1u : 0u };\n}\n\n// Arithmetic operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> pos_u(const value<BitsA> &a) {\n\treturn a.template zcast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> pos_s(const value<BitsA> &a) {\n\treturn a.template scast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> neg_u(const value<BitsA> &a) {\n\treturn a.template zcast<BitsY>().neg();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> neg_s(const value<BitsA> &a) {\n\treturn a.template scast<BitsY>().neg();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> add_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().add(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> add_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().add(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sub_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().sub(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sub_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().sub(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mul_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsM = BitsA >= BitsB ? BitsA : BitsB;\n\treturn a.template zcast<BitsM>().template mul<BitsY>(b.template zcast<BitsM>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mul_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().template mul<BitsY>(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nstd::pair<value<BitsY>, value<BitsY>> divmod_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t Bits = max(BitsY, max(BitsA, BitsB));\n\tvalue<Bits> quotient;\n\tvalue<Bits> remainder;\n\tvalue<Bits> dividend = a.template zext<Bits>();\n\tvalue<Bits> divisor = b.template zext<Bits>();\n\tstd::tie(quotient, remainder) = dividend.udivmod(divisor);\n\treturn {quotient.template trunc<BitsY>(), remainder.template trunc<BitsY>()};\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nstd::pair<value<BitsY>, value<BitsY>> divmod_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t Bits = max(BitsY, max(BitsA, BitsB));\n\tvalue<Bits> quotient;\n\tvalue<Bits> remainder;\n\tvalue<Bits> dividend = a.template sext<Bits>();\n\tvalue<Bits> divisor = b.template sext<Bits>();\n\tstd::tie(quotient, remainder) = dividend.sdivmod(divisor);\n\treturn {quotient.template trunc<BitsY>(), remainder.template trunc<BitsY>()};\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> div_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).first;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> div_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_ss<BitsY>(a, b).first;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mod_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).second;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mod_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_ss<BitsY>(a, b).second;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> modfloor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).second;\n}\n\n// GHDL Modfloor operator. Returns r=a mod b, such that r has the same sign as b and\n// a=b*N+r where N is some integer\n// In practical terms, when a and b have different signs and the remainder returned by divmod_ss is not 0\n// then return the remainder + b\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> modfloor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tvalue<BitsY> r;\n\tr = divmod_ss<BitsY>(a, b).second;\n\tif((b.is_neg() != a.is_neg()) && !r.is_zero())\n\t\treturn add_ss<BitsY>(b, r);\n\treturn r;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> divfloor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).first;\n}\n\n// Divfloor. Similar to above: returns q=a//b, where q has the sign of a*b and a=b*q+N.\n// In other words, returns (truncating) a/b, except if a and b have different signs\n// and there's non-zero remainder, subtract one more towards floor.\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> divfloor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tvalue<BitsY> q, r;\n\tstd::tie(q, r) = divmod_ss<BitsY>(a, b);\n\tif ((b.is_neg() != a.is_neg()) && !r.is_zero())\n\t\treturn sub_uu<BitsY>(q, value<1> { 1u });\n\treturn q;\n\n}\n\n// Memory helper\nstruct memory_index {\n\tbool valid;\n\tsize_t index;\n\n\ttemplate<size_t BitsAddr>\n\tmemory_index(const value<BitsAddr> &addr, size_t offset, size_t depth) {\n\t\tstatic_assert(value<BitsAddr>::chunks <= 1, \"memory address is too wide\");\n\t\tsize_t offset_index = addr.data[0];\n\n\t\tvalid = (offset_index >= offset && offset_index < offset + depth);\n\t\tindex = offset_index - offset;\n\t}\n};\n\n} // namespace cxxrtl_yosys\n\n#endif\n",
+                            "cxxrtl.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2019-2020  whitequark <whitequark@whitequark.org>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n// This file is included by the designs generated with `write_cxxrtl`. It is not used in Yosys itself.\n//\n// The CXXRTL support library implements compile time specialized arbitrary width arithmetics, as well as provides\n// composite lvalues made out of bit slices and concatenations of lvalues. This allows the `write_cxxrtl` pass\n// to perform a straightforward translation of RTLIL structures to readable C++, relying on the C++ compiler\n// to unwrap the abstraction and generate efficient code.\n\n#ifndef CXXRTL_H\n#define CXXRTL_H\n\n#include <cstddef>\n#include <cstdint>\n#include <cassert>\n#include <limits>\n#include <type_traits>\n#include <tuple>\n#include <vector>\n#include <map>\n#include <algorithm>\n#include <memory>\n#include <functional>\n#include <sstream>\n\n// `cxxrtl::debug_item` has to inherit from `cxxrtl_object` to satisfy strict aliasing requirements.\n#include <cxxrtl/capi/cxxrtl_capi.h>\n\n#ifndef __has_attribute\n#\tdefine __has_attribute(x) 0\n#endif\n\n// CXXRTL essentially uses the C++ compiler as a hygienic macro engine that feeds an instruction selector.\n// It generates a lot of specialized template functions with relatively large bodies that, when inlined\n// into the caller and (for those with loops) unrolled, often expose many new optimization opportunities.\n// Because of this, most of the CXXRTL runtime must be always inlined for best performance.\n#if __has_attribute(always_inline)\n#define CXXRTL_ALWAYS_INLINE inline __attribute__((__always_inline__))\n#else\n#define CXXRTL_ALWAYS_INLINE inline\n#endif\n// Conversely, some functions in the generated code are extremely large yet very cold, with both of these\n// properties being extreme enough to confuse C++ compilers into spending pathological amounts of time\n// on a futile (the code becomes worse) attempt to optimize the least important parts of code.\n#if __has_attribute(optnone)\n#define CXXRTL_EXTREMELY_COLD __attribute__((__optnone__))\n#elif __has_attribute(optimize)\n#define CXXRTL_EXTREMELY_COLD __attribute__((__optimize__(0)))\n#else\n#define CXXRTL_EXTREMELY_COLD\n#endif\n\n// CXXRTL uses assert() to check for C++ contract violations (which may result in e.g. undefined behavior\n// of the simulation code itself), and CXXRTL_ASSERT to check for RTL contract violations (which may at\n// most result in undefined simulation results).\n//\n// Though by default, CXXRTL_ASSERT() expands to assert(), it may be overridden e.g. when integrating\n// the simulation into another process that should survive violating RTL contracts.\n#ifndef CXXRTL_ASSERT\n#ifndef CXXRTL_NDEBUG\n#define CXXRTL_ASSERT(x) assert(x)\n#else\n#define CXXRTL_ASSERT(x)\n#endif\n#endif\n\nnamespace cxxrtl {\n\n// All arbitrary-width values in CXXRTL are backed by arrays of unsigned integers called chunks. The chunk size\n// is the same regardless of the value width to simplify manipulating values via FFI interfaces, e.g. driving\n// and introspecting the simulation in Python.\n//\n// It is practical to use chunk sizes between 32 bits and platform register size because when arithmetics on\n// narrower integer types is legalized by the C++ compiler, it inserts code to clear the high bits of the register.\n// However, (a) most of our operations do not change those bits in the first place because of invariants that are\n// invisible to the compiler, (b) we often operate on non-power-of-2 values and have to clear the high bits anyway.\n// Therefore, using relatively wide chunks and clearing the high bits explicitly and only when we know they may be\n// clobbered results in simpler generated code.\ntypedef uint32_t chunk_t;\ntypedef uint64_t wide_chunk_t;\n\ntemplate<typename T>\nstruct chunk_traits {\n\tstatic_assert(std::is_integral<T>::value && std::is_unsigned<T>::value,\n\t              \"chunk type must be an unsigned integral type\");\n\tusing type = T;\n\tstatic constexpr size_t bits = std::numeric_limits<T>::digits;\n\tstatic constexpr T mask = std::numeric_limits<T>::max();\n};\n\ntemplate<class T>\nstruct expr_base;\n\ntemplate<size_t Bits>\nstruct value : public expr_base<value<Bits>> {\n\tstatic constexpr size_t bits = Bits;\n\n\tusing chunk = chunk_traits<chunk_t>;\n\tstatic constexpr chunk::type msb_mask = (Bits % chunk::bits == 0) ? chunk::mask\n\t\t: chunk::mask >> (chunk::bits - (Bits % chunk::bits));\n\n\tstatic constexpr size_t chunks = (Bits + chunk::bits - 1) / chunk::bits;\n\tchunk::type data[chunks] = {};\n\n\tvalue() = default;\n\ttemplate<typename... Init>\n\texplicit constexpr value(Init ...init) : data{init...} {}\n\n\tvalue(const value<Bits> &) = default;\n\tvalue<Bits> &operator=(const value<Bits> &) = default;\n\n\tvalue(value<Bits> &&) = default;\n\tvalue<Bits> &operator=(value<Bits> &&) = default;\n\n\t// A (no-op) helper that forces the cast to value<>.\n\tCXXRTL_ALWAYS_INLINE\n\tconst value<Bits> &val() const {\n\t\treturn *this;\n\t}\n\n\tstd::string str() const {\n\t\tstd::stringstream ss;\n\t\tss << *this;\n\t\treturn ss.str();\n\t}\n\n\t// Conversion operations.\n\t//\n\t// These functions ensure that a conversion is never out of range, and should be always used, if at all\n\t// possible, instead of direct manipulation of the `data` member. For very large types, .slice() and\n\t// .concat() can be used to split them into more manageable parts.\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tIntegerT get() const {\n\t\tstatic_assert(std::numeric_limits<IntegerT>::is_integer && !std::numeric_limits<IntegerT>::is_signed,\n\t\t              \"get<T>() requires T to be an unsigned integral type\");\n\t\tstatic_assert(std::numeric_limits<IntegerT>::digits >= Bits,\n\t\t              \"get<T>() requires T to be at least as wide as the value is\");\n\t\tIntegerT result = 0;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult |= IntegerT(data[n]) << (n * chunk::bits);\n\t\treturn result;\n\t}\n\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tvoid set(IntegerT other) {\n\t\tstatic_assert(std::numeric_limits<IntegerT>::is_integer && !std::numeric_limits<IntegerT>::is_signed,\n\t\t              \"set<T>() requires T to be an unsigned integral type\");\n\t\tstatic_assert(std::numeric_limits<IntegerT>::digits >= Bits,\n\t\t              \"set<T>() requires the value to be at least as wide as T is\");\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tdata[n] = (other >> (n * chunk::bits)) & chunk::mask;\n\t}\n\n\t// Operations with compile-time parameters.\n\t//\n\t// These operations are used to implement slicing, concatenation, and blitting.\n\t// The trunc, zext and sext operations add or remove most significant bits (i.e. on the left);\n\t// the rtrunc and rzext operations add or remove least significant bits (i.e. on the right).\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> trunc() const {\n\t\tstatic_assert(NewBits <= Bits, \"trunc() may not increase width\");\n\t\tvalue<NewBits> result;\n\t\tfor (size_t n = 0; n < result.chunks; n++)\n\t\t\tresult.data[n] = data[n];\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> zext() const {\n\t\tstatic_assert(NewBits >= Bits, \"zext() may not decrease width\");\n\t\tvalue<NewBits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n];\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> sext() const {\n\t\tstatic_assert(NewBits >= Bits, \"sext() may not decrease width\");\n\t\tvalue<NewBits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n];\n\t\tif (is_neg()) {\n\t\t\tresult.data[chunks - 1] |= ~msb_mask;\n\t\t\tfor (size_t n = chunks; n < result.chunks; n++)\n\t\t\t\tresult.data[n] = chunk::mask;\n\t\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\t}\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> rtrunc() const {\n\t\tstatic_assert(NewBits <= Bits, \"rtrunc() may not increase width\");\n\t\tvalue<NewBits> result;\n\t\tconstexpr size_t shift_chunks = (Bits - NewBits) / chunk::bits;\n\t\tconstexpr size_t shift_bits   = (Bits - NewBits) % chunk::bits;\n\t\tchunk::type carry = 0;\n\t\tif (shift_chunks + result.chunks < chunks) {\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[shift_chunks + result.chunks] << (chunk::bits - shift_bits);\n\t\t}\n\t\tfor (size_t n = result.chunks; n > 0; n--) {\n\t\t\tresult.data[n - 1] = carry | (data[shift_chunks + n - 1] >> shift_bits);\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[shift_chunks + n - 1] << (chunk::bits - shift_bits);\n\t\t}\n\t\treturn result;\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> rzext() const {\n\t\tstatic_assert(NewBits >= Bits, \"rzext() may not decrease width\");\n\t\tvalue<NewBits> result;\n\t\tconstexpr size_t shift_chunks = (NewBits - Bits) / chunk::bits;\n\t\tconstexpr size_t shift_bits   = (NewBits - Bits) % chunk::bits;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tresult.data[shift_chunks + n] = (data[n] << shift_bits) | carry;\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[n] >> (chunk::bits - shift_bits);\n\t\t}\n\t\tif (shift_chunks + chunks < result.chunks)\n\t\t\tresult.data[shift_chunks + chunks] = carry;\n\t\treturn result;\n\t}\n\n\t// Bit blit operation, i.e. a partial read-modify-write.\n\ttemplate<size_t Stop, size_t Start>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<Bits> blit(const value<Stop - Start + 1> &source) const {\n\t\tstatic_assert(Stop >= Start, \"blit() may not reverse bit order\");\n\t\tconstexpr chunk::type start_mask = ~(chunk::mask << (Start % chunk::bits));\n\t\tconstexpr chunk::type stop_mask = (Stop % chunk::bits + 1 == chunk::bits) ? 0\n\t\t\t: (chunk::mask << (Stop % chunk::bits + 1));\n\t\tvalue<Bits> masked = *this;\n\t\tif (Start / chunk::bits == Stop / chunk::bits) {\n\t\t\tmasked.data[Start / chunk::bits] &= stop_mask | start_mask;\n\t\t} else {\n\t\t\tmasked.data[Start / chunk::bits] &= start_mask;\n\t\t\tfor (size_t n = Start / chunk::bits + 1; n < Stop / chunk::bits; n++)\n\t\t\t\tmasked.data[n] = 0;\n\t\t\tmasked.data[Stop / chunk::bits] &= stop_mask;\n\t\t}\n\t\tvalue<Bits> shifted = source\n\t\t\t.template rzext<Stop + 1>()\n\t\t\t.template zext<Bits>();\n\t\treturn masked.bit_or(shifted);\n\t}\n\n\t// Helpers for selecting extending or truncating operation depending on whether the result is wider or narrower\n\t// than the operand. In C++17 these can be replaced with `if constexpr`.\n\ttemplate<size_t NewBits, typename = void>\n\tstruct zext_cast {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template zext<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits>\n\tstruct zext_cast<NewBits, typename std::enable_if<(NewBits < Bits)>::type> {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template trunc<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits, typename = void>\n\tstruct sext_cast {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template sext<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits>\n\tstruct sext_cast<NewBits, typename std::enable_if<(NewBits < Bits)>::type> {\n\t\tCXXRTL_ALWAYS_INLINE\n\t\tvalue<NewBits> operator()(const value<Bits> &val) {\n\t\t\treturn val.template trunc<NewBits>();\n\t\t}\n\t};\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> zcast() const {\n\t\treturn zext_cast<NewBits>()(*this);\n\t}\n\n\ttemplate<size_t NewBits>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<NewBits> scast() const {\n\t\treturn sext_cast<NewBits>()(*this);\n\t}\n\n\t// Bit replication is far more efficient than the equivalent concatenation.\n\ttemplate<size_t Count>\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<Bits * Count> repeat() const {\n\t\tstatic_assert(Bits == 1, \"repeat() is implemented only for 1-bit values\");\n\t\treturn *this ? value<Bits * Count>().bit_not() : value<Bits * Count>();\n\t}\n\n\t// Operations with run-time parameters (offsets, amounts, etc).\n\t//\n\t// These operations are used for computations.\n\tbool bit(size_t offset) const {\n\t\treturn data[offset / chunk::bits] & (1 << (offset % chunk::bits));\n\t}\n\n\tvoid set_bit(size_t offset, bool value = true) {\n\t\tsize_t offset_chunks = offset / chunk::bits;\n\t\tsize_t offset_bits = offset % chunk::bits;\n\t\tdata[offset_chunks] &= ~(1 << offset_bits);\n\t\tdata[offset_chunks] |= value ? 1 << offset_bits : 0;\n\t}\n\n\texplicit operator bool() const {\n\t\treturn !is_zero();\n\t}\n\n\tbool is_zero() const {\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tif (data[n] != 0)\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\n\tbool is_neg() const {\n\t\treturn data[chunks - 1] & (1 << ((Bits - 1) % chunk::bits));\n\t}\n\n\tbool operator ==(const value<Bits> &other) const {\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tif (data[n] != other.data[n])\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\n\tbool operator !=(const value<Bits> &other) const {\n\t\treturn !(*this == other);\n\t}\n\n\tvalue<Bits> bit_not() const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = ~data[n];\n\t\tresult.data[chunks - 1] &= msb_mask;\n\t\treturn result;\n\t}\n\n\tvalue<Bits> bit_and(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n] & other.data[n];\n\t\treturn result;\n\t}\n\n\tvalue<Bits> bit_or(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n] | other.data[n];\n\t\treturn result;\n\t}\n\n\tvalue<Bits> bit_xor(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tfor (size_t n = 0; n < chunks; n++)\n\t\t\tresult.data[n] = data[n] ^ other.data[n];\n\t\treturn result;\n\t}\n\n\tvalue<Bits> update(const value<Bits> &val, const value<Bits> &mask) const {\n\t\treturn bit_and(mask.bit_not()).bit_or(val.bit_and(mask));\n\t}\n\n\ttemplate<size_t AmountBits>\n\tvalue<Bits> shl(const value<AmountBits> &amount) const {\n\t\t// Ensure our early return is correct by prohibiting values larger than 4 Gbit.\n\t\tstatic_assert(Bits <= chunk::mask, \"shl() of unreasonably large values is not supported\");\n\t\t// Detect shifts definitely large than Bits early.\n\t\tfor (size_t n = 1; n < amount.chunks; n++)\n\t\t\tif (amount.data[n] != 0)\n\t\t\t\treturn {};\n\t\t// Past this point we can use the least significant chunk as the shift size.\n\t\tsize_t shift_chunks = amount.data[0] / chunk::bits;\n\t\tsize_t shift_bits   = amount.data[0] % chunk::bits;\n\t\tif (shift_chunks >= chunks)\n\t\t\treturn {};\n\t\tvalue<Bits> result;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks - shift_chunks; n++) {\n\t\t\tresult.data[shift_chunks + n] = (data[n] << shift_bits) | carry;\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[n] >> (chunk::bits - shift_bits);\n\t\t}\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\ttemplate<size_t AmountBits, bool Signed = false>\n\tvalue<Bits> shr(const value<AmountBits> &amount) const {\n\t\t// Ensure our early return is correct by prohibiting values larger than 4 Gbit.\n\t\tstatic_assert(Bits <= chunk::mask, \"shr() of unreasonably large values is not supported\");\n\t\t// Detect shifts definitely large than Bits early.\n\t\tfor (size_t n = 1; n < amount.chunks; n++)\n\t\t\tif (amount.data[n] != 0)\n\t\t\t\treturn (Signed && is_neg()) ? value<Bits>().bit_not() : value<Bits>();\n\t\t// Past this point we can use the least significant chunk as the shift size.\n\t\tsize_t shift_chunks = amount.data[0] / chunk::bits;\n\t\tsize_t shift_bits   = amount.data[0] % chunk::bits;\n\t\tif (shift_chunks >= chunks)\n\t\t\treturn (Signed && is_neg()) ? value<Bits>().bit_not() : value<Bits>();\n\t\tvalue<Bits> result;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks - shift_chunks; n++) {\n\t\t\tresult.data[chunks - shift_chunks - 1 - n] = carry | (data[chunks - 1 - n] >> shift_bits);\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[chunks - 1 - n] << (chunk::bits - shift_bits);\n\t\t}\n\t\tif (Signed && is_neg()) {\n\t\t\tsize_t top_chunk_idx  = amount.data[0] > Bits ? 0 : (Bits - amount.data[0]) / chunk::bits;\n\t\t\tsize_t top_chunk_bits = amount.data[0] > Bits ? 0 : (Bits - amount.data[0]) % chunk::bits;\n\t\t\tfor (size_t n = top_chunk_idx + 1; n < chunks; n++)\n\t\t\t\tresult.data[n] = chunk::mask;\n\t\t\tif (amount.data[0] != 0)\n\t\t\t\tresult.data[top_chunk_idx] |= chunk::mask << top_chunk_bits;\n\t\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\t}\n\t\treturn result;\n\t}\n\n\ttemplate<size_t AmountBits>\n\tvalue<Bits> sshr(const value<AmountBits> &amount) const {\n\t\treturn shr<AmountBits, /*Signed=*/true>(amount);\n\t}\n\n\ttemplate<size_t ResultBits, size_t SelBits>\n\tvalue<ResultBits> bmux(const value<SelBits> &sel) const {\n\t\tstatic_assert(ResultBits << SelBits == Bits, \"invalid sizes used in bmux()\");\n\t\tsize_t amount = sel.data[0] * ResultBits;\n\t\tsize_t shift_chunks = amount / chunk::bits;\n\t\tsize_t shift_bits   = amount % chunk::bits;\n\t\tvalue<ResultBits> result;\n\t\tchunk::type carry = 0;\n\t\tif (ResultBits % chunk::bits + shift_bits > chunk::bits)\n\t\t\tcarry = data[result.chunks + shift_chunks] << (chunk::bits - shift_bits);\n\t\tfor (size_t n = 0; n < result.chunks; n++) {\n\t\t\tresult.data[result.chunks - 1 - n] = carry | (data[result.chunks + shift_chunks - 1 - n] >> shift_bits);\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[result.chunks + shift_chunks - 1 - n] << (chunk::bits - shift_bits);\n\t\t}\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\ttemplate<size_t ResultBits, size_t SelBits>\n\tvalue<ResultBits> demux(const value<SelBits> &sel) const {\n\t\tstatic_assert(Bits << SelBits == ResultBits, \"invalid sizes used in demux()\");\n\t\tsize_t amount = sel.data[0] * Bits;\n\t\tsize_t shift_chunks = amount / chunk::bits;\n\t\tsize_t shift_bits   = amount % chunk::bits;\n\t\tvalue<ResultBits> result;\n\t\tchunk::type carry = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tresult.data[shift_chunks + n] = (data[n] << shift_bits) | carry;\n\t\t\tcarry = (shift_bits == 0) ? 0\n\t\t\t\t: data[n] >> (chunk::bits - shift_bits);\n\t\t}\n\t\tif (Bits % chunk::bits + shift_bits > chunk::bits)\n\t\t\tresult.data[shift_chunks + chunks] = carry;\n\t\treturn result;\n\t}\n\n\tsize_t ctpop() const {\n\t\tsize_t count = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\t// This loop implements the population count idiom as recognized by LLVM and GCC.\n\t\t\tfor (chunk::type x = data[n]; x != 0; count++)\n\t\t\t\tx = x & (x - 1);\n\t\t}\n\t\treturn count;\n\t}\n\n\tsize_t ctlz() const {\n\t\tsize_t count = 0;\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tchunk::type x = data[chunks - 1 - n];\n\t\t\t// First add to `count` as if the chunk is zero\n\t\t\tconstexpr size_t msb_chunk_bits = Bits % chunk::bits != 0 ? Bits % chunk::bits : chunk::bits;\n\t\t\tcount += (n == 0 ? msb_chunk_bits : chunk::bits);\n\t\t\t// If the chunk isn't zero, correct the `count` value and return\n\t\t\tif (x != 0) {\n\t\t\t\tfor (; x != 0; count--)\n\t\t\t\t\tx >>= 1;\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\t\treturn count;\n\t}\n\n\ttemplate<bool Invert, bool CarryIn>\n\tstd::pair<value<Bits>, bool /*CarryOut*/> alu(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tbool carry = CarryIn;\n\t\tfor (size_t n = 0; n < result.chunks; n++) {\n\t\t\tresult.data[n] = data[n] + (Invert ? ~other.data[n] : other.data[n]) + carry;\n\t\t\tif (result.chunks - 1 == n)\n\t\t\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\t\tcarry = (result.data[n] <  data[n]) ||\n\t\t\t        (result.data[n] == data[n] && carry);\n\t\t}\n\t\treturn {result, carry};\n\t}\n\n\tvalue<Bits> add(const value<Bits> &other) const {\n\t\treturn alu</*Invert=*/false, /*CarryIn=*/false>(other).first;\n\t}\n\n\tvalue<Bits> sub(const value<Bits> &other) const {\n\t\treturn alu</*Invert=*/true, /*CarryIn=*/true>(other).first;\n\t}\n\n\tvalue<Bits> neg() const {\n\t\treturn value<Bits> { 0u }.sub(*this);\n\t}\n\n\tbool ucmp(const value<Bits> &other) const {\n\t\tbool carry;\n\t\tstd::tie(std::ignore, carry) = alu</*Invert=*/true, /*CarryIn=*/true>(other);\n\t\treturn !carry; // a.ucmp(b) ≡ a u< b\n\t}\n\n\tbool scmp(const value<Bits> &other) const {\n\t\tvalue<Bits> result;\n\t\tbool carry;\n\t\tstd::tie(result, carry) = alu</*Invert=*/true, /*CarryIn=*/true>(other);\n\t\tbool overflow = (is_neg() == !other.is_neg()) && (is_neg() != result.is_neg());\n\t\treturn result.is_neg() ^ overflow; // a.scmp(b) ≡ a s< b\n\t}\n\n\ttemplate<size_t ResultBits>\n\tvalue<ResultBits> mul(const value<Bits> &other) const {\n\t\tvalue<ResultBits> result;\n\t\twide_chunk_t wide_result[result.chunks + 1] = {};\n\t\tfor (size_t n = 0; n < chunks; n++) {\n\t\t\tfor (size_t m = 0; m < chunks && n + m < result.chunks; m++) {\n\t\t\t\twide_result[n + m] += wide_chunk_t(data[n]) * wide_chunk_t(other.data[m]);\n\t\t\t\twide_result[n + m + 1] += wide_result[n + m] >> chunk::bits;\n\t\t\t\twide_result[n + m] &= chunk::mask;\n\t\t\t}\n\t\t}\n\t\tfor (size_t n = 0; n < result.chunks; n++) {\n\t\t\tresult.data[n] = wide_result[n];\n\t\t}\n\t\tresult.data[result.chunks - 1] &= result.msb_mask;\n\t\treturn result;\n\t}\n\n\tstd::pair<value<Bits>, value<Bits>> udivmod(value<Bits> divisor) const {\n\t\tvalue<Bits> quotient;\n\t\tvalue<Bits> dividend = *this;\n\t\tif (dividend.ucmp(divisor))\n\t\t\treturn {/*quotient=*/value<Bits>{0u}, /*remainder=*/dividend};\n\t\tint64_t divisor_shift = divisor.ctlz() - dividend.ctlz();\n\t\tassert(divisor_shift >= 0);\n\t\tdivisor = divisor.shl(value<Bits>{(chunk::type) divisor_shift});\n\t\tfor (size_t step = 0; step <= divisor_shift; step++) {\n\t\t\tquotient = quotient.shl(value<Bits>{1u});\n\t\t\tif (!dividend.ucmp(divisor)) {\n\t\t\t\tdividend = dividend.sub(divisor);\n\t\t\t\tquotient.set_bit(0, true);\n\t\t\t}\n\t\t\tdivisor = divisor.shr(value<Bits>{1u});\n\t\t}\n\t\treturn {quotient, /*remainder=*/dividend};\n\t}\n\n\tstd::pair<value<Bits>, value<Bits>> sdivmod(const value<Bits> &other) const {\n\t\tvalue<Bits + 1> quotient;\n\t\tvalue<Bits + 1> remainder;\n\t\tvalue<Bits + 1> dividend = sext<Bits + 1>();\n\t\tvalue<Bits + 1> divisor = other.template sext<Bits + 1>();\n\t\tif (dividend.is_neg()) dividend = dividend.neg();\n\t\tif (divisor.is_neg()) divisor = divisor.neg();\n\t\tstd::tie(quotient, remainder) = dividend.udivmod(divisor);\n\t\tif (dividend.is_neg() != divisor.is_neg()) quotient = quotient.neg();\n\t\tif (dividend.is_neg()) remainder = remainder.neg();\n\t\treturn {quotient.template trunc<Bits>(), remainder.template trunc<Bits>()};\n\t}\n};\n\n// Expression template for a slice, usable as lvalue or rvalue, and composable with other expression templates here.\ntemplate<class T, size_t Stop, size_t Start>\nstruct slice_expr : public expr_base<slice_expr<T, Stop, Start>> {\n\tstatic_assert(Stop >= Start, \"slice_expr() may not reverse bit order\");\n\tstatic_assert(Start < T::bits && Stop < T::bits, \"slice_expr() must be within bounds\");\n\tstatic constexpr size_t bits = Stop - Start + 1;\n\n\tT &expr;\n\n\tslice_expr(T &expr) : expr(expr) {}\n\tslice_expr(const slice_expr<T, Stop, Start> &) = delete;\n\n\tCXXRTL_ALWAYS_INLINE\n\toperator value<bits>() const {\n\t\treturn static_cast<const value<T::bits> &>(expr)\n\t\t\t.template rtrunc<T::bits - Start>()\n\t\t\t.template trunc<bits>();\n\t}\n\n\tCXXRTL_ALWAYS_INLINE\n\tslice_expr<T, Stop, Start> &operator=(const value<bits> &rhs) {\n\t\t// Generic partial assignment implemented using a read-modify-write operation on the sliced expression.\n\t\texpr = static_cast<const value<T::bits> &>(expr)\n\t\t\t.template blit<Stop, Start>(rhs);\n\t\treturn *this;\n\t}\n\n\t// A helper that forces the cast to value<>, which allows deduction to work.\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<bits> val() const {\n\t\treturn static_cast<const value<bits> &>(*this);\n\t}\n};\n\n// Expression template for a concatenation, usable as lvalue or rvalue, and composable with other expression templates here.\ntemplate<class T, class U>\nstruct concat_expr : public expr_base<concat_expr<T, U>> {\n\tstatic constexpr size_t bits = T::bits + U::bits;\n\n\tT &ms_expr;\n\tU &ls_expr;\n\n\tconcat_expr(T &ms_expr, U &ls_expr) : ms_expr(ms_expr), ls_expr(ls_expr) {}\n\tconcat_expr(const concat_expr<T, U> &) = delete;\n\n\tCXXRTL_ALWAYS_INLINE\n\toperator value<bits>() const {\n\t\tvalue<bits> ms_shifted = static_cast<const value<T::bits> &>(ms_expr)\n\t\t\t.template rzext<bits>();\n\t\tvalue<bits> ls_extended = static_cast<const value<U::bits> &>(ls_expr)\n\t\t\t.template zext<bits>();\n\t\treturn ms_shifted.bit_or(ls_extended);\n\t}\n\n\tCXXRTL_ALWAYS_INLINE\n\tconcat_expr<T, U> &operator=(const value<bits> &rhs) {\n\t\tms_expr = rhs.template rtrunc<T::bits>();\n\t\tls_expr = rhs.template trunc<U::bits>();\n\t\treturn *this;\n\t}\n\n\t// A helper that forces the cast to value<>, which allows deduction to work.\n\tCXXRTL_ALWAYS_INLINE\n\tvalue<bits> val() const {\n\t\treturn static_cast<const value<bits> &>(*this);\n\t}\n};\n\n// Base class for expression templates, providing helper methods for operations that are valid on both rvalues and lvalues.\n//\n// Note that expression objects (slices and concatenations) constructed in this way should NEVER be captured because\n// they refer to temporaries that will, in general, only live until the end of the statement. For example, both of\n// these snippets perform use-after-free:\n//\n//    const auto &a = val.slice<7,0>().slice<1>();\n//    value<1> b = a;\n//\n//    auto &&c = val.slice<7,0>().slice<1>();\n//    c = value<1>{1u};\n//\n// An easy way to write code using slices and concatenations safely is to follow two simple rules:\n//   * Never explicitly name any type except `value<W>` or `const value<W> &`.\n//   * Never use a `const auto &` or `auto &&` in any such expression.\n// Then, any code that compiles will be well-defined.\ntemplate<class T>\nstruct expr_base {\n\ttemplate<size_t Stop, size_t Start = Stop>\n\tCXXRTL_ALWAYS_INLINE\n\tslice_expr<const T, Stop, Start> slice() const {\n\t\treturn {*static_cast<const T *>(this)};\n\t}\n\n\ttemplate<size_t Stop, size_t Start = Stop>\n\tCXXRTL_ALWAYS_INLINE\n\tslice_expr<T, Stop, Start> slice() {\n\t\treturn {*static_cast<T *>(this)};\n\t}\n\n\ttemplate<class U>\n\tCXXRTL_ALWAYS_INLINE\n\tconcat_expr<const T, typename std::remove_reference<const U>::type> concat(const U &other) const {\n\t\treturn {*static_cast<const T *>(this), other};\n\t}\n\n\ttemplate<class U>\n\tCXXRTL_ALWAYS_INLINE\n\tconcat_expr<T, typename std::remove_reference<U>::type> concat(U &&other) {\n\t\treturn {*static_cast<T *>(this), other};\n\t}\n};\n\ntemplate<size_t Bits>\nstd::ostream &operator<<(std::ostream &os, const value<Bits> &val) {\n\tauto old_flags = os.flags(std::ios::right);\n\tauto old_width = os.width(0);\n\tauto old_fill  = os.fill('0');\n\tos << val.bits << '\\'' << std::hex;\n\tfor (size_t n = val.chunks - 1; n != (size_t)-1; n--) {\n\t\tif (n == val.chunks - 1 && Bits % value<Bits>::chunk::bits != 0)\n\t\t\tos.width((Bits % value<Bits>::chunk::bits + 3) / 4);\n\t\telse\n\t\t\tos.width((value<Bits>::chunk::bits + 3) / 4);\n\t\tos << val.data[n];\n\t}\n\tos.fill(old_fill);\n\tos.width(old_width);\n\tos.flags(old_flags);\n\treturn os;\n}\n\ntemplate<size_t Bits>\nstruct value_formatted {\n\tconst value<Bits> &val;\n\tbool character;\n\tbool justify_left;\n\tchar padding;\n\tint width;\n\tint base;\n\tbool signed_;\n\tbool plus;\n\n\tvalue_formatted(const value<Bits> &val, bool character, bool justify_left, char padding, int width, int base, bool signed_, bool plus) :\n\t\tval(val), character(character), justify_left(justify_left), padding(padding), width(width), base(base), signed_(signed_), plus(plus) {}\n\tvalue_formatted(const value_formatted<Bits> &) = delete;\n\tvalue_formatted<Bits> &operator=(const value_formatted<Bits> &rhs) = delete;\n};\n\ntemplate<size_t Bits>\nstd::ostream &operator<<(std::ostream &os, const value_formatted<Bits> &vf)\n{\n\tvalue<Bits> val = vf.val;\n\n\tstd::string buf;\n\n\t// We might want to replace some of these bit() calls with direct\n\t// chunk access if it turns out to be slow enough to matter.\n\n\tif (!vf.character) {\n\t\tsize_t width = Bits;\n\t\tif (vf.base != 10) {\n\t\t\twidth = 0;\n\t\t\tfor (size_t index = 0; index < Bits; index++)\n\t\t\t\tif (val.bit(index))\n\t\t\t\t\twidth = index + 1;\n\t\t}\n\n\t\tif (vf.base == 2) {\n\t\t\tfor (size_t i = width; i > 0; i--)\n\t\t\t\tbuf += (val.bit(i - 1) ? '1' : '0');\n\t\t} else if (vf.base == 8 || vf.base == 16) {\n\t\t\tsize_t step = (vf.base == 16) ? 4 : 3;\n\t\t\tfor (size_t index = 0; index < width; index += step) {\n\t\t\t\tuint8_t value = val.bit(index) | (val.bit(index + 1) << 1) | (val.bit(index + 2) << 2);\n\t\t\t\tif (step == 4)\n\t\t\t\t\tvalue |= val.bit(index + 3) << 3;\n\t\t\t\tbuf += \"0123456789abcdef\"[value];\n\t\t\t}\n\t\t\tstd::reverse(buf.begin(), buf.end());\n\t\t} else if (vf.base == 10) {\n\t\t\tbool negative = vf.signed_ && val.is_neg();\n\t\t\tif (negative)\n\t\t\t\tval = val.neg();\n\t\t\tif (val.is_zero())\n\t\t\t\tbuf += '0';\n\t\t\twhile (!val.is_zero()) {\n\t\t\t\tvalue<Bits> quotient, remainder;\n\t\t\t\tif (Bits >= 4)\n\t\t\t\t\tstd::tie(quotient, remainder) = val.udivmod(value<Bits>{10u});\n\t\t\t\telse\n\t\t\t\t\tstd::tie(quotient, remainder) = std::make_pair(value<Bits>{0u}, val);\n\t\t\t\tbuf += '0' + remainder.template trunc<(Bits > 4 ? 4 : Bits)>().val().template get<uint8_t>();\n\t\t\t\tval = quotient;\n\t\t\t}\n\t\t\tif (negative || vf.plus)\n\t\t\t\tbuf += negative ? '-' : '+';\n\t\t\tstd::reverse(buf.begin(), buf.end());\n\t\t} else assert(false);\n\t} else {\n\t\tbuf.reserve(Bits/8);\n\t\tfor (int i = 0; i < Bits; i += 8) {\n\t\t\tchar ch = 0;\n\t\t\tfor (int j = 0; j < 8 && i + j < int(Bits); j++)\n\t\t\t\tif (val.bit(i + j))\n\t\t\t\t\tch |= 1 << j;\n\t\t\tif (ch != 0)\n\t\t\t\tbuf.append({ch});\n\t\t}\n\t\tstd::reverse(buf.begin(), buf.end());\n\t}\n\n\tassert(vf.width == 0 || vf.padding != '\\0');\n\tif (!vf.justify_left && buf.size() < vf.width) {\n\t\tsize_t pad_width = vf.width - buf.size();\n\t\tif (vf.padding == '0' && (buf.front() == '+' || buf.front() == '-')) {\n\t\t\tos << buf.front();\n\t\t\tbuf.erase(0, 1);\n\t\t}\n\t\tos << std::string(pad_width, vf.padding);\n\t}\n\tos << buf;\n\tif (vf.justify_left && buf.size() < vf.width)\n\t\tos << std::string(vf.width - buf.size(), vf.padding);\n\n\treturn os;\n}\n\n// An object that can be passed to a `commit()` method in order to produce a replay log of every state change in\n// the simulation.\nstruct observer {\n\t// Called when the `commit()` method for a wire is about to update the `chunks` chunks at `base` with `chunks` chunks\n\t// at `value` that have a different bit pattern. It is guaranteed that `chunks` is equal to the wire chunk count and\n\t// `base` points to the first chunk.\n\tvirtual void on_commit(size_t chunks, const chunk_t *base, const chunk_t *value) = 0;\n\n\t// Called when the `commit()` method for a memory is about to update the `chunks` chunks at `&base[chunks * index]`\n\t// with `chunks` chunks at `value` that have a different bit pattern. It is guaranteed that `chunks` is equal to\n\t// the memory element chunk count and `base` points to the first chunk of the first element of the memory.\n\tvirtual void on_commit(size_t chunks, const chunk_t *base, const chunk_t *value, size_t index) = 0;\n};\n\n// The `null_observer` class has the same interface as `observer`, but has no invocation overhead, since its methods\n// are final and have no implementation. This allows the observer feature to be zero-cost when not in use.\nstruct null_observer final: observer {\n\tvoid on_commit(size_t chunks, const chunk_t *base, const chunk_t *value) override {}\n\tvoid on_commit(size_t chunks, const chunk_t *base, const chunk_t *value, size_t index) override {}\n};\n\ntemplate<size_t Bits>\nstruct wire {\n\tstatic constexpr size_t bits = Bits;\n\n\tvalue<Bits> curr;\n\tvalue<Bits> next;\n\n\twire() = default;\n\texplicit constexpr wire(const value<Bits> &init) : curr(init), next(init) {}\n\ttemplate<typename... Init>\n\texplicit constexpr wire(Init ...init) : curr{init...}, next{init...} {}\n\n\t// Copying and copy-assigning values is natural. If, however, a value is replaced with a wire,\n\t// e.g. because a module is built with a different optimization level, then existing code could\n\t// unintentionally copy a wire instead, which would create a subtle but serious bug. To make sure\n\t// this doesn't happen, prohibit copying and copy-assigning wires.\n\twire(const wire<Bits> &) = delete;\n\twire<Bits> &operator=(const wire<Bits> &) = delete;\n\n\twire(wire<Bits> &&) = default;\n\twire<Bits> &operator=(wire<Bits> &&) = default;\n\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tIntegerT get() const {\n\t\treturn curr.template get<IntegerT>();\n\t}\n\n\ttemplate<class IntegerT>\n\tCXXRTL_ALWAYS_INLINE\n\tvoid set(IntegerT other) {\n\t\tnext.template set<IntegerT>(other);\n\t}\n\n\t// This method intentionally takes a mandatory argument (to make it more difficult to misuse in\n\t// black box implementations, leading to missed observer events). It is generic over its argument\n\t// to make sure the `on_commit` call is devirtualized. This is somewhat awkward but lets us keep\n\t// a single implementation for both this method and the one in `memory`.\n\ttemplate<class ObserverT>\n\tbool commit(ObserverT &observer) {\n\t\tif (curr != next) {\n\t\t\tobserver.on_commit(curr.chunks, curr.data, next.data);\n\t\t\tcurr = next;\n\t\t\treturn true;\n\t\t}\n\t\treturn false;\n\t}\n};\n\ntemplate<size_t Bits>\nstd::ostream &operator<<(std::ostream &os, const wire<Bits> &val) {\n\tos << val.curr;\n\treturn os;\n}\n\ntemplate<size_t Width>\nstruct memory {\n\tconst size_t depth;\n\tstd::unique_ptr<value<Width>[]> data;\n\n\texplicit memory(size_t depth) : depth(depth), data(new value<Width>[depth]) {}\n\n\tmemory(const memory<Width> &) = delete;\n\tmemory<Width> &operator=(const memory<Width> &) = delete;\n\n\tmemory(memory<Width> &&) = default;\n\tmemory<Width> &operator=(memory<Width> &&other) {\n\t\tassert(depth == other.depth);\n\t\tdata = std::move(other.data);\n\t\twrite_queue = std::move(other.write_queue);\n\t\treturn *this;\n\t}\n\n\t// An operator for direct memory reads. May be used at any time during the simulation.\n\tconst value<Width> &operator [](size_t index) const {\n\t\tassert(index < depth);\n\t\treturn data[index];\n\t}\n\n\t// An operator for direct memory writes. May only be used before the simulation is started. If used\n\t// after the simulation is started, the design may malfunction.\n\tvalue<Width> &operator [](size_t index) {\n\t\tassert(index < depth);\n\t\treturn data[index];\n\t}\n\n\t// A simple way to make a writable memory would be to use an array of wires instead of an array of values.\n\t// However, there are two significant downsides to this approach: first, it has large overhead (2× space\n\t// overhead, and O(depth) time overhead during commit); second, it does not simplify handling write port\n\t// priorities. Although in principle write ports could be ordered or conditionally enabled in generated\n\t// code based on their priorities and selected addresses, the feedback arc set problem is computationally\n\t// expensive, and the heuristic based algorithms are not easily modified to guarantee (rather than prefer)\n\t// a particular write port evaluation order.\n\t//\n\t// The approach used here instead is to queue writes into a buffer during the eval phase, then perform\n\t// the writes during the commit phase in the priority order. This approach has low overhead, with both space\n\t// and time proportional to the amount of write ports. Because virtually every memory in a practical design\n\t// has at most two write ports, linear search is used on every write, being the fastest and simplest approach.\n\tstruct write {\n\t\tsize_t index;\n\t\tvalue<Width> val;\n\t\tvalue<Width> mask;\n\t\tint priority;\n\t};\n\tstd::vector<write> write_queue;\n\n\tvoid update(size_t index, const value<Width> &val, const value<Width> &mask, int priority = 0) {\n\t\tassert(index < depth);\n\t\t// Queue up the write while keeping the queue sorted by priority.\n\t\twrite_queue.insert(\n\t\t\tstd::upper_bound(write_queue.begin(), write_queue.end(), priority,\n\t\t\t\t[](const int a, const write& b) { return a < b.priority; }),\n\t\t\twrite { index, val, mask, priority });\n\t}\n\n\t// See the note for `wire::commit()`.\n\ttemplate<class ObserverT>\n\tbool commit(ObserverT &observer) {\n\t\tbool changed = false;\n\t\tfor (const write &entry : write_queue) {\n\t\t\tvalue<Width> elem = data[entry.index];\n\t\t\telem = elem.update(entry.val, entry.mask);\n\t\t\tif (data[entry.index] != elem) {\n\t\t\t\tobserver.on_commit(value<Width>::chunks, data[0].data, elem.data, entry.index);\n\t\t\t\tchanged |= true;\n\t\t\t}\n\t\t\tdata[entry.index] = elem;\n\t\t}\n\t\twrite_queue.clear();\n\t\treturn changed;\n\t}\n};\n\nstruct metadata {\n\tconst enum {\n\t\tMISSING = 0,\n\t\tUINT   \t= 1,\n\t\tSINT   \t= 2,\n\t\tSTRING \t= 3,\n\t\tDOUBLE \t= 4,\n\t} value_type;\n\n\t// In debug mode, using the wrong .as_*() function will assert.\n\t// In release mode, using the wrong .as_*() function will safely return a default value.\n\tconst uint64_t    uint_value = 0;\n\tconst int64_t     sint_value = 0;\n\tconst std::string string_value = \"\";\n\tconst double      double_value = 0.0;\n\n\tmetadata() : value_type(MISSING) {}\n\tmetadata(uint64_t value) : value_type(UINT), uint_value(value) {}\n\tmetadata(int64_t value) : value_type(SINT), sint_value(value) {}\n\tmetadata(const std::string &value) : value_type(STRING), string_value(value) {}\n\tmetadata(const char *value) : value_type(STRING), string_value(value) {}\n\tmetadata(double value) : value_type(DOUBLE), double_value(value) {}\n\n\tmetadata(const metadata &) = default;\n\tmetadata &operator=(const metadata &) = delete;\n\n\tuint64_t as_uint() const {\n\t\tassert(value_type == UINT);\n\t\treturn uint_value;\n\t}\n\n\tint64_t as_sint() const {\n\t\tassert(value_type == SINT);\n\t\treturn sint_value;\n\t}\n\n\tconst std::string &as_string() const {\n\t\tassert(value_type == STRING);\n\t\treturn string_value;\n\t}\n\n\tdouble as_double() const {\n\t\tassert(value_type == DOUBLE);\n\t\treturn double_value;\n\t}\n};\n\ntypedef std::map<std::string, metadata> metadata_map;\n\n// Tag class to disambiguate values/wires and their aliases.\nstruct debug_alias {};\n\n// Tag declaration to disambiguate values and debug outlines.\nusing debug_outline = ::_cxxrtl_outline;\n\n// This structure is intended for consumption via foreign function interfaces, like Python's ctypes.\n// Because of this it uses a C-style layout that is easy to parse rather than more idiomatic C++.\n//\n// To avoid violating strict aliasing rules, this structure has to be a subclass of the one used\n// in the C API, or it would not be possible to cast between the pointers to these.\n//\n// The `attrs` member cannot be owned by this structure because a `cxxrtl_object` can be created\n// from external C code.\nstruct debug_item : ::cxxrtl_object {\n\t// Object types.\n\tenum : uint32_t {\n\t\tVALUE   = CXXRTL_VALUE,\n\t\tWIRE    = CXXRTL_WIRE,\n\t\tMEMORY  = CXXRTL_MEMORY,\n\t\tALIAS   = CXXRTL_ALIAS,\n\t\tOUTLINE = CXXRTL_OUTLINE,\n\t};\n\n\t// Object flags.\n\tenum : uint32_t {\n\t\tINPUT  = CXXRTL_INPUT,\n\t\tOUTPUT = CXXRTL_OUTPUT,\n\t\tINOUT  = CXXRTL_INOUT,\n\t\tDRIVEN_SYNC = CXXRTL_DRIVEN_SYNC,\n\t\tDRIVEN_COMB = CXXRTL_DRIVEN_COMB,\n\t\tUNDRIVEN    = CXXRTL_UNDRIVEN,\n\t};\n\n\tdebug_item(const ::cxxrtl_object &object) : cxxrtl_object(object) {}\n\n\ttemplate<size_t Bits>\n\tdebug_item(value<Bits> &item, size_t lsb_offset = 0, uint32_t flags_ = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = VALUE;\n\t\tflags   = flags_;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = item.data;\n\t\tnext    = item.data;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(const value<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = VALUE;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.data);\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(wire<Bits> &item, size_t lsb_offset = 0, uint32_t flags_ = 0) {\n\t\tstatic_assert(sizeof(item.curr) == value<Bits>::chunks * sizeof(chunk_t) &&\n\t\t              sizeof(item.next) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"wire<Bits> is not compatible with C layout\");\n\t\ttype    = WIRE;\n\t\tflags   = flags_;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = item.curr.data;\n\t\tnext    = item.next.data;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Width>\n\tdebug_item(memory<Width> &item, size_t zero_offset = 0) {\n\t\tstatic_assert(sizeof(item.data[0]) == value<Width>::chunks * sizeof(chunk_t),\n\t\t              \"memory<Width> is not compatible with C layout\");\n\t\ttype    = MEMORY;\n\t\tflags   = 0;\n\t\twidth   = Width;\n\t\tlsb_at  = 0;\n\t\tdepth   = item.depth;\n\t\tzero_at = zero_offset;\n\t\tcurr    = item.data ? item.data[0].data : nullptr;\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(debug_alias, const value<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = ALIAS;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.data);\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(debug_alias, const wire<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item.curr) == value<Bits>::chunks * sizeof(chunk_t) &&\n\t\t              sizeof(item.next) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"wire<Bits> is not compatible with C layout\");\n\t\ttype    = ALIAS;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.curr.data);\n\t\tnext    = nullptr;\n\t\toutline = nullptr;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits>\n\tdebug_item(debug_outline &group, const value<Bits> &item, size_t lsb_offset = 0) {\n\t\tstatic_assert(sizeof(item) == value<Bits>::chunks * sizeof(chunk_t),\n\t\t              \"value<Bits> is not compatible with C layout\");\n\t\ttype    = OUTLINE;\n\t\tflags   = DRIVEN_COMB;\n\t\twidth   = Bits;\n\t\tlsb_at  = lsb_offset;\n\t\tdepth   = 1;\n\t\tzero_at = 0;\n\t\tcurr    = const_cast<chunk_t*>(item.data);\n\t\tnext    = nullptr;\n\t\toutline = &group;\n\t\tattrs   = nullptr;\n\t}\n\n\ttemplate<size_t Bits, class IntegerT>\n\tIntegerT get() const {\n\t\tassert(width == Bits && depth == 1);\n\t\tvalue<Bits> item;\n\t\tstd::copy(curr, curr + value<Bits>::chunks, item.data);\n\t\treturn item.template get<IntegerT>();\n\t}\n\n\ttemplate<size_t Bits, class IntegerT>\n\tvoid set(IntegerT other) const {\n\t\tassert(width == Bits && depth == 1);\n\t\tvalue<Bits> item;\n\t\titem.template set<IntegerT>(other);\n\t\tstd::copy(item.data, item.data + value<Bits>::chunks, next);\n\t}\n};\nstatic_assert(std::is_standard_layout<debug_item>::value, \"debug_item is not compatible with C layout\");\n\n} // namespace cxxrtl\n\ntypedef struct _cxxrtl_attr_set {\n\tcxxrtl::metadata_map map;\n} *cxxrtl_attr_set;\n\nnamespace cxxrtl {\n\n// Representation of an attribute set in the C++ interface.\nusing debug_attrs = ::_cxxrtl_attr_set;\n\nstruct debug_items {\n\tstd::map<std::string, std::vector<debug_item>> table;\n\tstd::map<std::string, std::unique_ptr<debug_attrs>> attrs_table;\n\n\tvoid add(const std::string &name, debug_item &&item, metadata_map &&item_attrs = {}) {\n\t\tstd::unique_ptr<debug_attrs> &attrs = attrs_table[name];\n\t\tif (attrs.get() == nullptr)\n\t\t\tattrs = std::unique_ptr<debug_attrs>(new debug_attrs);\n\t\tfor (auto attr : item_attrs)\n\t\t\tattrs->map.insert(attr);\n\t\titem.attrs = attrs.get();\n\t\tstd::vector<debug_item> &parts = table[name];\n\t\tparts.emplace_back(item);\n\t\tstd::sort(parts.begin(), parts.end(),\n\t\t\t[](const debug_item &a, const debug_item &b) {\n\t\t\t\treturn a.lsb_at < b.lsb_at;\n\t\t\t});\n\t}\n\n\tsize_t count(const std::string &name) const {\n\t\tif (table.count(name) == 0)\n\t\t\treturn 0;\n\t\treturn table.at(name).size();\n\t}\n\n\tconst std::vector<debug_item> &parts_at(const std::string &name) const {\n\t\treturn table.at(name);\n\t}\n\n\tconst debug_item &at(const std::string &name) const {\n\t\tconst std::vector<debug_item> &parts = table.at(name);\n\t\tassert(parts.size() == 1);\n\t\treturn parts.at(0);\n\t}\n\n\tconst debug_item &operator [](const std::string &name) const {\n\t\treturn at(name);\n\t}\n\n\tconst metadata_map &attrs(const std::string &name) const {\n\t\treturn attrs_table.at(name)->map;\n\t}\n};\n\n// Tag class to disambiguate the default constructor used by the toplevel module that calls reset(),\n// and the constructor of interior modules that should not call it.\nstruct interior {};\n\nstruct module {\n\tmodule() {}\n\tvirtual ~module() {}\n\n\t// Modules with black boxes cannot be copied. Although not all designs include black boxes,\n\t// delete the copy constructor and copy assignment operator to make sure that any downstream\n\t// code that manipulates modules doesn't accidentally depend on their availability.\n\tmodule(const module &) = delete;\n\tmodule &operator=(const module &) = delete;\n\n\tmodule(module &&) = default;\n\tmodule &operator=(module &&) = default;\n\n\tvirtual void reset() = 0;\n\n\tvirtual bool eval() = 0;\n\tvirtual bool commit() = 0;\n\n\tunsigned int steps = 0;\n\n\tsize_t step() {\n\t\t++steps;\n\t\tsize_t deltas = 0;\n\t\tbool converged = false;\n\t\tdo {\n\t\t\tconverged = eval();\n\t\t\tdeltas++;\n\t\t} while (commit() && !converged);\n\t\treturn deltas;\n\t}\n\n\tvirtual void debug_info(debug_items &items, std::string path = \"\") {\n\t\t(void)items, (void)path;\n\t}\n};\n\n} // namespace cxxrtl\n\n// Internal structures used to communicate with the implementation of the C interface.\n\ntypedef struct _cxxrtl_toplevel {\n\tstd::unique_ptr<cxxrtl::module> module;\n} *cxxrtl_toplevel;\n\ntypedef struct _cxxrtl_outline {\n\tstd::function<void()> eval;\n} *cxxrtl_outline;\n\n// Definitions of internal Yosys cells. Other than the functions in this namespace, CXXRTL is fully generic\n// and indepenent of Yosys implementation details.\n//\n// The `write_cxxrtl` pass translates internal cells (cells with names that start with `$`) to calls of these\n// functions. All of Yosys arithmetic and logical cells perform sign or zero extension on their operands,\n// whereas basic operations on arbitrary width values require operands to be of the same width. These functions\n// bridge the gap by performing the necessary casts. They are named similar to `cell_A[B]`, where A and B are `u`\n// if the corresponding operand is unsigned, and `s` if it is signed.\nnamespace cxxrtl_yosys {\n\nusing namespace cxxrtl;\n\n// std::max isn't constexpr until C++14 for no particular reason (it's an oversight), so we define our own.\ntemplate<class T>\nCXXRTL_ALWAYS_INLINE\nconstexpr T max(const T &a, const T &b) {\n\treturn a > b ? a : b;\n}\n\n// Logic operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> logic_not(const value<BitsA> &a) {\n\treturn value<BitsY> { a ? 0u : 1u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> logic_and(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn value<BitsY> { (bool(a) && bool(b)) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> logic_or(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn value<BitsY> { (bool(a) || bool(b)) ? 1u : 0u };\n}\n\n// Reduction operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_and(const value<BitsA> &a) {\n\treturn value<BitsY> { a.bit_not().is_zero() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_or(const value<BitsA> &a) {\n\treturn value<BitsY> { a ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_xor(const value<BitsA> &a) {\n\treturn value<BitsY> { (a.ctpop() % 2) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_xnor(const value<BitsA> &a) {\n\treturn value<BitsY> { (a.ctpop() % 2) ? 0u : 1u };\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> reduce_bool(const value<BitsA> &a) {\n\treturn value<BitsY> { a ? 1u : 0u };\n}\n\n// Bitwise operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> not_u(const value<BitsA> &a) {\n\treturn a.template zcast<BitsY>().bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> not_s(const value<BitsA> &a) {\n\treturn a.template scast<BitsY>().bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> and_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_and(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> and_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_and(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> or_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_or(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> or_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_or(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_xor(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_xor(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xnor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().bit_xor(b.template zcast<BitsY>()).bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> xnor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().bit_xor(b.template scast<BitsY>()).bit_not();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shl_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shl_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshl_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshl_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().shl(b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shr_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.shr(b).template zcast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shr_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.shr(b).template scast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshr_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.shr(b).template zcast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sshr_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.sshr(b).template scast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shr_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shr_su<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_us(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn b.is_neg() ? shl_uu<BitsY>(a, b.template sext<BitsB + 1>().neg()) : shr_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shift_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn b.is_neg() ? shl_su<BitsY>(a, b.template sext<BitsB + 1>().neg()) : shr_su<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_su(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_su<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_us(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_us<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> shiftx_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn shift_ss<BitsY>(a, b);\n}\n\n// Comparison operations\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eq_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template zext<BitsExt>() == b.template zext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eq_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template sext<BitsExt>() == b.template sext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ne_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template zext<BitsExt>() != b.template zext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ne_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY>{ a.template sext<BitsExt>() != b.template sext<BitsExt>() ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eqx_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn eq_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> eqx_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn eq_ss<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> nex_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn ne_uu<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> nex_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn ne_ss<BitsY>(a, b);\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> gt_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { b.template zext<BitsExt>().ucmp(a.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> gt_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { b.template sext<BitsExt>().scmp(a.template sext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ge_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !a.template zext<BitsExt>().ucmp(b.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> ge_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !a.template sext<BitsExt>().scmp(b.template sext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> lt_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { a.template zext<BitsExt>().ucmp(b.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> lt_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { a.template sext<BitsExt>().scmp(b.template sext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> le_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !b.template zext<BitsExt>().ucmp(a.template zext<BitsExt>()) ? 1u : 0u };\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> le_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsExt = max(BitsA, BitsB);\n\treturn value<BitsY> { !b.template sext<BitsExt>().scmp(a.template sext<BitsExt>()) ? 1u : 0u };\n}\n\n// Arithmetic operations\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> pos_u(const value<BitsA> &a) {\n\treturn a.template zcast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> pos_s(const value<BitsA> &a) {\n\treturn a.template scast<BitsY>();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> neg_u(const value<BitsA> &a) {\n\treturn a.template zcast<BitsY>().neg();\n}\n\ntemplate<size_t BitsY, size_t BitsA>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> neg_s(const value<BitsA> &a) {\n\treturn a.template scast<BitsY>().neg();\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> add_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().add(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> add_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().add(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sub_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template zcast<BitsY>().sub(b.template zcast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> sub_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().sub(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mul_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t BitsM = BitsA >= BitsB ? BitsA : BitsB;\n\treturn a.template zcast<BitsM>().template mul<BitsY>(b.template zcast<BitsM>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mul_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn a.template scast<BitsY>().template mul<BitsY>(b.template scast<BitsY>());\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nstd::pair<value<BitsY>, value<BitsY>> divmod_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t Bits = max(BitsY, max(BitsA, BitsB));\n\tvalue<Bits> quotient;\n\tvalue<Bits> remainder;\n\tvalue<Bits> dividend = a.template zext<Bits>();\n\tvalue<Bits> divisor = b.template zext<Bits>();\n\tstd::tie(quotient, remainder) = dividend.udivmod(divisor);\n\treturn {quotient.template trunc<BitsY>(), remainder.template trunc<BitsY>()};\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nstd::pair<value<BitsY>, value<BitsY>> divmod_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tconstexpr size_t Bits = max(BitsY, max(BitsA, BitsB));\n\tvalue<Bits> quotient;\n\tvalue<Bits> remainder;\n\tvalue<Bits> dividend = a.template sext<Bits>();\n\tvalue<Bits> divisor = b.template sext<Bits>();\n\tstd::tie(quotient, remainder) = dividend.sdivmod(divisor);\n\treturn {quotient.template trunc<BitsY>(), remainder.template trunc<BitsY>()};\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> div_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).first;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> div_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_ss<BitsY>(a, b).first;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mod_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).second;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> mod_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_ss<BitsY>(a, b).second;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> modfloor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).second;\n}\n\n// GHDL Modfloor operator. Returns r=a mod b, such that r has the same sign as b and\n// a=b*N+r where N is some integer\n// In practical terms, when a and b have different signs and the remainder returned by divmod_ss is not 0\n// then return the remainder + b\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> modfloor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tvalue<BitsY> r;\n\tr = divmod_ss<BitsY>(a, b).second;\n\tif((b.is_neg() != a.is_neg()) && !r.is_zero())\n\t\treturn add_ss<BitsY>(b, r);\n\treturn r;\n}\n\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> divfloor_uu(const value<BitsA> &a, const value<BitsB> &b) {\n\treturn divmod_uu<BitsY>(a, b).first;\n}\n\n// Divfloor. Similar to above: returns q=a//b, where q has the sign of a*b and a=b*q+N.\n// In other words, returns (truncating) a/b, except if a and b have different signs\n// and there's non-zero remainder, subtract one more towards floor.\ntemplate<size_t BitsY, size_t BitsA, size_t BitsB>\nCXXRTL_ALWAYS_INLINE\nvalue<BitsY> divfloor_ss(const value<BitsA> &a, const value<BitsB> &b) {\n\tvalue<BitsY> q, r;\n\tstd::tie(q, r) = divmod_ss<BitsY>(a, b);\n\tif ((b.is_neg() != a.is_neg()) && !r.is_zero())\n\t\treturn sub_uu<BitsY>(q, value<1> { 1u });\n\treturn q;\n\n}\n\n// Memory helper\nstruct memory_index {\n\tbool valid;\n\tsize_t index;\n\n\ttemplate<size_t BitsAddr>\n\tmemory_index(const value<BitsAddr> &addr, size_t offset, size_t depth) {\n\t\tstatic_assert(value<BitsAddr>::chunks <= 1, \"memory address is too wide\");\n\t\tsize_t offset_index = addr.data[0];\n\n\t\tvalid = (offset_index >= offset && offset_index < offset + depth);\n\t\tindex = offset_index - offset;\n\t}\n};\n\n} // namespace cxxrtl_yosys\n\n#endif\n",
                             "cxxrtl_vcd.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2020  whitequark <whitequark@whitequark.org>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#ifndef CXXRTL_VCD_H\n#define CXXRTL_VCD_H\n\n#include <cxxrtl/cxxrtl.h>\n\nnamespace cxxrtl {\n\nclass vcd_writer {\n\tstruct variable {\n\t\tsize_t ident;\n\t\tsize_t width;\n\t\tchunk_t *curr;\n\t\tsize_t cache_offset;\n\t\tdebug_outline *outline;\n\t\tbool *outline_warm;\n\t};\n\n\tstd::vector<std::string> current_scope;\n\tstd::map<debug_outline*, bool> outlines;\n\tstd::vector<variable> variables;\n\tstd::vector<chunk_t> cache;\n\tstd::map<chunk_t*, size_t> aliases;\n\tbool streaming = false;\n\n\tvoid emit_timescale(unsigned number, const std::string &unit) {\n\t\tassert(!streaming);\n\t\tassert(number == 1 || number == 10 || number == 100);\n\t\tassert(unit == \"s\" || unit == \"ms\" || unit == \"us\" ||\n\t\t       unit == \"ns\" || unit == \"ps\" || unit == \"fs\");\n\t\tbuffer += \"$timescale \" + std::to_string(number) + \" \" + unit + \" $end\\n\";\n\t}\n\n\tvoid emit_scope(const std::vector<std::string> &scope) {\n\t\tassert(!streaming);\n\t\twhile (current_scope.size() > scope.size() ||\n\t\t       (current_scope.size() > 0 &&\n\t\t\tcurrent_scope[current_scope.size() - 1] != scope[current_scope.size() - 1])) {\n\t\t\tbuffer += \"$upscope $end\\n\";\n\t\t\tcurrent_scope.pop_back();\n\t\t}\n\t\twhile (current_scope.size() < scope.size()) {\n\t\t\tbuffer += \"$scope module \" + scope[current_scope.size()] + \" $end\\n\";\n\t\t\tcurrent_scope.push_back(scope[current_scope.size()]);\n\t\t}\n\t}\n\n\tvoid emit_ident(size_t ident) {\n\t\tdo {\n\t\t\tbuffer += '!' + ident % 94; // \"base94\"\n\t\t\tident /= 94;\n\t\t} while (ident != 0);\n\t}\n\n\tvoid emit_name(const std::string &name) {\n\t\tfor (char c : name) {\n\t\t\tif (c == ':') {\n\t\t\t\t// Due to a bug, GTKWave cannot parse a colon in the variable name, causing the VCD file\n\t\t\t\t// to be unreadable. It cannot be escaped either, so replace it with the sideways colon.\n\t\t\t\tbuffer += \"..\";\n\t\t\t} else {\n\t\t\t\tbuffer += c;\n\t\t\t}\n\t\t}\n\t}\n\n\tvoid emit_var(const variable &var, const std::string &type, const std::string &name,\n\t              size_t lsb_at, bool multipart) {\n\t\tassert(!streaming);\n\t\tbuffer += \"$var \" + type + \" \" + std::to_string(var.width) + \" \";\n\t\temit_ident(var.ident);\n\t\tbuffer += \" \";\n\t\temit_name(name);\n\t\tif (multipart || name.back() == ']' || lsb_at != 0) {\n\t\t\tif (var.width == 1)\n\t\t\t\tbuffer += \" [\" + std::to_string(lsb_at) + \"]\";\n\t\t\telse\n\t\t\t\tbuffer += \" [\" + std::to_string(lsb_at + var.width - 1) + \":\" + std::to_string(lsb_at) + \"]\";\n\t\t}\n\t\tbuffer += \" $end\\n\";\n\t}\n\n\tvoid emit_enddefinitions() {\n\t\tassert(!streaming);\n\t\tbuffer += \"$enddefinitions $end\\n\";\n\t\tstreaming = true;\n\t}\n\n\tvoid emit_time(uint64_t timestamp) {\n\t\tassert(streaming);\n\t\tbuffer += \"#\" + std::to_string(timestamp) + \"\\n\";\n\t}\n\n\tvoid emit_scalar(const variable &var) {\n\t\tassert(streaming);\n\t\tassert(var.width == 1);\n\t\tbuffer += (*var.curr ? '1' : '0');\n\t\temit_ident(var.ident);\n\t\tbuffer += '\\n';\n\t}\n\n\tvoid emit_vector(const variable &var) {\n\t\tassert(streaming);\n\t\tbuffer += 'b';\n\t\tfor (size_t bit = var.width - 1; bit != (size_t)-1; bit--) {\n\t\t\tbool bit_curr = var.curr[bit / (8 * sizeof(chunk_t))] & (1 << (bit % (8 * sizeof(chunk_t))));\n\t\t\tbuffer += (bit_curr ? '1' : '0');\n\t\t}\n\t\tbuffer += ' ';\n\t\temit_ident(var.ident);\n\t\tbuffer += '\\n';\n\t}\n\n\tvoid reset_outlines() {\n\t\tfor (auto &outline_it : outlines)\n\t\t\toutline_it.second = /*warm=*/(outline_it.first == nullptr);\n\t}\n\n\tvariable &register_variable(size_t width, chunk_t *curr, bool constant = false, debug_outline *outline = nullptr) {\n\t\tif (aliases.count(curr)) {\n\t\t\treturn variables[aliases[curr]];\n\t\t} else {\n\t\t\tauto outline_it = outlines.emplace(outline, /*warm=*/(outline == nullptr)).first;\n\t\t\tconst size_t chunks = (width + (sizeof(chunk_t) * 8 - 1)) / (sizeof(chunk_t) * 8);\n\t\t\taliases[curr] = variables.size();\n\t\t\tif (constant) {\n\t\t\t\tvariables.emplace_back(variable { variables.size(), width, curr, (size_t)-1, outline_it->first, &outline_it->second });\n\t\t\t} else {\n\t\t\t\tvariables.emplace_back(variable { variables.size(), width, curr, cache.size(), outline_it->first, &outline_it->second });\n\t\t\t\tcache.insert(cache.end(), &curr[0], &curr[chunks]);\n\t\t\t}\n\t\t\treturn variables.back();\n\t\t}\n\t}\n\n\tbool test_variable(const variable &var) {\n\t\tif (var.cache_offset == (size_t)-1)\n\t\t\treturn false; // constant\n\t\tif (!*var.outline_warm) {\n\t\t\tvar.outline->eval();\n\t\t\t*var.outline_warm = true;\n\t\t}\n\t\tconst size_t chunks = (var.width + (sizeof(chunk_t) * 8 - 1)) / (sizeof(chunk_t) * 8);\n\t\tif (std::equal(&var.curr[0], &var.curr[chunks], &cache[var.cache_offset])) {\n\t\t\treturn false;\n\t\t} else {\n\t\t\tstd::copy(&var.curr[0], &var.curr[chunks], &cache[var.cache_offset]);\n\t\t\treturn true;\n\t\t}\n\t}\n\n\tstatic std::vector<std::string> split_hierarchy(const std::string &hier_name) {\n\t\tstd::vector<std::string> hierarchy;\n\t\tsize_t prev = 0;\n\t\twhile (true) {\n\t\t\tsize_t curr = hier_name.find_first_of(' ', prev);\n\t\t\tif (curr == std::string::npos) {\n\t\t\t\thierarchy.push_back(hier_name.substr(prev));\n\t\t\t\tbreak;\n\t\t\t} else {\n\t\t\t\thierarchy.push_back(hier_name.substr(prev, curr - prev));\n\t\t\t\tprev = curr + 1;\n\t\t\t}\n\t\t}\n\t\treturn hierarchy;\n\t}\n\npublic:\n\tstd::string buffer;\n\n\tvoid timescale(unsigned number, const std::string &unit) {\n\t\temit_timescale(number, unit);\n\t}\n\n\tvoid add(const std::string &hier_name, const debug_item &item, bool multipart = false) {\n\t\tstd::vector<std::string> scope = split_hierarchy(hier_name);\n\t\tstd::string name = scope.back();\n\t\tscope.pop_back();\n\n\t\temit_scope(scope);\n\t\tswitch (item.type) {\n\t\t\t// Not the best naming but oh well...\n\t\t\tcase debug_item::VALUE:\n\t\t\t\temit_var(register_variable(item.width, item.curr, /*constant=*/item.next == nullptr),\n\t\t\t\t         \"wire\", name, item.lsb_at, multipart);\n\t\t\t\tbreak;\n\t\t\tcase debug_item::WIRE:\n\t\t\t\temit_var(register_variable(item.width, item.curr),\n\t\t\t\t         \"reg\", name, item.lsb_at, multipart);\n\t\t\t\tbreak;\n\t\t\tcase debug_item::MEMORY: {\n\t\t\t\tconst size_t stride = (item.width + (sizeof(chunk_t) * 8 - 1)) / (sizeof(chunk_t) * 8);\n\t\t\t\tfor (size_t index = 0; index < item.depth; index++) {\n\t\t\t\t\tchunk_t *nth_curr = &item.curr[stride * index];\n\t\t\t\t\tstd::string nth_name = name + '[' + std::to_string(index) + ']';\n\t\t\t\t\temit_var(register_variable(item.width, nth_curr),\n\t\t\t\t\t         \"reg\", nth_name, item.lsb_at, multipart);\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tcase debug_item::ALIAS:\n\t\t\t\t// Like VALUE, but, even though `item.next == nullptr` always holds, the underlying value\n\t\t\t\t// can actually change, and must be tracked. In most cases the VCD identifier will be\n\t\t\t\t// unified with the aliased reg, but we should handle the case where only the alias is\n\t\t\t\t// added to the VCD writer, too.\n\t\t\t\temit_var(register_variable(item.width, item.curr),\n\t\t\t\t         \"wire\", name, item.lsb_at, multipart);\n\t\t\t\tbreak;\n\t\t\tcase debug_item::OUTLINE:\n\t\t\t\temit_var(register_variable(item.width, item.curr, /*constant=*/false, item.outline),\n\t\t\t\t         \"wire\", name, item.lsb_at, multipart);\n\t\t\t\tbreak;\n\t\t}\n\t}\n\n\ttemplate<class Filter>\n\tvoid add(const debug_items &items, const Filter &filter) {\n\t\t// `debug_items` is a map, so the items are already sorted in an order optimal for emitting\n\t\t// VCD scope sections.\n\t\tfor (auto &it : items.table)\n\t\t\tfor (auto &part : it.second)\n\t\t\t\tif (filter(it.first, part))\n\t\t\t\t\tadd(it.first, part, it.second.size() > 1);\n\t}\n\n\tvoid add(const debug_items &items) {\n\t\tthis->add(items, [](const std::string &, const debug_item &) {\n\t\t\treturn true;\n\t\t});\n\t}\n\n\tvoid add_without_memories(const debug_items &items) {\n\t\tthis->add(items, [](const std::string &, const debug_item &item) {\n\t\t\treturn item.type != debug_item::MEMORY;\n\t\t});\n\t}\n\n\tvoid sample(uint64_t timestamp) {\n\t\tbool first_sample = !streaming;\n\t\tif (first_sample) {\n\t\t\temit_scope({});\n\t\t\temit_enddefinitions();\n\t\t}\n\t\treset_outlines();\n\t\temit_time(timestamp);\n\t\tfor (auto var : variables)\n\t\t\tif (test_variable(var) || first_sample) {\n\t\t\t\tif (var.width == 1)\n\t\t\t\t\temit_scalar(var);\n\t\t\t\telse\n\t\t\t\t\temit_vector(var);\n\t\t\t}\n\t}\n};\n\n}\n\n#endif\n",
                         },
                     },
@@ -162,7 +162,7 @@ export const filesystem = {
                 "ffinit.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2020  Marcelina Kościelnicka <mwk@0x04.net>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted, provided that the above\n *  copyright notice and this permission notice appear in all copies.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#ifndef FFINIT_H\n#define FFINIT_H\n\n#include \"kernel/yosys.h\"\n#include \"kernel/sigtools.h\"\n\nYOSYS_NAMESPACE_BEGIN\n\nstruct FfInitVals\n{\n\tconst SigMap *sigmap;\n\tdict<SigBit, std::pair<State,SigBit>> initbits;\n\n\tvoid set(const SigMap *sigmap_, RTLIL::Module *module)\n\t{\n\t\tsigmap = sigmap_;\n\t\tinitbits.clear();\n\t\tfor (auto wire : module->wires())\n\t\t{\n\t\t\tif (wire->attributes.count(ID::init) == 0)\n\t\t\t\tcontinue;\n\n\t\t\tSigSpec wirebits = (*sigmap)(wire);\n\t\t\tConst initval = wire->attributes.at(ID::init);\n\n\t\t\tfor (int i = 0; i < GetSize(wirebits) && i < GetSize(initval); i++)\n\t\t\t{\n\t\t\t\tSigBit bit = wirebits[i];\n\t\t\t\tState val = initval[i];\n\n\t\t\t\tif (val != State::S0 && val != State::S1 && bit.wire != nullptr)\n\t\t\t\t\tcontinue;\n\n\t\t\t\tif (initbits.count(bit)) {\n\t\t\t\t\tif (initbits.at(bit).first != val)\n\t\t\t\t\t\tlog_error(\"Conflicting init values for signal %s (%s = %s != %s).\\n\",\n\t\t\t\t\t\t\t\tlog_signal(bit), log_signal(SigBit(wire, i)),\n\t\t\t\t\t\t\t\tlog_signal(val), log_signal(initbits.at(bit).first));\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\n\t\t\t\tinitbits[bit] = std::make_pair(val,SigBit(wire,i));\n\t\t\t}\n\t\t}\n\t}\n\n\tRTLIL::State operator()(RTLIL::SigBit bit) const\n\t{\n\t\tauto it = initbits.find((*sigmap)(bit));\n\t\tif (it != initbits.end())\n\t\t\treturn it->second.first;\n\t\telse\n\t\t\treturn State::Sx;\n\t}\n\n\tRTLIL::Const operator()(const RTLIL::SigSpec &sig) const\n\t{\n\t\tRTLIL::Const res;\n\t\tfor (auto bit : sig)\n\t\t\tres.bits.push_back((*this)(bit));\n\t\treturn res;\n\t}\n\n\tvoid set_init(RTLIL::SigBit bit, RTLIL::State val)\n\t{\n\t\tSigBit mbit = (*sigmap)(bit);\n\t\tSigBit abit = bit;\n\t\tauto it = initbits.find(mbit);\n\t\tif (it != initbits.end())\n\t\t\tabit = it->second.second;\n\t\telse if (val == State::Sx)\n\t\t\treturn;\n\t\tlog_assert(abit.wire);\n\t\tinitbits[mbit] = std::make_pair(val,abit);\n\t\tauto it2 = abit.wire->attributes.find(ID::init);\n\t\tif (it2 != abit.wire->attributes.end()) {\n\t\t\tit2->second[abit.offset] = val;\n\t\t\tif (it2->second.is_fully_undef())\n\t\t\t\tabit.wire->attributes.erase(it2);\n\t\t} else if (val != State::Sx) {\n\t\t\tConst cval(State::Sx, GetSize(abit.wire));\n\t\t\tcval[abit.offset] = val;\n\t\t\tabit.wire->attributes[ID::init] = cval;\n\t\t}\n\t}\n\n\tvoid set_init(const RTLIL::SigSpec &sig, RTLIL::Const val)\n\t{\n\t\tlog_assert(GetSize(sig) == GetSize(val));\n\t\tfor (int i = 0; i < GetSize(sig); i++)\n\t\t\tset_init(sig[i], val[i]);\n\t}\n\n\tvoid remove_init(RTLIL::SigBit bit)\n\t{\n\t\tset_init(bit, State::Sx);\n\t}\n\n\tvoid remove_init(const RTLIL::SigSpec &sig)\n\t{\n\t\tfor (auto bit : sig)\n\t\t\tremove_init(bit);\n\t}\n\n\tvoid clear()\n\t{\n\t\tinitbits.clear();\n\t}\n\n\tFfInitVals (const SigMap *sigmap, RTLIL::Module *module)\n\t{\n\t\tset(sigmap, module);\n\t}\n\n\tFfInitVals () {}\n};\n\n\nYOSYS_NAMESPACE_END\n\n#endif\n",
                 "ffmerge.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2021  Marcelina Kościelnicka <mwk@0x04.net>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted, provided that the above\n *  copyright notice and this permission notice appear in all copies.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#ifndef FFMERGE_H\n#define FFMERGE_H\n\n#include \"kernel/ffinit.h\"\n#include \"kernel/ff.h\"\n\nYOSYS_NAMESPACE_BEGIN\n\n// A helper class for passes that want to merge FFs on the input or output\n// of a cell into the cell itself.\n//\n// The procedure is:\n//\n// 1. Construct this class (at beginning of processing for a given module).\n// 2. For every considered cell:\n//\n//    a. Call find_output_ff for every considered output.\n//    b. Call find_input_ff for every considered input.\n//    c. Look at the FF description returned (if any) from each call, reject\n//       results that cannot be merged into given cell for any reason.\n//       If both inputs and outputs are being merged, take care of FF bits that\n//       are returned in both input and output results (a FF bit cannot be\n//       merged to both).  Decide on the final set of FF bits to merge.\n//    d. Call remove_output_ff for every find_output_ff result that will be used\n//       for merging.  This removes the actual FF bits from design and from index.\n//    e. Call mark_input_ff for every find_input_ff result that will be used\n//       for merging.  This updates the index disallowing further usage of these\n//       FF bits for output FF merging, if they were eligible before.  The actual\n//       FF bits are still left in the design and can be merged into other inputs.\n//       If the FF bits are not otherwise used, they will be removed by later\n//       opt passes.\n//    f. Merge the FFs into the cell.\n//\n// Note that, if both inputs and outputs are being considered for merging in\n// a single pass, the result may be nondeterministic (depending on cell iteration\n// order) because a given FF bit could be eligible for both input and output merge,\n// perhaps in different cells.  For this reason, it may be a good idea to separate\n// input and output merging.\n\nstruct FfMergeHelper\n{\n\tconst SigMap *sigmap;\n\tRTLIL::Module *module;\n\tFfInitVals *initvals;\n\n\tdict<SigBit, std::pair<Cell*, int>> dff_driver;\n\tdict<SigBit, pool<std::pair<Cell*, int>>> dff_sink;\n\tdict<SigBit, int> sigbit_users_count;\n\n\t// Returns true if all bits in sig are completely unused.\n\tbool is_output_unused(RTLIL::SigSpec sig);\n\n\t// Finds the FF to merge into a given cell output.  Takes sig, which\n\t// is the current cell output — it will be the sig_d of the found FF.\n\t// If found, returns true, and fills the two output arguments.\n\t//\n\t// For every bit of sig, this function finds a FF bit that has\n\t// the same sig_d, and fills the output FfData according to the FF\n\t// bits found.  This function will only consider FF bits that are\n\t// the only user of the given sig bits — if any bit in sig is used\n\t// by anything other than a single FF, this function will return false.\n\t//\n\t// The returned FfData structure does not correspond to any actual FF\n\t// cell in the design — it is the amalgamation of extracted FF bits,\n\t// possibly coming from several FF cells.\n\t//\n\t// If some of the bits in sig have no users at all, this function\n\t// will accept them as well (and fill returned FfData with dummy values\n\t// for the given bit, effectively synthesizing an unused FF bit of the\n\t// appropriate type).  However, if all bits in sig are completely\n\t// unused, this function will fail and return false (having no idea\n\t// what kind of FF to produce) — use the above helper if that case\n\t// is important to handle.\n\t//\n\t// Note that this function does not remove the FF bits returned from\n\t// the design — this is so that the caller can decide whether to accept\n\t// this FF for merging or not.  If the result is accepted,\n\t// remove_output_ff should be called on the second output argument.\n\tbool find_output_ff(RTLIL::SigSpec sig, FfData &ff, pool<std::pair<Cell *, int>> &bits);\n\n\t// Like above, but returns a FF to merge into a given cell input.  Takes\n\t// sig_q, which is the current cell input — it will search for FFs with\n\t// matching sig_q.\n\t//\n\t// As opposed to find_output_ff, this function doesn't care about usage\n\t// counts, and may return FF bits that also have other fanout.  This\n\t// should not be a problem for input FF merging.\n\t//\n\t// As a special case, if some of the bits in sig_q are constant, this\n\t// function will accept them as well, by synthesizing in-place\n\t// a constant-input FF bit (with matching initial value and reset value).\n\t// However, this will not work if the input is all-constant — if the caller\n\t// cares about this case, it needs to check for it explicitely.\n\tbool find_input_ff(RTLIL::SigSpec sig, FfData &ff, pool<std::pair<Cell *, int>> &bits);\n\n\t// To be called on find_output_ff result that will be merged.  This\n\t// marks the given FF bits as used up (and not to be considered for\n\t// further merging as inputs), and reconnects their Q ports to a dummy\n\t// wire (since the wire previously connected there will now be driven\n\t// by the merged-to cell instead).\n\tvoid remove_output_ff(const pool<std::pair<Cell *, int>> &bits);\n\n\t// To be called on find_input_ff result that will be merged.  This\n\t// marks the given FF bits as used, and disallows merging them as\n\t// outputs.  They can, however, still be merged as inputs again\n\t// (perhaps for another cell).\n\tvoid mark_input_ff(const pool<std::pair<Cell *, int>> &bits);\n\n\tvoid set(FfInitVals *initvals_, RTLIL::Module *module_);\n\n\tvoid clear();\n\n\tFfMergeHelper(FfInitVals *initvals, RTLIL::Module *module) {\n\t\tset(initvals, module);\n\t}\n\n\tFfMergeHelper() {}\n};\n\nYOSYS_NAMESPACE_END\n\n#endif\n",
                 "fmt.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2020  whitequark <whitequark@whitequark.org>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted, provided that the above\n *  copyright notice and this permission notice appear in all copies.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#ifndef FMT_H\n#define FMT_H\n\n#include \"kernel/yosys.h\"\n\nYOSYS_NAMESPACE_BEGIN\n\n// Verilog format argument, such as the arguments in:\n//   $display(\"foo %d bar %01x\", 4'b0, $signed(2'b11))\nstruct VerilogFmtArg {\n\tenum {\n\t\tSTRING  = 0,\n\t\tINTEGER = 1,\n\t\tTIME    = 2,\n\t} type;\n\n\t// All types\n\tstd::string filename;\n\tunsigned first_line;\n\n\t// STRING type\n\tstd::string str;\n\n\t// INTEGER type\n\tRTLIL::SigSpec sig;\n\tbool signed_ = false;\n\n\t// TIME type\n\tbool realtime = false;\n};\n\n// RTLIL format part, such as the substitutions in:\n//   \"foo {4:> 4du} bar {2:<01hs}\"\nstruct FmtPart {\n\tenum {\n\t\tSTRING  \t= 0,\n\t\tINTEGER \t= 1,\n\t\tCHARACTER = 2,\n\t\tTIME    \t= 3,\n\t} type;\n\n\t// STRING type\n\tstd::string str;\n\n\t// INTEGER/CHARACTER types\n\tRTLIL::SigSpec sig;\n\n\t// INTEGER/CHARACTER/TIME types\n\tenum {\n\t\tRIGHT\t= 0,\n\t\tLEFT\t= 1,\n\t} justify = RIGHT;\n\tchar padding = '\\0';\n\tsize_t width = 0;\n\t\n\t// INTEGER type\n\tunsigned base = 10;\n\tbool signed_ = false;\n\tbool plus = false;\n\n\t// TIME type\n\tbool realtime = false;\n};\n\nstruct Fmt {\npublic:\n\tstd::vector<FmtPart> parts;\n\n\tvoid append_string(const std::string &str);\n\n\tvoid parse_rtlil(const RTLIL::Cell *cell);\n\tvoid emit_rtlil(RTLIL::Cell *cell) const;\n\n\tvoid parse_verilog(const std::vector<VerilogFmtArg> &args, bool sformat_like, int default_base, RTLIL::IdString task_name, RTLIL::IdString module_name);\n\tstd::vector<VerilogFmtArg> emit_verilog() const;\n\n\tvoid emit_cxxrtl(std::ostream &f, std::function<void(const RTLIL::SigSpec &)> emit_sig) const;\n\n\tstd::string render() const;\n\nprivate:\n\tvoid apply_verilog_automatic_sizing_and_add(FmtPart &part);\n};\n\nYOSYS_NAMESPACE_END\n\n#endif\n",
-                "hashlib.h": "// This is free and unencumbered software released into the public domain.\n//\n// Anyone is free to copy, modify, publish, use, compile, sell, or\n// distribute this software, either in source code form or as a compiled\n// binary, for any purpose, commercial or non-commercial, and by any\n// means.\n\n// -------------------------------------------------------\n// Written by Claire Xenia Wolf <claire@yosyshq.com> in 2014\n// -------------------------------------------------------\n\n#ifndef HASHLIB_H\n#define HASHLIB_H\n\n#include <stdexcept>\n#include <algorithm>\n#include <string>\n#include <vector>\n\nnamespace hashlib {\n\nconst int hashtable_size_trigger = 2;\nconst int hashtable_size_factor = 3;\n\n// The XOR version of DJB2\ninline unsigned int mkhash(unsigned int a, unsigned int b) {\n\treturn ((a << 5) + a) ^ b;\n}\n\n// traditionally 5381 is used as starting value for the djb2 hash\nconst unsigned int mkhash_init = 5381;\n\n// The ADD version of DJB2\n// (use this version for cache locality in b)\ninline unsigned int mkhash_add(unsigned int a, unsigned int b) {\n\treturn ((a << 5) + a) + b;\n}\n\ninline unsigned int mkhash_xorshift(unsigned int a) {\n\tif (sizeof(a) == 4) {\n\t\ta ^= a << 13;\n\t\ta ^= a >> 17;\n\t\ta ^= a << 5;\n\t} else if (sizeof(a) == 8) {\n\t\ta ^= a << 13;\n\t\ta ^= a >> 7;\n\t\ta ^= a << 17;\n\t} else\n\t\tthrow std::runtime_error(\"mkhash_xorshift() only implemented for 32 bit and 64 bit ints\");\n\treturn a;\n}\n\ntemplate<typename T> struct hash_ops {\n\tstatic inline bool cmp(const T &a, const T &b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(const T &a) {\n\t\treturn a.hash();\n\t}\n};\n\nstruct hash_int_ops {\n\ttemplate<typename T>\n\tstatic inline bool cmp(T a, T b) {\n\t\treturn a == b;\n\t}\n};\n\ntemplate<> struct hash_ops<bool> : hash_int_ops\n{\n\tstatic inline unsigned int hash(bool a) {\n\t\treturn a ? 1 : 0;\n\t}\n};\ntemplate<> struct hash_ops<int32_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(int32_t a) {\n\t\treturn a;\n\t}\n};\ntemplate<> struct hash_ops<int64_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(int64_t a) {\n\t\treturn mkhash((unsigned int)(a), (unsigned int)(a >> 32));\n\t}\n};\ntemplate<> struct hash_ops<uint32_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(uint32_t a) {\n\t\treturn a;\n\t}\n};\ntemplate<> struct hash_ops<uint64_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(uint64_t a) {\n\t\treturn mkhash((unsigned int)(a), (unsigned int)(a >> 32));\n\t}\n};\n\ntemplate<> struct hash_ops<std::string> {\n\tstatic inline bool cmp(const std::string &a, const std::string &b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(const std::string &a) {\n\t\tunsigned int v = 0;\n\t\tfor (auto c : a)\n\t\t\tv = mkhash(v, c);\n\t\treturn v;\n\t}\n};\n\ntemplate<typename P, typename Q> struct hash_ops<std::pair<P, Q>> {\n\tstatic inline bool cmp(std::pair<P, Q> a, std::pair<P, Q> b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(std::pair<P, Q> a) {\n\t\treturn mkhash(hash_ops<P>::hash(a.first), hash_ops<Q>::hash(a.second));\n\t}\n};\n\ntemplate<typename... T> struct hash_ops<std::tuple<T...>> {\n\tstatic inline bool cmp(std::tuple<T...> a, std::tuple<T...> b) {\n\t\treturn a == b;\n\t}\n\ttemplate<size_t I = 0>\n\tstatic inline typename std::enable_if<I == sizeof...(T), unsigned int>::type hash(std::tuple<T...>) {\n\t\treturn mkhash_init;\n\t}\n\ttemplate<size_t I = 0>\n\tstatic inline typename std::enable_if<I != sizeof...(T), unsigned int>::type hash(std::tuple<T...> a) {\n\t\ttypedef hash_ops<typename std::tuple_element<I, std::tuple<T...>>::type> element_ops_t;\n\t\treturn mkhash(hash<I+1>(a), element_ops_t::hash(std::get<I>(a)));\n\t}\n};\n\ntemplate<typename T> struct hash_ops<std::vector<T>> {\n\tstatic inline bool cmp(std::vector<T> a, std::vector<T> b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(std::vector<T> a) {\n\t\tunsigned int h = mkhash_init;\n\t\tfor (auto k : a)\n\t\t\th = mkhash(h, hash_ops<T>::hash(k));\n\t\treturn h;\n\t}\n};\n\nstruct hash_cstr_ops {\n\tstatic inline bool cmp(const char *a, const char *b) {\n\t\tfor (int i = 0; a[i] || b[i]; i++)\n\t\t\tif (a[i] != b[i])\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\tstatic inline unsigned int hash(const char *a) {\n\t\tunsigned int hash = mkhash_init;\n\t\twhile (*a)\n\t\t\thash = mkhash(hash, *(a++));\n\t\treturn hash;\n\t}\n};\n\nstruct hash_ptr_ops {\n\tstatic inline bool cmp(const void *a, const void *b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(const void *a) {\n\t\treturn (uintptr_t)a;\n\t}\n};\n\nstruct hash_obj_ops {\n\tstatic inline bool cmp(const void *a, const void *b) {\n\t\treturn a == b;\n\t}\n\ttemplate<typename T>\n\tstatic inline unsigned int hash(const T *a) {\n\t\treturn a ? a->hash() : 0;\n\t}\n};\n\ntemplate<typename T>\ninline unsigned int mkhash(const T &v) {\n\treturn hash_ops<T>().hash(v);\n}\n\ninline int hashtable_size(int min_size)\n{\n\tstatic std::vector<int> zero_and_some_primes = {\n\t\t0, 23, 29, 37, 47, 59, 79, 101, 127, 163, 211, 269, 337, 431, 541, 677,\n\t\t853, 1069, 1361, 1709, 2137, 2677, 3347, 4201, 5261, 6577, 8231, 10289,\n\t\t12889, 16127, 20161, 25219, 31531, 39419, 49277, 61603, 77017, 96281,\n\t\t120371, 150473, 188107, 235159, 293957, 367453, 459317, 574157, 717697,\n\t\t897133, 1121423, 1401791, 1752239, 2190299, 2737937, 3422429, 4278037,\n\t\t5347553, 6684443, 8355563, 10444457, 13055587, 16319519, 20399411,\n\t\t25499291, 31874149, 39842687, 49803361, 62254207, 77817767, 97272239,\n\t\t121590311, 151987889, 189984863, 237481091, 296851369, 371064217\n\t};\n\n\tfor (auto p : zero_and_some_primes)\n\t\tif (p >= min_size) return p;\n\n\tif (sizeof(int) == 4)\n\t\tthrow std::length_error(\"hash table exceeded maximum size.\\nDesign is likely too large for yosys to handle, if possible try not to flatten the design.\");\n\n\tfor (auto p : zero_and_some_primes)\n\t\tif (100129 * p > min_size) return 100129 * p;\n\n\tthrow std::length_error(\"hash table exceeded maximum size.\");\n}\n\ntemplate<typename K, typename T, typename OPS = hash_ops<K>> class dict;\ntemplate<typename K, int offset = 0, typename OPS = hash_ops<K>> class idict;\ntemplate<typename K, typename OPS = hash_ops<K>> class pool;\ntemplate<typename K, typename OPS = hash_ops<K>> class mfp;\n\ntemplate<typename K, typename T, typename OPS>\nclass dict\n{\n\tstruct entry_t\n\t{\n\t\tstd::pair<K, T> udata;\n\t\tint next;\n\n\t\tentry_t() { }\n\t\tentry_t(const std::pair<K, T> &udata, int next) : udata(udata), next(next) { }\n\t\tentry_t(std::pair<K, T> &&udata, int next) : udata(std::move(udata)), next(next) { }\n\t\tbool operator<(const entry_t &other) const { return udata.first < other.udata.first; }\n\t};\n\n\tstd::vector<int> hashtable;\n\tstd::vector<entry_t> entries;\n\tOPS ops;\n\n#ifdef NDEBUG\n\tstatic inline void do_assert(bool) { }\n#else\n\tstatic inline void do_assert(bool cond) {\n\t\tif (!cond) throw std::runtime_error(\"dict<> assert failed.\");\n\t}\n#endif\n\n\tint do_hash(const K &key) const\n\t{\n\t\tunsigned int hash = 0;\n\t\tif (!hashtable.empty())\n\t\t\thash = ops.hash(key) % (unsigned int)(hashtable.size());\n\t\treturn hash;\n\t}\n\n\tvoid do_rehash()\n\t{\n\t\thashtable.clear();\n\t\thashtable.resize(hashtable_size(entries.capacity() * hashtable_size_factor), -1);\n\n\t\tfor (int i = 0; i < int(entries.size()); i++) {\n\t\t\tdo_assert(-1 <= entries[i].next && entries[i].next < int(entries.size()));\n\t\t\tint hash = do_hash(entries[i].udata.first);\n\t\t\tentries[i].next = hashtable[hash];\n\t\t\thashtable[hash] = i;\n\t\t}\n\t}\n\n\tint do_erase(int index, int hash)\n\t{\n\t\tdo_assert(index < int(entries.size()));\n\t\tif (hashtable.empty() || index < 0)\n\t\t\treturn 0;\n\n\t\tint k = hashtable[hash];\n\t\tdo_assert(0 <= k && k < int(entries.size()));\n\n\t\tif (k == index) {\n\t\t\thashtable[hash] = entries[index].next;\n\t\t} else {\n\t\t\twhile (entries[k].next != index) {\n\t\t\t\tk = entries[k].next;\n\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t}\n\t\t\tentries[k].next = entries[index].next;\n\t\t}\n\n\t\tint back_idx = entries.size()-1;\n\n\t\tif (index != back_idx)\n\t\t{\n\t\t\tint back_hash = do_hash(entries[back_idx].udata.first);\n\n\t\t\tk = hashtable[back_hash];\n\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\n\t\t\tif (k == back_idx) {\n\t\t\t\thashtable[back_hash] = index;\n\t\t\t} else {\n\t\t\t\twhile (entries[k].next != back_idx) {\n\t\t\t\t\tk = entries[k].next;\n\t\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t\t}\n\t\t\t\tentries[k].next = index;\n\t\t\t}\n\n\t\t\tentries[index] = std::move(entries[back_idx]);\n\t\t}\n\n\t\tentries.pop_back();\n\n\t\tif (entries.empty())\n\t\t\thashtable.clear();\n\n\t\treturn 1;\n\t}\n\n\tint do_lookup(const K &key, int &hash) const\n\t{\n\t\tif (hashtable.empty())\n\t\t\treturn -1;\n\n\t\tif (entries.size() * hashtable_size_trigger > hashtable.size()) {\n\t\t\t((dict*)this)->do_rehash();\n\t\t\thash = do_hash(key);\n\t\t}\n\n\t\tint index = hashtable[hash];\n\n\t\twhile (index >= 0 && !ops.cmp(entries[index].udata.first, key)) {\n\t\t\tindex = entries[index].next;\n\t\t\tdo_assert(-1 <= index && index < int(entries.size()));\n\t\t}\n\n\t\treturn index;\n\t}\n\n\tint do_insert(const K &key, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(std::pair<K, T>(key, T()), -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(key);\n\t\t} else {\n\t\t\tentries.emplace_back(std::pair<K, T>(key, T()), hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\n\tint do_insert(const std::pair<K, T> &value, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(value, -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(value.first);\n\t\t} else {\n\t\t\tentries.emplace_back(value, hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\n\tint do_insert(std::pair<K, T> &&rvalue, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tauto key = rvalue.first;\n\t\t\tentries.emplace_back(std::forward<std::pair<K, T>>(rvalue), -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(key);\n\t\t} else {\n\t\t\tentries.emplace_back(std::forward<std::pair<K, T>>(rvalue), hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\npublic:\n\tclass const_iterator\n\t{\n\t\tfriend class dict;\n\tprotected:\n\t\tconst dict *ptr;\n\t\tint index;\n\t\tconst_iterator(const dict *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef std::pair<K, T> value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef std::pair<K, T>* pointer;\n\t\ttypedef std::pair<K, T>& reference;\n\t\tconst_iterator() { }\n\t\tconst_iterator operator++() { index--; return *this; }\n\t\tconst_iterator operator+=(int amt) { index -= amt; return *this; }\n\t\tbool operator<(const const_iterator &other) const { return index > other.index; }\n\t\tbool operator==(const const_iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const const_iterator &other) const { return index != other.index; }\n\t\tconst std::pair<K, T> &operator*() const { return ptr->entries[index].udata; }\n\t\tconst std::pair<K, T> *operator->() const { return &ptr->entries[index].udata; }\n\t};\n\n\tclass iterator\n\t{\n\t\tfriend class dict;\n\tprotected:\n\t\tdict *ptr;\n\t\tint index;\n\t\titerator(dict *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef std::pair<K, T> value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef std::pair<K, T>* pointer;\n\t\ttypedef std::pair<K, T>& reference;\n\t\titerator() { }\n\t\titerator operator++() { index--; return *this; }\n\t\titerator operator+=(int amt) { index -= amt; return *this; }\n\t\tbool operator<(const iterator &other) const { return index > other.index; }\n\t\tbool operator==(const iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const iterator &other) const { return index != other.index; }\n\t\tstd::pair<K, T> &operator*() { return ptr->entries[index].udata; }\n\t\tstd::pair<K, T> *operator->() { return &ptr->entries[index].udata; }\n\t\tconst std::pair<K, T> &operator*() const { return ptr->entries[index].udata; }\n\t\tconst std::pair<K, T> *operator->() const { return &ptr->entries[index].udata; }\n\t\toperator const_iterator() const { return const_iterator(ptr, index); }\n\t};\n\n\tdict()\n\t{\n\t}\n\n\tdict(const dict &other)\n\t{\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t}\n\n\tdict(dict &&other)\n\t{\n\t\tswap(other);\n\t}\n\n\tdict &operator=(const dict &other) {\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t\treturn *this;\n\t}\n\n\tdict &operator=(dict &&other) {\n\t\tclear();\n\t\tswap(other);\n\t\treturn *this;\n\t}\n\n\tdict(const std::initializer_list<std::pair<K, T>> &list)\n\t{\n\t\tfor (auto &it : list)\n\t\t\tinsert(it);\n\t}\n\n\ttemplate<class InputIterator>\n\tdict(InputIterator first, InputIterator last)\n\t{\n\t\tinsert(first, last);\n\t}\n\n\ttemplate<class InputIterator>\n\tvoid insert(InputIterator first, InputIterator last)\n\t{\n\t\tfor (; first != last; ++first)\n\t\t\tinsert(*first);\n\t}\n\n\tstd::pair<iterator, bool> insert(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(key, hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> insert(const std::pair<K, T> &value)\n\t{\n\t\tint hash = do_hash(value.first);\n\t\tint i = do_lookup(value.first, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(value, hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> insert(std::pair<K, T> &&rvalue)\n\t{\n\t\tint hash = do_hash(rvalue.first);\n\t\tint i = do_lookup(rvalue.first, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::forward<std::pair<K, T>>(rvalue), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K const &key, T const &value)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(key, value), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K const &key, T &&rvalue)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(key, std::forward<T>(rvalue)), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K &&rkey, T const &value)\n\t{\n\t\tint hash = do_hash(rkey);\n\t\tint i = do_lookup(rkey, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(std::forward<K>(rkey), value), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K &&rkey, T &&rvalue)\n\t{\n\t\tint hash = do_hash(rkey);\n\t\tint i = do_lookup(rkey, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(std::forward<K>(rkey), std::forward<T>(rvalue)), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tint erase(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint index = do_lookup(key, hash);\n\t\treturn do_erase(index, hash);\n\t}\n\n\titerator erase(iterator it)\n\t{\n\t\tint hash = do_hash(it->first);\n\t\tdo_erase(it.index, hash);\n\t\treturn ++it;\n\t}\n\n\tint count(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 ? 0 : 1;\n\t}\n\n\tint count(const K &key, const_iterator it) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 || i > it.index ? 0 : 1;\n\t}\n\n\titerator find(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn iterator(this, i);\n\t}\n\n\tconst_iterator find(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn const_iterator(this, i);\n\t}\n\n\tT& at(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\tthrow std::out_of_range(\"dict::at()\");\n\t\treturn entries[i].udata.second;\n\t}\n\n\tconst T& at(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\tthrow std::out_of_range(\"dict::at()\");\n\t\treturn entries[i].udata.second;\n\t}\n\n\tconst T& at(const K &key, const T &defval) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn defval;\n\t\treturn entries[i].udata.second;\n\t}\n\n\tT& operator[](const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\ti = do_insert(std::pair<K, T>(key, T()), hash);\n\t\treturn entries[i].udata.second;\n\t}\n\n\ttemplate<typename Compare = std::less<K>>\n\tvoid sort(Compare comp = Compare())\n\t{\n\t\tstd::sort(entries.begin(), entries.end(), [comp](const entry_t &a, const entry_t &b){ return comp(b.udata.first, a.udata.first); });\n\t\tdo_rehash();\n\t}\n\n\tvoid swap(dict &other)\n\t{\n\t\thashtable.swap(other.hashtable);\n\t\tentries.swap(other.entries);\n\t}\n\n\tbool operator==(const dict &other) const {\n\t\tif (size() != other.size())\n\t\t\treturn false;\n\t\tfor (auto &it : entries) {\n\t\t\tauto oit = other.find(it.udata.first);\n\t\t\tif (oit == other.end() || !(oit->second == it.udata.second))\n\t\t\t\treturn false;\n\t\t}\n\t\treturn true;\n\t}\n\n\tbool operator!=(const dict &other) const {\n\t\treturn !operator==(other);\n\t}\n\n\tunsigned int hash() const {\n\t\tunsigned int h = mkhash_init;\n\t\tfor (auto &entry : entries) {\n\t\t\th ^= hash_ops<K>::hash(entry.udata.first);\n\t\t\th ^= hash_ops<T>::hash(entry.udata.second);\n\t\t}\n\t\treturn h;\n\t}\n\n\tvoid reserve(size_t n) { entries.reserve(n); }\n\tsize_t size() const { return entries.size(); }\n\tbool empty() const { return entries.empty(); }\n\tvoid clear() { hashtable.clear(); entries.clear(); }\n\n\titerator begin() { return iterator(this, int(entries.size())-1); }\n\titerator element(int n) { return iterator(this, int(entries.size())-1-n); }\n\titerator end() { return iterator(nullptr, -1); }\n\n\tconst_iterator begin() const { return const_iterator(this, int(entries.size())-1); }\n\tconst_iterator element(int n) const { return const_iterator(this, int(entries.size())-1-n); }\n\tconst_iterator end() const { return const_iterator(nullptr, -1); }\n};\n\ntemplate<typename K, typename OPS>\nclass pool\n{\n\ttemplate<typename, int, typename> friend class idict;\n\nprotected:\n\tstruct entry_t\n\t{\n\t\tK udata;\n\t\tint next;\n\n\t\tentry_t() { }\n\t\tentry_t(const K &udata, int next) : udata(udata), next(next) { }\n\t\tentry_t(K &&udata, int next) : udata(std::move(udata)), next(next) { }\n\t};\n\n\tstd::vector<int> hashtable;\n\tstd::vector<entry_t> entries;\n\tOPS ops;\n\n#ifdef NDEBUG\n\tstatic inline void do_assert(bool) { }\n#else\n\tstatic inline void do_assert(bool cond) {\n\t\tif (!cond) throw std::runtime_error(\"pool<> assert failed.\");\n\t}\n#endif\n\n\tint do_hash(const K &key) const\n\t{\n\t\tunsigned int hash = 0;\n\t\tif (!hashtable.empty())\n\t\t\thash = ops.hash(key) % (unsigned int)(hashtable.size());\n\t\treturn hash;\n\t}\n\n\tvoid do_rehash()\n\t{\n\t\thashtable.clear();\n\t\thashtable.resize(hashtable_size(entries.capacity() * hashtable_size_factor), -1);\n\n\t\tfor (int i = 0; i < int(entries.size()); i++) {\n\t\t\tdo_assert(-1 <= entries[i].next && entries[i].next < int(entries.size()));\n\t\t\tint hash = do_hash(entries[i].udata);\n\t\t\tentries[i].next = hashtable[hash];\n\t\t\thashtable[hash] = i;\n\t\t}\n\t}\n\n\tint do_erase(int index, int hash)\n\t{\n\t\tdo_assert(index < int(entries.size()));\n\t\tif (hashtable.empty() || index < 0)\n\t\t\treturn 0;\n\n\t\tint k = hashtable[hash];\n\t\tif (k == index) {\n\t\t\thashtable[hash] = entries[index].next;\n\t\t} else {\n\t\t\twhile (entries[k].next != index) {\n\t\t\t\tk = entries[k].next;\n\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t}\n\t\t\tentries[k].next = entries[index].next;\n\t\t}\n\n\t\tint back_idx = entries.size()-1;\n\n\t\tif (index != back_idx)\n\t\t{\n\t\t\tint back_hash = do_hash(entries[back_idx].udata);\n\n\t\t\tk = hashtable[back_hash];\n\t\t\tif (k == back_idx) {\n\t\t\t\thashtable[back_hash] = index;\n\t\t\t} else {\n\t\t\t\twhile (entries[k].next != back_idx) {\n\t\t\t\t\tk = entries[k].next;\n\t\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t\t}\n\t\t\t\tentries[k].next = index;\n\t\t\t}\n\n\t\t\tentries[index] = std::move(entries[back_idx]);\n\t\t}\n\n\t\tentries.pop_back();\n\n\t\tif (entries.empty())\n\t\t\thashtable.clear();\n\n\t\treturn 1;\n\t}\n\n\tint do_lookup(const K &key, int &hash) const\n\t{\n\t\tif (hashtable.empty())\n\t\t\treturn -1;\n\n\t\tif (entries.size() * hashtable_size_trigger > hashtable.size()) {\n\t\t\t((pool*)this)->do_rehash();\n\t\t\thash = do_hash(key);\n\t\t}\n\n\t\tint index = hashtable[hash];\n\n\t\twhile (index >= 0 && !ops.cmp(entries[index].udata, key)) {\n\t\t\tindex = entries[index].next;\n\t\t\tdo_assert(-1 <= index && index < int(entries.size()));\n\t\t}\n\n\t\treturn index;\n\t}\n\n\tint do_insert(const K &value, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(value, -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(value);\n\t\t} else {\n\t\t\tentries.emplace_back(value, hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\n\tint do_insert(K &&rvalue, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(std::forward<K>(rvalue), -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(rvalue);\n\t\t} else {\n\t\t\tentries.emplace_back(std::forward<K>(rvalue), hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\npublic:\n\tclass const_iterator\n\t{\n\t\tfriend class pool;\n\tprotected:\n\t\tconst pool *ptr;\n\t\tint index;\n\t\tconst_iterator(const pool *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef K value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef K* pointer;\n\t\ttypedef K& reference;\n\t\tconst_iterator() { }\n\t\tconst_iterator operator++() { index--; return *this; }\n\t\tbool operator==(const const_iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const const_iterator &other) const { return index != other.index; }\n\t\tconst K &operator*() const { return ptr->entries[index].udata; }\n\t\tconst K *operator->() const { return &ptr->entries[index].udata; }\n\t};\n\n\tclass iterator\n\t{\n\t\tfriend class pool;\n\tprotected:\n\t\tpool *ptr;\n\t\tint index;\n\t\titerator(pool *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef K value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef K* pointer;\n\t\ttypedef K& reference;\n\t\titerator() { }\n\t\titerator operator++() { index--; return *this; }\n\t\tbool operator==(const iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const iterator &other) const { return index != other.index; }\n\t\tK &operator*() { return ptr->entries[index].udata; }\n\t\tK *operator->() { return &ptr->entries[index].udata; }\n\t\tconst K &operator*() const { return ptr->entries[index].udata; }\n\t\tconst K *operator->() const { return &ptr->entries[index].udata; }\n\t\toperator const_iterator() const { return const_iterator(ptr, index); }\n\t};\n\n\tpool()\n\t{\n\t}\n\n\tpool(const pool &other)\n\t{\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t}\n\n\tpool(pool &&other)\n\t{\n\t\tswap(other);\n\t}\n\n\tpool &operator=(const pool &other) {\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t\treturn *this;\n\t}\n\n\tpool &operator=(pool &&other) {\n\t\tclear();\n\t\tswap(other);\n\t\treturn *this;\n\t}\n\n\tpool(const std::initializer_list<K> &list)\n\t{\n\t\tfor (auto &it : list)\n\t\t\tinsert(it);\n\t}\n\n\ttemplate<class InputIterator>\n\tpool(InputIterator first, InputIterator last)\n\t{\n\t\tinsert(first, last);\n\t}\n\n\ttemplate<class InputIterator>\n\tvoid insert(InputIterator first, InputIterator last)\n\t{\n\t\tfor (; first != last; ++first)\n\t\t\tinsert(*first);\n\t}\n\n\tstd::pair<iterator, bool> insert(const K &value)\n\t{\n\t\tint hash = do_hash(value);\n\t\tint i = do_lookup(value, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(value, hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> insert(K &&rvalue)\n\t{\n\t\tint hash = do_hash(rvalue);\n\t\tint i = do_lookup(rvalue, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::forward<K>(rvalue), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\ttemplate<typename... Args>\n\tstd::pair<iterator, bool> emplace(Args&&... args)\n\t{\n\t\treturn insert(K(std::forward<Args>(args)...));\n\t}\n\n\tint erase(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint index = do_lookup(key, hash);\n\t\treturn do_erase(index, hash);\n\t}\n\n\titerator erase(iterator it)\n\t{\n\t\tint hash = do_hash(*it);\n\t\tdo_erase(it.index, hash);\n\t\treturn ++it;\n\t}\n\n\tint count(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 ? 0 : 1;\n\t}\n\n\tint count(const K &key, const_iterator it) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 || i > it.index ? 0 : 1;\n\t}\n\n\titerator find(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn iterator(this, i);\n\t}\n\n\tconst_iterator find(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn const_iterator(this, i);\n\t}\n\n\tbool operator[](const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i >= 0;\n\t}\n\n\ttemplate<typename Compare = std::less<K>>\n\tvoid sort(Compare comp = Compare())\n\t{\n\t\tstd::sort(entries.begin(), entries.end(), [comp](const entry_t &a, const entry_t &b){ return comp(b.udata, a.udata); });\n\t\tdo_rehash();\n\t}\n\n\tK pop()\n\t{\n\t\titerator it = begin();\n\t\tK ret = *it;\n\t\terase(it);\n\t\treturn ret;\n\t}\n\n\tvoid swap(pool &other)\n\t{\n\t\thashtable.swap(other.hashtable);\n\t\tentries.swap(other.entries);\n\t}\n\n\tbool operator==(const pool &other) const {\n\t\tif (size() != other.size())\n\t\t\treturn false;\n\t\tfor (auto &it : entries)\n\t\t\tif (!other.count(it.udata))\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\n\tbool operator!=(const pool &other) const {\n\t\treturn !operator==(other);\n\t}\n\n\tunsigned int hash() const {\n\t\tunsigned int hashval = mkhash_init;\n\t\tfor (auto &it : entries)\n\t\t\thashval ^= ops.hash(it.udata);\n\t\treturn hashval;\n\t}\n\n\tvoid reserve(size_t n) { entries.reserve(n); }\n\tsize_t size() const { return entries.size(); }\n\tbool empty() const { return entries.empty(); }\n\tvoid clear() { hashtable.clear(); entries.clear(); }\n\n\titerator begin() { return iterator(this, int(entries.size())-1); }\n\titerator element(int n) { return iterator(this, int(entries.size())-1-n); }\n\titerator end() { return iterator(nullptr, -1); }\n\n\tconst_iterator begin() const { return const_iterator(this, int(entries.size())-1); }\n\tconst_iterator element(int n) const { return const_iterator(this, int(entries.size())-1-n); }\n\tconst_iterator end() const { return const_iterator(nullptr, -1); }\n};\n\ntemplate<typename K, int offset, typename OPS>\nclass idict\n{\n\tpool<K, OPS> database;\n\npublic:\n\tclass const_iterator\n\t{\n\t\tfriend class idict;\n\tprotected:\n\t\tconst idict &container;\n\t\tint index;\n\t\tconst_iterator(const idict &container, int index) : container(container), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef K value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef K* pointer;\n\t\ttypedef K& reference;\n\t\tconst_iterator() { }\n\t\tconst_iterator operator++() { index++; return *this; }\n\t\tbool operator==(const const_iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const const_iterator &other) const { return index != other.index; }\n\t\tconst K &operator*() const { return container[index]; }\n\t\tconst K *operator->() const { return &container[index]; }\n\t};\n\n\tint operator()(const K &key)\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\ti = database.do_insert(key, hash);\n\t\treturn i + offset;\n\t}\n\n\tint at(const K &key) const\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\tthrow std::out_of_range(\"idict::at()\");\n\t\treturn i + offset;\n\t}\n\n\tint at(const K &key, int defval) const\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn defval;\n\t\treturn i + offset;\n\t}\n\n\tint count(const K &key) const\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\treturn i < 0 ? 0 : 1;\n\t}\n\n\tvoid expect(const K &key, int i)\n\t{\n\t\tint j = (*this)(key);\n\t\tif (i != j)\n\t\t\tthrow std::out_of_range(\"idict::expect()\");\n\t}\n\n\tconst K &operator[](int index) const\n\t{\n\t\treturn database.entries.at(index - offset).udata;\n\t}\n\n\tvoid swap(idict &other)\n\t{\n\t\tdatabase.swap(other.database);\n\t}\n\n\tvoid reserve(size_t n) { database.reserve(n); }\n\tsize_t size() const { return database.size(); }\n\tbool empty() const { return database.empty(); }\n\tvoid clear() { database.clear(); }\n\n\tconst_iterator begin() const { return const_iterator(*this, offset); }\n\tconst_iterator element(int n) const { return const_iterator(*this, n); }\n\tconst_iterator end() const { return const_iterator(*this, offset + size()); }\n};\n\ntemplate<typename K, typename OPS>\nclass mfp\n{\n\tmutable idict<K, 0, OPS> database;\n\tmutable std::vector<int> parents;\n\npublic:\n\ttypedef typename idict<K, 0, OPS>::const_iterator const_iterator;\n\n\tint operator()(const K &key) const\n\t{\n\t\tint i = database(key);\n\t\tparents.resize(database.size(), -1);\n\t\treturn i;\n\t}\n\n\tconst K &operator[](int index) const\n\t{\n\t\treturn database[index];\n\t}\n\n\tint ifind(int i) const\n\t{\n\t\tint p = i, k = i;\n\n\t\twhile (parents[p] != -1)\n\t\t\tp = parents[p];\n\n\t\twhile (k != p) {\n\t\t\tint next_k = parents[k];\n\t\t\tparents[k] = p;\n\t\t\tk = next_k;\n\t\t}\n\n\t\treturn p;\n\t}\n\n\tvoid imerge(int i, int j)\n\t{\n\t\ti = ifind(i);\n\t\tj = ifind(j);\n\n\t\tif (i != j)\n\t\t\tparents[i] = j;\n\t}\n\n\tvoid ipromote(int i)\n\t{\n\t\tint k = i;\n\n\t\twhile (k != -1) {\n\t\t\tint next_k = parents[k];\n\t\t\tparents[k] = i;\n\t\t\tk = next_k;\n\t\t}\n\n\t\tparents[i] = -1;\n\t}\n\n\tint lookup(const K &a) const\n\t{\n\t\treturn ifind((*this)(a));\n\t}\n\n\tconst K &find(const K &a) const\n\t{\n\t\tint i = database.at(a, -1);\n\t\tif (i < 0)\n\t\t\treturn a;\n\t\treturn (*this)[ifind(i)];\n\t}\n\n\tvoid merge(const K &a, const K &b)\n\t{\n\t\timerge((*this)(a), (*this)(b));\n\t}\n\n\tvoid promote(const K &a)\n\t{\n\t\tint i = database.at(a, -1);\n\t\tif (i >= 0)\n\t\t\tipromote(i);\n\t}\n\n\tvoid swap(mfp &other)\n\t{\n\t\tdatabase.swap(other.database);\n\t\tparents.swap(other.parents);\n\t}\n\n\tvoid reserve(size_t n) { database.reserve(n); }\n\tsize_t size() const { return database.size(); }\n\tbool empty() const { return database.empty(); }\n\tvoid clear() { database.clear(); parents.clear(); }\n\n\tconst_iterator begin() const { return database.begin(); }\n\tconst_iterator element(int n) const { return database.element(n); }\n\tconst_iterator end() const { return database.end(); }\n};\n\n} /* namespace hashlib */\n\n#endif\n",
+                "hashlib.h": "// This is free and unencumbered software released into the public domain.\n//\n// Anyone is free to copy, modify, publish, use, compile, sell, or\n// distribute this software, either in source code form or as a compiled\n// binary, for any purpose, commercial or non-commercial, and by any\n// means.\n\n// -------------------------------------------------------\n// Written by Claire Xenia Wolf <claire@yosyshq.com> in 2014\n// -------------------------------------------------------\n\n#ifndef HASHLIB_H\n#define HASHLIB_H\n\n#include <stdexcept>\n#include <algorithm>\n#include <string>\n#include <vector>\n\nnamespace hashlib {\n\nconst int hashtable_size_trigger = 2;\nconst int hashtable_size_factor = 3;\n\n// The XOR version of DJB2\ninline unsigned int mkhash(unsigned int a, unsigned int b) {\n\treturn ((a << 5) + a) ^ b;\n}\n\n// traditionally 5381 is used as starting value for the djb2 hash\nconst unsigned int mkhash_init = 5381;\n\n// The ADD version of DJB2\n// (use this version for cache locality in b)\ninline unsigned int mkhash_add(unsigned int a, unsigned int b) {\n\treturn ((a << 5) + a) + b;\n}\n\ninline unsigned int mkhash_xorshift(unsigned int a) {\n\tif (sizeof(a) == 4) {\n\t\ta ^= a << 13;\n\t\ta ^= a >> 17;\n\t\ta ^= a << 5;\n\t} else if (sizeof(a) == 8) {\n\t\ta ^= a << 13;\n\t\ta ^= a >> 7;\n\t\ta ^= a << 17;\n\t} else\n\t\tthrow std::runtime_error(\"mkhash_xorshift() only implemented for 32 bit and 64 bit ints\");\n\treturn a;\n}\n\ntemplate<typename T> struct hash_ops {\n\tstatic inline bool cmp(const T &a, const T &b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(const T &a) {\n\t\treturn a.hash();\n\t}\n};\n\nstruct hash_int_ops {\n\ttemplate<typename T>\n\tstatic inline bool cmp(T a, T b) {\n\t\treturn a == b;\n\t}\n};\n\ntemplate<> struct hash_ops<bool> : hash_int_ops\n{\n\tstatic inline unsigned int hash(bool a) {\n\t\treturn a ? 1 : 0;\n\t}\n};\ntemplate<> struct hash_ops<int32_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(int32_t a) {\n\t\treturn a;\n\t}\n};\ntemplate<> struct hash_ops<int64_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(int64_t a) {\n\t\treturn mkhash((unsigned int)(a), (unsigned int)(a >> 32));\n\t}\n};\ntemplate<> struct hash_ops<uint32_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(uint32_t a) {\n\t\treturn a;\n\t}\n};\ntemplate<> struct hash_ops<uint64_t> : hash_int_ops\n{\n\tstatic inline unsigned int hash(uint64_t a) {\n\t\treturn mkhash((unsigned int)(a), (unsigned int)(a >> 32));\n\t}\n};\n\ntemplate<> struct hash_ops<std::string> {\n\tstatic inline bool cmp(const std::string &a, const std::string &b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(const std::string &a) {\n\t\tunsigned int v = 0;\n\t\tfor (auto c : a)\n\t\t\tv = mkhash(v, c);\n\t\treturn v;\n\t}\n};\n\ntemplate<typename P, typename Q> struct hash_ops<std::pair<P, Q>> {\n\tstatic inline bool cmp(std::pair<P, Q> a, std::pair<P, Q> b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(std::pair<P, Q> a) {\n\t\treturn mkhash(hash_ops<P>::hash(a.first), hash_ops<Q>::hash(a.second));\n\t}\n};\n\ntemplate<typename... T> struct hash_ops<std::tuple<T...>> {\n\tstatic inline bool cmp(std::tuple<T...> a, std::tuple<T...> b) {\n\t\treturn a == b;\n\t}\n\ttemplate<size_t I = 0>\n\tstatic inline typename std::enable_if<I == sizeof...(T), unsigned int>::type hash(std::tuple<T...>) {\n\t\treturn mkhash_init;\n\t}\n\ttemplate<size_t I = 0>\n\tstatic inline typename std::enable_if<I != sizeof...(T), unsigned int>::type hash(std::tuple<T...> a) {\n\t\ttypedef hash_ops<typename std::tuple_element<I, std::tuple<T...>>::type> element_ops_t;\n\t\treturn mkhash(hash<I+1>(a), element_ops_t::hash(std::get<I>(a)));\n\t}\n};\n\ntemplate<typename T> struct hash_ops<std::vector<T>> {\n\tstatic inline bool cmp(std::vector<T> a, std::vector<T> b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(std::vector<T> a) {\n\t\tunsigned int h = mkhash_init;\n\t\tfor (auto k : a)\n\t\t\th = mkhash(h, hash_ops<T>::hash(k));\n\t\treturn h;\n\t}\n};\n\nstruct hash_cstr_ops {\n\tstatic inline bool cmp(const char *a, const char *b) {\n\t\tfor (int i = 0; a[i] || b[i]; i++)\n\t\t\tif (a[i] != b[i])\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\tstatic inline unsigned int hash(const char *a) {\n\t\tunsigned int hash = mkhash_init;\n\t\twhile (*a)\n\t\t\thash = mkhash(hash, *(a++));\n\t\treturn hash;\n\t}\n};\n\nstruct hash_ptr_ops {\n\tstatic inline bool cmp(const void *a, const void *b) {\n\t\treturn a == b;\n\t}\n\tstatic inline unsigned int hash(const void *a) {\n\t\treturn (uintptr_t)a;\n\t}\n};\n\nstruct hash_obj_ops {\n\tstatic inline bool cmp(const void *a, const void *b) {\n\t\treturn a == b;\n\t}\n\ttemplate<typename T>\n\tstatic inline unsigned int hash(const T *a) {\n\t\treturn a ? a->hash() : 0;\n\t}\n};\n\ntemplate<typename T>\ninline unsigned int mkhash(const T &v) {\n\treturn hash_ops<T>().hash(v);\n}\n\ninline int hashtable_size(int min_size)\n{\n\tstatic std::vector<int> zero_and_some_primes = {\n\t\t0, 23, 29, 37, 47, 59, 79, 101, 127, 163, 211, 269, 337, 431, 541, 677,\n\t\t853, 1069, 1361, 1709, 2137, 2677, 3347, 4201, 5261, 6577, 8231, 10289,\n\t\t12889, 16127, 20161, 25219, 31531, 39419, 49277, 61603, 77017, 96281,\n\t\t120371, 150473, 188107, 235159, 293957, 367453, 459317, 574157, 717697,\n\t\t897133, 1121423, 1401791, 1752239, 2190299, 2737937, 3422429, 4278037,\n\t\t5347553, 6684443, 8355563, 10444457, 13055587, 16319519, 20399411,\n\t\t25499291, 31874149, 39842687, 49803361, 62254207, 77817767, 97272239,\n\t\t121590311, 151987889, 189984863, 237481091, 296851369, 371064217\n\t};\n\n\tfor (auto p : zero_and_some_primes)\n\t\tif (p >= min_size) return p;\n\n\tif (sizeof(int) == 4)\n\t\tthrow std::length_error(\"hash table exceeded maximum size.\\nDesign is likely too large for yosys to handle, if possible try not to flatten the design.\");\n\n\tfor (auto p : zero_and_some_primes)\n\t\tif (100129 * p > min_size) return 100129 * p;\n\n\tthrow std::length_error(\"hash table exceeded maximum size.\");\n}\n\ntemplate<typename K, typename T, typename OPS = hash_ops<K>> class dict;\ntemplate<typename K, int offset = 0, typename OPS = hash_ops<K>> class idict;\ntemplate<typename K, typename OPS = hash_ops<K>> class pool;\ntemplate<typename K, typename OPS = hash_ops<K>> class mfp;\n\ntemplate<typename K, typename T, typename OPS>\nclass dict\n{\n\tstruct entry_t\n\t{\n\t\tstd::pair<K, T> udata;\n\t\tint next;\n\n\t\tentry_t() { }\n\t\tentry_t(const std::pair<K, T> &udata, int next) : udata(udata), next(next) { }\n\t\tentry_t(std::pair<K, T> &&udata, int next) : udata(std::move(udata)), next(next) { }\n\t\tbool operator<(const entry_t &other) const { return udata.first < other.udata.first; }\n\t};\n\n\tstd::vector<int> hashtable;\n\tstd::vector<entry_t> entries;\n\tOPS ops;\n\n#ifdef NDEBUG\n\tstatic inline void do_assert(bool) { }\n#else\n\tstatic inline void do_assert(bool cond) {\n\t\tif (!cond) throw std::runtime_error(\"dict<> assert failed.\");\n\t}\n#endif\n\n\tint do_hash(const K &key) const\n\t{\n\t\tunsigned int hash = 0;\n\t\tif (!hashtable.empty())\n\t\t\thash = ops.hash(key) % (unsigned int)(hashtable.size());\n\t\treturn hash;\n\t}\n\n\tvoid do_rehash()\n\t{\n\t\thashtable.clear();\n\t\thashtable.resize(hashtable_size(entries.capacity() * hashtable_size_factor), -1);\n\n\t\tfor (int i = 0; i < int(entries.size()); i++) {\n\t\t\tdo_assert(-1 <= entries[i].next && entries[i].next < int(entries.size()));\n\t\t\tint hash = do_hash(entries[i].udata.first);\n\t\t\tentries[i].next = hashtable[hash];\n\t\t\thashtable[hash] = i;\n\t\t}\n\t}\n\n\tint do_erase(int index, int hash)\n\t{\n\t\tdo_assert(index < int(entries.size()));\n\t\tif (hashtable.empty() || index < 0)\n\t\t\treturn 0;\n\n\t\tint k = hashtable[hash];\n\t\tdo_assert(0 <= k && k < int(entries.size()));\n\n\t\tif (k == index) {\n\t\t\thashtable[hash] = entries[index].next;\n\t\t} else {\n\t\t\twhile (entries[k].next != index) {\n\t\t\t\tk = entries[k].next;\n\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t}\n\t\t\tentries[k].next = entries[index].next;\n\t\t}\n\n\t\tint back_idx = entries.size()-1;\n\n\t\tif (index != back_idx)\n\t\t{\n\t\t\tint back_hash = do_hash(entries[back_idx].udata.first);\n\n\t\t\tk = hashtable[back_hash];\n\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\n\t\t\tif (k == back_idx) {\n\t\t\t\thashtable[back_hash] = index;\n\t\t\t} else {\n\t\t\t\twhile (entries[k].next != back_idx) {\n\t\t\t\t\tk = entries[k].next;\n\t\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t\t}\n\t\t\t\tentries[k].next = index;\n\t\t\t}\n\n\t\t\tentries[index] = std::move(entries[back_idx]);\n\t\t}\n\n\t\tentries.pop_back();\n\n\t\tif (entries.empty())\n\t\t\thashtable.clear();\n\n\t\treturn 1;\n\t}\n\n\tint do_lookup(const K &key, int &hash) const\n\t{\n\t\tif (hashtable.empty())\n\t\t\treturn -1;\n\n\t\tif (entries.size() * hashtable_size_trigger > hashtable.size()) {\n\t\t\t((dict*)this)->do_rehash();\n\t\t\thash = do_hash(key);\n\t\t}\n\n\t\tint index = hashtable[hash];\n\n\t\twhile (index >= 0 && !ops.cmp(entries[index].udata.first, key)) {\n\t\t\tindex = entries[index].next;\n\t\t\tdo_assert(-1 <= index && index < int(entries.size()));\n\t\t}\n\n\t\treturn index;\n\t}\n\n\tint do_insert(const K &key, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(std::pair<K, T>(key, T()), -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(key);\n\t\t} else {\n\t\t\tentries.emplace_back(std::pair<K, T>(key, T()), hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\n\tint do_insert(const std::pair<K, T> &value, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(value, -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(value.first);\n\t\t} else {\n\t\t\tentries.emplace_back(value, hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\n\tint do_insert(std::pair<K, T> &&rvalue, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tauto key = rvalue.first;\n\t\t\tentries.emplace_back(std::forward<std::pair<K, T>>(rvalue), -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(key);\n\t\t} else {\n\t\t\tentries.emplace_back(std::forward<std::pair<K, T>>(rvalue), hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\npublic:\n\tclass const_iterator\n\t{\n\t\tfriend class dict;\n\tprotected:\n\t\tconst dict *ptr;\n\t\tint index;\n\t\tconst_iterator(const dict *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef std::pair<K, T> value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef std::pair<K, T>* pointer;\n\t\ttypedef std::pair<K, T>& reference;\n\t\tconst_iterator() { }\n\t\tconst_iterator operator++() { index--; return *this; }\n\t\tconst_iterator operator+=(int amt) { index -= amt; return *this; }\n\t\tbool operator<(const const_iterator &other) const { return index > other.index; }\n\t\tbool operator==(const const_iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const const_iterator &other) const { return index != other.index; }\n\t\tconst std::pair<K, T> &operator*() const { return ptr->entries[index].udata; }\n\t\tconst std::pair<K, T> *operator->() const { return &ptr->entries[index].udata; }\n\t};\n\n\tclass iterator\n\t{\n\t\tfriend class dict;\n\tprotected:\n\t\tdict *ptr;\n\t\tint index;\n\t\titerator(dict *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef std::pair<K, T> value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef std::pair<K, T>* pointer;\n\t\ttypedef std::pair<K, T>& reference;\n\t\titerator() { }\n\t\titerator operator++() { index--; return *this; }\n\t\titerator operator+=(int amt) { index -= amt; return *this; }\n\t\tbool operator<(const iterator &other) const { return index > other.index; }\n\t\tbool operator==(const iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const iterator &other) const { return index != other.index; }\n\t\tstd::pair<K, T> &operator*() { return ptr->entries[index].udata; }\n\t\tstd::pair<K, T> *operator->() { return &ptr->entries[index].udata; }\n\t\tconst std::pair<K, T> &operator*() const { return ptr->entries[index].udata; }\n\t\tconst std::pair<K, T> *operator->() const { return &ptr->entries[index].udata; }\n\t\toperator const_iterator() const { return const_iterator(ptr, index); }\n\t};\n\n\tconstexpr dict()\n\t{\n\t}\n\n\tdict(const dict &other)\n\t{\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t}\n\n\tdict(dict &&other)\n\t{\n\t\tswap(other);\n\t}\n\n\tdict &operator=(const dict &other) {\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t\treturn *this;\n\t}\n\n\tdict &operator=(dict &&other) {\n\t\tclear();\n\t\tswap(other);\n\t\treturn *this;\n\t}\n\n\tdict(const std::initializer_list<std::pair<K, T>> &list)\n\t{\n\t\tfor (auto &it : list)\n\t\t\tinsert(it);\n\t}\n\n\ttemplate<class InputIterator>\n\tdict(InputIterator first, InputIterator last)\n\t{\n\t\tinsert(first, last);\n\t}\n\n\ttemplate<class InputIterator>\n\tvoid insert(InputIterator first, InputIterator last)\n\t{\n\t\tfor (; first != last; ++first)\n\t\t\tinsert(*first);\n\t}\n\n\tstd::pair<iterator, bool> insert(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(key, hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> insert(const std::pair<K, T> &value)\n\t{\n\t\tint hash = do_hash(value.first);\n\t\tint i = do_lookup(value.first, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(value, hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> insert(std::pair<K, T> &&rvalue)\n\t{\n\t\tint hash = do_hash(rvalue.first);\n\t\tint i = do_lookup(rvalue.first, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::forward<std::pair<K, T>>(rvalue), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K const &key, T const &value)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(key, value), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K const &key, T &&rvalue)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(key, std::forward<T>(rvalue)), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K &&rkey, T const &value)\n\t{\n\t\tint hash = do_hash(rkey);\n\t\tint i = do_lookup(rkey, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(std::forward<K>(rkey), value), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> emplace(K &&rkey, T &&rvalue)\n\t{\n\t\tint hash = do_hash(rkey);\n\t\tint i = do_lookup(rkey, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::make_pair(std::forward<K>(rkey), std::forward<T>(rvalue)), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tint erase(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint index = do_lookup(key, hash);\n\t\treturn do_erase(index, hash);\n\t}\n\n\titerator erase(iterator it)\n\t{\n\t\tint hash = do_hash(it->first);\n\t\tdo_erase(it.index, hash);\n\t\treturn ++it;\n\t}\n\n\tint count(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 ? 0 : 1;\n\t}\n\n\tint count(const K &key, const_iterator it) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 || i > it.index ? 0 : 1;\n\t}\n\n\titerator find(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn iterator(this, i);\n\t}\n\n\tconst_iterator find(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn const_iterator(this, i);\n\t}\n\n\tT& at(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\tthrow std::out_of_range(\"dict::at()\");\n\t\treturn entries[i].udata.second;\n\t}\n\n\tconst T& at(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\tthrow std::out_of_range(\"dict::at()\");\n\t\treturn entries[i].udata.second;\n\t}\n\n\tconst T& at(const K &key, const T &defval) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn defval;\n\t\treturn entries[i].udata.second;\n\t}\n\n\tT& operator[](const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\ti = do_insert(std::pair<K, T>(key, T()), hash);\n\t\treturn entries[i].udata.second;\n\t}\n\n\ttemplate<typename Compare = std::less<K>>\n\tvoid sort(Compare comp = Compare())\n\t{\n\t\tstd::sort(entries.begin(), entries.end(), [comp](const entry_t &a, const entry_t &b){ return comp(b.udata.first, a.udata.first); });\n\t\tdo_rehash();\n\t}\n\n\tvoid swap(dict &other)\n\t{\n\t\thashtable.swap(other.hashtable);\n\t\tentries.swap(other.entries);\n\t}\n\n\tbool operator==(const dict &other) const {\n\t\tif (size() != other.size())\n\t\t\treturn false;\n\t\tfor (auto &it : entries) {\n\t\t\tauto oit = other.find(it.udata.first);\n\t\t\tif (oit == other.end() || !(oit->second == it.udata.second))\n\t\t\t\treturn false;\n\t\t}\n\t\treturn true;\n\t}\n\n\tbool operator!=(const dict &other) const {\n\t\treturn !operator==(other);\n\t}\n\n\tunsigned int hash() const {\n\t\tunsigned int h = mkhash_init;\n\t\tfor (auto &entry : entries) {\n\t\t\th ^= hash_ops<K>::hash(entry.udata.first);\n\t\t\th ^= hash_ops<T>::hash(entry.udata.second);\n\t\t}\n\t\treturn h;\n\t}\n\n\tvoid reserve(size_t n) { entries.reserve(n); }\n\tsize_t size() const { return entries.size(); }\n\tbool empty() const { return entries.empty(); }\n\tvoid clear() { hashtable.clear(); entries.clear(); }\n\n\titerator begin() { return iterator(this, int(entries.size())-1); }\n\titerator element(int n) { return iterator(this, int(entries.size())-1-n); }\n\titerator end() { return iterator(nullptr, -1); }\n\n\tconst_iterator begin() const { return const_iterator(this, int(entries.size())-1); }\n\tconst_iterator element(int n) const { return const_iterator(this, int(entries.size())-1-n); }\n\tconst_iterator end() const { return const_iterator(nullptr, -1); }\n};\n\ntemplate<typename K, typename OPS>\nclass pool\n{\n\ttemplate<typename, int, typename> friend class idict;\n\nprotected:\n\tstruct entry_t\n\t{\n\t\tK udata;\n\t\tint next;\n\n\t\tentry_t() { }\n\t\tentry_t(const K &udata, int next) : udata(udata), next(next) { }\n\t\tentry_t(K &&udata, int next) : udata(std::move(udata)), next(next) { }\n\t};\n\n\tstd::vector<int> hashtable;\n\tstd::vector<entry_t> entries;\n\tOPS ops;\n\n#ifdef NDEBUG\n\tstatic inline void do_assert(bool) { }\n#else\n\tstatic inline void do_assert(bool cond) {\n\t\tif (!cond) throw std::runtime_error(\"pool<> assert failed.\");\n\t}\n#endif\n\n\tint do_hash(const K &key) const\n\t{\n\t\tunsigned int hash = 0;\n\t\tif (!hashtable.empty())\n\t\t\thash = ops.hash(key) % (unsigned int)(hashtable.size());\n\t\treturn hash;\n\t}\n\n\tvoid do_rehash()\n\t{\n\t\thashtable.clear();\n\t\thashtable.resize(hashtable_size(entries.capacity() * hashtable_size_factor), -1);\n\n\t\tfor (int i = 0; i < int(entries.size()); i++) {\n\t\t\tdo_assert(-1 <= entries[i].next && entries[i].next < int(entries.size()));\n\t\t\tint hash = do_hash(entries[i].udata);\n\t\t\tentries[i].next = hashtable[hash];\n\t\t\thashtable[hash] = i;\n\t\t}\n\t}\n\n\tint do_erase(int index, int hash)\n\t{\n\t\tdo_assert(index < int(entries.size()));\n\t\tif (hashtable.empty() || index < 0)\n\t\t\treturn 0;\n\n\t\tint k = hashtable[hash];\n\t\tif (k == index) {\n\t\t\thashtable[hash] = entries[index].next;\n\t\t} else {\n\t\t\twhile (entries[k].next != index) {\n\t\t\t\tk = entries[k].next;\n\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t}\n\t\t\tentries[k].next = entries[index].next;\n\t\t}\n\n\t\tint back_idx = entries.size()-1;\n\n\t\tif (index != back_idx)\n\t\t{\n\t\t\tint back_hash = do_hash(entries[back_idx].udata);\n\n\t\t\tk = hashtable[back_hash];\n\t\t\tif (k == back_idx) {\n\t\t\t\thashtable[back_hash] = index;\n\t\t\t} else {\n\t\t\t\twhile (entries[k].next != back_idx) {\n\t\t\t\t\tk = entries[k].next;\n\t\t\t\t\tdo_assert(0 <= k && k < int(entries.size()));\n\t\t\t\t}\n\t\t\t\tentries[k].next = index;\n\t\t\t}\n\n\t\t\tentries[index] = std::move(entries[back_idx]);\n\t\t}\n\n\t\tentries.pop_back();\n\n\t\tif (entries.empty())\n\t\t\thashtable.clear();\n\n\t\treturn 1;\n\t}\n\n\tint do_lookup(const K &key, int &hash) const\n\t{\n\t\tif (hashtable.empty())\n\t\t\treturn -1;\n\n\t\tif (entries.size() * hashtable_size_trigger > hashtable.size()) {\n\t\t\t((pool*)this)->do_rehash();\n\t\t\thash = do_hash(key);\n\t\t}\n\n\t\tint index = hashtable[hash];\n\n\t\twhile (index >= 0 && !ops.cmp(entries[index].udata, key)) {\n\t\t\tindex = entries[index].next;\n\t\t\tdo_assert(-1 <= index && index < int(entries.size()));\n\t\t}\n\n\t\treturn index;\n\t}\n\n\tint do_insert(const K &value, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(value, -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(value);\n\t\t} else {\n\t\t\tentries.emplace_back(value, hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\n\tint do_insert(K &&rvalue, int &hash)\n\t{\n\t\tif (hashtable.empty()) {\n\t\t\tentries.emplace_back(std::forward<K>(rvalue), -1);\n\t\t\tdo_rehash();\n\t\t\thash = do_hash(rvalue);\n\t\t} else {\n\t\t\tentries.emplace_back(std::forward<K>(rvalue), hashtable[hash]);\n\t\t\thashtable[hash] = entries.size() - 1;\n\t\t}\n\t\treturn entries.size() - 1;\n\t}\n\npublic:\n\tclass const_iterator\n\t{\n\t\tfriend class pool;\n\tprotected:\n\t\tconst pool *ptr;\n\t\tint index;\n\t\tconst_iterator(const pool *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef K value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef K* pointer;\n\t\ttypedef K& reference;\n\t\tconst_iterator() { }\n\t\tconst_iterator operator++() { index--; return *this; }\n\t\tbool operator==(const const_iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const const_iterator &other) const { return index != other.index; }\n\t\tconst K &operator*() const { return ptr->entries[index].udata; }\n\t\tconst K *operator->() const { return &ptr->entries[index].udata; }\n\t};\n\n\tclass iterator\n\t{\n\t\tfriend class pool;\n\tprotected:\n\t\tpool *ptr;\n\t\tint index;\n\t\titerator(pool *ptr, int index) : ptr(ptr), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef K value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef K* pointer;\n\t\ttypedef K& reference;\n\t\titerator() { }\n\t\titerator operator++() { index--; return *this; }\n\t\tbool operator==(const iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const iterator &other) const { return index != other.index; }\n\t\tK &operator*() { return ptr->entries[index].udata; }\n\t\tK *operator->() { return &ptr->entries[index].udata; }\n\t\tconst K &operator*() const { return ptr->entries[index].udata; }\n\t\tconst K *operator->() const { return &ptr->entries[index].udata; }\n\t\toperator const_iterator() const { return const_iterator(ptr, index); }\n\t};\n\n\tconstexpr pool()\n\t{\n\t}\n\n\tpool(const pool &other)\n\t{\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t}\n\n\tpool(pool &&other)\n\t{\n\t\tswap(other);\n\t}\n\n\tpool &operator=(const pool &other) {\n\t\tentries = other.entries;\n\t\tdo_rehash();\n\t\treturn *this;\n\t}\n\n\tpool &operator=(pool &&other) {\n\t\tclear();\n\t\tswap(other);\n\t\treturn *this;\n\t}\n\n\tpool(const std::initializer_list<K> &list)\n\t{\n\t\tfor (auto &it : list)\n\t\t\tinsert(it);\n\t}\n\n\ttemplate<class InputIterator>\n\tpool(InputIterator first, InputIterator last)\n\t{\n\t\tinsert(first, last);\n\t}\n\n\ttemplate<class InputIterator>\n\tvoid insert(InputIterator first, InputIterator last)\n\t{\n\t\tfor (; first != last; ++first)\n\t\t\tinsert(*first);\n\t}\n\n\tstd::pair<iterator, bool> insert(const K &value)\n\t{\n\t\tint hash = do_hash(value);\n\t\tint i = do_lookup(value, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(value, hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\tstd::pair<iterator, bool> insert(K &&rvalue)\n\t{\n\t\tint hash = do_hash(rvalue);\n\t\tint i = do_lookup(rvalue, hash);\n\t\tif (i >= 0)\n\t\t\treturn std::pair<iterator, bool>(iterator(this, i), false);\n\t\ti = do_insert(std::forward<K>(rvalue), hash);\n\t\treturn std::pair<iterator, bool>(iterator(this, i), true);\n\t}\n\n\ttemplate<typename... Args>\n\tstd::pair<iterator, bool> emplace(Args&&... args)\n\t{\n\t\treturn insert(K(std::forward<Args>(args)...));\n\t}\n\n\tint erase(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint index = do_lookup(key, hash);\n\t\treturn do_erase(index, hash);\n\t}\n\n\titerator erase(iterator it)\n\t{\n\t\tint hash = do_hash(*it);\n\t\tdo_erase(it.index, hash);\n\t\treturn ++it;\n\t}\n\n\tint count(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 ? 0 : 1;\n\t}\n\n\tint count(const K &key, const_iterator it) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i < 0 || i > it.index ? 0 : 1;\n\t}\n\n\titerator find(const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn iterator(this, i);\n\t}\n\n\tconst_iterator find(const K &key) const\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn end();\n\t\treturn const_iterator(this, i);\n\t}\n\n\tbool operator[](const K &key)\n\t{\n\t\tint hash = do_hash(key);\n\t\tint i = do_lookup(key, hash);\n\t\treturn i >= 0;\n\t}\n\n\ttemplate<typename Compare = std::less<K>>\n\tvoid sort(Compare comp = Compare())\n\t{\n\t\tstd::sort(entries.begin(), entries.end(), [comp](const entry_t &a, const entry_t &b){ return comp(b.udata, a.udata); });\n\t\tdo_rehash();\n\t}\n\n\tK pop()\n\t{\n\t\titerator it = begin();\n\t\tK ret = *it;\n\t\terase(it);\n\t\treturn ret;\n\t}\n\n\tvoid swap(pool &other)\n\t{\n\t\thashtable.swap(other.hashtable);\n\t\tentries.swap(other.entries);\n\t}\n\n\tbool operator==(const pool &other) const {\n\t\tif (size() != other.size())\n\t\t\treturn false;\n\t\tfor (auto &it : entries)\n\t\t\tif (!other.count(it.udata))\n\t\t\t\treturn false;\n\t\treturn true;\n\t}\n\n\tbool operator!=(const pool &other) const {\n\t\treturn !operator==(other);\n\t}\n\n\tunsigned int hash() const {\n\t\tunsigned int hashval = mkhash_init;\n\t\tfor (auto &it : entries)\n\t\t\thashval ^= ops.hash(it.udata);\n\t\treturn hashval;\n\t}\n\n\tvoid reserve(size_t n) { entries.reserve(n); }\n\tsize_t size() const { return entries.size(); }\n\tbool empty() const { return entries.empty(); }\n\tvoid clear() { hashtable.clear(); entries.clear(); }\n\n\titerator begin() { return iterator(this, int(entries.size())-1); }\n\titerator element(int n) { return iterator(this, int(entries.size())-1-n); }\n\titerator end() { return iterator(nullptr, -1); }\n\n\tconst_iterator begin() const { return const_iterator(this, int(entries.size())-1); }\n\tconst_iterator element(int n) const { return const_iterator(this, int(entries.size())-1-n); }\n\tconst_iterator end() const { return const_iterator(nullptr, -1); }\n};\n\ntemplate<typename K, int offset, typename OPS>\nclass idict\n{\n\tpool<K, OPS> database;\n\npublic:\n\tclass const_iterator\n\t{\n\t\tfriend class idict;\n\tprotected:\n\t\tconst idict &container;\n\t\tint index;\n\t\tconst_iterator(const idict &container, int index) : container(container), index(index) { }\n\tpublic:\n\t\ttypedef std::forward_iterator_tag iterator_category;\n\t\ttypedef K value_type;\n\t\ttypedef ptrdiff_t difference_type;\n\t\ttypedef K* pointer;\n\t\ttypedef K& reference;\n\t\tconst_iterator() { }\n\t\tconst_iterator operator++() { index++; return *this; }\n\t\tbool operator==(const const_iterator &other) const { return index == other.index; }\n\t\tbool operator!=(const const_iterator &other) const { return index != other.index; }\n\t\tconst K &operator*() const { return container[index]; }\n\t\tconst K *operator->() const { return &container[index]; }\n\t};\n\n\tconstexpr idict()\n\t{\n\t}\n\n\tint operator()(const K &key)\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\ti = database.do_insert(key, hash);\n\t\treturn i + offset;\n\t}\n\n\tint at(const K &key) const\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\tthrow std::out_of_range(\"idict::at()\");\n\t\treturn i + offset;\n\t}\n\n\tint at(const K &key, int defval) const\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\tif (i < 0)\n\t\t\treturn defval;\n\t\treturn i + offset;\n\t}\n\n\tint count(const K &key) const\n\t{\n\t\tint hash = database.do_hash(key);\n\t\tint i = database.do_lookup(key, hash);\n\t\treturn i < 0 ? 0 : 1;\n\t}\n\n\tvoid expect(const K &key, int i)\n\t{\n\t\tint j = (*this)(key);\n\t\tif (i != j)\n\t\t\tthrow std::out_of_range(\"idict::expect()\");\n\t}\n\n\tconst K &operator[](int index) const\n\t{\n\t\treturn database.entries.at(index - offset).udata;\n\t}\n\n\tvoid swap(idict &other)\n\t{\n\t\tdatabase.swap(other.database);\n\t}\n\n\tvoid reserve(size_t n) { database.reserve(n); }\n\tsize_t size() const { return database.size(); }\n\tbool empty() const { return database.empty(); }\n\tvoid clear() { database.clear(); }\n\n\tconst_iterator begin() const { return const_iterator(*this, offset); }\n\tconst_iterator element(int n) const { return const_iterator(*this, n); }\n\tconst_iterator end() const { return const_iterator(*this, offset + size()); }\n};\n\ntemplate<typename K, typename OPS>\nclass mfp\n{\n\tmutable idict<K, 0, OPS> database;\n\tmutable std::vector<int> parents;\n\npublic:\n\ttypedef typename idict<K, 0, OPS>::const_iterator const_iterator;\n\n\tconstexpr mfp()\n\t{\n\t}\n\n\tint operator()(const K &key) const\n\t{\n\t\tint i = database(key);\n\t\tparents.resize(database.size(), -1);\n\t\treturn i;\n\t}\n\n\tconst K &operator[](int index) const\n\t{\n\t\treturn database[index];\n\t}\n\n\tint ifind(int i) const\n\t{\n\t\tint p = i, k = i;\n\n\t\twhile (parents[p] != -1)\n\t\t\tp = parents[p];\n\n\t\twhile (k != p) {\n\t\t\tint next_k = parents[k];\n\t\t\tparents[k] = p;\n\t\t\tk = next_k;\n\t\t}\n\n\t\treturn p;\n\t}\n\n\tvoid imerge(int i, int j)\n\t{\n\t\ti = ifind(i);\n\t\tj = ifind(j);\n\n\t\tif (i != j)\n\t\t\tparents[i] = j;\n\t}\n\n\tvoid ipromote(int i)\n\t{\n\t\tint k = i;\n\n\t\twhile (k != -1) {\n\t\t\tint next_k = parents[k];\n\t\t\tparents[k] = i;\n\t\t\tk = next_k;\n\t\t}\n\n\t\tparents[i] = -1;\n\t}\n\n\tint lookup(const K &a) const\n\t{\n\t\treturn ifind((*this)(a));\n\t}\n\n\tconst K &find(const K &a) const\n\t{\n\t\tint i = database.at(a, -1);\n\t\tif (i < 0)\n\t\t\treturn a;\n\t\treturn (*this)[ifind(i)];\n\t}\n\n\tvoid merge(const K &a, const K &b)\n\t{\n\t\timerge((*this)(a), (*this)(b));\n\t}\n\n\tvoid promote(const K &a)\n\t{\n\t\tint i = database.at(a, -1);\n\t\tif (i >= 0)\n\t\t\tipromote(i);\n\t}\n\n\tvoid swap(mfp &other)\n\t{\n\t\tdatabase.swap(other.database);\n\t\tparents.swap(other.parents);\n\t}\n\n\tvoid reserve(size_t n) { database.reserve(n); }\n\tsize_t size() const { return database.size(); }\n\tbool empty() const { return database.empty(); }\n\tvoid clear() { database.clear(); parents.clear(); }\n\n\tconst_iterator begin() const { return database.begin(); }\n\tconst_iterator element(int n) const { return database.element(n); }\n\tconst_iterator end() const { return database.end(); }\n};\n\n} /* namespace hashlib */\n\n#endif\n",
                 "json.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2022  Jannis Harder <jix@yosyshq.com> <me@jix.one>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted, provided that the above\n *  copyright notice and this permission notice appear in all copies.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#ifndef JSON_H\n#define JSON_H\n\n#include \"kernel/yosys.h\"\n#include \"libs/json11/json11.hpp\"\n#include <functional>\n\nYOSYS_NAMESPACE_BEGIN\n\nusing json11::Json;\n\nclass PrettyJson\n{\n    enum Scope {\n        VALUE,\n        OBJECT_FIRST,\n        OBJECT,\n        ARRAY_FIRST,\n        ARRAY,\n    };\n\n    struct Target {\n        virtual void emit(const char *data) = 0;\n        virtual void flush() {};\n        virtual ~Target() {};\n    };\n\n    std::string newline_indent = \"\\n\";\n    std::vector<std::unique_ptr<Target>> targets;\n    std::vector<Scope> state = {VALUE};\n    int compact_depth = INT_MAX;\npublic:\n\n    void emit_to_log();\n    void append_to_string(std::string &target);\n    bool write_to_file(const std::string &path);\n\n    bool active() { return !targets.empty(); }\n\n    void compact() { compact_depth = GetSize(state); }\n\n    void line(bool space_if_inline = true);\n    void raw(const char *raw_json);\n    void flush();\n    void begin_object();\n    void begin_array();\n    void end_object();\n    void end_array();\n    void name(const char *name);\n    void begin_value();\n    void end_value();\n    void value_json(const Json &value);\n    void value(unsigned int value) { value_json(Json((int)value)); }\n    template<typename T>\n    void value(T &&value) { value_json(Json(std::forward<T>(value))); };\n\n    void entry_json(const char *name, const Json &value);\n    void entry(const char *name, unsigned int value) { entry_json(name, Json((int)value)); }\n    template<typename T>\n    void entry(const char *name, T &&value) { entry_json(name, Json(std::forward<T>(value))); };\n\n    template<typename T>\n    void object(const T &&values)\n    {\n        begin_object();\n        for (auto &item : values)\n            entry(item.first, item.second);\n        end_object();\n    }\n\n    template<typename T>\n    void array(const T &&values)\n    {\n        begin_object();\n        for (auto &item : values)\n            value(item);\n        end_object();\n    }\n};\n\n\n\nYOSYS_NAMESPACE_END\n\n#endif\n",
                 "log.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2012  Claire Xenia Wolf <claire@yosyshq.com>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted, provided that the above\n *  copyright notice and this permission notice appear in all copies.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#include \"kernel/yosys.h\"\n\n#ifndef LOG_H\n#define LOG_H\n\n#include <time.h>\n\n#include <regex>\n#define YS_REGEX_COMPILE(param) std::regex(param, \\\n\t\t\t\tstd::regex_constants::nosubs | \\\n\t\t\t\tstd::regex_constants::optimize | \\\n\t\t\t\tstd::regex_constants::egrep)\n#define YS_REGEX_COMPILE_WITH_SUBS(param) std::regex(param, \\\n\t\t\t\tstd::regex_constants::optimize | \\\n\t\t\t\tstd::regex_constants::egrep)\n\n#if defined(_WIN32)\n#  include <intrin.h>\n#else\n#  include <sys/time.h>\n#  include <sys/resource.h>\n#  if defined(__unix__) || (defined(__APPLE__) && defined(__MACH__))\n#    include <signal.h>\n#  endif\n#endif\n\n#if defined(_MSC_VER)\n// At least this is not in MSVC++ 2013.\n#  define __PRETTY_FUNCTION__ __FUNCTION__\n#endif\n\n// from libs/sha1/sha1.h\nclass SHA1;\n\nYOSYS_NAMESPACE_BEGIN\n\n#define S__LINE__sub2(x) #x\n#define S__LINE__sub1(x) S__LINE__sub2(x)\n#define S__LINE__ S__LINE__sub1(__LINE__)\n\n// YS_DEBUGTRAP is a macro that is functionally equivalent to a breakpoint\n// if the platform provides such functionality, and does nothing otherwise.\n// If no debugger is attached, it starts a just-in-time debugger if available,\n// and crashes the process otherwise.\n#if defined(_WIN32)\n# define YS_DEBUGTRAP __debugbreak()\n#else\n# ifndef __has_builtin\n// __has_builtin is a GCC/Clang extension; on a different compiler (or old enough GCC/Clang)\n// that does not have it, using __has_builtin(...) is a syntax error.\n#  define __has_builtin(x) 0\n# endif\n# if __has_builtin(__builtin_debugtrap)\n#  define YS_DEBUGTRAP __builtin_debugtrap()\n# elif defined(__unix__) || (defined(__APPLE__) && defined(__MACH__))\n#  define YS_DEBUGTRAP raise(SIGTRAP)\n# else\n#  define YS_DEBUGTRAP do {} while(0)\n# endif\n#endif\n\n// YS_DEBUGTRAP_IF_DEBUGGING is a macro that is functionally equivalent to a breakpoint\n// if a debugger is attached, and does nothing otherwise.\n#if defined(_WIN32)\n# define YS_DEBUGTRAP_IF_DEBUGGING do { if (IsDebuggerPresent()) DebugBreak(); } while(0)\n# elif defined(__unix__) || (defined(__APPLE__) && defined(__MACH__))\n// There is no reliable (or portable) *nix equivalent of IsDebuggerPresent(). However,\n// debuggers will stop when SIGTRAP is raised, even if the action is set to ignore.\n# define YS_DEBUGTRAP_IF_DEBUGGING do { \\\n\t\tauto old = signal(SIGTRAP, SIG_IGN); raise(SIGTRAP); signal(SIGTRAP, old); \\\n\t} while(0)\n#else\n# define YS_DEBUGTRAP_IF_DEBUGGING do {} while(0)\n#endif\n\nstruct log_cmd_error_exception { };\n\nextern std::vector<FILE*> log_files;\nextern std::vector<std::ostream*> log_streams;\nextern std::vector<std::string> log_scratchpads;\nextern std::map<std::string, std::set<std::string>> log_hdump;\nextern std::vector<std::regex> log_warn_regexes, log_nowarn_regexes, log_werror_regexes;\nextern std::set<std::string> log_warnings, log_experimentals, log_experimentals_ignored;\nextern int log_warnings_count;\nextern int log_warnings_count_noexpect;\nextern bool log_expect_no_warnings;\nextern bool log_hdump_all;\nextern FILE *log_errfile;\nextern SHA1 *log_hasher;\n\nextern bool log_time;\nextern bool log_error_stderr;\nextern bool log_cmd_error_throw;\nextern bool log_quiet_warnings;\nextern int log_verbose_level;\nextern string log_last_error;\nextern void (*log_error_atexit)();\n\nextern int log_make_debug;\nextern int log_force_debug;\nextern int log_debug_suppressed;\n\nvoid logv(const char *format, va_list ap);\nvoid logv_header(RTLIL::Design *design, const char *format, va_list ap);\nvoid logv_warning(const char *format, va_list ap);\nvoid logv_warning_noprefix(const char *format, va_list ap);\n[[noreturn]] void logv_error(const char *format, va_list ap);\n[[noreturn]] void logv_file_error(const string &filename, int lineno, const char *format, va_list ap);\n\nvoid log(const char *format, ...)  YS_ATTRIBUTE(format(printf, 1, 2));\nvoid log_header(RTLIL::Design *design, const char *format, ...) YS_ATTRIBUTE(format(printf, 2, 3));\nvoid log_warning(const char *format, ...) YS_ATTRIBUTE(format(printf, 1, 2));\nvoid log_experimental(const char *format, ...) YS_ATTRIBUTE(format(printf, 1, 2));\n\nvoid set_verific_logging(void (*cb)(int msg_type, const char *message_id, const char* file_path, unsigned int left_line, unsigned int left_col, unsigned int right_line, unsigned int right_col, const char *msg));\nextern void (*log_verific_callback)(int msg_type, const char *message_id, const char* file_path, unsigned int left_line, unsigned int left_col, unsigned int right_line, unsigned int right_col, const char *msg);\n\n// Log with filename to report a problem in a source file.\nvoid log_file_warning(const std::string &filename, int lineno, const char *format, ...) YS_ATTRIBUTE(format(printf, 3, 4));\nvoid log_file_info(const std::string &filename, int lineno, const char *format, ...) YS_ATTRIBUTE(format(printf, 3, 4));\n\nvoid log_warning_noprefix(const char *format, ...) YS_ATTRIBUTE(format(printf, 1, 2));\n[[noreturn]] void log_error(const char *format, ...) YS_ATTRIBUTE(format(printf, 1, 2));\n[[noreturn]] void log_file_error(const string &filename, int lineno, const char *format, ...) YS_ATTRIBUTE(format(printf, 3, 4));\n[[noreturn]] void log_cmd_error(const char *format, ...) YS_ATTRIBUTE(format(printf, 1, 2));\n\n#ifndef NDEBUG\nstatic inline bool ys_debug(int n = 0) { if (log_force_debug) return true; log_debug_suppressed += n; return false; }\n#else\nstatic inline bool ys_debug(int = 0) { return false; }\n#endif\n#  define log_debug(...) do { if (ys_debug(1)) log(__VA_ARGS__); } while (0)\n\nstatic inline void log_suppressed() {\n\tif (log_debug_suppressed && !log_make_debug) {\n\t\tlog(\"<suppressed ~%d debug messages>\\n\", log_debug_suppressed);\n\t\tlog_debug_suppressed = 0;\n\t}\n}\n\nstruct LogMakeDebugHdl {\n\tbool status = false;\n\tLogMakeDebugHdl(bool start_on = false) {\n\t\tif (start_on)\n\t\t\ton();\n\t}\n\t~LogMakeDebugHdl() {\n\t\toff();\n\t}\n\tvoid on() {\n\t\tif (status) return;\n\t\tstatus=true;\n\t\tlog_make_debug++;\n\t}\n\tvoid off_silent() {\n\t\tif (!status) return;\n\t\tstatus=false;\n\t\tlog_make_debug--;\n\t}\n\tvoid off() {\n\t\toff_silent();\n\t}\n};\n\nvoid log_spacer();\nvoid log_push();\nvoid log_pop();\n\nvoid log_backtrace(const char *prefix, int levels);\nvoid log_reset_stack();\nvoid log_flush();\n\nstruct LogExpectedItem\n{\n\tLogExpectedItem(const std::regex &pat, int expected) :\n\t\t\tpattern(pat), expected_count(expected), current_count(0) {}\n\tLogExpectedItem() : expected_count(0), current_count(0) {}\n\n\tstd::regex pattern;\n\tint expected_count;\n\tint current_count;\n};\n\nextern dict<std::string, LogExpectedItem> log_expect_log, log_expect_warning, log_expect_error;\nvoid log_check_expected();\n\nconst char *log_signal(const RTLIL::SigSpec &sig, bool autoint = true);\nconst char *log_const(const RTLIL::Const &value, bool autoint = true);\nconst char *log_id(const RTLIL::IdString &id);\n\ntemplate<typename T> static inline const char *log_id(T *obj, const char *nullstr = nullptr) {\n\tif (nullstr && obj == nullptr)\n\t\treturn nullstr;\n\treturn log_id(obj->name);\n}\n\nvoid log_module(RTLIL::Module *module, std::string indent = \"\");\nvoid log_cell(RTLIL::Cell *cell, std::string indent = \"\");\nvoid log_wire(RTLIL::Wire *wire, std::string indent = \"\");\n\n#ifndef NDEBUG\nstatic inline void log_assert_worker(bool cond, const char *expr, const char *file, int line) {\n\tif (!cond) log_error(\"Assert `%s' failed in %s:%d.\\n\", expr, file, line);\n}\n#  define log_assert(_assert_expr_) YOSYS_NAMESPACE_PREFIX log_assert_worker(_assert_expr_, #_assert_expr_, __FILE__, __LINE__)\n#else\n#  define log_assert(_assert_expr_) do { if (0) { (void)(_assert_expr_); } } while(0)\n#endif\n\n#define log_abort() YOSYS_NAMESPACE_PREFIX log_error(\"Abort in %s:%d.\\n\", __FILE__, __LINE__)\n#define log_ping() YOSYS_NAMESPACE_PREFIX log(\"-- %s:%d %s --\\n\", __FILE__, __LINE__, __PRETTY_FUNCTION__)\n\n\n// ---------------------------------------------------\n// This is the magic behind the code coverage counters\n// ---------------------------------------------------\n\n#if defined(YOSYS_ENABLE_COVER) && (defined(__linux__) || defined(__FreeBSD__))\n\n#define cover(_id) do { \\\n    static CoverData __d __attribute__((section(\"yosys_cover_list\"), aligned(1), used)) = { __FILE__, __FUNCTION__, _id, __LINE__, 0 }; \\\n    __d.counter++; \\\n} while (0)\n\nstruct CoverData {\n\tconst char *file, *func, *id;\n\tint line, counter;\n} YS_ATTRIBUTE(packed);\n\n// this two symbols are created by the linker for the \"yosys_cover_list\" ELF section\nextern \"C\" struct CoverData __start_yosys_cover_list[];\nextern \"C\" struct CoverData __stop_yosys_cover_list[];\n\nextern dict<std::string, std::pair<std::string, int>> extra_coverage_data;\n\nvoid cover_extra(std::string parent, std::string id, bool increment = true);\ndict<std::string, std::pair<std::string, int>> get_coverage_data();\n\n#define cover_list(_id, ...) do { cover(_id); \\\n\tstd::string r = cover_list_worker(_id, __VA_ARGS__); \\\n\tlog_assert(r.empty()); \\\n} while (0)\n\nstatic inline std::string cover_list_worker(std::string, std::string last) {\n\treturn last;\n}\n\ntemplate<typename... T>\nstd::string cover_list_worker(std::string prefix, std::string first, T... rest) {\n\tstd::string selected = cover_list_worker(prefix, rest...);\n\tcover_extra(prefix, prefix + \".\" + first, first == selected);\n\treturn first == selected ? \"\" : selected;\n}\n\n#else\n#  define cover(...) do { } while (0)\n#  define cover_list(...) do { } while (0)\n#endif\n\n\n// ------------------------------------------------------------\n// everything below this line are utilities for troubleshooting\n// ------------------------------------------------------------\n\n// simple timer for performance measurements\n// toggle the '#if 1' to get a baseline for the performance penalty added by the measurement\nstruct PerformanceTimer\n{\n#if 1\n\tint64_t total_ns;\n\n\tPerformanceTimer() {\n\t\ttotal_ns = 0;\n\t}\n\n\tstatic int64_t query() {\n#  ifdef _WIN32\n\t\treturn 0;\n#  elif defined(RUSAGE_SELF)\n\t\tstruct rusage rusage;\n\t\tint64_t t = 0;\n\t\tfor (int who : {RUSAGE_SELF, RUSAGE_CHILDREN}) {\n\t\t\tif (getrusage(who, &rusage) == -1) {\n\t\t\t\tlog_cmd_error(\"getrusage failed!\\n\");\n\t\t\t\tlog_abort();\n\t\t\t}\n\t\t\tt += 1000000000ULL * (int64_t) rusage.ru_utime.tv_sec + (int64_t) rusage.ru_utime.tv_usec * 1000ULL;\n\t\t\tt += 1000000000ULL * (int64_t) rusage.ru_stime.tv_sec + (int64_t) rusage.ru_stime.tv_usec * 1000ULL;\n\t\t}\n\t\treturn t;\n#  else\n#    error \"Don't know how to measure per-process CPU time. Need alternative method (times()/clocks()/gettimeofday()?).\"\n#  endif\n\t}\n\n\tvoid reset() {\n\t\ttotal_ns = 0;\n\t}\n\n\tvoid begin() {\n\t\ttotal_ns -= query();\n\t}\n\n\tvoid end() {\n\t\ttotal_ns += query();\n\t}\n\n\tfloat sec() const {\n\t\treturn total_ns * 1e-9f;\n\t}\n#else\n\tstatic int64_t query() { return 0; }\n\tvoid reset() { }\n\tvoid begin() { }\n\tvoid end() { }\n\tfloat sec() const { return 0; }\n#endif\n};\n\n// simple API for quickly dumping values when debugging\n\nstatic inline void log_dump_val_worker(short v) { log(\"%d\", v); }\nstatic inline void log_dump_val_worker(unsigned short v) { log(\"%u\", v); }\nstatic inline void log_dump_val_worker(int v) { log(\"%d\", v); }\nstatic inline void log_dump_val_worker(unsigned int v) { log(\"%u\", v); }\nstatic inline void log_dump_val_worker(long int v) { log(\"%ld\", v); }\nstatic inline void log_dump_val_worker(unsigned long int v) { log(\"%lu\", v); }\n#ifndef _WIN32\nstatic inline void log_dump_val_worker(long long int v) { log(\"%lld\", v); }\nstatic inline void log_dump_val_worker(unsigned long long int v) { log(\"%lld\", v); }\n#endif\nstatic inline void log_dump_val_worker(char c) { log(c >= 32 && c < 127 ? \"'%c'\" : \"'\\\\x%02x'\", c); }\nstatic inline void log_dump_val_worker(unsigned char c) { log(c >= 32 && c < 127 ? \"'%c'\" : \"'\\\\x%02x'\", c); }\nstatic inline void log_dump_val_worker(bool v) { log(\"%s\", v ? \"true\" : \"false\"); }\nstatic inline void log_dump_val_worker(double v) { log(\"%f\", v); }\nstatic inline void log_dump_val_worker(char *v) { log(\"%s\", v); }\nstatic inline void log_dump_val_worker(const char *v) { log(\"%s\", v); }\nstatic inline void log_dump_val_worker(std::string v) { log(\"%s\", v.c_str()); }\nstatic inline void log_dump_val_worker(PerformanceTimer p) { log(\"%f seconds\", p.sec()); }\nstatic inline void log_dump_args_worker(const char *p) { log_assert(*p == 0); }\nvoid log_dump_val_worker(RTLIL::IdString v);\nvoid log_dump_val_worker(RTLIL::SigSpec v);\nvoid log_dump_val_worker(RTLIL::State v);\n\ntemplate<typename K, typename T, typename OPS> static inline void log_dump_val_worker(dict<K, T, OPS> &v);\ntemplate<typename K, typename OPS> static inline void log_dump_val_worker(pool<K, OPS> &v);\ntemplate<typename K> static inline void log_dump_val_worker(std::vector<K> &v);\ntemplate<typename T> static inline void log_dump_val_worker(T *ptr);\n\ntemplate<typename K, typename T, typename OPS>\nstatic inline void log_dump_val_worker(dict<K, T, OPS> &v) {\n\tlog(\"{\");\n\tbool first = true;\n\tfor (auto &it : v) {\n\t\tlog(first ? \" \" : \", \");\n\t\tlog_dump_val_worker(it.first);\n\t\tlog(\": \");\n\t\tlog_dump_val_worker(it.second);\n\t\tfirst = false;\n\t}\n\tlog(\" }\");\n}\n\ntemplate<typename K, typename OPS>\nstatic inline void log_dump_val_worker(pool<K, OPS> &v) {\n\tlog(\"{\");\n\tbool first = true;\n\tfor (auto &it : v) {\n\t\tlog(first ? \" \" : \", \");\n\t\tlog_dump_val_worker(it);\n\t\tfirst = false;\n\t}\n\tlog(\" }\");\n}\n\ntemplate<typename K>\nstatic inline void log_dump_val_worker(std::vector<K> &v) {\n\tlog(\"{\");\n\tbool first = true;\n\tfor (auto &it : v) {\n\t\tlog(first ? \" \" : \", \");\n\t\tlog_dump_val_worker(it);\n\t\tfirst = false;\n\t}\n\tlog(\" }\");\n}\n\ntemplate<typename T>\nstatic inline void log_dump_val_worker(T *ptr) { log(\"%p\", ptr); }\n\ntemplate<typename T, typename ... Args>\nvoid log_dump_args_worker(const char *p, T first, Args ... args)\n{\n\tint next_p_state = 0;\n\tconst char *next_p = p;\n\twhile (*next_p && (next_p_state != 0 || *next_p != ',')) {\n\t\tif (*next_p == '\"')\n\t\t\tdo {\n\t\t\t\tnext_p++;\n\t\t\t\twhile (*next_p == '\\\\' && *(next_p + 1))\n\t\t\t\t\tnext_p += 2;\n\t\t\t} while (*next_p && *next_p != '\"');\n\t\tif (*next_p == '\\'') {\n\t\t\tnext_p++;\n\t\t\tif (*next_p == '\\\\')\n\t\t\t\tnext_p++;\n\t\t\tif (*next_p)\n\t\t\t\tnext_p++;\n\t\t}\n\t\tif (*next_p == '(' || *next_p == '[' || *next_p == '{')\n\t\t\tnext_p_state++;\n\t\tif ((*next_p == ')' || *next_p == ']' || *next_p == '}') && next_p_state > 0)\n\t\t\tnext_p_state--;\n\t\tnext_p++;\n\t}\n\tlog(\"\\n\\t%.*s => \", int(next_p - p), p);\n\tif (*next_p == ',')\n\t\tnext_p++;\n\twhile (*next_p == ' ' || *next_p == '\\t' || *next_p == '\\r' || *next_p == '\\n')\n\t\tnext_p++;\n\tlog_dump_val_worker(first);\n\tlog_dump_args_worker(next_p, args ...);\n}\n\n#define log_dump(...) do { \\\n\tlog(\"DEBUG DUMP IN %s AT %s:%d:\", __PRETTY_FUNCTION__, __FILE__, __LINE__); \\\n\tlog_dump_args_worker(#__VA_ARGS__, __VA_ARGS__); \\\n\tlog(\"\\n\"); \\\n} while (0)\n\nYOSYS_NAMESPACE_END\n\n#endif\n",
                 "macc.h": "/*\n *  yosys -- Yosys Open SYnthesis Suite\n *\n *  Copyright (C) 2012  Claire Xenia Wolf <claire@yosyshq.com>\n *\n *  Permission to use, copy, modify, and/or distribute this software for any\n *  purpose with or without fee is hereby granted, provided that the above\n *  copyright notice and this permission notice appear in all copies.\n *\n *  THE SOFTWARE IS PROVIDED \"AS IS\" AND THE AUTHOR DISCLAIMS ALL WARRANTIES\n *  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF\n *  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR\n *  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES\n *  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN\n *  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF\n *  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.\n *\n */\n\n#ifndef MACC_H\n#define MACC_H\n\n#include \"kernel/yosys.h\"\n\nYOSYS_NAMESPACE_BEGIN\n\nstruct Macc\n{\n\tstruct port_t {\n\t\tRTLIL::SigSpec in_a, in_b;\n\t\tbool is_signed, do_subtract;\n\t};\n\n\tstd::vector<port_t> ports;\n\tRTLIL::SigSpec bit_ports;\n\n\tvoid optimize(int width)\n\t{\n\t\tstd::vector<port_t> new_ports;\n\t\tRTLIL::SigSpec new_bit_ports;\n\t\tRTLIL::Const off(0, width);\n\n\t\tfor (auto &port : ports)\n\t\t{\n\t\t\tif (GetSize(port.in_a) == 0 && GetSize(port.in_b) == 0)\n\t\t\t\tcontinue;\n\n\t\t\tif (GetSize(port.in_a) < GetSize(port.in_b))\n\t\t\t\tstd::swap(port.in_a, port.in_b);\n\n\t\t\tif (GetSize(port.in_a) == 1 && GetSize(port.in_b) == 0 && !port.is_signed && !port.do_subtract) {\n\t\t\t\tbit_ports.append(port.in_a);\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\tif (port.in_a.is_fully_const() && port.in_b.is_fully_const()) {\n\t\t\t\tRTLIL::Const v = port.in_a.as_const();\n\t\t\t\tif (GetSize(port.in_b))\n\t\t\t\t\tv = const_mul(v, port.in_b.as_const(), port.is_signed, port.is_signed, width);\n\t\t\t\tif (port.do_subtract)\n\t\t\t\t\toff = const_sub(off, v, port.is_signed, port.is_signed, width);\n\t\t\t\telse\n\t\t\t\t\toff = const_add(off, v, port.is_signed, port.is_signed, width);\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\tif (port.is_signed) {\n\t\t\t\twhile (GetSize(port.in_a) > 1 && port.in_a[GetSize(port.in_a)-1] == port.in_a[GetSize(port.in_a)-2])\n\t\t\t\t\tport.in_a.remove(GetSize(port.in_a)-1);\n\t\t\t\twhile (GetSize(port.in_b) > 1 && port.in_b[GetSize(port.in_b)-1] == port.in_b[GetSize(port.in_b)-2])\n\t\t\t\t\tport.in_b.remove(GetSize(port.in_b)-1);\n\t\t\t} else {\n\t\t\t\twhile (GetSize(port.in_a) > 1 && port.in_a[GetSize(port.in_a)-1] == State::S0)\n\t\t\t\t\tport.in_a.remove(GetSize(port.in_a)-1);\n\t\t\t\twhile (GetSize(port.in_b) > 1 && port.in_b[GetSize(port.in_b)-1] == State::S0)\n\t\t\t\t\tport.in_b.remove(GetSize(port.in_b)-1);\n\t\t\t}\n\n\t\t\tnew_ports.push_back(port);\n\t\t}\n\n\t\tfor (auto &bit : bit_ports)\n\t\t\tif (bit == State::S1)\n\t\t\t\toff = const_add(off, RTLIL::Const(1, width), false, false, width);\n\t\t\telse if (bit != State::S0)\n\t\t\t\tnew_bit_ports.append(bit);\n\n\t\tif (off.as_bool()) {\n\t\t\tport_t port;\n\t\t\tport.in_a = off;\n\t\t\tport.is_signed = false;\n\t\t\tport.do_subtract = false;\n\t\t\tnew_ports.push_back(port);\n\t\t}\n\n\t\tnew_ports.swap(ports);\n\t\tbit_ports = new_bit_ports;\n\t}\n\n\tvoid from_cell(RTLIL::Cell *cell)\n\t{\n\t\tRTLIL::SigSpec port_a = cell->getPort(ID::A);\n\n\t\tports.clear();\n\t\tbit_ports = cell->getPort(ID::B);\n\n\t\tstd::vector<RTLIL::State> config_bits = cell->getParam(ID::CONFIG).bits;\n\t\tint config_cursor = 0;\n\n\t\tint config_width = cell->getParam(ID::CONFIG_WIDTH).as_int();\n\t\tlog_assert(GetSize(config_bits) >= config_width);\n\n\t\tint num_bits = 0;\n\t\tif (config_bits[config_cursor++] == State::S1) num_bits |= 1;\n\t\tif (config_bits[config_cursor++] == State::S1) num_bits |= 2;\n\t\tif (config_bits[config_cursor++] == State::S1) num_bits |= 4;\n\t\tif (config_bits[config_cursor++] == State::S1) num_bits |= 8;\n\n\t\tint port_a_cursor = 0;\n\t\twhile (port_a_cursor < GetSize(port_a))\n\t\t{\n\t\t\tlog_assert(config_cursor + 2 + 2*num_bits <= config_width);\n\n\t\t\tport_t this_port;\n\t\t\tthis_port.is_signed = config_bits[config_cursor++] == State::S1;\n\t\t\tthis_port.do_subtract = config_bits[config_cursor++] == State::S1;\n\n\t\t\tint size_a = 0;\n\t\t\tfor (int i = 0; i < num_bits; i++)\n\t\t\t\tif (config_bits[config_cursor++] == State::S1)\n\t\t\t\t\tsize_a |= 1 << i;\n\n\t\t\tthis_port.in_a = port_a.extract(port_a_cursor, size_a);\n\t\t\tport_a_cursor += size_a;\n\n\t\t\tint size_b = 0;\n\t\t\tfor (int i = 0; i < num_bits; i++)\n\t\t\t\tif (config_bits[config_cursor++] == State::S1)\n\t\t\t\t\tsize_b |= 1 << i;\n\n\t\t\tthis_port.in_b = port_a.extract(port_a_cursor, size_b);\n\t\t\tport_a_cursor += size_b;\n\n\t\t\tif (size_a || size_b)\n\t\t\t\tports.push_back(this_port);\n\t\t}\n\n\t\tlog_assert(config_cursor == config_width);\n\t\tlog_assert(port_a_cursor == GetSize(port_a));\n\t}\n\n\tvoid to_cell(RTLIL::Cell *cell) const\n\t{\n\t\tRTLIL::SigSpec port_a;\n\t\tstd::vector<RTLIL::State> config_bits;\n\t\tint max_size = 0, num_bits = 0;\n\n\t\tfor (auto &port : ports) {\n\t\t\tmax_size = max(max_size, GetSize(port.in_a));\n\t\t\tmax_size = max(max_size, GetSize(port.in_b));\n\t\t}\n\n\t\twhile (max_size)\n\t\t\tnum_bits++, max_size /= 2;\n\n\t\tlog_assert(num_bits < 16);\n\t\tconfig_bits.push_back(num_bits & 1 ? State::S1 : State::S0);\n\t\tconfig_bits.push_back(num_bits & 2 ? State::S1 : State::S0);\n\t\tconfig_bits.push_back(num_bits & 4 ? State::S1 : State::S0);\n\t\tconfig_bits.push_back(num_bits & 8 ? State::S1 : State::S0);\n\n\t\tfor (auto &port : ports)\n\t\t{\n\t\t\tif (GetSize(port.in_a) == 0)\n\t\t\t\tcontinue;\n\n\t\t\tconfig_bits.push_back(port.is_signed ? State::S1 : State::S0);\n\t\t\tconfig_bits.push_back(port.do_subtract ? State::S1 : State::S0);\n\n\t\t\tint size_a = GetSize(port.in_a);\n\t\t\tfor (int i = 0; i < num_bits; i++)\n\t\t\t\tconfig_bits.push_back(size_a & (1 << i) ? State::S1 : State::S0);\n\n\t\t\tint size_b = GetSize(port.in_b);\n\t\t\tfor (int i = 0; i < num_bits; i++)\n\t\t\t\tconfig_bits.push_back(size_b & (1 << i) ? State::S1 : State::S0);\n\n\t\t\tport_a.append(port.in_a);\n\t\t\tport_a.append(port.in_b);\n\t\t}\n\n\t\tcell->setPort(ID::A, port_a);\n\t\tcell->setPort(ID::B, bit_ports);\n\t\tcell->setParam(ID::CONFIG, config_bits);\n\t\tcell->setParam(ID::CONFIG_WIDTH, GetSize(config_bits));\n\t\tcell->setParam(ID::A_WIDTH, GetSize(port_a));\n\t\tcell->setParam(ID::B_WIDTH, GetSize(bit_ports));\n\t}\n\n\tbool eval(RTLIL::Const &result) const\n\t{\n\t\tfor (auto &bit : result.bits)\n\t\t\tbit = State::S0;\n\n\t\tfor (auto &port : ports)\n\t\t{\n\t\t\tif (!port.in_a.is_fully_const() || !port.in_b.is_fully_const())\n\t\t\t\treturn false;\n\n\t\t\tRTLIL::Const summand;\n\t\t\tif (GetSize(port.in_b) == 0)\n\t\t\t\tsummand = const_pos(port.in_a.as_const(), port.in_b.as_const(), port.is_signed, port.is_signed, GetSize(result));\n\t\t\telse\n\t\t\t\tsummand = const_mul(port.in_a.as_const(), port.in_b.as_const(), port.is_signed, port.is_signed, GetSize(result));\n\n\t\t\tif (port.do_subtract)\n\t\t\t\tresult = const_sub(result, summand, port.is_signed, port.is_signed, GetSize(result));\n\t\t\telse\n\t\t\t\tresult = const_add(result, summand, port.is_signed, port.is_signed, GetSize(result));\n\t\t}\n\n\t\tfor (auto bit : bit_ports) {\n\t\t\tif (bit.wire)\n\t\t\t\treturn false;\n\t\t\tresult = const_add(result, bit.data, false, false, GetSize(result));\n\t\t}\n\n\t\treturn true;\n\t}\n\n\tMacc(RTLIL::Cell *cell = nullptr)\n\t{\n\t\tif (cell != nullptr)\n\t\t\tfrom_cell(cell);\n\t}\n};\n\nYOSYS_NAMESPACE_END\n\n#endif\n",