Speeding up C implementation

Switch from Sarwate's algorithm to byte-order free Slicing-by-8.
The specialized so files for the built-in CRCs by digest-crc contain only the initialization functions of the library and the table data.
Each table data file is generated by `extconf.rb`.

User-defined CRC classes are also accelerated since a generic `Digest::CRC#update` is provided.
However, the `Digest::CRC::POLYNOMIAL` constants must be overridden.

Also, at build time, the environment variable "RUBY_DIGEST_CRC_ENABLE_SLICING_BY_16=1" can be used to enable "Slicing-by-16" for a further slight speedup.

Author: dearblue

.gitignore

@@ -6,6 +6,7 @@ pkg
 /ext/digest/crc*/Makefile
 /ext/digest/crc*/mkmf.log
 /ext/digest/crc*/extconf.h
+/ext/digest/crc*/*_table.h
 /ext/digest/crc*/*.o
 /ext/digest/crc*/*.so
 /spec/integration/docker/digest-crc.gem

README.md

@@ -117,6 +117,8 @@ module Digest
 
     WIDTH = 32
 
+    POLYNOMIAL = 0x04c11db7
+
     REFLECT_INPUT = true
 
     INIT_CRC = 0xffffffff
@@ -130,6 +132,21 @@ module Digest
 end
 ```
 
+### Notes on using with Ractor
+
+* The `Digest::CRC#update` method in Ruby requires `YourCRC::TABLE` to be
+  in the "frozen" state. This is because if `YourCRC::TABLE` is mutable,
+  it can only be referenced by the main Ractor.
+* The `Digest::CRC#update` method by C must first be called in the main
+  Ractor. This is because the internal tables must be initialized from the
+  main Ractor.
+
+  ```ruby
+  # To digest empty characters in the main Ractor
+  # for initialization of the internal tables.
+  YourCRC.digest("")
+  ```
+
 ## Benchmarks
 
 ### Ruby 2.7.4 (pure Ruby)

Rakefile

@@ -20,6 +20,7 @@ end
 require 'rake/clean'
 CLEAN.include('ext/digest/crc*/extconf.h')
 CLEAN.include('ext/digest/crc*/Makefile')
+CLEAN.include('ext/digest/crc*/*_table.h')
 CLEAN.include('ext/digest/crc*/*.o')
 CLEAN.include('ext/digest/crc*/*.so')
 

ext/digest/Rakefile

@@ -17,6 +17,8 @@ CRCS  = Dir['crc*']
 DLEXT = MakeMakefile::CONFIG['DLEXT']
 
 CRCS.each do |crc|
+  next unless File.file?(File.join(crc, "extconf.rb"))
+
   crc_ext = "#{crc}_ext"
 
   file "#{crc}/Makefile" => "#{crc}/extconf.rb" do

ext/digest/crc.h

@@ -0,0 +1,70 @@
+#ifndef RUBY_DIGEST_CRC_H
+#define RUBY_DIGEST_CRC_H 1
+
+#include <ruby.h>
+#include <stddef.h>
+#include <stdint.h>
+
+enum crc_input_direction {
+	crc_normal_input  = 0,
+	crc_reflect_input = 1
+};
+
+enum crc_allocation_type {
+	crc_alloc_static  = 0,
+	crc_alloc_heap    = 1
+};
+
+struct crc_model
+{
+	int                      bitwidth      : 8;
+	enum crc_input_direction reflect_input : 1;
+	enum crc_allocation_type table_heap    : 1;
+	uint64_t                 polynomial;
+	const void               *table;
+};
+
+/*
+ * This structure is materialized in each shared object file that defines
+ * a CRC.  Therefore, instead of checking the structure address, the
+ * `wrap_struct_name` field is simply checked.  This is to avoid link errors
+ * at build time.
+ */
+static const rb_data_type_t crc_model_const_type = {
+	"digest-crc:model-const",
+	{
+		NULL,
+		NULL,
+		NULL,
+		NULL
+	}
+};
+
+static inline void ruby_digest_crc_model_implant(VALUE crc_class, const struct crc_model *model)
+{
+	ID id_model = rb_intern("digest-crc-model");
+	VALUE obj = rb_data_typed_object_wrap(rb_cObject, (void *)(uintptr_t)model, &crc_model_const_type);
+	rb_obj_freeze(obj);
+	rb_ivar_set(crc_class, id_model, obj);
+}
+
+#if !defined(HAVE_RB_EXT_RACTOR_SAFE)
+static inline void ruby_digest_crc_ensure_ractor_main(const char *mesg)
+{
+	(void)mesg;
+}
+#else
+#include <ruby/ractor.h>
+
+static inline void ruby_digest_crc_ensure_ractor_main(const char *mesg)
+{
+	if (rb_funcallv(rb_cRactor, rb_intern("main"), 0, NULL) !=
+	    rb_funcallv(rb_cRactor, rb_intern("current"), 0, NULL)) {
+		VALUE ractor = rb_const_get(rb_cObject, rb_intern("Ractor"));
+		VALUE ractor_error = rb_const_get(ractor, rb_intern("Error"));
+		rb_raise(ractor_error, "%s", mesg);
+	}
+}
+#endif
+
+#endif // RUBY_DIGEST_CRC_H