Use BLAKE3 from LLVM (11% faster)

Results are the same:

  upstream: SELECT hex(BLAKE3('bar')): F2E897EED7D206CD855D441598FA521ABC75AA96953E97C030C9612C30C1293D
  llvm:     SELECT hex(BLAKE3('bar')): F2E897EED7D206CD855D441598FA521ABC75AA96953E97C030C9612C30C1293D

Query for benchmark:

    SELECT ignore(BLAKE3(materialize('Lorem ipsum dolor sit amet, consectetur adipiscing elit'))) FROM numbers(1000000000) FORMAT `Null`

  upstream           : Elapsed: 2.494 sec. Processed 31.13 million rows, 249.08 MB (12.48 million rows/s., 99.86 MB/s.)
  upstream + rust lto:                                                             ~14.00 million rows/s
  llvm               : Elapsed: 3.053 sec. Processed 43.24 million rows, 345.88 MB (14.16 million rows/s., 113.28 MB/s.)

And note, that now, integrating_rust_libraries.md became deprecated.

P.S. LLVM implementation had been choosen over Rust + LTO, because there
are issues with linking multiple Rust libraries together with LTO:
- https://alanwu.space/post/symbol-hygiene/
- rust-lang/rust#44322

Author: azat

rust/BLAKE3/CMakeLists.txt

@@ -67,5 +67,4 @@ function(add_rust_subdirectory src)
         VERBATIM)
 endfunction()
 
-add_rust_subdirectory (BLAKE3)
 add_rust_subdirectory (skim)

rust/BLAKE3/Cargo.lock

@@ -4,6 +4,7 @@
 #include <farmhash.h>
 #include <metrohash.h>
 #include <wyhash.h>
+#include <llvm/ADT/StringRef.h>
 #include <MurmurHash2.h>
 #include <MurmurHash3.h>
 
@@ -15,10 +16,6 @@
 #endif
 #include <xxhash.h>
 
-#if USE_BLAKE3
-#    include <blake3.h>
-#endif
-
 #include <Common/SipHash.h>
 #include <Common/typeid_cast.h>
 #include <Common/safe_cast.h>
@@ -57,6 +54,10 @@
 #include <base/bit_cast.h>
 #include <base/unaligned.h>
 
+#if USE_BLAKE3
+#include <llvm-c/blake3.h>
+#endif
+
 namespace DB
 {
 
@@ -819,18 +820,10 @@ struct ImplBLAKE3
     #else
     static void apply(const char * begin, const size_t size, unsigned char* out_char_data)
     {
-        #if defined(MEMORY_SANITIZER)
-            auto err_msg = blake3_apply_shim_msan_compat(begin, safe_cast<uint32_t>(size), out_char_data);
-            __msan_unpoison(out_char_data, length);
-        #else
-            auto err_msg = blake3_apply_shim(begin, safe_cast<uint32_t>(size), out_char_data);
-        #endif
-        if (err_msg != nullptr)
-        {
-            auto err_st = std::string(err_msg);
-            blake3_free_char_pointer(err_msg);
-            throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Function returned error message: {}", err_st);
-        }
+        llvm_blake3_hasher hasher;
+        llvm_blake3_hasher_init(&hasher);
+        llvm_blake3_hasher_update(&hasher, begin, size);
+        llvm_blake3_hasher_finalize(&hasher, out_char_data, length);
     }
     #endif
 };

rust/BLAKE3/Cargo.toml

@@ -19,9 +19,6 @@ endif()
 if (TARGET ch_contrib::rdkafka)
     set(USE_RDKAFKA 1)
 endif()
-if (TARGET ch_rust::blake3)
-    set(USE_BLAKE3 1)
-endif()
 if (TARGET ch_rust::skim)
     set(USE_SKIM 1)
 endif()
@@ -99,6 +96,7 @@ if (TARGET ch_contrib::ulid)
 endif()
 if (TARGET ch_contrib::llvm)
     set(USE_EMBEDDED_COMPILER 1)
+    set(USE_BLAKE3 1)
 endif()
 if (TARGET ch_contrib::unixodbc)
     set(USE_ODBC 1)