digest-blake3 1.4.1.0 → 1.5.1.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c1fde08d1bb8e1e082e30d543fff61c7bbad77a577109d400883c733c72404df
-  data.tar.gz: e8705dcc092f4cfe63d95d9d837233717a3aeea0f994e2fc49498dae947c7a42
+  metadata.gz: 2bf10e44aaa74a31f9a334b67ecfadfeda7f31a4d3055bd48c5f1a8609e53661
+  data.tar.gz: 53072abb4749ecdfd6748360fc33d39789078413078858953bffcd1ae1cfcdaf
 SHA512:
-  metadata.gz: ca517b08b45f06d098a2938f9cf321a85069d0f59861606fa3ddb44f192d54ca6c8a3dc8607a7704692b81c0c695f1fe9d98bf14fc0068273b165b2772375047
-  data.tar.gz: c295b09c4cbbe7f840bfe4088d522b9e2ca362ed9118962b23dd7756b5b4c916bd974b6132e65bb08ea375d83e5fe5b5281f29fdb40be44b03f6da9451bfbfac
+  metadata.gz: b93a9bdf8b7f2fa4986090e466dfe6c0661d6e9e2de7864cb83846f314a07f198ce16bb6289edb50cf1a7333463136049d61c5ee6fa3db9c8c4855e229fd93a7
+  data.tar.gz: df226266cb38882b121c401d074ad84559265751c897fccda004a8f3d230c2e67215ee6086cfac13fc7043316955511179de4c024dd9d4b5f9048a948ea52080

data/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    digest-blake3 (1.3.3.1)
+    digest-blake3 (1.4.1.0)
 
 GEM
   remote: https://rubygems.org/

data/ext/digest/blake3/blake3.c

@@ -341,21 +341,24 @@ INLINE void compress_subtree_to_parent_node(
   size_t num_cvs = blake3_compress_subtree_wide(input, input_len, key,
                                                 chunk_counter, flags, cv_array);
   assert(num_cvs <= MAX_SIMD_DEGREE_OR_2);
-
-  // If MAX_SIMD_DEGREE is greater than 2 and there's enough input,
+  // The following loop never executes when MAX_SIMD_DEGREE_OR_2 is 2, because
+  // as we just asserted, num_cvs will always be <=2 in that case. But GCC
+  // (particularly GCC 8.5) can't tell that it never executes, and if NDEBUG is
+  // set then it emits incorrect warnings here. We tried a few different
+  // hacks to silence these, but in the end our hacks just produced different
+  // warnings (see https://github.com/BLAKE3-team/BLAKE3/pull/380). Out of
+  // desperation, we ifdef out this entire loop when we know it's not needed.
+#if MAX_SIMD_DEGREE_OR_2 > 2
+  // If MAX_SIMD_DEGREE_OR_2 is greater than 2 and there's enough input,
   // compress_subtree_wide() returns more than 2 chaining values. Condense
   // them into 2 by forming parent nodes repeatedly.
   uint8_t out_array[MAX_SIMD_DEGREE_OR_2 * BLAKE3_OUT_LEN / 2];
-  // The second half of this loop condition is always true, and we just
-  // asserted it above. But GCC can't tell that it's always true, and if NDEBUG
-  // is set on platforms where MAX_SIMD_DEGREE_OR_2 == 2, GCC emits spurious
-  // warnings here. GCC 8.5 is particularly sensitive, so if you're changing
-  // this code, test it against that version.
-  while (num_cvs > 2 && num_cvs <= MAX_SIMD_DEGREE_OR_2) {
+  while (num_cvs > 2) {
     num_cvs =
         compress_parents_parallel(cv_array, num_cvs, key, flags, out_array);
     memcpy(cv_array, out_array, num_cvs * BLAKE3_OUT_LEN);
   }
+#endif
   memcpy(out, cv_array, 2 * BLAKE3_OUT_LEN);
 }
 

data/ext/digest/blake3/blake3.h

@@ -30,7 +30,7 @@
 extern "C" {
 #endif
 
-#define BLAKE3_VERSION_STRING "1.4.1"
+#define BLAKE3_VERSION_STRING "1.5.1"
 #define BLAKE3_KEY_LEN 32
 #define BLAKE3_OUT_LEN 32
 #define BLAKE3_BLOCK_LEN 64

data/ext/digest/blake3/blake3_dispatch.c

@@ -6,6 +6,7 @@
 
 #if defined(IS_X86)
 #if defined(_MSC_VER)
+#include <Windows.h>
 #include <intrin.h>
 #elif defined(__GNUC__)
 #include <immintrin.h>
@@ -14,6 +15,32 @@
 #endif
 #endif
 
+#if !defined(BLAKE3_ATOMICS)
+#if defined(__has_include)
+#if __has_include(<stdatomic.h>) && !defined(_MSC_VER)
+#define BLAKE3_ATOMICS 1
+#else
+#define BLAKE3_ATOMICS 0
+#endif /* __has_include(<stdatomic.h>) && !defined(_MSC_VER) */
+#else
+#define BLAKE3_ATOMICS 0
+#endif /* defined(__has_include) */
+#endif /* BLAKE3_ATOMICS */
+
+#if BLAKE3_ATOMICS
+#define ATOMIC_INT _Atomic int
+#define ATOMIC_LOAD(x) x
+#define ATOMIC_STORE(x, y) x = y
+#elif defined(_MSC_VER)
+#define ATOMIC_INT LONG
+#define ATOMIC_LOAD(x) InterlockedOr(&x, 0)
+#define ATOMIC_STORE(x, y) InterlockedExchange(&x, y)
+#else
+#define ATOMIC_INT int
+#define ATOMIC_LOAD(x) x
+#define ATOMIC_STORE(x, y) x = y
+#endif
+
 #define MAYBE_UNUSED(x) (void)((x))
 
 #if defined(IS_X86)
@@ -76,7 +103,7 @@ enum cpu_feature {
 #if !defined(BLAKE3_TESTING)
 static /* Allow the variable to be controlled manually for testing */
 #endif
-    enum cpu_feature g_cpu_features = UNDEFINED;
+    ATOMIC_INT g_cpu_features = UNDEFINED;
 
 #if !defined(BLAKE3_TESTING)
 static
@@ -84,14 +111,16 @@ static
     enum cpu_feature
     get_cpu_features(void) {
 
-  if (g_cpu_features != UNDEFINED) {
-    return g_cpu_features;
+  /* If TSAN detects a data race here, try compiling with -DBLAKE3_ATOMICS=1 */
+  enum cpu_feature features = ATOMIC_LOAD(g_cpu_features);
+  if (features != UNDEFINED) {
+    return features;
   } else {
 #if defined(IS_X86)
     uint32_t regs[4] = {0};
     uint32_t *eax = &regs[0], *ebx = &regs[1], *ecx = &regs[2], *edx = &regs[3];
     (void)edx;
-    enum cpu_feature features = 0;
+    features = 0;
     cpuid(regs, 0);
     const int max_id = *eax;
     cpuid(regs, 1);
@@ -124,7 +153,7 @@ static
         }
       }
     }
-    g_cpu_features = features;
+    ATOMIC_STORE(g_cpu_features, features);
     return features;
 #else
     /* How to detect NEON? */

data/ext/digest/blake3/blake3_impl.h

@@ -51,7 +51,11 @@ enum blake3_flags {
 #if !defined(BLAKE3_USE_NEON) 
   // If BLAKE3_USE_NEON not manually set, autodetect based on AArch64ness
   #if defined(IS_AARCH64)
-    #define BLAKE3_USE_NEON 1
+    #if defined(__ARM_BIG_ENDIAN)
+      #define BLAKE3_USE_NEON 0
+    #else
+      #define BLAKE3_USE_NEON 1
+    #endif
   #else
     #define BLAKE3_USE_NEON 0
   #endif

data/ext/digest/blake3/blake3_neon.c

@@ -10,14 +10,12 @@
 
 INLINE uint32x4_t loadu_128(const uint8_t src[16]) {
   // vld1q_u32 has alignment requirements. Don't use it.
-  uint32x4_t x;
-  memcpy(&x, src, 16);
-  return x;
+  return vreinterpretq_u32_u8(vld1q_u8(src));
 }
 
 INLINE void storeu_128(uint32x4_t src, uint8_t dest[16]) {
   // vst1q_u32 has alignment requirements. Don't use it.
-  memcpy(dest, &src, 16);
+  vst1q_u8(dest, vreinterpretq_u8_u32(src));
 }
 
 INLINE uint32x4_t add_128(uint32x4_t a, uint32x4_t b) {

data/lib/digest/blake3/version.rb

@@ -2,6 +2,6 @@ require 'digest'
 
 module Digest
   class BLAKE3 < Base
-    VERSION = "1.4.1.0"
+    VERSION = "1.5.1.0"
   end
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: digest-blake3
 version: !ruby/object:Gem::Version
-  version: 1.4.1.0
+  version: 1.5.1.0
 platform: ruby
 authors:
 - Will Bryant
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2023-09-17 00:00:00.000000000 Z
+date: 2024-07-10 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: bundler