From: Raihaan Shouhell <raihaanhimself@gmail.com>
Date: Sat, 13 Apr 2024 18:14:46 +0000 (+0800)
Subject: bump: Upgrade to blake3 1.5.1 (#1428)
X-Git-Tag: v4.10~49
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=f5f1e429428260df65ffbe5e26150d4077d4a129;p=thirdparty%2Fccache.git

bump: Upgrade to blake3 1.5.1 (#1428)

Update our dependencies
---

diff --git a/LICENSE.adoc b/LICENSE.adoc
index 322881a9..076fe54c 100644
--- a/LICENSE.adoc
+++ b/LICENSE.adoc
@@ -50,7 +50,7 @@ under less restrictive terms.
 
 === src/third_party/blake3/blake3_*
 
-This is a subset of https://github.com/BLAKE3-team/BLAKE3[BLAKE3] 1.5.0 with the
+This is a subset of https://github.com/BLAKE3-team/BLAKE3[BLAKE3] 1.5.1 with the
 following license:
 
 ----
diff --git a/src/third_party/blake3/blake3.c b/src/third_party/blake3/blake3.c
index 692f4b02..1b44c719 100644
--- a/src/third_party/blake3/blake3.c
+++ b/src/third_party/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);
 }
 
diff --git a/src/third_party/blake3/blake3.h b/src/third_party/blake3/blake3.h
index f694dcf2..48284e50 100644
--- a/src/third_party/blake3/blake3.h
+++ b/src/third_party/blake3/blake3.h
@@ -30,7 +30,7 @@
 extern "C" {
 #endif
 
-#define BLAKE3_VERSION_STRING "1.5.0"
+#define BLAKE3_VERSION_STRING "1.5.1"
 #define BLAKE3_KEY_LEN 32
 #define BLAKE3_OUT_LEN 32
 #define BLAKE3_BLOCK_LEN 64
diff --git a/src/third_party/blake3/blake3_dispatch.c b/src/third_party/blake3/blake3_dispatch.c
index af6c3dad..c9abc13f 100644
--- a/src/third_party/blake3/blake3_dispatch.c
+++ b/src/third_party/blake3/blake3_dispatch.c
@@ -4,9 +4,12 @@
 
 #include "blake3_impl.h"
 
-#if defined(IS_X86)
 #if defined(_MSC_VER)
 #include <Windows.h>
+#endif
+
+#if defined(IS_X86)
+#if defined(_MSC_VER)
 #include <intrin.h>
 #elif defined(__GNUC__)
 #include <immintrin.h>
diff --git a/src/third_party/blake3/blake3_impl.h b/src/third_party/blake3/blake3_impl.h
index beab5cf5..98611c31 100644
--- a/src/third_party/blake3/blake3_impl.h
+++ b/src/third_party/blake3/blake3_impl.h
@@ -28,7 +28,7 @@ enum blake3_flags {
 #define INLINE static inline __attribute__((always_inline))
 #endif
 
-#if defined(__x86_64__) || defined(_M_X64) 
+#if (defined(__x86_64__) || defined(_M_X64)) && !defined(_M_ARM64EC)
 #define IS_X86
 #define IS_X86_64
 #endif
@@ -38,7 +38,7 @@ enum blake3_flags {
 #define IS_X86_32
 #endif
 
-#if defined(__aarch64__) || defined(_M_ARM64)
+#if defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
 #define IS_AARCH64
 #endif
 
diff --git a/src/third_party/blake3/blake3_neon.c b/src/third_party/blake3/blake3_neon.c
index 8a818fc7..90bdd572 100644
--- a/src/third_party/blake3/blake3_neon.c
+++ b/src/third_party/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) {