[Support/BLAKE3] Rename blake3_* -> llvm_blake3_* to avoid symbol collisions

[jbms]

Summary

While some blake3 symbols are already prefixed, a number of symbols with hidden visibility have been left without an llvm_ prefix. This results in symbol collisions when statically linking llvm into a binary that also uses the external blake3 library.

[MaskRay]

LGTM in spirit, it matches what we did to blake3.c and other libraries (e.g. Support/regexec.c) How do you verify that all blake3 non-local symbols are now prefixed?
I think this is a good candidate for the release/16.x branch (https://github.com/llvm/llvm-project/issues?q=is%3Aopen+is%3Aissue+milestone%3A%22LLVM+16.0.0+Release%22)

edit: I can verify with llvm-nm -jgU /tmp/Rel/lib/libLLVMSupport.a | grep '^blake3'

[MaskRay]

Can you fix blake3_neon.c for aarch64 as well?

[jbms]

In D143981#4125145, @MaskRay wrote:

Can you fix blake3_neon.c for aarch64 as well?

I believe that is already fixed by the #define in blake3_impl.h

[akyrtzi]

.s files can use the preprocessor as well. I'd recommend to add a file that do the renames using #define, like in blake3_impl.h, and include it in all the .s files.
That way the changes to the original files is minimized, which is useful for pulling in future changes from the original repo.

[jbms]

In D143981#4125191, @akyrtzi wrote:

.s files can use the preprocessor as well. I'd recommend to add a file that do the renames using #define, like in blake3_impl.h, and include it in all the .s files.
That way the changes to the original files is minimized, which is useful for pulling in future changes from the original repo.

The MSVC ".asm" files aren't preprocessed, though. Given that, it may be simpler to just use a sed script to apply the renames to all the assembly files when importing (which is what I used to generate the patch in the first place):

sed -i -e 's/\b\(_\?\)blake3_/\1llvm_blake3_/g' *.S *.asm

[akyrtzi]

In D143981#4125246, @jbms wrote:

The MSVC ".asm" files aren't preprocessed, though.

Good point, for now let's use #defines for .S files and rename inline for the .asm files; later on we could consider setting up manual preprocessing via CMake for the .asm files as well.
Essentially my suggestion is to pull the renaming #defines out of blake3_impl.h and into a separate file, then have it included from blake3_impl.h and the .S files. The .asm files get renames inline for now.

[jbms]

In D143981#4126218, @akyrtzi wrote:

In D143981#4125246, @jbms wrote:

The MSVC ".asm" files aren't preprocessed, though.

Good point, for now let's use #defines for .S files and rename inline for the .asm files; later on we could consider setting up manual preprocessing via CMake for the .asm files as well.
Essentially my suggestion is to pull the renaming #defines out of blake3_impl.h and into a separate file, then have it included from blake3_impl.h and the .S files. The .asm files get renames inline for now.

Done.

[MaskRay]

In D143981#4126218, @akyrtzi wrote:

In D143981#4125246, @jbms wrote:

The MSVC ".asm" files aren't preprocessed, though.

Good point, for now let's use #defines for .S files and rename inline for the .asm files; later on we could consider setting up manual preprocessing via CMake for the .asm files as well.
Essentially my suggestion is to pull the renaming #defines out of blake3_impl.h and into a separate file, then have it included from blake3_impl.h and the .S files. The .asm files get renames inline for now.

This plan looks good to me. I am fine with changing .asm files. If we use a CMake approach, gn and bazel (unsupported build systems) users need to port this logic as well...

[MaskRay]

@mstorsjo Will you be kindly to test this on Windows to see that there is no more external blake3_* definitions? :)

[mstorsjo]

In D143981#4126902, @MaskRay wrote:

@mstorsjo Will you be kindly to test this on Windows to see that there is no more external blake3_* definitions? :)

This built fine in both mingw/x86_64 and msvc/x86_64 configs.

[akyrtzi]

llvm/lib/Support/BLAKE3/blake3_impl.h

35

Could you exclude the whitespace changes from this file?

[akyrtzi]

Thank you for the fix! 🙇🏻‍♂️

[MaskRay]

In D143981#4127091, @mstorsjo wrote:

In D143981#4126902, @MaskRay wrote:

@mstorsjo Will you be kindly to test this on Windows to see that there is no more external blake3_* definitions? :)

This built fine in both mingw/x86_64 and msvc/x86_64 configs.

Thanks for verification!

[MaskRay]

In D143981#4125145, @MaskRay wrote:

Can you fix blake3_neon.c for aarch64 as well?

I confirm that this renames all blake3_* external symbols on Linux aarch64 when neon is enabled.

llvm/lib/Support/BLAKE3/blake3_avx2_x86-64_unix.S

 #if defined(__x86_64__)
 
+#include "llvm_blake3_prefix.h"
+
 #if defined(__ELF__) && defined(__linux__)
 .section .note.GNU-stack,"",%progbits
 #endif
 
 #if defined(__ELF__) && defined(__CET__) && defined(__has_include)
 #if __has_include(<cet.h>)
 #include <cet.h>
 #endif

llvm/lib/Support/BLAKE3/blake3_avx2_x86-64_windows_gnu.S

+#include "llvm_blake3_prefix.h"
+
 .intel_syntax noprefix
 .global _blake3_hash_many_avx2
 .global blake3_hash_many_avx2
 .section .text
         .p2align  6
 _blake3_hash_many_avx2:
 blake3_hash_many_avx2:
         push    r15

llvm/lib/Support/BLAKE3/blake3_avx2_x86-64_windows_msvc.asm

-public _blake3_hash_many_avx2
-public blake3_hash_many_avx2
+public _llvm_blake3_hash_many_avx2
+public llvm_blake3_hash_many_avx2
 
 _TEXT   SEGMENT ALIGN(16) 'CODE'
 
 ALIGN   16
-blake3_hash_many_avx2 PROC
-_blake3_hash_many_avx2 PROC
+llvm_blake3_hash_many_avx2 PROC
+_llvm_blake3_hash_many_avx2 PROC
         push    r15
         push    r14
         push    r13
         push    r12
         push    rsi
         push    rdi
         push    rbx
         push    rbp
endroundloop1:
         vpxor   xmm1, xmm1, xmm3
         mov     eax, r13d
         cmp     rdx, r15
         jne     innerloop1
         vmovdqu xmmword ptr [rbx], xmm0
         vmovdqu xmmword ptr [rbx+10H], xmm1
         jmp     unwind
 
-_blake3_hash_many_avx2 ENDP
-blake3_hash_many_avx2 ENDP
+_llvm_blake3_hash_many_avx2 ENDP
+llvm_blake3_hash_many_avx2 ENDP
 _TEXT ENDS
 
 _RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
 ALIGN   64
 ADD0:
         dd 0, 1, 2, 3, 4, 5, 6, 7
 
 ADD1:

llvm/lib/Support/BLAKE3/blake3_avx512_x86-64_unix.S

 #if defined(__x86_64__)
 
+#include "llvm_blake3_prefix.h"
+
 #if defined(__ELF__) && defined(__linux__)
 .section .note.GNU-stack,"",%progbits
 #endif
 
 #if defined(__ELF__) && defined(__CET__) && defined(__has_include)
 #if __has_include(<cet.h>)
 #include <cet.h>
 #endif

llvm/lib/Support/BLAKE3/blake3_avx512_x86-64_windows_gnu.S

+#include "llvm_blake3_prefix.h"
+
 .intel_syntax noprefix
 
 .global _blake3_hash_many_avx512
 .global blake3_hash_many_avx512
 .global blake3_compress_in_place_avx512
 .global _blake3_compress_in_place_avx512
 .global blake3_compress_xof_avx512
 .global _blake3_compress_xof_avx512

llvm/lib/Support/BLAKE3/blake3_avx512_x86-64_windows_msvc.asm

-public _blake3_hash_many_avx512
-public blake3_hash_many_avx512
-public blake3_compress_in_place_avx512
-public _blake3_compress_in_place_avx512
-public blake3_compress_xof_avx512
-public _blake3_compress_xof_avx512
+public _llvm_blake3_hash_many_avx512
+public llvm_blake3_hash_many_avx512
+public llvm_blake3_compress_in_place_avx512
+public _llvm_blake3_compress_in_place_avx512
+public llvm_blake3_compress_xof_avx512
+public _llvm_blake3_compress_xof_avx512
 
 _TEXT   SEGMENT ALIGN(16) 'CODE'
 
 ALIGN   16
-blake3_hash_many_avx512 PROC
-_blake3_hash_many_avx512 PROC
+llvm_blake3_hash_many_avx512 PROC
+_llvm_blake3_hash_many_avx512 PROC
         push    r15
         push    r14
         push    r13
         push    r12
         push    rdi
         push    rsi
         push    rbx
         push    rbp
endroundloop1:
         vpxor   xmm1, xmm1, xmm3
         mov     eax, r13d
         cmp     rdx, r15
         jne     innerloop1
         vmovdqu xmmword ptr [rbx], xmm0
         vmovdqu xmmword ptr [rbx+10H], xmm1
         jmp     unwind
 
-_blake3_hash_many_avx512 ENDP
-blake3_hash_many_avx512 ENDP
+_llvm_blake3_hash_many_avx512 ENDP
+llvm_blake3_hash_many_avx512 ENDP
 
 ALIGN 16
-blake3_compress_in_place_avx512 PROC
-_blake3_compress_in_place_avx512 PROC
+llvm_blake3_compress_in_place_avx512 PROC
+_llvm_blake3_compress_in_place_avx512 PROC
         sub     rsp, 72
         vmovdqa xmmword ptr [rsp], xmm6
         vmovdqa xmmword ptr [rsp+10H], xmm7
         vmovdqa xmmword ptr [rsp+20H], xmm8
         vmovdqa xmmword ptr [rsp+30H], xmm9
         vmovdqu xmm0, xmmword ptr [rcx]
         vmovdqu xmm1, xmmword ptr [rcx+10H]
         movzx   eax, byte ptr [rsp+70H]
@@:
         vmovdqu xmmword ptr [rcx], xmm0
         vmovdqu xmmword ptr [rcx+10H], xmm1
         vmovdqa xmm6, xmmword ptr [rsp]
         vmovdqa xmm7, xmmword ptr [rsp+10H]
         vmovdqa xmm8, xmmword ptr [rsp+20H]
         vmovdqa xmm9, xmmword ptr [rsp+30H]
         add     rsp, 72
         ret
-_blake3_compress_in_place_avx512 ENDP
-blake3_compress_in_place_avx512 ENDP
+_llvm_blake3_compress_in_place_avx512 ENDP
+llvm_blake3_compress_in_place_avx512 ENDP
 
 ALIGN 16
-blake3_compress_xof_avx512 PROC
-_blake3_compress_xof_avx512 PROC
+llvm_blake3_compress_xof_avx512 PROC
+_llvm_blake3_compress_xof_avx512 PROC
         sub     rsp, 72
         vmovdqa xmmword ptr [rsp], xmm6
         vmovdqa xmmword ptr [rsp+10H], xmm7
         vmovdqa xmmword ptr [rsp+20H], xmm8
         vmovdqa xmmword ptr [rsp+30H], xmm9
         vmovdqu xmm0, xmmword ptr [rcx]
         vmovdqu xmm1, xmmword ptr [rcx+10H]
         movzx   eax, byte ptr [rsp+70H]
@@:
         vmovdqu xmmword ptr [r10+20H], xmm2
         vmovdqu xmmword ptr [r10+30H], xmm3
         vmovdqa xmm6, xmmword ptr [rsp]
         vmovdqa xmm7, xmmword ptr [rsp+10H]
         vmovdqa xmm8, xmmword ptr [rsp+20H]
         vmovdqa xmm9, xmmword ptr [rsp+30H]
         add     rsp, 72
         ret
-_blake3_compress_xof_avx512 ENDP
-blake3_compress_xof_avx512 ENDP
+_llvm_blake3_compress_xof_avx512 ENDP
+llvm_blake3_compress_xof_avx512 ENDP
 
 _TEXT ENDS
 
 _RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
 ALIGN   64
 INDEX0:
         dd    0,  1,  2,  3, 16, 17, 18, 19
         dd    8,  9, 10, 11, 24, 25, 26, 27

llvm/lib/Support/BLAKE3/blake3_impl.h

 #ifndef BLAKE3_IMPL_H
 #define BLAKE3_IMPL_H
 
 #include <assert.h>
 #include <stdbool.h>
 #include <stddef.h>
 #include <stdint.h>
 #include <string.h>
 
 #include "llvm-c/blake3.h"
 // For \p LLVM_LIBRARY_VISIBILITY
 #include "llvm/Support/Compiler.h"
 
-// Remove the 'llvm_' prefix for the rest of the internal implementation.
-#define BLAKE3_VERSION_STRING LLVM_BLAKE3_VERSION_STRING
-#define BLAKE3_KEY_LEN LLVM_BLAKE3_KEY_LEN
-#define BLAKE3_OUT_LEN LLVM_BLAKE3_OUT_LEN
-#define BLAKE3_BLOCK_LEN LLVM_BLAKE3_BLOCK_LEN
-#define BLAKE3_CHUNK_LEN LLVM_BLAKE3_CHUNK_LEN
-#define BLAKE3_MAX_DEPTH LLVM_BLAKE3_MAX_DEPTH
-#define blake3_hasher llvm_blake3_hasher
-#define blake3_chunk_state llvm_blake3_chunk_state
+#include "llvm_blake3_prefix.h"
 
 // internal flags
 enum blake3_flags {
   CHUNK_START         = 1 << 0,
   CHUNK_END           = 1 << 1,
   PARENT              = 1 << 2,
   ROOT                = 1 << 3,
   KEYED_HASH          = 1 << 4,
   DERIVE_KEY_CONTEXT  = 1 << 5,
   DERIVE_KEY_MATERIAL = 1 << 6,
 };
 
 // This C implementation tries to support recent versions of GCC, Clang, and
 // MSVC.
 #if defined(_MSC_VER)
 #define INLINE static __forceinline
 #else
 #define INLINE static inline __attribute__((always_inline))
 #endif
 
 #if defined(__x86_64__) || defined(_M_X64) 
 #define IS_X86
 #define IS_X86_64
 #endif
 
 #if defined(__i386__) || defined(_M_IX86)
 #define IS_X86
 #define IS_X86_32
 #endif

llvm/lib/Support/BLAKE3/blake3_sse2_x86-64_unix.S

 #if defined(__x86_64__)
 
+#include "llvm_blake3_prefix.h"
+
 #if defined(__ELF__) && defined(__linux__)
 .section .note.GNU-stack,"",%progbits
 #endif
 
 #if defined(__ELF__) && defined(__CET__) && defined(__has_include)
 #if __has_include(<cet.h>)
 #include <cet.h>
 #endif

llvm/lib/Support/BLAKE3/blake3_sse2_x86-64_windows_gnu.S

+#include "llvm_blake3_prefix.h"
+
 .intel_syntax noprefix
 .global blake3_hash_many_sse2
 .global _blake3_hash_many_sse2
 .global blake3_compress_in_place_sse2
 .global _blake3_compress_in_place_sse2
 .global blake3_compress_xof_sse2
 .global _blake3_compress_xof_sse2
 .section .text

llvm/lib/Support/BLAKE3/blake3_sse2_x86-64_windows_msvc.asm

-public _blake3_hash_many_sse2
-public blake3_hash_many_sse2
-public blake3_compress_in_place_sse2
-public _blake3_compress_in_place_sse2
-public blake3_compress_xof_sse2
-public _blake3_compress_xof_sse2
+public _llvm_blake3_hash_many_sse2
+public llvm_blake3_hash_many_sse2
+public llvm_blake3_compress_in_place_sse2
+public _llvm_blake3_compress_in_place_sse2
+public llvm_blake3_compress_xof_sse2
+public _llvm_blake3_compress_xof_sse2
 
 _TEXT   SEGMENT ALIGN(16) 'CODE'
 
 ALIGN   16
-blake3_hash_many_sse2 PROC
-_blake3_hash_many_sse2 PROC
+llvm_blake3_hash_many_sse2 PROC
+_llvm_blake3_hash_many_sse2 PROC
         push    r15
         push    r14
         push    r13
         push    r12
         push    rsi
         push    rdi
         push    rbx
         push    rbp
endroundloop1:
         pxor    xmm0, xmm2
         pxor    xmm1, xmm3
         mov     eax, r13d
         cmp     rdx, r15
         jne     innerloop1
         movups  xmmword ptr [rbx], xmm0
         movups  xmmword ptr [rbx+10H], xmm1
         jmp     unwind
-_blake3_hash_many_sse2 ENDP
-blake3_hash_many_sse2 ENDP
+_llvm_blake3_hash_many_sse2 ENDP
+llvm_blake3_hash_many_sse2 ENDP
 
-blake3_compress_in_place_sse2 PROC
-_blake3_compress_in_place_sse2 PROC
+llvm_blake3_compress_in_place_sse2 PROC
+_llvm_blake3_compress_in_place_sse2 PROC
         sub     rsp, 120
         movdqa  xmmword ptr [rsp], xmm6
         movdqa  xmmword ptr [rsp+10H], xmm7
         movdqa  xmmword ptr [rsp+20H], xmm8
         movdqa  xmmword ptr [rsp+30H], xmm9
         movdqa  xmmword ptr [rsp+40H], xmm11
         movdqa  xmmword ptr [rsp+50H], xmm14
         movdqa  xmmword ptr [rsp+60H], xmm15
@@:
         movdqa  xmm7, xmmword ptr [rsp+10H]
         movdqa  xmm8, xmmword ptr [rsp+20H]
         movdqa  xmm9, xmmword ptr [rsp+30H]
         movdqa  xmm11, xmmword ptr [rsp+40H]
         movdqa  xmm14, xmmword ptr [rsp+50H]
         movdqa  xmm15, xmmword ptr [rsp+60H]
         add     rsp, 120
         ret
-_blake3_compress_in_place_sse2 ENDP
-blake3_compress_in_place_sse2 ENDP
+_llvm_blake3_compress_in_place_sse2 ENDP
+llvm_blake3_compress_in_place_sse2 ENDP
 
 ALIGN 16
-blake3_compress_xof_sse2 PROC
-_blake3_compress_xof_sse2 PROC
+llvm_blake3_compress_xof_sse2 PROC
+_llvm_blake3_compress_xof_sse2 PROC
         sub     rsp, 120
         movdqa  xmmword ptr [rsp], xmm6
         movdqa  xmmword ptr [rsp+10H], xmm7
         movdqa  xmmword ptr [rsp+20H], xmm8
         movdqa  xmmword ptr [rsp+30H], xmm9
         movdqa  xmmword ptr [rsp+40H], xmm11
         movdqa  xmmword ptr [rsp+50H], xmm14
         movdqa  xmmword ptr [rsp+60H], xmm15
@@:
         movdqa  xmm7, xmmword ptr [rsp+10H]
         movdqa  xmm8, xmmword ptr [rsp+20H]
         movdqa  xmm9, xmmword ptr [rsp+30H]
         movdqa  xmm11, xmmword ptr [rsp+40H]
         movdqa  xmm14, xmmword ptr [rsp+50H]
         movdqa  xmm15, xmmword ptr [rsp+60H]
         add     rsp, 120
         ret
-_blake3_compress_xof_sse2 ENDP
-blake3_compress_xof_sse2 ENDP
+_llvm_blake3_compress_xof_sse2 ENDP
+llvm_blake3_compress_xof_sse2 ENDP
 
 _TEXT ENDS
 
 
 _RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
 ALIGN   64
 BLAKE3_IV:
         dd 6A09E667H, 0BB67AE85H, 3C6EF372H, 0A54FF53AH

llvm/lib/Support/BLAKE3/blake3_sse41_x86-64_unix.S

 #if defined(__x86_64__)
 
+#include "llvm_blake3_prefix.h"
+
 #if defined(__ELF__) && defined(__linux__)
 .section .note.GNU-stack,"",%progbits
 #endif
 
 #if defined(__ELF__) && defined(__CET__) && defined(__has_include)
 #if __has_include(<cet.h>)
 #include <cet.h>
 #endif

llvm/lib/Support/BLAKE3/blake3_sse41_x86-64_windows_gnu.S

+#include "llvm_blake3_prefix.h"
+
 .intel_syntax noprefix
 .global blake3_hash_many_sse41
 .global _blake3_hash_many_sse41
 .global blake3_compress_in_place_sse41
 .global _blake3_compress_in_place_sse41
 .global blake3_compress_xof_sse41
 .global _blake3_compress_xof_sse41
 .section .text

llvm/lib/Support/BLAKE3/blake3_sse41_x86-64_windows_msvc.asm

-public _blake3_hash_many_sse41
-public blake3_hash_many_sse41
-public blake3_compress_in_place_sse41
-public _blake3_compress_in_place_sse41
-public blake3_compress_xof_sse41
-public _blake3_compress_xof_sse41
+public _llvm_blake3_hash_many_sse41
+public llvm_blake3_hash_many_sse41
+public llvm_blake3_compress_in_place_sse41
+public _llvm_blake3_compress_in_place_sse41
+public llvm_blake3_compress_xof_sse41
+public _llvm_blake3_compress_xof_sse41
 
 _TEXT   SEGMENT ALIGN(16) 'CODE'
 
 ALIGN   16
-blake3_hash_many_sse41 PROC
-_blake3_hash_many_sse41 PROC
+llvm_blake3_hash_many_sse41 PROC
+_llvm_blake3_hash_many_sse41 PROC
         push    r15
         push    r14
         push    r13
         push    r12
         push    rsi
         push    rdi
         push    rbx
         push    rbp
endroundloop1:
         pxor    xmm0, xmm2
         pxor    xmm1, xmm3
         mov     eax, r13d
         cmp     rdx, r15
         jne     innerloop1
         movups  xmmword ptr [rbx], xmm0
         movups  xmmword ptr [rbx+10H], xmm1
         jmp     unwind
-_blake3_hash_many_sse41 ENDP
-blake3_hash_many_sse41 ENDP
+_llvm_blake3_hash_many_sse41 ENDP
+llvm_blake3_hash_many_sse41 ENDP
 
-blake3_compress_in_place_sse41 PROC
-_blake3_compress_in_place_sse41 PROC
+llvm_blake3_compress_in_place_sse41 PROC
+_llvm_blake3_compress_in_place_sse41 PROC
         sub     rsp, 120
         movdqa  xmmword ptr [rsp], xmm6
         movdqa  xmmword ptr [rsp+10H], xmm7
         movdqa  xmmword ptr [rsp+20H], xmm8
         movdqa  xmmword ptr [rsp+30H], xmm9
         movdqa  xmmword ptr [rsp+40H], xmm11
         movdqa  xmmword ptr [rsp+50H], xmm14
         movdqa  xmmword ptr [rsp+60H], xmm15
@@:
         movdqa  xmm7, xmmword ptr [rsp+10H]
         movdqa  xmm8, xmmword ptr [rsp+20H]
         movdqa  xmm9, xmmword ptr [rsp+30H]
         movdqa  xmm11, xmmword ptr [rsp+40H]
         movdqa  xmm14, xmmword ptr [rsp+50H]
         movdqa  xmm15, xmmword ptr [rsp+60H]
         add     rsp, 120
         ret
-_blake3_compress_in_place_sse41 ENDP
-blake3_compress_in_place_sse41 ENDP
+_llvm_blake3_compress_in_place_sse41 ENDP
+llvm_blake3_compress_in_place_sse41 ENDP
 
 ALIGN 16
-blake3_compress_xof_sse41 PROC
-_blake3_compress_xof_sse41 PROC
+llvm_blake3_compress_xof_sse41 PROC
+_llvm_blake3_compress_xof_sse41 PROC
         sub     rsp, 120
         movdqa  xmmword ptr [rsp], xmm6
         movdqa  xmmword ptr [rsp+10H], xmm7
         movdqa  xmmword ptr [rsp+20H], xmm8
         movdqa  xmmword ptr [rsp+30H], xmm9
         movdqa  xmmword ptr [rsp+40H], xmm11
         movdqa  xmmword ptr [rsp+50H], xmm14
         movdqa  xmmword ptr [rsp+60H], xmm15
@@:
         movdqa  xmm7, xmmword ptr [rsp+10H]
         movdqa  xmm8, xmmword ptr [rsp+20H]
         movdqa  xmm9, xmmword ptr [rsp+30H]
         movdqa  xmm11, xmmword ptr [rsp+40H]
         movdqa  xmm14, xmmword ptr [rsp+50H]
         movdqa  xmm15, xmmword ptr [rsp+60H]
         add     rsp, 120
         ret
-_blake3_compress_xof_sse41 ENDP
-blake3_compress_xof_sse41 ENDP
+_llvm_blake3_compress_xof_sse41 ENDP
+llvm_blake3_compress_xof_sse41 ENDP
 
 _TEXT ENDS
 
 
 _RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
 ALIGN   64
 BLAKE3_IV:
         dd 6A09E667H, 0BB67AE85H, 3C6EF372H, 0A54FF53AH

llvm/lib/Support/BLAKE3/llvm_blake3_prefix.h

+#ifndef LLVM_BLAKE3_PREFIX_H
+#define LLVM_BLAKE3_PREFIX_H
+
+#define BLAKE3_VERSION_STRING LLVM_BLAKE3_VERSION_STRING
+#define BLAKE3_KEY_LEN LLVM_BLAKE3_KEY_LEN
+#define BLAKE3_OUT_LEN LLVM_BLAKE3_OUT_LEN
+#define BLAKE3_BLOCK_LEN LLVM_BLAKE3_BLOCK_LEN
+#define BLAKE3_CHUNK_LEN LLVM_BLAKE3_CHUNK_LEN
+#define BLAKE3_MAX_DEPTH LLVM_BLAKE3_MAX_DEPTH
+#define blake3_hasher llvm_blake3_hasher
+#define blake3_chunk_state llvm_blake3_chunk_state
+#define blake3_compress_in_place llvm_blake3_compress_in_place
+#define blake3_compress_xof llvm_blake3_compress_xof
+#define blake3_hash_many llvm_blake3_hash_many
+#define blake3_simd_degree llvm_blake3_simd_degree
+#define blake3_compress_in_place_portable llvm_blake3_compress_in_place_portable
+#define blake3_compress_xof_portable llvm_blake3_compress_xof_portable
+#define blake3_hash_many_portable llvm_blake3_hash_many_portable
+#define blake3_compress_in_place_sse2 llvm_blake3_compress_in_place_sse2
+#define _blake3_compress_in_place_sse2 _llvm_blake3_compress_in_place_sse2
+#define blake3_compress_xof_sse2 llvm_blake3_compress_xof_sse2
+#define _blake3_compress_xof_sse2 _llvm_blake3_compress_xof_sse2
+#define blake3_hash_many_sse2 llvm_blake3_hash_many_sse2
+#define _blake3_hash_many_sse2 _llvm_blake3_hash_many_sse2
+#define blake3_compress_in_place_sse41 llvm_blake3_compress_in_place_sse41
+#define _blake3_compress_in_place_sse41 _llvm_blake3_compress_in_place_sse41
+#define blake3_compress_xof_sse41 llvm_blake3_compress_xof_sse41
+#define _blake3_compress_xof_sse41 _llvm_blake3_compress_xof_sse41
+#define blake3_hash_many_sse41 llvm_blake3_hash_many_sse41
+#define _blake3_hash_many_sse41 _llvm_blake3_hash_many_sse41
+#define blake3_hash_many_avx2 llvm_blake3_hash_many_avx2
+#define _blake3_hash_many_avx2 _llvm_blake3_hash_many_avx2
+#define blake3_compress_in_place_avx512 llvm_blake3_compress_in_place_avx512
+#define _blake3_compress_in_place_avx512 _llvm_blake3_compress_in_place_avx512
+#define blake3_compress_xof_avx512 llvm_blake3_compress_xof_avx512
+#define _blake3_compress_xof_avx512 _llvm_blake3_compress_xof_avx512
+#define blake3_hash_many_avx512 llvm_blake3_hash_many_avx512
+#define _blake3_hash_many_avx512 _llvm_blake3_hash_many_avx512
+#define blake3_hash_many_neon llvm_blake3_hash_many_neon
+
+#endif /* LLVM_BLAKE3_PREFIX_H */