Merge mainline llama.cpp (#3)

* Merging mainline - WIP

* Merging mainline - WIP

AVX2 and CUDA appear to work.
CUDA performance seems slightly (~1-2%) lower as it is so often
the case with llama.cpp/ggml after some "improvements" have been made.

* Merging mainline - fix Metal

* Remove check

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>

1 files changed, 24 insertions, 0 deletions

diff --git a/examples/gguf-hash/deps/sha256/sha256.h b/examples/gguf-hash/deps/sha256/sha256.h
new file mode 100644
index 00000000..21657e66
--- /dev/null
+++ b/examples/gguf-hash/deps/sha256/sha256.h
@@ -0,0 +1,24 @@
+/* Sha256.h -- SHA-256 Hash
+2010-06-11 : Igor Pavlov : Public domain */
+
+#ifndef __CRYPTO_SHA256_H
+#define __CRYPTO_SHA256_H
+
+#include <stdlib.h>
+#include <stdint.h>
+
+#define SHA256_DIGEST_SIZE 32
+
+typedef struct sha256_t
+{
+  uint32_t state[8];
+  uint64_t count;
+  unsigned char buffer[64];
+} sha256_t;
+
+void sha256_init(sha256_t *p);
+void sha256_update(sha256_t *p, const unsigned char *data, size_t size);
+void sha256_final(sha256_t *p, unsigned char *digest);
+void sha256_hash(unsigned char *buf, const unsigned char *data, size_t size);
+
+#endif