From df030ac51c05045af88f95f9b453d14c835c925a Mon Sep 17 00:00:00 2001
From: Manuel Barrio Linares <mbarriolinares@gmail.com>
Date: Sun, 15 Feb 2026 22:10:53 -0300
Subject: [PATCH] optimize `blake3` using a runtime for loop

instruction count went from 60k to 9k
no difference in speed for -O2 or higher
---
 benchmarks/stdlib/hash/blake3.c3 | 57 ++++++++++++++++++++++++++++++++
 lib/std/hash/blake3.c3           | 12 +++----
 2 files changed, 62 insertions(+), 7 deletions(-)
 create mode 100644 benchmarks/stdlib/hash/blake3.c3

diff --git a/benchmarks/stdlib/hash/blake3.c3 b/benchmarks/stdlib/hash/blake3.c3
new file mode 100644
index 000000000..b1bfdd2c6
--- /dev/null
+++ b/benchmarks/stdlib/hash/blake3.c3
@@ -0,0 +1,57 @@
+module blake3_bench;
+
+fn void initialize_bench() @init
+{
+	set_benchmark_warmup_iterations(3);
+	set_benchmark_max_iterations(128);
+
+    input = mem::alloc_array(char, BUFSZ);
+    input[:BUFSZ] = (char[]){ [0..BUFSZ-1] = 0xA5 }[..];
+    input_slice = input[:BUFSZ];
+}
+
+fn void teardown_bench() @finalizer
+{
+    mem::free(input);
+    input = null;
+}
+
+char* input;
+char[] input_slice;
+const usz BUFSZ = 1024 * 1024;
+
+module blake3_bench @benchmark;
+
+import std::hash;
+
+fn void blake3_hash()
+{
+    runtime::@start_benchmark();
+    char[*] myset = blake3::hash(input_slice);
+    runtime::@end_benchmark();
+    mem::zero_volatile(myset[..]);
+}
+
+fn void compared_with_sha256()
+{
+    runtime::@start_benchmark();
+    char[*] myset = sha256::hash(input_slice);
+    runtime::@end_benchmark();
+    mem::zero_volatile(myset[..]);
+}
+
+fn void compared_with_sha512()
+{
+    runtime::@start_benchmark();
+    char[*] myset = sha512::hash(input_slice);
+    runtime::@end_benchmark();
+    mem::zero_volatile(myset[..]);
+}
+
+fn void compared_with_whirlpool()
+{
+    runtime::@start_benchmark();
+    char[*] myset = whirlpool::hash(input_slice);
+    runtime::@end_benchmark();
+    mem::zero_volatile(myset[..]);
+}
diff --git a/lib/std/hash/blake3.c3 b/lib/std/hash/blake3.c3
index 29946680e..3c3986ccd 100644
--- a/lib/std/hash/blake3.c3
+++ b/lib/std/hash/blake3.c3
@@ -606,13 +606,11 @@ fn void compress_pre(uint[] state, uint[] cv, char[BLOCK_SIZE] block, usz block_
 	state[13] = (uint)(counter >> 32);
 	state[14] = (uint)block_len;
 	state[15] = (uint)flags;
-	@round(state, &block_words[0], 0);
-	@round(state, &block_words[0], 1);
-	@round(state, &block_words[0], 2);
-	@round(state, &block_words[0], 3);
-	@round(state, &block_words[0], 4);
-	@round(state, &block_words[0], 5);
-	@round(state, &block_words[0], 6);
+
+	for (int i = 0; i < 7; i++)
+	{
+		@round(state, &block_words[0], (usz)i);
+	}
 }
 
 macro compress_in_place(uint[] cv, char[BLOCK_SIZE] block, usz block_len, ulong counter, char flags) @local