using trick

Author: lemire

benchmark/Benchmark.cs

@@ -153,8 +153,8 @@ public Config()
         }
         // Parameters and variables for real data
         [Params(
-                @"data/email/" //,
-                //@"data/dns/swedenzonebase.txt"
+                @"data/email/" //,
+                               //@"data/dns/swedenzonebase.txt"
                 )]
 #pragma warning disable CA1051
         public string? FileName;

src/Base64AVX2UTF8.cs

@@ -246,7 +246,9 @@ private static unsafe void CopyBlock(Block64* b, byte* output)
             private static unsafe void Base64DecodeBlockSafe(byte* outPtr, Block64* b)
             {
                 Base64Decode(outPtr, b->chunk0);
-                Span<byte> buffer = stackalloc byte[32];
+                // Should be:
+                // Span<byte> buffer = stackalloc byte[32];
+                Span<byte> buffer = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
 
                 // Safe memory copy for the last part of the data
                 fixed (byte* bufferStart = buffer)
@@ -296,10 +298,12 @@ private static unsafe void Base64DecodeBlock(byte* output, Block64* block)
             private static unsafe void Base64DecodeBlockSafe(byte* outPtr, byte* srcPtr)
             {
                 Base64Decode(outPtr, Avx2.LoadVector256(srcPtr));
-                Base64Decode(outPtr + 24, Avx2.LoadVector256(srcPtr + 32));
-                Span<byte> buffer = stackalloc byte[32];
+                // should be:
+                // Span<byte> buffer = stackalloc byte[32];
+                Span<byte> buffer = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
                 fixed (byte* bufferPtr = buffer)
                 {
+                    Base64Decode(bufferPtr, Avx2.LoadVector256(srcPtr + 32));
                     // Copy only the first 12 bytes of the decoded fourth block into the output buffer, offset by 36 bytes.
                     // This step is necessary because the fourth block may not need all 16 bytes if it contains padding characters.
                     Buffer.MemoryCopy(bufferPtr, outPtr + 24, 24, 24);

src/Base64SSEUTF8.cs

@@ -200,7 +200,9 @@ private static unsafe void Base64DecodeBlockSafe(byte* outPtr, Block64* b)
                 Base64Decode(outPtr, b->chunk0);
                 Base64Decode(outPtr + 12, b->chunk1);
                 Base64Decode(outPtr + 24, b->chunk2);
-                Span<byte> buffer = stackalloc byte[16];
+                // should be:
+                //Span<byte> buffer = stackalloc byte[16];
+                Span<byte> buffer = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
 
                 // Safe memory copy for the last part of the data
                 fixed (byte* bufferStart = buffer)
@@ -258,7 +260,9 @@ private static unsafe void Base64DecodeBlockSafe(byte* outPtr, byte* srcPtr)
                 Base64Decode(outPtr + 12, Sse2.LoadVector128(srcPtr + 16));
                 Base64Decode(outPtr + 24, Sse2.LoadVector128(srcPtr + 32));
                 Vector128<byte> tempBlock = Sse2.LoadVector128(srcPtr + 48);
-                Span<byte> buffer = stackalloc byte[16];
+                // should be:
+                // Span<byte> buffer = stackalloc byte[16];
+                Span<byte> buffer = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
                 fixed (byte* bufferPtr = buffer)
                 {
                     Base64Decode(bufferPtr, tempBlock);