Merge branch 'main' into base64-encode-wrapcol

python · serhiy-storchaka · Jan 14, 2026 · Dec 26, 2025 · Dec 27, 2025 · Dec 27, 2025
commit bb5ff5755617da17ffe3e828b893c0e5fe4bc277
@@ -881,6 +881,13 @@ Optimizations
   (Contributed by Chris Eibl, Ken Jin, and Brandt Bucher in :gh:`143068`.
   Special thanks to the MSVC team including Hulon Jenkins.)
 
+base64 & binascii
+-----------------
+
+* CPython's underlying base64 implementation now encodes 2x faster and decodes 3x
+  faster thanks to simple CPU pipelining optimizations.
+  (Contributed by Gregory P. Smith and Serhiy Storchaka in :gh:`143262`.)
+
 csv
 ---
 

diff --git a/Modules/binascii.c b/Modules/binascii.c
@@ -659,9 +659,6 @@ binascii_b2a_base64_impl(PyObject *module, Py_buffer *data, size_t wrapcol,
                          int newline)
 /*[clinic end generated code: output=2edc7311a9515eac input=2ee4214e6d489e2e]*/
 {
-    int leftbits = 0;
-    unsigned int leftchar = 0;
-
     const unsigned char *bin_data = data->buf;
     Py_ssize_t bin_len = data->len;
     assert(bin_len >= 0);
@@ -703,18 +700,24 @@ binascii_b2a_base64_impl(PyObject *module, Py_buffer *data, size_t wrapcol,
     ascii_data += (fast_bytes / 3) * 4;
     bin_len -= fast_bytes;
 
-        /* See if there are 6-bit groups ready */
-        while ( leftbits >= 6 ) {
-            unsigned char this_ch = (leftchar >> (leftbits-6)) & 0x3f;
-            leftbits -= 6;
-            *ascii_data++ = table_b2a_base64[this_ch];
-        }
-    }
-    if ( leftbits == 2 ) {
-        *ascii_data++ = table_b2a_base64[(leftchar&3) << 4];
+    /* Handle remaining 0-2 bytes */
+    if (bin_len == 1) {
+        /* 1 byte remaining: produces 2 base64 chars + 2 padding */
+        unsigned int val = bin_data[0];
+        *ascii_data++ = table_b2a_base64[(val >> 2) & 0x3f];
+        *ascii_data++ = table_b2a_base64[(val << 4) & 0x3f];
         *ascii_data++ = BASE64_PAD;
         *ascii_data++ = BASE64_PAD;
     }
+    else if (bin_len == 2) {
+        /* 2 bytes remaining: produces 3 base64 chars + 1 padding */
+        unsigned int val = ((unsigned int)bin_data[0] << 8) | bin_data[1];
+        *ascii_data++ = table_b2a_base64[(val >> 10) & 0x3f];
+        *ascii_data++ = table_b2a_base64[(val >> 4) & 0x3f];
+        *ascii_data++ = table_b2a_base64[(val << 2) & 0x3f];
+        *ascii_data++ = BASE64_PAD;
+    }
+
     if (wrapcol) {
         unsigned char *start = PyBytesWriter_GetData(writer);
         ascii_data = start + wraplines(start, ascii_data - start, wrapcol);