[libc] Add a smaller b36_char_to_int (#180841)

For ASCII systems, b36_char_to_int gets compiled into a jump table. That
jump table ends up being pretty large because it covers the range from
'0' (48) to 'z' (122). On size-constrained systems that can assume
ASCII, this patch provides a new flag: LIBC_CONF_CTYPE_SMALLER_ASCII
that forces a smaller implementation that doesn't compile into a jump
table.

libc/cmake/modules/LLVMLibCCompileOptionRules.cmake

@@ -147,6 +147,10 @@ function(_get_compile_options_from_config output_var)
     endif()
   endif()
 
+  if(LIBC_CONF_CTYPE_SMALLER_ASCII)
+    list(APPEND config_options "-DLIBC_COPT_CTYPE_SMALLER_ASCII")
+  endif()
+
   if(LIBC_CONF_PRINTF_DISABLE_WIDE)
     list(APPEND config_options "-DLIBC_COPT_PRINTF_DISABLE_WIDE")
   endif()

libc/config/baremetal/config.json

@@ -70,5 +70,10 @@
     "LIBC_CONF_ENABLE_STRONG_STACK_PROTECTOR": {
       "value": false
     }
+  },
+  "ctype": {
+    "LIBC_CONF_CTYPE_SMALLER_ASCII": {
+      "value": true
+    }
   }
 }

libc/config/config.json

@@ -99,6 +99,12 @@
       "doc": "Inserts prefetch for write instructions (PREFETCHW) for memset on x86 to recover performance when hardware prefetcher is disabled."
     }
   },
+  "ctype": {
+    "LIBC_CONF_CTYPE_SMALLER_ASCII": {
+      "value": false,
+      "doc": "Shrinks b36_char_to_int by assuming the character encoding is ASCII."
+    }
+  },
   "codegen": {
     "LIBC_CONF_KEEP_FRAME_POINTER": {
       "value": true,

libc/docs/configure.rst

@@ -28,6 +28,8 @@ to learn about the defaults for your platform and target.
 * **"codegen" options**
     - ``LIBC_CONF_ENABLE_STRONG_STACK_PROTECTOR``: Enable -fstack-protector-strong to defend against stack smashing attack.
     - ``LIBC_CONF_KEEP_FRAME_POINTER``: Keep frame pointer in functions for better debugging experience.
+* **"ctype" options**
+    - ``LIBC_CONF_CTYPE_SMALLER_ASCII``: Shrinks b36_char_to_int by assuming the character encoding is ASCII.
 * **"errno" options**
     - ``LIBC_CONF_ERRNO_MODE``: The implementation used for errno, acceptable values are LIBC_ERRNO_MODE_DEFAULT, LIBC_ERRNO_MODE_UNDEFINED, LIBC_ERRNO_MODE_THREAD_LOCAL, LIBC_ERRNO_MODE_SHARED, LIBC_ERRNO_MODE_EXTERNAL, and LIBC_ERRNO_MODE_SYSTEM_INLINE.
 * **"fenv" options**

libc/src/__support/ctype_utils.h

@@ -371,6 +371,7 @@ LIBC_INLINE constexpr bool isalnum(char ch) {
   }
 }
 
+#ifndef LIBC_COPT_CTYPE_SMALLER_ASCII
 LIBC_INLINE constexpr int b36_char_to_int(char ch) {
   switch (ch) {
   case '0':
@@ -475,6 +476,19 @@ LIBC_INLINE constexpr int b36_char_to_int(char ch) {
     return 0;
   }
 }
+#else  // LIBC_COPT_SMALL_ASCII_CTYPE
+// This version assumes ASCII for the tolower, but generates smaller code since
+// the switch version of this function ends up with a table. This should only be
+// used when the target is known to be ASCII.
+LIBC_INLINE constexpr int b36_char_to_int(char ch) {
+  if (ch >= '0' && ch <= '9')
+    return ch - '0';
+  char ch_unsafe_lower = ch | 32;
+  if (ch_unsafe_lower >= 'a' && ch_unsafe_lower <= 'z')
+    return ch_unsafe_lower - 'a' + 10;
+  return 0;
+}
+#endif // LIBC_COPT_SMALL_ASCII_CTYPE
 
 LIBC_INLINE constexpr char int_to_b36_char(int num) {
   // Can't actually use LIBC_ASSERT here because it depends on integer_to_string