Use mpz_sizeinbase(gmpnum_base, 2) - 1

That is equivalent to `floor(log2(Z_LVAL_P(base_arg)))` and as such
more suitable[1].

Since this now fails gmp_pow_fpe.phpt, we use an int there.

[1] <php#16884 (comment)>

Author: cmb69

ext/gmp/gmp.c

@@ -1363,7 +1363,7 @@ ZEND_FUNCTION(gmp_pow)
 		mpz_ptr gmpnum_base;
 		FETCH_GMP_ZVAL(gmpnum_base, base_arg, temp_base, 1);
 		INIT_GMP_RETVAL(gmpnum_result);
-		if ((mpz_sizeinbase(gmpnum_base, 2) * exp) > max_bits) {
+		if (((mpz_sizeinbase(gmpnum_base, 2) - 1) * exp) > max_bits) {
 			FREE_GMP_TEMP(temp_base);
 			zend_value_error("base and exponent overflow");
 			RETURN_THROWS();

ext/gmp/tests/gmp_pow_fpe.phpt

@@ -23,7 +23,7 @@ try {
 	echo $e->getMessage() . PHP_EOL;
 }
 try {
-    gmp_pow(gmp_init(PHP_INT_MAX), 256);
+    gmp_pow(PHP_INT_MAX, 256);
 } catch (\ValueError $e) {
 	echo $e->getMessage();
 }