From c7bfd1e3e00f11b9cc4ab24324ce2e4d783c3bbd Mon Sep 17 00:00:00 2001
From: Cliff Click <cliffc@acm.org>
Date: Wed, 4 Sep 2024 10:15:45 -0700
Subject: Add README

---
 chapter14/README.md | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 65 insertions(+), 1 deletion(-)

(limited to 'chapter14')

diff --git a/chapter14/README.md b/chapter14/README.md
index 5fc41b3b..a42c37e2 100644
--- a/chapter14/README.md
+++ b/chapter14/README.md
@@ -1,3 +1,67 @@
-# Chapter 14: Arrays
+# Chapter 14: Narrow types
+
+## Narrow or sub-word types
+
+As a prelude to arrays, especially arrays of bytes common in all networking
+codes, Simple needs some way to manipulate sub-word integer types.
+Computations are always done in 64 bits for both integer and float ops, but
+will be truncated (rounded) on a variable or field assignment, and (sign)
+extended on a load.
+
+```java
+i8 x = 123456789;
+if( x != 21 ) 
+  return false;
+return true;
+```
+
+The variable `x` is limited to 8 bits signed.  Assigning the vary large value
+`123456789` (hex `0x75BCD15`) simply truncates to 8 bits yielding `21 (0x15)`.
+The value is sign-extended on a load.
+
+```java
+i16 x = 123456789;
+if( x != -13035 ) 
+  return false;
+return true;
+```
+
+The variable `x` is limited to 16 bits signed.  Assigning the vary large value
+`0x75BCD15` simply truncates to 16 bits yielding `0xCD15`.  The value is
+sign-extended on a load yielding `-13035 (0xFFFF FFFF FFFF CD15)`.
+
+```java
+u16 x = 123456789;
+if( x != 52501 ) 
+  return false;
+return true;
+```
+
+The variable `x` is limited to 16 bits unsigned.  Assigning the vary large value
+`0x75BCD15` simply truncates to 16 bits yielding `0xCD15`.  The value is
+zero-extended on a load yielding `52501 (0x0000 0000 0000 CD15)`.
+
+Similarly, floats can be limited to IEEE754 32-bit standard format and will be
+rounded when assigned, and inflated to 64-bits when loaded.
+
+
+## Type Implementation
+
+The TypeInteger class is reworked to support a full range of min/max values.
+At this time, only some power-of-2 sized ranges are exposed but the optimizer
+internally supports all ranges.  The MEET operation takes the min-of-mins and
+max-of-maxes.  The DUAL operation inverts the min and max.
+
+The TypeFloat class is also reworked to support 32-bit and 64-bit sizes.
+
+
+## Nodes
+
+Added this Chapter are some bit-manipulation Nodes representing common hardware
+ops to mask, shift, sign-extend and round bits.  The truncation and extension
+logic uses these.  They are otherwise very similar to the previous
+`Add/Sub/Mul/Div` Nodes but with slightly different constant math and
+slightly different `idealize()` calls.
+
 
 
-- 
cgit v1.2.3