From: Luke Kenneth Casson Leighton <lkcl@lkcl.net>
Date: Thu, 27 Apr 2023 09:54:59 +0000 (+0100)
Subject: add chacha20 introduction to REMAP Indexing
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=966e0c4179a2d9aed05583dd9a07e3442db84ad3;p=libreriscv.git

add chacha20 introduction to REMAP Indexing
---

diff --git a/openpower/sv/cookbook/chacha20.mdwn b/openpower/sv/cookbook/chacha20.mdwn
index 0a825a326..fb473eac0 100644
--- a/openpower/sv/cookbook/chacha20.mdwn
+++ b/openpower/sv/cookbook/chacha20.mdwn
@@ -130,6 +130,42 @@ Let's list the additions only:
     x9  = x9 + x14
 ```
     
+## Introduction to REMAP Indexing
+
+REMAP Indexing performs any arbitrary re-positioning of elements.
+Where normally any other Vector Processor would only be able to do a
+sequential element-level series of operations, and if re-ordering
+of the elements is required use a special re-ordering instruction,
+SVP64 can *in-place* reorder elements on *any* instruction, using
+the REMAP subsystem.
+
+Most of the REMAP systems are simple fixed-hardware Deterministic
+Schedules, but there is one that is general-purpose: Indexing.  It
+requires specifying a group of GPRs (or indices packed into GPRs)
+that are to be used as the offsets.
+
+This is a normal Simple-V operation:
+
+```
+    for i in range(VL):
+        GPR[RT+i] = OPERATION(GPR[RA+i])
+```
+
+This is what happens when REMAP is enabled with Indexing:
+
+```
+    def REMAP(SVSHAPE, i):
+        return GPR(SVSHAPE.GPR + i)
+    for i in range(VL):
+        idx_rt = REMAP(SVSHAPE0, i)
+        idx_ra = REMAP(SVSHAPE1, i)
+        GPR[RT+idx_rt] = OPERATION(GPR[RA+idx_ra])
+```
+
+In this way we can literally jump about, pretty much anywhere in
+the register file, according to a Schedule that is determined by
+the programmer.
+
 ## Introduction to Vertical-First Mode
 
 We're going to use Vertical-First mode (VF) to implement this, so we