From 273f688b8a2b04a569b49f42610255d8abb159a0 Mon Sep 17 00:00:00 2001 From: Luke Kenneth Casson Leighton Date: Mon, 25 Jul 2022 19:32:29 +0100 Subject: [PATCH] add warm words explaining Scalar part III --- openpower/simple_v_spec.tex | 41 +++++++++++++++++++++++++++++++++++++ 1 file changed, 41 insertions(+) diff --git a/openpower/simple_v_spec.tex b/openpower/simple_v_spec.tex index f5c4cbb8f..b3fbc12a0 100644 --- a/openpower/simple_v_spec.tex +++ b/openpower/simple_v_spec.tex @@ -208,6 +208,7 @@ For questions, comments, and clarification, please contact the following: \tableofcontents +% Part II \part{Scalable Vectors for the Power ISA} @@ -268,8 +269,47 @@ For questions, comments, and clarification, please contact the following: \end{appendices} +% Part III \part{Scalar Instructions} +\chapter*{Preamble}{} + +As explained in the Simple-V introduction +these are all intentionally and specifically Scalar instructions. +They have with almost no exceptions been specifically crafted to +have a justification for their inclusion in the Power ISA as Scalar +instructions purely on their own merit. + +\begin{itemize} + \item The biginteger multiply-and-add instruction is similar + to Intel's mulx in that it produces a pair of results. + \item Javascript(tm) rounding is present in ARM as fjcvtzs + and would save an astounding 35 instructions with 5 branches. + \item Whilst there exist CR bit manipulation and copying + instructions there are no CR Field maniulation instructions, + putting pressure on GPRs if several CR fits need to be analysed. +\end{itemize} + +All of these have nothing to do with Simple-V at all. + +That said: by a wonderful coincidence, should they be included, then +Simple-V's capabilities increase significantly. For example the CRweird +instructions combined with the bitmanip instructions, alongside +Vectorised Rc=1 turn CR Fields into +extremely powerful Predicate masks. + +The clean and clear separation between Vectorisation Prefix and Scalar +Suffix is what makes it possible for both Scalar-only and Scalable-Vectors +to benefit. It also makes proposal much easier, as there is no +inter-dependence. + +It is however important to note that the rationale for these instructions +comes from a more general-purpose moderen computing paradigm that is +outside of IBM's much more focussed and specialist traditional customer +base. We deeply respect IBM's curator role of the Power ISA of the past 25 +years as much as we appreciate their courage in transferring that role +to the OpenPOWER Foundation ISA Working Group. + \chapter{SV Vector ops}\hypertarget{svux2fvector_ops}{} \input{tex_out/vector_ops.tex} \chapter{CR Weird ops}\hypertarget{svux2fcr_int_predication}{} @@ -303,6 +343,7 @@ For questions, comments, and clarification, please contact the following: \input{tex_out/pseudocode_svfixedarith.tex} \end{appendices} +% Part IV \part{Scalar Power ISA pseudocode} \backmatter % temporary fix for too many appenfices %\setcounter{chapter}{0} -- 2.30.2