From ebf3891935d9d51d402cafca770700a34f797f2f Mon Sep 17 00:00:00 2001 From: lkcl Date: Sat, 24 Sep 2022 21:29:37 +0100 Subject: [PATCH] --- openpower/sv/overview/discussion.mdwn | 21 +++++++++++++++++---- 1 file changed, 17 insertions(+), 4 deletions(-) diff --git a/openpower/sv/overview/discussion.mdwn b/openpower/sv/overview/discussion.mdwn index b4287ba28..d9d001235 100644 --- a/openpower/sv/overview/discussion.mdwn +++ b/openpower/sv/overview/discussion.mdwn @@ -16,14 +16,18 @@ possibility of MSB0 numbering being potentially applicable to: * register numbers * element numbers +In other words, when communicating in a Specification there are **four** +possible areas for confusion. + Then there is the completely separate issue of whether: * Memory should be the sole location where LE/BE byteswapping should - be applicable, Architecturally -* Both the register file *and* Memory are, both, Architecturally, - LE/BE reorderable and thus **direct and explicit** control over + be applicable, Architecturally. +* Both the register file *and* Memory should both be, Architecturally, + LE/BE reorderable with the implication that + **direct and explicit** control over reordering over Memory **and the bytes inside regfile data** - is provided at the ISA Level. + needs provided at the ISA Level. Simple-V has made the decision that providing programmers with explicit control over the byte-ordering when data is transferred in and out of @@ -37,6 +41,15 @@ that how the Hardware is implemented is entirely down to the implementor, but that as far as *software* running on that hardware is concerned, "Architecturally" everything appears as described. +In other words once loaded, software may be written that +considers arithmetic values in a uniform environment regardless +of whether `MSR.BE`, which is considered to affect **Memory only**, +is set or unset. If byteswapping is required it may be performed +explicitly, by either loading to/from Memory or by performing Simple-V +byte-level element-width override "Reverse Gear" element walking, +or using Matrix REMAP with Dimensional Inversion, or any number of +methods. + In the overview page is this: | byte0 | byte1 | byte2 | byte3 | byte4 | byte5 | byte6 | byte7 | -- 2.30.2