From 64e56576b25e33ea2f3083f0107b404084cc2b74 Mon Sep 17 00:00:00 2001 From: Luke Kenneth Casson Leighton Date: Thu, 13 Apr 2023 12:32:09 +0100 Subject: [PATCH] ha! noticed that LD-ST-Update-Shifted-Postinc is far less than anticipated, there is no point shifting an Immediate for example therefore the numbers reduce down to "manageable" levels and could well be proposed for inclusion in EXT2xx --- openpower/sv/rfc/ls012.mdwn | 29 ++++++++++----------- openpower/sv/rfc/ls012/optable.csv | 41 ++++++++++-------------------- 2 files changed, 27 insertions(+), 43 deletions(-) diff --git a/openpower/sv/rfc/ls012.mdwn b/openpower/sv/rfc/ls012.mdwn index 74e63f16d..68dd582c6 100644 --- a/openpower/sv/rfc/ls012.mdwn +++ b/openpower/sv/rfc/ls012.mdwn @@ -80,8 +80,8 @@ Audio/Visual, High-Performance Compute, GPU workloads and DSP. * 4 - INT<->FP mv [[ls006]] * 19 - GPR LD/ST-PostIncrement-Update (saves hugely in hot-loops) [[ls011]] * ~12 - FPR LD/ST-PostIncrement-Update (ditto) [[ls011]] -* 19 - GPR LD/ST-Shifted-PostIncrement-Update (saves hugely in hot-loops) [[ls011]] -* ~12 - FPR LD/ST-Shifted-PostIncrement-Update (ditto) [[ls011]] +* 11 - GPR LD/ST-Shifted-PostIncrement-Update (saves hugely in hot-loops) [[ls011]] +* ~4 - FPR LD/ST-Shifted-PostIncrement-Update (ditto) [[ls011]] * 26 - GPR LD/ST-Shifted (again saves hugely in hot-loops) [[ls004]] * 11 - FPR LD/ST-Shifted (ditto) [[ls004]] * 2 - Float-Load-Immediate (always saves one LD L1/2/3 D-Cache op) [[ls002]] @@ -381,11 +381,10 @@ Power ISA up a level. Where things begin to get more than a little hairy is if both Post-Increment *and* Shifted are included. If SVP64 keeps one single bit (/pi) dedicated in the `RM.Mode` field then this -problem ges away, at the cost of reducing SVP64's effectiveness, -but at least a stunning **24** Primary Opcodes (there are only -32 in EXT2xx) would not disappear overnight. -Mostly the Post-Increment-and-Shifted set are included to illustrate -the options and have a formal record of the evluation, for Due Diligence. +problem ges away, at the cost of reducing SVP64's effectiveness. +However again, given that even the Shifted-Post-Increment +instructions are all 9-bit XO it is not outside the realm of +possibility to include them in EXT2xx. ## Shift-and-add (and LD/ST Indexed-Shift) @@ -401,12 +400,6 @@ Being a 10-bit XO it would be somewhat punitive to place these in EXT2xx when their whole purpose and value is to reduce binary size in Address offset computation, thus they are best placed in EXT0xx. -Also included because it is important to see the quantity of instructions: -LD/ST-Indexed-Shifted. Across Update variants, Byte-reverse variants, -Arithmetic and FP, the total is a slightly-eye-watering **37** -instructions, only ameliorated by the fact that they are all 9-bit XO. -When it comes to Shifted-Postincrement the number of Primary Opcodes -needed in EXT2xx comes to 24 which is most of them. The upside as far as adding them is concerned is that existing hardware will already have amalgamated pipelines with very few actual back-end (Micro-Coded) internal operations (likely just two: one load, one store). @@ -416,9 +409,13 @@ is not hard. *(Readers unfamiliar with Micro-coding should look at the Microwatt VHDL source code)* -When it comes to LD/ST-Shifted-Postincrement the sheer number particularly -Primary Opcodes needed in EXT2xx makes for a compelling case to prioritise -Shift-and-Add. +Also included because it is important to see the quantity of instructions: +LD/ST-Indexed-Shifted. Across Update variants, Byte-reverse variants, +Arithmetic and FP, the total is a slightly-eye-watering **37** +instructions, only ameliorated by the fact that they are all 9-bit XO. +Even when adding the Post-Increment-Shifted group it is still only +52 9-bit XO instructions, which is not unreasonable to consider (in +EXT2xx). \newpage{} diff --git a/openpower/sv/rfc/ls012/optable.csv b/openpower/sv/rfc/ls012/optable.csv index dfcf382f4..c98a6266a 100644 --- a/openpower/sv/rfc/ls012/optable.csv +++ b/openpower/sv/rfc/ls012/optable.csv @@ -29,34 +29,21 @@ stfsup, ls011, high, PO, yes, EXT2xx, no, isa/pifixedstore, 2R1W stfdupx, ls011, high, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W stfsupx, ls011, high, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W # LD/ST-Shifted-Postincrement -lbzusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedload, 1R2W -lbzuspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -lhzusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedload, 1R2W -lhzuspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -lhausp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedload, 1R2W -lhauspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -lwzusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedload, 1R2W -lwzuspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -lwauspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -ldusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedload, 1R2W -lduspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -stbusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedstore, 2R1W -stbuspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W -sthusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedstore, 2R1W -sthuspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W -stwusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedstore, 2R1W -stwuspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W -stdusp, ls011, low, PO, yes, EXT2xx, no, isa/pifixedstore, 2R1W -stduspx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W +lbzuspx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +lhzuspx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +lhauspx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +lwzuspx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +lwauspx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +lduspx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +stbuspx, ls011, med, 10, yes, EXT2xx, no, ls011, 3R1W +sthuspx, ls011, med, 10, yes, EXT2xx, no, ls011, 3R1W +stwuspx, ls011, med, 10, yes, EXT2xx, no, ls011, 3R1W +stduspx, ls011, med, 10, yes, EXT2xx, no, ls011, 3R1W # FP LD/ST-Shifted-Postincrement -lfdups, ls011, low, PO, yes, EXT2xx, no, isa/pifixedload, 1R2W -lfsups, ls011, low, PO, yes, EXT2xx, no, isa/pifixedload, 1R2W -lfdupsx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -lsdupsx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedload, 2R2W -stfdups, ls011, low, PO, yes, EXT2xx, no, isa/pifixedstore, 2R1W -stfsups, ls011, low, PO, yes, EXT2xx, no, isa/pifixedstore, 2R1W -stfdupsx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W -stfsupsx, ls011, med, 10, yes, EXT2xx, no, isa/pifixedstore, 3R1W +lfdupsx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +lsdupsx, ls011, med, 10, yes, EXT2xx, no, ls011, 2R2W +stfdupsx, ls011, med, 10, yes, EXT2xx, no, ls011, 3R1W +stfsupsx, ls011, med, 10, yes, EXT2xx, no, ls011, 3R1W # LD/ST-Index-Shifted (w/Update) lbzsx, ls004, high, 9, yes, EXT0xx, no, ls004, 2R1W lbzusx, ls004, high, 9, yes, EXT0xx, no, ls004, 2R2W -- 2.30.2