[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Patch] PVM low-level cleanup and optimization, good version [Was: R
|
From: |
Jose E. Marchesi |
|
Subject: |
Re: [Patch] PVM low-level cleanup and optimization, good version [Was: Re: [Patch, first version] PVM low-level cleanup and optimization] |
|
Date: |
Sun, 01 Dec 2019 13:14:59 +0100 |
|
User-agent: |
Gnus/5.13 (Gnus v5.13) Emacs/26.1 (gnu/linux) |
Hi Luca.
diff --git a/src/pkl-asm.pks b/src/pkl-asm.pks
index 5fef03d..7e07047 100644
--- a/src/pkl-asm.pks
+++ b/src/pkl-asm.pks
@@ -58,8 +58,7 @@
mgetsiz ; WCLS MCLS IOS OFF EBOUND VAL SBOUND
swap ; WCLS MCLS IOS OFF EBOUND SBOUND VAL
mgetm ; WCLS MCLS IOS OFF EBOUND SBOUND VAL
MCLS
- swap ; WCLS MCLS IOS OFF EBOUND SBOUND MCLS
VAL
- drop ; WCLS MCLS IOS OFF EBOUND SBOUND MCLS
+ nip ; WCLS MCLS IOS OFF EBOUND SBOUND MCLS
This is OK.
call ; WCLS MCLS NVAL
swap ; WCLS NVAL MCLS
msetm ; WCLS NVAL
@@ -88,7 +87,7 @@
nrot ; VAL IOS OFF VAL [WCLS]
fromr ; VAL IOS OFF VAL WCLS
call ; VAL null
- push null ; VAL null null
+ dup ; VAL null null
Likewise.
.label:
drop ; VAL (VAL|null)
drop ; VAL
@@ -156,11 +155,9 @@
over ; A B A
over ; A B A B
.loop:
- bz @type, .endloop ; ... A B
mod @type ; ... A B A%B
- rot ; ... B A%B A
- drop ; ... B A%B
- ba .loop
+ slide 1, 2 ; ... B A%B
+ bnz @type, .loop ; ... B A%B
.endloop:
drop ; A B GCD
.end
I don't really like this new version of gcd.
After pondering about it, I think I don't want to adopt slide and bulge
as PVM instructions. They are too complicated for my taste.
@@ -184,10 +181,9 @@
rot ; OFF1 OFF1M OFF2
ogetm ; OFF1 OFF1M OFF2 OFF2M
rot ; OFF1 OFF2 OFF2M OFF1M
- swap ; OFF1 OFF2 OFF1M OFF2M
add @base_type
- nip2 ; OFF1 OFF2 (OFF1M+OFF2M)
- push #unit ; OFF1 OFF2 (OFF1M+OFF2M) UNIT
+ nip2 ; OFF1 OFF2 (OFF2M+OFF1M)
+ push #unit ; OFF1 OFF2 (OFF2M+OFF1M) UNIT
mko ; OFF1 OFF2 OFFR
.end
Nice, this is OK.
@@ -208,8 +204,7 @@
ogetm ; OFF2 OFF1 OFF1M
rot ; OFF1 OFF1M OFF2
ogetm ; OFF1 OFF1M OFF2 OFF2M
- rot ; OFF1 OFF2 OFF2M OFF1M
- swap ; OFF1 OFF2 OFF1M OFF2M
+ quake ; OFF1 OFF2 OFF1M OFF2M
sub @base_type
nip2 ; OFF1 OFF2 (OFF1M+OFF2M)
push #unit ; OFF1 OFF2 (OFF1M+OFF2M) UNIT
OK.
@@ -236,8 +231,7 @@
nip2 ; OFF (OFFM*VAL)
swap ; (OFFM*VAL) OFF
ogetu ; (OFFM*VAL) OFF UNIT
- rot ; OFF UNIT (OFFM*VAL)
- swap ; OFF (OFFM*VAL) UNIT
+ quake ; OFF (OFFM*VAL) UNIT
mko ; OFF OFFR
fromr ; OFF OFFR VAL
swap ; OFF VAL OFFR
OK.
@@ -259,8 +253,7 @@
ogetm ; OFF2 OFF1 OFF1M
rot ; OFF1 OFF1M OFF2
ogetm ; OFF1 OFF1M OFF2 OFF2M
- rot ; OFF1 OFF2 OFF2M OFF1M
- swap ; OFF1 OFF2 OFF1M OFF2M
+ quake ; OFF1 OFF2 OFF1M OFF2M
div @base_type
nip2 ; OFF1 OFF2 (OFF1M/OFF2M)
.end
OK.
@@ -283,8 +276,7 @@
ogetm ; OFF2 OFF1 OFF1M
rot ; OFF1 OFF1M OFF2
ogetm ; OFF1 OFF1M OFF2 OFF2M
- rot ; OFF1 OFF2 OFF2M OFF1M
- swap ; OFF1 OFF2 OFF1M OFF2M
+ quake ; OFF1 OFF2 OFF1M OFF2M
mod @base_type
nip2 ; OFF1 OFF2 (OFF1M%OFF2M)
push #unit ; OFF1 OFF2 (OFF1M%OFF2M) UNIT
OK.
@@ -351,11 +343,9 @@
pushvar $array ; ... NULL IDX ARR
swap ; ... NULL ARR IDX
aref ; ... NULL ARR IDX EVAL
- rot ; ... NULL IDX EVAL ARR
- drop ; ... NULL IDX EVAL
+ slide 1, 2 ; ... NULL IDX EVAL
pushvar $from ; ... NULL IDX EVAL FROM
- rot ; ... NULL EVAL FROM IDX
- swap ; ... NULL EVAL IDX FROM
+ quake ; ... NULL EVAL IDX FROM
sublu
nip2 ; ... NULL EVAL (IDX-FROM)
swap ; ... NULL (IDX-FROM) EVAL
OK for the quake part.
@@ -535,8 +525,7 @@
ogetm ; OFF2 OFF1 OFF1M
rot ; OFF1 OFF1M OFF2
ogetm ; OFF1 OFF1M OFF2 OFF2M
- rot ; OFF1 OFF2 OFF2M OFF1M
- swap ; OFF1 OFF2 OFF1M OFF2M
+ quake ; OFF1 OFF2 OFF1M OFF2M
cdiv @type
nip2 ; OFF1 OFF2 (OFF1M/^OFF2M)
.end
OK.
@@ -585,19 +574,13 @@
push ulong<64>1
addlu ; SEL ELEM VAL IDX 1UL (IDX+1UL) [ARR
NRES]
nip2 ; SEL ELEM VAL NIDX [ARR NRES]
- rot ; SEL VAL NIDX ELEM [ARR NRES]
- drop ; SEL VAL NIDX [ARR NREGS]
- nrot ; NIDX SEL VAL [ARR NREGS]
- swap ; NIDX VAL SEL [ARR NRES]
- rot ; VAL SEL NIDX [ARR NRES]
+ slide 1, 2 ; SEL VAL NIDX [ARR NRES]
+ quake ; VAL SEL NIDX [ARR NRES]
ba .loop
OK for quake, not for slide.
.foundit:
tor ; SEL ELEM VAL IDX [ARR NRES]
- rot ; SEL VAL IDX ELEM [ARR NRES]
- drop ; SEL VAL IDX [ARR NRES]
- tor ; SEL VAL [ARR NRES IDX]
- swap ; VAL SEL [ARR NRES IDX]
- fromr ; VAL SEL IDX [ARR NRES]
+ slide 1, 2 ; SEL VAL IDX [ARR NRES]
+ quake ; VAL SEL IDX [ARR NRES]
dup ; VAL SEL IDX IDX [ARR NRES]
Likewise.
.endloop:
drop ; VAL SEL IDX [ARR RES]
@@ -625,10 +608,8 @@
;;; AST node with the type of the result.
.macro bconc #op2_type_size @op1_type @op2_type @res_type
- dup ; OP1 OP2 OP2
- rot ; OP2 OP2 OP1
- dup ; OP2 OP2 OP1 OP1
- rot ; OP2 OP1 OP1 OP2
+ tuck ; OP2 OP1 OP2
+ bulge 1 ; OP2 OP1 OP1 OP2
;; Convert the second operand to the result type.
nton @op2_type, @res_type ; ... OP1 OP2 OP2C
nip ; ... OP1 OP2C
OK to tuck, not to bulge.
diff --git a/src/pkl-gen.pks b/src/pkl-gen.pks
index bbc7453..9120323 100644
--- a/src/pkl-gen.pks
+++ b/src/pkl-gen.pks
@@ -94,8 +94,7 @@
ogetm ; OFF SBOUND SBOUNDM
swap ; OFF SBOUNDM SBOUND
ogetu ; OFF SBOUNDM SBOUND SBOUNDU
- swap ; OFF SBOUNDM SBOUNDU SBOUND
- drop ; OFF SOBUNDM SBOUNDU
+ nip ; OFF SOBUNDM SBOUNDU
mullu ; OFF SBOUNDM SBOUNDU (SBOUNDM*SBOUNDU)
nip2 ; OFF (SBOUNDM*SBOUNDU)
popvar $sboundm ; OFF
OK.
@@ -330,8 +329,7 @@
ogetm ; OFF SBOUND SBOUNDM
swap ; OFF SBOUNDM SBOUND
ogetu ; OFF SBOUNDM SBOUND SBOUNDU
- swap ; OFF SBOUNDM SBOUNDU SBOUND
- drop ; OFF SOBUNDM SBOUNDU
+ nip ; OFF SOBUNDM SBOUNDU
mullu ; OFF SBOUNDM SBOUNDU (SBOUNDM*SBOUNDU)
nip2 ; OFF (SBOUNDM*SBOUNDU)
popvar $sboundm ; OFF
OK.
@@ -556,8 +554,7 @@
.macro off_plus_sizeof
swap ; OFF VAL
siz ; OFF VAL ESIZ
- rot ; VAL ESIZ OFF
- swap ; VAL OFF ESIZ
+ quake ; VAL OFF ESIZ
ogetm ; VAL OFF ESIZ ESIZM
nip ; VAL OFF ESIZM
swap ; VAL ESIZM OFF
OK.
diff --git a/src/pkl-insn.def b/src/pkl-insn.def
index cd8a769..80013c2 100644
--- a/src/pkl-insn.def
+++ b/src/pkl-insn.def
@@ -55,6 +55,8 @@ PKL_DEF_INSN (PKL_INSN_DUP, "", "dup")
PKL_DEF_INSN (PKL_INSN_OVER, "", "over")
PKL_DEF_INSN (PKL_INSN_ROT, "", "rot")
PKL_DEF_INSN (PKL_INSN_NROT, "", "nrot")
+PKL_DEF_INSN (PKL_INSN_TUCK, "", "tuck")
+PKL_DEF_INSN (PKL_INSN_QUAKE, "", "quake")
OK.
PKL_DEF_INSN (PKL_INSN_SAVER, "r", "saver")
PKL_DEF_INSN (PKL_INSN_RESTORER, "r", "restorer")
PKL_DEF_INSN (PKL_INSN_TOR, "", "tor")
@@ -62,6 +64,8 @@ PKL_DEF_INSN (PKL_INSN_FROMR, "", "fromr")
PKL_DEF_INSN (PKL_INSN_ATR, "", "atr")
PKL_DEF_INSN (PKL_INSN_REVN, "n", "revn")
+PKL_DEF_INSN (PKL_INSN_BULGE, "n", "bulge")
+PKL_DEF_INSN (PKL_INSN_SLIDE, "nn", "slide")
No bulge nor slide pls.
+instruction rot () # ( A B C -- B C A )
+ code
+ JITTER_ROT_STACK ();
end
end
Ah yes, much better :)
instruction nrot () # ( A B C -- C A B )
code
- pvm_val a, b, c;
+ JITTER_MROT_STACK ();
+ end
+end
Likewise.
+instruction tuck () # ( A B -- B A B )
+ code
+ JITTER_TUCK_STACK ();
+ end
+end
OK.
+# Remove JITTER_ARGU0 non-top elements from the stack, of which the
deepest is
+# at depth JITTER_ARGU1 (where the top is at depth 0).
+instruction slide (?n 1 2 popf_printer, ?n 2 3 popf_printer)
+ code
+ JITTER_SLIDE_STACK (JITTER_ARGU0, JITTER_ARGU1);
end
end
Nope too complicated.
-instruction revn (?n popf_printer) # ( VAL{N} -- VAL{N} )
+instruction revn (?n 3 4 popf_printer) # ( VAL{N} -- VAL{N} )
code
- JITTER_REVERSE_STACK (JITTER_ARGN0);
+ JITTER_REVERSE_STACK (JITTER_ARGU0);
end
end
Hm why this? JITTER_ARGU0?
+rule swap-over-to-tuck rewrite
+ swap; over
+into
+ tuck
+end
OK.
+# A B -- A A B
+rule over-swap-to-bulge1 rewrite
+ over; swap
+into
+ bulge 1
+end
Nope.
- Re: [Patch] PVM low-level cleanup and optimization, good version [Was: Re: [Patch, first version] PVM low-level cleanup and optimization],
Jose E. Marchesi <=
- Re: [Patch] PVM low-level cleanup and optimization, good version [Was: Re: [Patch, first version] PVM low-level cleanup and optimization], Luca Saiu, 2019/12/01
- Re: [Patch] PVM low-level cleanup and optimization, good version [Was: Re: [Patch, first version] PVM low-level cleanup and optimization], Jose E. Marchesi, 2019/12/01
- Re: [Patch] PVM low-level cleanup and optimization, good version [Was: Re: [Patch, first version] PVM low-level cleanup and optimization], Luca Saiu, 2019/12/01
- Re: [Patch] PVM low-level cleanup and optimization, good version [Was: Re: [Patch, first version] PVM low-level cleanup and optimization], Jose E. Marchesi, 2019/12/01
- Re: [Patch] PVM low-level cleanup and optimization, good version [Was: Re: [Patch, first version] PVM low-level cleanup and optimization], Luca Saiu, 2019/12/01