powerpc backend fixes, remove some redundant vops
Slava Pestov
parent
5866613d3e
commit
976cb891df
27
CHANGES.html
27
CHANGES.html
|
|
@ -14,23 +14,44 @@
|
|||
<li>New basic block optimizer performs more aggressive dead load and store elimination.</li>
|
||||
<li>Stack shuffles are compiled more efficiently.</li>
|
||||
<li>Pushing literals on either side of a stack shuffle is now compiled more efficiently.</li>
|
||||
<li>Tail-recursive inlined words are compiled in a new way, saving a few instructions.</li>
|
||||
</ul>
|
||||
|
||||
</li>
|
||||
|
||||
<li>Sequences:
|
||||
<li>Collections:
|
||||
|
||||
<ul>
|
||||
<li>Faster <code>map</code>, <code>2each</code> and <code>2map</code>.</li>
|
||||
<li>Arrays are now better supported and should be used instead of vectors where resizing is not desired.</li>
|
||||
<li>Some new sequence words that do not bounds check: <code>nth-unsafe</code> and <code>set-nth-unsafe</code>. These should only be used in special circumstances, such as inner loops (<code>each</code>, <code>map</code> and so on use them).</li>
|
||||
<li>Hashtables did not obey the rule that equal objects must have equal hashcodes, so using hashtables as hashtable keys did not work.</li>
|
||||
</li>
|
||||
|
||||
<li>Everything else:
|
||||
|
||||
<ul>
|
||||
<li>Easier exception handling. The <code>cleanup ( try cleanup -- )</code> word encapsulates the following idiom:
|
||||
<pre>
|
||||
[ A ] [ B rethrow ] catch
|
||||
[ A ] [ B ] cleanup
|
||||
</pre>
|
||||
The <code>recover ( try recover -- )</code> word encapsulates the following idiom:
|
||||
<pre>
|
||||
[ A ] [ [ B ] when* ] catch
|
||||
[ A ] [ B ] recover
|
||||
</pre>
|
||||
The <code>catch ( try -- error/f )</code> word no longer takes a quotation that receives the error caught; rather, it just pushes the error that was caught on the stack. That is, the old way:
|
||||
<pre>
|
||||
[ A ] [ B ] catch
|
||||
</pre>
|
||||
Becomes:
|
||||
<pre>
|
||||
[ A ] catch B
|
||||
</pre>
|
||||
However, most uses of <code>catch</code> can be replaced by <code>cleanup</code> and <code>recover</code>.</li>
|
||||
<li>The distinct <code>t</code> type is gone. Now, the <code>t</code> object is just a symbol.</li>
|
||||
<li>Hashtables did not obey the rule that equal objects must have equal hashcodes, so using hashtables as hashtable keys did not work.</li>
|
||||
<li>A new <code>with-server ( port ident quot -- )</code> combinator takes care of listening on a network socket, logging client connections, spawning client handler threads, and error handling. The quotation is called for each client, in a new scope with the client socket as the default stream.</li>
|
||||
<li>New <code>1+</code>, <code>1-</code> words are a bit faster than <code>1 +</code> and <code>1 -</code> since there is one less generic dispatch involved.</li>
|
||||
</ul>
|
||||
|
||||
</ul>
|
||||
|
|
|
|||
|
|
@ -1,6 +1,21 @@
|
|||
- nodes: lazily create history, class/literal map hashes
|
||||
- 5 car gives wrong error
|
||||
- compile interruption checks
|
||||
- check that set-datastack and set-callstack compile correctly in the
|
||||
face of optimization
|
||||
- [ 2cdr < ] sort -- sig11
|
||||
- slice: if sequence or seq start is changed, abstraction violation
|
||||
- array>tuple: check length and class
|
||||
- code walker & exceptions
|
||||
- floating point intrinsics
|
||||
- fix fixnum/mod overflow on PowerPC
|
||||
- intrinsic char-slot set-char-slot
|
||||
- split: return vectors
|
||||
- vectors: ensure its ok with bignum indices
|
||||
- nth-unsafe and set-nth-unsafe; ensure it works with bignum indices
|
||||
- set-path: iterative
|
||||
- parse-command-line: no unswons of cli args
|
||||
- investigate if rehashing on startup is really necessary
|
||||
- remove word transfer hack in bootstrap
|
||||
|
||||
+ ui:
|
||||
|
||||
|
|
@ -35,7 +50,6 @@
|
|||
http://clozure.com/cgi-bin/viewcvs.cgi/ccl/lisp-kernel/lisp-exceptions.c?rev=1.9&content-type=text/vnd.viewcvs-markup
|
||||
|
||||
- http keep alive, and range get
|
||||
- code walker & exceptions
|
||||
|
||||
+ ffi:
|
||||
|
||||
|
|
@ -55,10 +69,7 @@
|
|||
- [ ] [ throw ] ifte ==> should raise 'unbalanced branches' error
|
||||
- declare slot types for built-ins
|
||||
- remove dead code after a 'throw'
|
||||
- floating point intrinsics
|
||||
- flushing optimization
|
||||
- fix fixnum/mod overflow on PowerPC
|
||||
- intrinsic char-slot set-char-slot
|
||||
- [ [ dup call ] dup call ] infer hangs
|
||||
- the invalid recursion form case needs to be fixed, for inlines too
|
||||
- recursion is iffy; if the stack at the recursive call doesn't match
|
||||
|
|
@ -66,11 +77,8 @@
|
|||
|
||||
+ sequences:
|
||||
|
||||
- split: return vectors
|
||||
- specialized arrays
|
||||
- instances: do not use make-list
|
||||
- vectors: ensure its ok with bignum indices
|
||||
- nth-unsafe and set-nth-unsafe; ensure it works with bignum indices
|
||||
- >c/c>: vector stack
|
||||
- search: slow
|
||||
- vectorize >n, n>, (get)
|
||||
|
|
@ -91,10 +99,6 @@
|
|||
- there is a problem with hashcodes of words and bootstrapping
|
||||
- delegating generic words with a non-standard picker
|
||||
- code gc
|
||||
- set-path: iterative
|
||||
- parse-command-line: no unswons of cli args
|
||||
- investigate if rehashing on startup is really necessary
|
||||
- remove word transfer hack in bootstrap
|
||||
|
||||
+ i/o:
|
||||
|
||||
|
|
|
|||
|
|
@ -30,19 +30,17 @@ M: %inc-d simplify-stack* ( vop -- ) d-height accum-height ;
|
|||
|
||||
M: %inc-r simplify-stack* ( vop -- ) r-height accum-height ;
|
||||
|
||||
: update-ds ( vop -- )
|
||||
GENERIC: update-loc ( loc -- )
|
||||
|
||||
M: ds-loc update-loc
|
||||
dup ds-loc-n d-height get - swap set-ds-loc-n ;
|
||||
|
||||
: update-cs ( vop -- )
|
||||
M: cs-loc update-loc
|
||||
dup cs-loc-n r-height get - swap set-cs-loc-n ;
|
||||
|
||||
M: %peek-d simplify-stack* ( vop -- ) 0 vop-in update-ds ;
|
||||
M: %peek simplify-stack* ( vop -- ) 0 vop-in update-loc ;
|
||||
|
||||
M: %peek-r simplify-stack* ( vop -- ) 0 vop-in update-cs ;
|
||||
|
||||
M: %replace-d simplify-stack* ( vop -- ) 0 vop-out update-ds ;
|
||||
|
||||
M: %replace-r simplify-stack* ( vop -- ) 0 vop-out update-cs ;
|
||||
M: %replace simplify-stack* ( vop -- ) 0 vop-out update-loc ;
|
||||
|
||||
: simplify-stack ( block -- )
|
||||
#! Combine all %inc-d/%inc-r into two final ones.
|
||||
|
|
@ -106,14 +104,10 @@ M: %inc-r trim-dead* ( tail vop -- ) simplify-inc drop ;
|
|||
: redundant-peek? ( vop -- ? )
|
||||
dup 0 vop-in swap 0 vop-out vreg-contents get hash = ;
|
||||
|
||||
: ?dead-peek ( tail vop -- )
|
||||
M: %peek trim-dead* ( tail vop -- )
|
||||
dup redundant-peek? >r tuck live-load? not r> or
|
||||
[ dup remember-peek dup , ] unless drop ;
|
||||
|
||||
M: %peek-d trim-dead* ( tail vop -- ) ?dead-peek ;
|
||||
|
||||
M: %peek-r trim-dead* ( tail vop -- ) ?dead-peek ;
|
||||
|
||||
: redundant-replace? ( vop -- ? )
|
||||
dup 0 vop-out swap 0 vop-in vreg-contents get hash = ;
|
||||
|
||||
|
|
@ -128,14 +122,10 @@ M: %peek-r trim-dead* ( tail vop -- ) ?dead-peek ;
|
|||
dup 0 vop-out forget-stack-loc
|
||||
dup 0 vop-out swap 0 vop-in vreg-contents get set-hash ;
|
||||
|
||||
: ?dead-replace ( tail vop -- )
|
||||
M: %replace trim-dead* ( tail vop -- )
|
||||
dup redundant-replace? >r tuck live-load? not r> or
|
||||
[ dup remember-replace dup , ] unless drop ;
|
||||
|
||||
M: %replace-d trim-dead* ( tail vop -- ) ?dead-replace ;
|
||||
|
||||
M: %replace-r trim-dead* ( tail vop -- ) ?dead-replace ;
|
||||
|
||||
: ?dead-literal dup forget-vregs tuck live-load? ?, ;
|
||||
|
||||
M: %immediate trim-dead* ( tail vop -- ) ?dead-literal ;
|
||||
|
|
|
|||
|
|
@ -72,7 +72,7 @@ M: %untag generate-node ( vop -- )
|
|||
|
||||
M: %dispatch generate-node ( vop -- )
|
||||
0 <vreg> check-src
|
||||
3 3 2 SLWI
|
||||
3 3 1 SRAWI
|
||||
! The value 24 is a magic number. It is the length of the
|
||||
! instruction sequence that follows to be generated.
|
||||
0 1 rel-address compiled-offset 24 + 4 LOAD32
|
||||
|
|
@ -86,9 +86,11 @@ M: %type generate-node ( vop -- )
|
|||
<label> "f" set
|
||||
<label> "end" set
|
||||
! Get the tag
|
||||
3 4 tag-mask ANDI
|
||||
3 5 tag-mask ANDI
|
||||
! Tag the tag
|
||||
5 4 tag-fixnum
|
||||
! Compare with object tag number (3).
|
||||
0 4 object-tag CMPI
|
||||
0 5 object-tag CMPI
|
||||
! Jump if the object doesn't store type info in its header
|
||||
"end" get BNE
|
||||
! It does store type info in its header
|
||||
|
|
@ -97,13 +99,14 @@ M: %type generate-node ( vop -- )
|
|||
"f" get BEQ
|
||||
! The pointer is not equal to 3. Load the object header.
|
||||
4 3 object-tag neg LWZ
|
||||
4 4 3 SRAWI
|
||||
4 4 untag
|
||||
"end" get B
|
||||
"f" get save-xt
|
||||
! The pointer is equal to 3. Load F_TYPE (9).
|
||||
f type 4 LI
|
||||
f type tag-bits shift 4 LI
|
||||
"end" get save-xt
|
||||
3 4 MR ;
|
||||
|
||||
M: %tag generate-node ( vop -- )
|
||||
dup 0 vop-in v>operand swap 0 vop-out v>operand tag-mask ANDI ;
|
||||
dup 0 vop-in v>operand swap 0 vop-out v>operand
|
||||
[ tag-mask ANDI ] keep dup tag-fixnum ;
|
||||
|
|
|
|||
|
|
@ -3,8 +3,10 @@
|
|||
IN: compiler-backend
|
||||
USING: assembler compiler errors kernel math memory words ;
|
||||
|
||||
: ds-op cell * neg 14 swap ;
|
||||
: cs-op cell * neg 15 swap ;
|
||||
GENERIC: loc>operand
|
||||
|
||||
M: ds-loc loc>operand ds-loc-n cell * neg 14 swap ;
|
||||
M: cs-loc loc>operand cs-loc-n cell * neg 15 swap ;
|
||||
|
||||
M: %immediate generate-node ( vop -- )
|
||||
dup 0 vop-in address swap 0 vop-out v>operand LOAD ;
|
||||
|
|
@ -15,20 +17,14 @@ M: %immediate generate-node ( vop -- )
|
|||
M: %indirect generate-node ( vop -- )
|
||||
dup 0 vop-out v>operand swap 0 vop-in load-indirect ;
|
||||
|
||||
M: %peek-d generate-node ( vop -- )
|
||||
dup 0 vop-out v>operand swap 0 vop-in ds-op LWZ ;
|
||||
M: %peek generate-node ( vop -- )
|
||||
dup 0 vop-out v>operand swap 0 vop-in loc>operand LWZ ;
|
||||
|
||||
M: %replace-d generate-node ( vop -- )
|
||||
dup 1 vop-in v>operand swap 0 vop-in ds-op STW ;
|
||||
M: %replace generate-node ( vop -- )
|
||||
dup 0 vop-in v>operand swap 0 vop-out loc>operand STW ;
|
||||
|
||||
M: %inc-d generate-node ( vop -- )
|
||||
14 14 rot 0 vop-in cell * ADDI ;
|
||||
|
||||
M: %inc-r generate-node ( vop -- )
|
||||
15 15 rot 0 vop-in cell * ADDI ;
|
||||
|
||||
M: %peek-r generate-node ( vop -- )
|
||||
dup 0 vop-out v>operand swap 0 vop-in cs-op LWZ ;
|
||||
|
||||
M: %replace-r generate-node ( vop -- )
|
||||
dup 1 vop-in v>operand swap 0 vop-in cs-op STW ;
|
||||
|
|
|
|||
|
|
@ -119,21 +119,27 @@ C: %end-dispatch make-vop ;
|
|||
: %end-dispatch empty-vop <%end-dispatch> ;
|
||||
|
||||
! stack operations
|
||||
TUPLE: %peek-d ;
|
||||
C: %peek-d make-vop ;
|
||||
TUPLE: %peek ;
|
||||
C: %peek make-vop ;
|
||||
|
||||
M: %peek basic-block? drop t ;
|
||||
|
||||
: %peek-d ( vreg n -- vop )
|
||||
<ds-loc> swap <vreg> src/dest-vop <%peek-d> ;
|
||||
<ds-loc> swap <vreg> src/dest-vop <%peek> ;
|
||||
|
||||
M: %peek-d basic-block? drop t ;
|
||||
: %peek-r ( vreg n -- vop )
|
||||
<cs-loc> swap <vreg> src/dest-vop <%peek> ;
|
||||
|
||||
TUPLE: %replace-d ;
|
||||
C: %replace-d make-vop ;
|
||||
TUPLE: %replace ;
|
||||
C: %replace make-vop ;
|
||||
|
||||
M: %replace basic-block? drop t ;
|
||||
|
||||
: %replace-d ( vreg n -- vop )
|
||||
<ds-loc> src/dest-vop <%replace-d> ;
|
||||
<ds-loc> src/dest-vop <%replace> ;
|
||||
|
||||
M: %replace-d basic-block? drop t ;
|
||||
: %replace-r ( vreg n -- vop )
|
||||
<cs-loc> src/dest-vop <%replace> ;
|
||||
|
||||
TUPLE: %inc-d ;
|
||||
C: %inc-d make-vop ;
|
||||
|
|
@ -141,30 +147,6 @@ C: %inc-d make-vop ;
|
|||
|
||||
M: %inc-d basic-block? drop t ;
|
||||
|
||||
TUPLE: %immediate ;
|
||||
C: %immediate make-vop ;
|
||||
|
||||
: %immediate ( vreg obj -- vop )
|
||||
swap <vreg> src/dest-vop <%immediate> ;
|
||||
|
||||
M: %immediate basic-block? drop t ;
|
||||
|
||||
TUPLE: %peek-r ;
|
||||
C: %peek-r make-vop ;
|
||||
|
||||
: %peek-r ( vreg n -- vop )
|
||||
<cs-loc> swap <vreg> src/dest-vop <%peek-r> ;
|
||||
|
||||
M: %peek-r basic-block? drop t ;
|
||||
|
||||
TUPLE: %replace-r ;
|
||||
C: %replace-r make-vop ;
|
||||
|
||||
: %replace-r ( vreg n -- vop )
|
||||
<cs-loc> src/dest-vop <%replace-r> ;
|
||||
|
||||
M: %replace-r basic-block? drop t ;
|
||||
|
||||
TUPLE: %inc-r ;
|
||||
|
||||
C: %inc-r make-vop ;
|
||||
|
|
@ -173,6 +155,14 @@ C: %inc-r make-vop ;
|
|||
|
||||
M: %inc-r basic-block? drop t ;
|
||||
|
||||
TUPLE: %immediate ;
|
||||
C: %immediate make-vop ;
|
||||
|
||||
: %immediate ( vreg obj -- vop )
|
||||
swap <vreg> src/dest-vop <%immediate> ;
|
||||
|
||||
M: %immediate basic-block? drop t ;
|
||||
|
||||
: in-1 0 0 %peek-d , ;
|
||||
: in-2 0 1 %peek-d , 1 0 %peek-d , ;
|
||||
: in-3 0 2 %peek-d , 1 1 %peek-d , 2 0 %peek-d , ;
|
||||
|
|
|
|||
|
|
@ -5,20 +5,17 @@ USING: alien assembler compiler inference kernel lists math
|
|||
memory sequences words ;
|
||||
|
||||
: reg-stack ( reg n -- op ) cell * neg 2list ;
|
||||
: ds-op ( ds-loc -- op ) ds-loc-n ESI swap reg-stack ;
|
||||
: cs-op ( cs-loc -- op ) cs-loc-n EBX swap reg-stack ;
|
||||
|
||||
: (%peek) dup 0 vop-out v>operand swap 0 vop-in ;
|
||||
GENERIC: loc>operand
|
||||
|
||||
M: %peek-d generate-node ( vop -- ) (%peek) ds-op MOV ;
|
||||
M: ds-loc loc>operand ds-loc-n ESI swap reg-stack ;
|
||||
M: cs-loc loc>operand cs-loc-n EBX swap reg-stack ;
|
||||
|
||||
M: %peek-r generate-node ( vop -- ) (%peek) cs-op MOV ;
|
||||
M: %peek generate-node ( vop -- )
|
||||
dup 0 vop-out v>operand swap 0 vop-in loc>operand swap MOV ;
|
||||
|
||||
: (%replace) dup 0 vop-in v>operand swap 0 vop-out ;
|
||||
|
||||
M: %replace-d generate-node ( vop -- ) (%replace) ds-op swap MOV ;
|
||||
|
||||
M: %replace-r generate-node ( vop -- ) (%replace) cs-op swap MOV ;
|
||||
M: %replace generate-node ( vop -- )
|
||||
dup 0 vop-in v>operand swap 0 vop-out loc>operand swap MOV ;
|
||||
|
||||
: (%inc) swap 0 vop-in cell * dup 0 > [ ADD ] [ neg SUB ] ifte ;
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue