Performance improvements to make struct-arrays benchmark faster

- improved optimization of ##unbox-any-c-ptr on ##box-displaced-alien; convert it to ##unbox-c-ptr where possible using class info stored in the ##bda instruction
- make fcos, fsin, etc inline again; everything in math.libm inline again, except for fsqrt which is an intrinsic
- convert min and max on floats to float-min and float-max
- make min and max not inline, so that the above can work
- struct-arrays: rice a bit so that more fixnums come up

basis/compiler/cfg/builder/builder-tests.factor

@@ -185,3 +185,9 @@ IN: compiler.cfg.builder.tests
     [ 1000 [ ] times ]
     [ [ ##peek? ] [ ##replace? ] bi or ] contains-insn?
 ] unit-test
+
+[ f t ] [
+    [ { fixnum simple-alien } declare <displaced-alien> 0 alien-cell ]
+    [ [ ##unbox-any-c-ptr? ] contains-insn? ]
+    [ [ ##slot-imm? ] contains-insn? ] bi
+] unit-test

basis/compiler/cfg/hats/hats.factor

@@ -43,6 +43,8 @@ IN: compiler.cfg.hats
 : ^^sub-float ( src1 src2 -- dst ) ^^r2 ##sub-float ; inline
 : ^^mul-float ( src1 src2 -- dst ) ^^r2 ##mul-float ; inline
 : ^^div-float ( src1 src2 -- dst ) ^^r2 ##div-float ; inline
+: ^^max-float ( src1 src2 -- dst ) ^^r2 ##max-float ; inline
+: ^^min-float ( src1 src2 -- dst ) ^^r2 ##min-float ; inline
 : ^^sqrt ( src -- dst ) ^^r1 ##sqrt ; inline
 : ^^float>integer ( src -- dst ) ^^r1 ##float>integer ; inline
 : ^^integer>float ( src -- dst ) ^^r1 ##integer>float ; inline
@@ -51,7 +53,8 @@ IN: compiler.cfg.hats
 : ^^allot-array ( n -- dst ) 2 + cells array ^^allot ; inline
 : ^^allot-byte-array ( n -- dst ) 2 cells + byte-array ^^allot ; inline
 : ^^box-alien ( src -- dst ) ^^r1 next-vreg ##box-alien ; inline
-: ^^box-displaced-alien ( base displacement -- dst ) ^^r2 next-vreg ##box-displaced-alien ; inline
+: ^^box-displaced-alien ( base displacement base-class -- dst )
+    ^^r3 [ next-vreg ] dip ##box-displaced-alien ; inline
 : ^^unbox-alien ( src -- dst ) ^^r1 ##unbox-alien ; inline
 : ^^unbox-c-ptr ( src class -- dst ) ^^r2 next-vreg ##unbox-c-ptr ;
 : ^^alien-unsigned-1 ( src -- dst ) ^^r1 ##alien-unsigned-1 ; inline

basis/compiler/cfg/instructions/instructions.factor

@@ -106,6 +106,8 @@ INSN: ##add-float < ##commutative ;
 INSN: ##sub-float < ##binary ;
 INSN: ##mul-float < ##commutative ;
 INSN: ##div-float < ##binary ;
+INSN: ##min-float < ##binary ;
+INSN: ##max-float < ##binary ;
 INSN: ##sqrt < ##unary ;
 
 ! Float/integer conversion
@@ -118,7 +120,7 @@ INSN: ##unbox-float < ##unary ;
 INSN: ##unbox-any-c-ptr < ##unary/temp ;
 INSN: ##box-float < ##unary/temp ;
 INSN: ##box-alien < ##unary/temp ;
-INSN: ##box-displaced-alien < ##binary temp ;
+INSN: ##box-displaced-alien < ##binary temp base-class ;
 
 : ##unbox-f ( dst src -- ) drop 0 ##load-immediate ;
 : ##unbox-byte-array ( dst src -- ) byte-array-offset ##add-imm ;
@@ -263,6 +265,8 @@ UNION: output-float-insn
     ##sub-float
     ##mul-float
     ##div-float
+    ##min-float
+    ##max-float
     ##sqrt
     ##integer>float
     ##unbox-float
@@ -275,6 +279,8 @@ UNION: input-float-insn
     ##sub-float
     ##mul-float
     ##div-float
+    ##min-float
+    ##max-float
     ##sqrt
     ##float>integer
     ##box-float

basis/compiler/cfg/intrinsics/alien/alien.factor

@@ -14,10 +14,11 @@ IN: compiler.cfg.intrinsics.alien
     } 1&& ;
 
 : emit-<displaced-alien> ( node -- )
-    dup emit-<displaced-alien>?
+    dup emit-<displaced-alien>? [
-    [ drop 2inputs [ ^^untag-fixnum ] dip ^^box-displaced-alien ds-push ]
+        [ 2inputs [ ^^untag-fixnum ] dip ] dip
-    [ emit-primitive ]
+        node-input-infos second class>>
-    if ;
+        ^^box-displaced-alien ds-push
+    ] [ emit-primitive ] if ;
 
 : (prepare-alien-accessor-imm) ( class offset -- offset-vreg )
     ds-drop [ ds-pop swap ^^unbox-c-ptr ] dip ^^add-imm ;

basis/compiler/cfg/intrinsics/intrinsics.factor

@@ -21,6 +21,7 @@ QUALIFIED: strings.private
 QUALIFIED: classes.tuple.private
 QUALIFIED: math.private
 QUALIFIED: math.integers.private
+QUALIFIED: math.floats.private
 QUALIFIED: math.libm
 IN: compiler.cfg.intrinsics
 
@@ -98,6 +99,12 @@ IN: compiler.cfg.intrinsics
 : enable-fsqrt ( -- )
     \ math.libm:fsqrt t "intrinsic" set-word-prop ;
 
+: enable-float-min/max ( -- )
+    {
+        math.floats.private:float-min
+        math.floats.private:float-max
+    } [ t "intrinsic" set-word-prop ] each ;
+
 : enable-fixnum-log2 ( -- )
     \ math.integers.private:fixnum-log2 t "intrinsic" set-word-prop ;
 
@@ -136,6 +143,8 @@ IN: compiler.cfg.intrinsics
         { \ math.private:float= [ drop cc= emit-float-comparison ] }
         { \ math.private:float>fixnum [ drop emit-float>fixnum ] }
         { \ math.private:fixnum>float [ drop emit-fixnum>float ] }
+        { \ math.floats.private:float-min [ drop [ ^^min-float ] emit-float-op ] }
+        { \ math.floats.private:float-max [ drop [ ^^max-float ] emit-float-op ] }
         { \ math.libm:fsqrt [ drop emit-fsqrt ] }
         { \ slots.private:slot [ emit-slot ] }
         { \ slots.private:set-slot [ emit-set-slot ] }

basis/compiler/cfg/two-operand/two-operand.factor

@@ -39,7 +39,9 @@ UNION: two-operand-insn
     ##add-float
     ##sub-float
     ##mul-float
-    ##div-float ;
+    ##div-float
+    ##min-float
+    ##max-float ;
 
 GENERIC: convert-two-operand* ( insn -- )
 

basis/compiler/cfg/value-numbering/expressions/expressions.factor

@@ -12,6 +12,7 @@ TUPLE: commutative-expr < binary-expr ;
 TUPLE: compare-expr < binary-expr cc ;
 TUPLE: constant-expr < expr value ;
 TUPLE: reference-expr < expr value ;
+TUPLE: box-displaced-alien-expr < expr displacement base base-class ;
 
 : <constant> ( constant -- expr )
     f swap constant-expr boa ; inline
@@ -85,6 +86,14 @@ M: ##compare-imm >expr compare-imm>expr ;
 
 M: ##compare-float >expr compare>expr ;
 
+M: ##box-displaced-alien >expr
+    {
+        [ class ]
+        [ src1>> vreg>vn ]
+        [ src2>> vreg>vn ]
+        [ base-class>> ]
+    } cleave box-displaced-alien-expr boa ;
+
 M: ##flushable >expr drop next-input-expr ;
 
 : init-expressions ( -- )

basis/compiler/cfg/value-numbering/rewrite/rewrite.factor

@@ -354,18 +354,18 @@ M: ##sar rewrite \ ##sar-imm rewrite-arithmetic ;
 : box-displaced-alien? ( expr -- ? )
     op>> \ ##box-displaced-alien eq? ;
 
-! ##box-displaced-alien f 1 2 3
+! ##box-displaced-alien f 1 2 3 <class>
-! ##unbox-any-c-ptr 4 1
+! ##unbox-c-ptr 4 1 <class>
 ! =>
-! ##box-displaced-alien f 1 2 3
+! ##box-displaced-alien f 1 2 3 <class>
-! ##unbox-any-c-ptr 5 3
+! ##unbox-c-ptr 5 3 <class>
 ! ##add 4 5 2
 
 :: rewrite-unbox-displaced-alien ( insn expr -- insns )
     [
         next-vreg :> temp
-        temp expr in2>> vn>vreg insn temp>> ##unbox-any-c-ptr
+        temp expr base>> vn>vreg expr base-class>> insn temp>> ##unbox-c-ptr
-        insn dst>> temp expr in1>> vn>vreg ##add
+        insn dst>> temp expr displacement>> vn>vreg ##add
     ] { } make ;
 
 M: ##unbox-any-c-ptr rewrite

basis/compiler/cfg/value-numbering/simplify/simplify.factor

@@ -87,12 +87,6 @@ M: unary-expr simplify*
         [ 2drop f ]
     } cond ; inline
 
-: simplify-box-displaced-alien ( expr -- vn/expr/f )
-    >binary-expr< {
-        { [ over expr-zero? ] [ nip ] }
-        [ 2drop f ]
-    } cond ;
-
 M: binary-expr simplify*
     dup op>> {
         { \ ##add [ simplify-add ] }
@@ -113,10 +107,15 @@ M: binary-expr simplify*
         { \ ##sar-imm [ simplify-shr ] }
         { \ ##shl [ simplify-shl ] }
         { \ ##shl-imm [ simplify-shl ] }
-        { \ ##box-displaced-alien [ simplify-box-displaced-alien ] }
         [ 2drop f ]
     } case ;
 
+M: box-displaced-alien-expr simplify*
+    [ base>> ] [ displacement>> ] bi {
+        { [ dup expr-zero? ] [ drop ] }
+        [ 2drop f ]
+    } cond ;
+
 M: expr simplify* drop f ;
 
 : simplify ( expr -- vn )

basis/compiler/cfg/value-numbering/value-numbering-tests.factor

@@ -4,7 +4,7 @@ cpu.architecture tools.test kernel math combinators.short-circuit
 accessors sequences compiler.cfg.predecessors locals compiler.cfg.dce
 compiler.cfg.ssa.destruction compiler.cfg.loop-detection
 compiler.cfg.representations compiler.cfg assocs vectors arrays
-layouts namespaces ;
+layouts namespaces alien ;
 IN: compiler.cfg.value-numbering.tests
 
 : trim-temps ( insns -- insns )
@@ -877,7 +877,7 @@ cell 8 = [
     {
         T{ ##peek f 0 D 0 }
         T{ ##load-immediate f 2 16 }
-        T{ ##box-displaced-alien f 1 2 0 }
+        T{ ##box-displaced-alien f 1 2 0 c-ptr }
         T{ ##unbox-any-c-ptr f 4 0 }
         T{ ##add-imm f 3 4 16 }
     }
@@ -885,7 +885,7 @@ cell 8 = [
     {
         T{ ##peek f 0 D 0 }
         T{ ##load-immediate f 2 16 }
-        T{ ##box-displaced-alien f 1 2 0 }
+        T{ ##box-displaced-alien f 1 2 0 c-ptr }
         T{ ##unbox-any-c-ptr f 3 1 }
     } value-numbering-step
 ] unit-test
@@ -896,7 +896,7 @@ cell 8 = [
     {
         T{ ##box-alien f 0 1 }
         T{ ##load-immediate f 2 16 }
-        T{ ##box-displaced-alien f 3 2 0 }
+        T{ ##box-displaced-alien f 3 2 0 c-ptr }
         T{ ##copy f 5 1 any-rep }
         T{ ##add-imm f 4 5 16 }
     }
@@ -904,7 +904,7 @@ cell 8 = [
     {
         T{ ##box-alien f 0 1 }
         T{ ##load-immediate f 2 16 }
-        T{ ##box-displaced-alien f 3 2 0 }
+        T{ ##box-displaced-alien f 3 2 0 c-ptr }
         T{ ##unbox-any-c-ptr f 4 3 }
     } value-numbering-step
 ] unit-test
@@ -922,7 +922,7 @@ cell 8 = [
     {
         T{ ##peek f 0 D 0 }
         T{ ##load-immediate f 2 0 }
-        T{ ##box-displaced-alien f 3 2 0 }
+        T{ ##box-displaced-alien f 3 2 0 c-ptr }
         T{ ##replace f 3 D 1 }
     } value-numbering-step
 ] unit-test

basis/compiler/codegen/codegen.factor

@@ -169,6 +169,8 @@ M: ##add-float generate-insn dst/src1/src2 %add-float ;
 M: ##sub-float generate-insn dst/src1/src2 %sub-float ;
 M: ##mul-float generate-insn dst/src1/src2 %mul-float ;
 M: ##div-float generate-insn dst/src1/src2 %div-float ;
+M: ##min-float generate-insn dst/src1/src2 %min-float ;
+M: ##max-float generate-insn dst/src1/src2 %max-float ;
 
 M: ##sqrt generate-insn dst/src %sqrt ;
 

basis/compiler/tests/float.factor

@@ -83,3 +83,8 @@ IN: compiler.tests.float
 [ f ] [ 3.0 [ dup 0.0 float= swap -0.0 float= or ] compile-call ] unit-test
 
 [ 315 315.0 ] [ 313 [ 2 fixnum+fast dup fixnum>float ] compile-call ] unit-test
+
+[ 17.5 ] [ -11.3 17.5 [ float-max ] compile-call ] unit-test
+[ 17.5 ] [ 17.5 -11.3 [ float-max ] compile-call ] unit-test
+[ -11.3 ] [ -11.3 17.5 [ float-min ] compile-call ] unit-test
+[ -11.3 ] [ 17.5 -11.3 [ float-min ] compile-call ] unit-test

basis/compiler/tree/propagation/known-words/known-words.factor

@@ -1,9 +1,9 @@
-! Copyright (C) 2008 Slava Pestov.
+! Copyright (C) 2008, 2009 Slava Pestov.
 ! See http://factorcode.org/license.txt for BSD license.
 USING: kernel effects accessors math math.private
-math.integers.private math.partial-dispatch math.intervals
+math.integers.private math.floats.private math.partial-dispatch
-math.parser math.order math.functions math.libm layouts words
+math.intervals math.parser math.order math.functions math.libm
-sequences sequences.private arrays assocs classes
+layouts words sequences sequences.private arrays assocs classes
 classes.algebra combinators generic.math splitting fry locals
 classes.tuple alien.accessors classes.tuple.private
 slots.private definitions strings.private vectors hashtables
@@ -303,3 +303,8 @@ generic-comparison-ops [
 flog fpow fsqrt facosh fasinh fatanh } [
     { float } "default-output-classes" set-word-prop
 ] each
+
+{ float-min float-max } [
+    [ { float float } "input-classes" set-word-prop ]
+    [ { float } "default-output-classes" set-word-prop ] bi
+] each

basis/compiler/tree/propagation/transforms/transforms.factor

@@ -3,7 +3,7 @@
 USING: kernel sequences words fry generic accessors classes.tuple
 classes classes.algebra definitions stack-checker.state quotations
 classes.tuple.private math math.partial-dispatch math.private
-math.intervals layouts math.order vectors hashtables
+math.intervals math.floats.private layouts math.order vectors hashtables
 combinators effects generalizations assocs sets
 combinators.short-circuit sequences.private locals
 stack-checker namespaces compiler.tree.propagation.info ;
@@ -79,6 +79,16 @@ IN: compiler.tree.propagation.transforms
     ] [ f ] if
 ] "custom-inlining" set-word-prop
 
+{
+    { min [ float-min ] }
+    { max [ float-max ] }
+} [
+    '[
+        in-d>> first2 [ value-info class>> float class<= ] both?
+        [ _ ] [ f ] if
+    ] "custom-inlining" set-word-prop
+] assoc-each
+
 ! Generate more efficient code for common idiom
 \ clone [
     in-d>> first value-info literal>> {

basis/cpu/architecture/architecture.factor

@@ -110,6 +110,8 @@ HOOK: %add-float cpu ( dst src1 src2 -- )
 HOOK: %sub-float cpu ( dst src1 src2 -- )
 HOOK: %mul-float cpu ( dst src1 src2 -- )
 HOOK: %div-float cpu ( dst src1 src2 -- )
+HOOK: %min-float cpu ( dst src1 src2 -- )
+HOOK: %max-float cpu ( dst src1 src2 -- )
 HOOK: %sqrt cpu ( dst src -- )
 
 HOOK: %integer>float cpu ( dst src -- )

basis/cpu/x86/32/32.factor

@@ -303,8 +303,7 @@ USING: cpu.x86.features cpu.x86.features.private ;
     "Checking if your CPU supports SSE2..." print flush
     sse2? [
         " - yes" print
-        enable-float-intrinsics
+        enable-sse2
-        enable-fsqrt
         [
             sse2? [
                 "This image was built to use SSE2, which your CPU does not support." print

basis/cpu/x86/64/64.factor

@@ -202,8 +202,7 @@ M: x86.64 %callback-value ( ctype -- )
 enable-alien-4-intrinsics
 
 ! SSE2 is always available on x86-64.
-enable-float-intrinsics
+enable-sse2
-enable-fsqrt
 
 USE: vocabs.loader
 

basis/cpu/x86/x86.factor

@@ -203,6 +203,8 @@ M: x86 %add-float nip ADDSD ;
 M: x86 %sub-float nip SUBSD ;
 M: x86 %mul-float nip MULSD ;
 M: x86 %div-float nip DIVSD ;
+M: x86 %min-float nip MINSD ;
+M: x86 %max-float nip MAXSD ;
 M: x86 %sqrt SQRTSD ;
 
 M: x86 %integer>float CVTSI2SD ;
@@ -572,3 +574,8 @@ M: x86 small-enough? ( n -- ? )
     #! stack frame set up, and we want to read the frame
     #! set up by the caller.
     stack-frame get total-size>> + stack@ ;
+
+: enable-sse2 ( -- )
+    enable-float-intrinsics
+    enable-fsqrt
+    enable-float-min/max ;

basis/math/libm/libm.factor

@@ -4,53 +4,54 @@ USING: alien ;
 IN: math.libm
 
 : facos ( x -- y )
-    "double" "libm" "acos" { "double" } alien-invoke ;
+    "double" "libm" "acos" { "double" } alien-invoke ; inline
 
 : fasin ( x -- y )
-    "double" "libm" "asin" { "double" } alien-invoke ;
+    "double" "libm" "asin" { "double" } alien-invoke ; inline
 
 : fatan ( x -- y )
-    "double" "libm" "atan" { "double" } alien-invoke ;
+    "double" "libm" "atan" { "double" } alien-invoke ; inline
 
 : fatan2 ( x y -- z )
-    "double" "libm" "atan2" { "double" "double" } alien-invoke ;
+    "double" "libm" "atan2" { "double" "double" } alien-invoke ; inline
 
 : fcos ( x -- y )
-    "double" "libm" "cos" { "double" } alien-invoke ;
+    "double" "libm" "cos" { "double" } alien-invoke ; inline
 
 : fsin ( x -- y )
-    "double" "libm" "sin" { "double" } alien-invoke ;
+    "double" "libm" "sin" { "double" } alien-invoke ; inline
 
 : ftan ( x -- y )
-    "double" "libm" "tan" { "double" } alien-invoke ;
+    "double" "libm" "tan" { "double" } alien-invoke ; inline
 
 : fcosh ( x -- y )
-    "double" "libm" "cosh" { "double" } alien-invoke ;
+    "double" "libm" "cosh" { "double" } alien-invoke ; inline
 
 : fsinh ( x -- y )
-    "double" "libm" "sinh" { "double" } alien-invoke ;
+    "double" "libm" "sinh" { "double" } alien-invoke ; inline
 
 : ftanh ( x -- y )
-    "double" "libm" "tanh" { "double" } alien-invoke ;
+    "double" "libm" "tanh" { "double" } alien-invoke ; inline
 
 : fexp ( x -- y )
-    "double" "libm" "exp" { "double" } alien-invoke ;
+    "double" "libm" "exp" { "double" } alien-invoke ; inline
 
 : flog ( x -- y )
-    "double" "libm" "log" { "double" } alien-invoke ;
+    "double" "libm" "log" { "double" } alien-invoke ; inline
 
 : fpow ( x y -- z )
-    "double" "libm" "pow" { "double" "double" } alien-invoke ;
+    "double" "libm" "pow" { "double" "double" } alien-invoke ; inline
 
+! Don't inline fsqrt -- its an intrinsic!
 : fsqrt ( x -- y )
     "double" "libm" "sqrt" { "double" } alien-invoke ;
     
 ! Windows doesn't have these...
 : facosh ( x -- y )
-    "double" "libm" "acosh" { "double" } alien-invoke ;
+    "double" "libm" "acosh" { "double" } alien-invoke ; inline
 
 : fasinh ( x -- y )
-    "double" "libm" "asinh" { "double" } alien-invoke ;
+    "double" "libm" "asinh" { "double" } alien-invoke ; inline
 
 : fatanh ( x -- y )
-    "double" "libm" "atanh" { "double" } alien-invoke ;
+    "double" "libm" "atanh" { "double" } alien-invoke ; inline

basis/struct-arrays/struct-arrays.factor

@@ -17,7 +17,7 @@ M: struct-array length length>> ; inline
 M: struct-array byte-length [ length>> ] [ element-size>> ] bi * ; inline
 
 : (nth-ptr) ( i struct-array -- alien )
-    [ element-size>> * ] [ underlying>> ] bi <displaced-alien> ; inline
+    [ element-size>> * >fixnum ] [ underlying>> ] bi <displaced-alien> ; inline
 
 M: struct-array nth-unsafe
     [ (nth-ptr) ] [ class>> dup struct-class? ] bi [ memory>struct ] [ drop ] if ; inline
@@ -26,7 +26,7 @@ M: struct-array set-nth-unsafe
     [ (nth-ptr) swap ] [ element-size>> ] bi memcpy ; inline
 
 M: struct-array new-sequence
-    [ element-size>> [ * <byte-array> ] 2keep ]
+    [ element-size>> [ * (byte-array) ] 2keep ]
     [ class>> ] bi struct-array boa ; inline
 
 M: struct-array resize ( n seq -- newseq )

core/math/floats/floats.factor

@@ -3,6 +3,9 @@
 USING: kernel math math.private ;
 IN: math.floats.private
 
+: float-min ( x y -- z ) [ float< ] 2keep ? ;
+: float-max ( x y -- z ) [ float> ] 2keep ? ;
+
 M: fixnum >float fixnum>float ; inline
 M: bignum >float bignum>float ; inline
 

core/math/order/order.factor

@@ -32,8 +32,8 @@ M: real after? ( obj1 obj2 -- ? ) > ; inline
 M: real before=? ( obj1 obj2 -- ? ) <= ; inline
 M: real after=? ( obj1 obj2 -- ? ) >= ; inline
 
-: min ( x y -- z ) [ before? ] most ; inline
+: min ( x y -- z ) [ before? ] most ;
-: max ( x y -- z ) [ after? ] most ; inline
+: max ( x y -- z ) [ after? ] most ;
 : clamp ( x min max -- y ) [ max ] dip min ; inline
 
 : between? ( x y z -- ? )