Bitfields

doc/ref/integers.xml

@@ -174,5 +174,20 @@ random integers and random choices from lists.
 <#Include Label="RandomSource">
 
 </Section>
+<!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -->
+<Section Label="Bitfields">
+  <Heading>Bitfields</Heading>
+Bitfields are a low-level feature intended to support efficient 
+subdivision of immediate integers into bitfields of various widths.
+This is typically useful in implementing space-efficient and/or
+cache-efficient data structures. This feature should be used with care
+because (<E>inter alia</E>) it has different limitations on 32-bit and 64-bit
+architectures.
+
+<#Include Label="MakeBitfields">
+<#Include Label="BuildBitfields">
+
+</Section>
+
 </Chapter>
 

doc/ref/makedocreldata.g

@@ -28,6 +28,7 @@ GAPInfo.ManualDataRef:= rec(
     "../../lib/attr.gd",
     "../../lib/basis.gd",
     "../../lib/basismut.gd",
+    "../../lib/bitfields.gd",
     "../../lib/boolean.g",
     "../../lib/clas.gd",
     "../../lib/cmdledit.g",

lib/bitfields.gd

@@ -0,0 +1,119 @@
+#############################################################################
+##
+#W  bitfields.gd                  GAP library                   Steve Linton
+##
+##
+#Y  Copyright (C) 2017 The GAP Group
+##
+##  This file declares the operations for bitfields
+##
+
+
+#############################################################################
+##
+#F  MakeBitfields(<width>[, <width>[, <width>...]]])
+##  <#GAPDoc Label="MakeBitfields">
+##  <ManSection>
+##  <Func Name="MakeBitfields" Arg='width....'/>
+##
+##  <Description>
+##
+##  This function sets up the machinery for a set of bitfields of the given
+##  widths. All bitfield values are treated as unsigned. 
+##  The total of the widths must not exceed 60 bits on 64-bit architecture
+##  or 28 bits on a 32-bit architecture. For performance 
+##  reasons some checks that one might wish to do are ommitted. In particular, 
+##  the builder and setter functions do not check if the value[s] passed to them are 
+##  negative or too large (unless &GAP; is specially compiled for debugging).
+##  Behaviour when such arguments are passed is undefined.
+##
+##  You can tell which type of architecture you are running on by acccessing 
+##  <C>GAPInfo.BytesPerVariable</C> which is 8 on 64-bits and 4 on 32. 
+##
+##
+##  The return value  when <M>n</M> widths are given is a record whose fields are 
+##  <List>
+##  <Mark><C>widths</C></Mark> <Item>a copy of the arguments, for convenience,</Item>
+##  <Mark><C>getters</C></Mark> <Item> a list of <M>n</M> functions of one
+##  argument each of which extracts one of the fields from an immediate integer</Item>
+##  <Mark><C>setters</C></Mark> <Item> a list of <M>n</M> functions each taking
+##  two arguments: a packed value and a new value for one of its fields and
+##  returning a new packed value. The
+##  <M>i</M>th function returned the new packed value in which the <M>i</M>th
+##  field has been replaced by the new value.
+##   Note that this does NOT modify the original packed value.</Item>
+##
+##  </List>
+## 
+##  Two additional fields may be present if any of the field widths is
+##  one. Each is a list and only has entried bound in the positions
+##  corresponding to the width 1 fields.
+##  <List> <Mark><C>booleanGetters</C></Mark> <Item>if the <M>i</M>th position of
+##  this list is set, it contains a function which extracts the <M>i</M>th
+##  field (which will have width one) and returns <C>true</C> if it contains 1
+##  and <C>false</C> if it contains 0</Item>
+##  <Mark><C>booleanSetters</C></Mark> <Item>if the <M>i</M>th position of
+##  this list is set, it contains a function of two arguments. The first
+##  argument is a packed value, the second is <C>true</C> or <C>false</C>. It returns a
+##  new packed value in which the <M>i</M>th field is set to 1 if the second
+##  argument was <C>true</C> and 0 if it was <C>false</C>. Behaviour for any
+##  other value is undefined.</Item></List>
+##  </Description>
+##  </ManSection>
+##  <#/GAPDoc>
+
+DeclareGlobalFunction( "MakeBitfields" ); 
+
+##  
+#############################################################################
+##
+#F  BuildBitfields( <widths>[, <val1>, [<val2>...]])
+##  <#GAPDoc Label="BuildBitfields">
+##  <ManSection>
+##  <Func Name="BuildBitfields" Arg='widths,....'/>
+##  <Description>
+##
+##  This function takes one or more argument. It's first argument is a list
+##  of field widths, as found in the <C>widths</C> entry of a record returned
+##  by <C>MakeBitfields</C>. The remaining arguments are unsigned integer
+##  values, equal in number to the entries of the list of field widths. It returns
+##  a small integer in which those entries are packed into bitfields of the
+##  given widths. The first entry occupies the least significant bits. 
+##
+##  
+
+DeclareGlobalFunction("BuildBitfields");
+
+##  
+##  <Example><![CDATA[
+##  gap> bf := MakeBitfields(1,2,3);
+##  rec( booleanGetters := [ function( data ) ... end ], 
+##    booleanSetters := [ function( data, val ) ... end ], 
+##    getters := [ function( data ) ... end, function( data ) ... end, 
+##        function( data ) ... end ], 
+##    setters := [ function( data, val ) ... end, function( data, val ) ... end, 
+##        function( data, val ) ... end ], widths := [ 1, 2, 3 ] )
+##  gap> x := BuildBitfields(bf.widths,0,3,5);
+##  46
+##  gap> bf.getters[3](x);
+##  5
+##  gap> y := bf.setters[1](x,1);
+##  47  
+##  gap> x;
+##  46
+##  gap> bf.booleanGetters[1](x);
+##  false
+##  gap> bf.booleanGetters[1](y);
+##  true
+##  ]]></Example>
+##  </Description>
+##  </ManSection>
+##  <#/GAPDoc>
+##
+
+
+
+#############################################################################
+##
+#E
+

lib/bitfields.gi

@@ -0,0 +1,2 @@
+InstallGlobalFunction(MakeBitfields, MAKE_BITFIELDS);
+InstallGlobalFunction(BuildBitfields, BUILD_BITFIELDS);

lib/read3.g

@@ -12,6 +12,7 @@ ReadLib( "extrset.gd"  );
 ReadLib( "extuset.gd"  );
 
 ReadLib( "dict.gd"     );
+ReadLib( "bitfields.gd" );
 
 ReadLib( "mapping.gd"  );
 ReadLib( "mapphomo.gd" );

lib/read5.g

@@ -149,6 +149,7 @@ ReadLib( "grppcatr.gi" );
 ReadLib( "grppcnrm.gi" );
 
 # files dealing with trees and hash tables
+ReadLib( "bitfields.gi" );
 ReadLib( "dict.gi"  );
 ReadLib( "dicthf.gi"  );
 

src/intfuncs.c

@@ -615,6 +615,207 @@ Obj FuncINT_STRING ( Obj self, Obj string )
     return IntStringInternal(string);
 }
 
+/****************************************************************************
+**
+*F SmallInt Bitfield operations
+*
+* The goal here it to set up a division of the usable bits in a small
+* integer into fields which can be accessed very quickly from GAP
+* level and quickly and conveniently from C. The purpose is to allow
+* implementation of data structures that divide up the bits within a
+* word without having to make them entirely opaque to the GAP level or
+* ridiculously slow
+*
+* The API is defined in lib/bitfields.gd and works by providing the
+* user with a collection of functions to get and set fields and
+* assemble an entire word.
+*
+* These functions are constructed here and have special handlers. The
+* information the handlers need about the size and position of the
+* bitfields are stored in some of the fields of the function header
+* which are not normally used for kernel functions. Specifically, we
+* use the NLOC_FUNC and FEXS_FUNC fields, which we alias as
+* MASK_BITFIELD_FUNC and OFFSET_BITFIELD_FUNC.
+*
+* For fields of size 1 we also offer Boolean setters and getters which
+* accept and return True for 1 and False for 0. This makes for much
+* nicer code on the GAP side.
+*
+*/
+
+
+static inline UInt MASK_BITFIELD_FUNC(Obj func)
+{
+  return NLOC_FUNC(func);
+}
+
+static inline void SET_MASK_BITFIELD_FUNC(Obj func, UInt mask)
+{
+  SET_NLOC_FUNC(func, mask);
+}
+
+static inline UInt OFFSET_BITFIELD_FUNC(Obj func)
+{
+  GAP_ASSERT(IS_INTOBJ(FEXS_FUNC(func)));
+  return INT_INTOBJ(FEXS_FUNC(func));
+}
+
+static inline void SET_OFFFSET_BITFIELD_FUNC(Obj func, UInt offset)
+{
+  return SET_FEXS_FUNC(func, INTOBJ_INT(offset));		    
+}
+
+static Obj DoFieldGetter(Obj self, Obj data)
+{
+    UInt mask = MASK_BITFIELD_FUNC(self);
+    UInt offset = OFFSET_BITFIELD_FUNC(self);
+    if (!IS_INTOBJ(data))
+        ErrorMayQuit("Field getter: argument must be small integer", 0, 0);
+    UInt x = INT_INTOBJ(data);
+    return INTOBJ_INT((x & mask) >> offset);
+}
+
+static Obj DoFieldSetter(Obj self, Obj data, Obj val)
+{
+    UInt mask = MASK_BITFIELD_FUNC(self);
+    UInt offset = OFFSET_BITFIELD_FUNC(self);
+    if (!ARE_INTOBJS(data, val))
+        ErrorMayQuit("Field Setter: both arguments must be small integers", 0,
+                     0);
+    UInt x = INT_INTOBJ(data);
+    UInt y = INT_INTOBJ(val);
+    return INTOBJ_INT((x & ~mask) | (y << offset));
+}
+
+static Obj DoBooleanFieldGetter(Obj self, Obj data)
+{
+  UInt mask = MASK_BITFIELD_FUNC(self);
+    if (!IS_INTOBJ(data))
+        ErrorMayQuit("Boolean Field getter: argument must be small integer", 0, 0);
+    UInt x = INT_INTOBJ(data);
+    return (x & mask) ? True : False;
+}
+
+static Obj DoBooleanFieldSetter(Obj self, Obj data, Obj val)
+{
+    UInt mask = MASK_BITFIELD_FUNC(self);
+    if (!IS_INTOBJ(data))
+        ErrorMayQuit("Boolean Field Setter: data must be small integer", 0,
+                     0);
+    UInt x = INT_INTOBJ(data);
+    if (val == True)
+        x |= mask;
+    else if (val == False)
+        x &= ~mask;
+    else
+        ErrorMayQuit("Boolean Field Setter: value must be true or false", 0,
+                     0);
+    return INTOBJ_INT(x);
+}
+
+
+static Obj FuncBUILD_BITFIELDS(Obj self, Obj args)
+{
+    GAP_ASSERT(IS_PLIST(args));
+    GAP_ASSERT(LEN_PLIST(args) >= 1 && ELM_PLIST(args, 1));
+    Obj widths = ELM_PLIST(args, 1);
+    if (!IS_LIST(widths))
+        ErrorMayQuit("Fields builder: first argument must be list of widths",
+                     0, 0);
+    UInt nfields = LEN_LIST(widths);
+    if (LEN_PLIST(args) != nfields + 1)
+        ErrorMayQuit(
+            "Fields builder: number of values must match number of widths", 0,
+            0);
+    UInt x = 0;
+    UInt i;
+    for (i = nfields; i > 0; i--) {
+        GAP_ASSERT(ISB_LIST(widths, i));
+        Obj y = ELM_LIST(widths, i);
+        x <<= INT_INTOBJ(y);
+        GAP_ASSERT(ELM_PLIST(args, i + 1));
+        Obj z = ELM_PLIST(args, i + 1);
+        if (!IS_INTOBJ(z))
+            ErrorMayQuit("Fields builder: values must be small integers", 0,
+                         0);
+        GAP_ASSERT(INT_INTOBJ(z) < (1 << INT_INTOBJ(y)));
+        x |= INT_INTOBJ(z);
+    }
+    return INTOBJ_INT(x);
+}
+
+
+Obj FuncMAKE_BITFIELDS(Obj self, Obj widths)
+{
+    if (!IS_LIST(widths))
+        ErrorMayQuit("MAKE_BITFIELDS: widths must be a list", 0, 0);
+    UInt nfields = LEN_LIST(widths);
+    UInt starts[nfields + 1];
+    starts[0] = 0;
+    for (UInt i = 1; i <= nfields; i++) {
+        Obj o = ELM_LIST(widths, i);
+        if (!IS_INTOBJ(o))
+            ErrorMayQuit("MAKE_BITFIELDS: widths must be small integers", 0,
+                         0);
+        UInt width = INT_INTOBJ(o);
+        starts[i] = starts[i - 1] + width;
+    }
+    if (starts[nfields] > 8 * sizeof(UInt))
+        ErrorMayQuit("MAKE_BITFIELDS: total widths too large", 0, 0);
+
+    Obj  setters = NEW_PLIST(T_PLIST_DENSE + IMMUTABLE, nfields);
+    Obj  getters = NEW_PLIST(T_PLIST_DENSE + IMMUTABLE, nfields);
+    Obj  bsetters = NEW_PLIST(T_PLIST + IMMUTABLE, nfields);
+    UInt bslen = 0;
+    Obj  bgetters = NEW_PLIST(T_PLIST + IMMUTABLE, nfields);
+    for (UInt i = 1; i <= nfields; i++) {
+        UInt mask = (1L << starts[i]) - (1L << starts[i - 1]);
+        Obj s = NewFunctionC("<field setter>", 2, "data, val", DoFieldSetter);
+        SET_MASK_BITFIELD_FUNC(s, mask);
+        SET_OFFFSET_BITFIELD_FUNC(s, starts[i - 1]);
+        SET_ELM_PLIST(setters, i, s);
+        CHANGED_BAG(setters);
+        Obj g = NewFunctionC("<field getter>", 1, "data", DoFieldGetter);
+        SET_MASK_BITFIELD_FUNC(g, mask);
+        SET_OFFFSET_BITFIELD_FUNC(g, starts[i - 1]);
+        SET_ELM_PLIST(getters, i, g);
+        CHANGED_BAG(getters);
+        if (starts[i] - starts[i - 1] == 1) {
+            s = NewFunctionC("<boolean field setter>", 2, "data, val",
+                             DoBooleanFieldSetter);
+            SET_MASK_BITFIELD_FUNC(s, mask);
+            SET_OFFFSET_BITFIELD_FUNC(s, starts[i - 1]);
+            SET_ELM_PLIST(bsetters, i, s);
+            CHANGED_BAG(bsetters);
+            bslen = i;
+            g = NewFunctionC("<boolean field getter>", 1, "data",
+                             DoBooleanFieldGetter);
+            SET_MASK_BITFIELD_FUNC(g, mask);
+            SET_OFFFSET_BITFIELD_FUNC(g, starts[i - 1]);
+            SET_ELM_PLIST(bgetters, i, g);
+            CHANGED_BAG(bgetters);
+        }
+    }
+
+    SET_LEN_PLIST(setters, nfields);
+    SET_LEN_PLIST(getters, nfields);
+    SET_LEN_PLIST(bsetters, bslen);
+    SET_LEN_PLIST(bgetters, bslen);
+
+    Obj ms = NEW_PREC(5);
+    AssPRec(ms, RNamName("widths"), CopyObj(widths, 0));
+    AssPRec(ms, RNamName("getters"), getters);
+    AssPRec(ms, RNamName("setters"), setters);
+    if (bslen > 0) {
+        AssPRec(ms, RNamName("booleanGetters"), bgetters);
+        AssPRec(ms, RNamName("booleanSetters"), bsetters);
+    }
+    SortPRecRNam(ms, 0);
+    RetypeBag(ms, T_PREC + IMMUTABLE);
+    return ms;
+}
+
+
 /****************************************************************************
 **
 *F * * * * * * * * * * * * * initialize package * * * * * * * * * * * * * * *
@@ -632,6 +833,8 @@ static StructGVarFunc GVarFuncs [] = {
     GVAR_FUNC(SIZE_OBJ, 1, "obj"),
     GVAR_FUNC(InitRandomMT, 1, "initstr"),
     GVAR_FUNC(INT_STRING, 1, "string"),
+    GVAR_FUNC(MAKE_BITFIELDS, -1, "widths"),
+    GVAR_FUNC(BUILD_BITFIELDS, -2, "widths..."),
     { 0, 0, 0, 0, 0 }
 
 };
@@ -645,6 +848,10 @@ static Int InitKernel (
     StructInitInfo *    module )
 {
 
+    InitHandlerFunc(DoFieldSetter, "field-setter");
+    InitHandlerFunc(DoFieldGetter, "field-getter");
+    InitHandlerFunc(DoBooleanFieldSetter, "boolean-field-setter");
+    InitHandlerFunc(DoBooleanFieldGetter, "boolean-field-getter");
 
     /* init filters and functions                                          */
     InitHdlrFuncsFromTable( GVarFuncs );