FUNCTION static Word_t j__udyGetMemActive(
Pjp_t Pjp) // top of subtree.
{
Word_t offset; // in a branch.
Word_t Bytes = 0; // actual bytes used at this level.
Word_t IdxSz; // bytes per index in leaves
switch (JU_JPTYPE(Pjp))
{
case cJU_JPBRANCH_L2:
case cJU_JPBRANCH_L3:
#ifdef JU_64BIT
case cJU_JPBRANCH_L4:
case cJU_JPBRANCH_L5:
case cJU_JPBRANCH_L6:
case cJU_JPBRANCH_L7:
#endif
case cJU_JPBRANCH_L:
{
Pjbl_t Pjbl = P_JBL(Pjp->jp_Addr);
for (offset = 0; offset < (Pjbl->jbl_NumJPs); ++offset)
Bytes += j__udyGetMemActive((Pjbl->jbl_jp) + offset);
return(Bytes + sizeof(jbl_t));
}
case cJU_JPBRANCH_B2:
case cJU_JPBRANCH_B3:
#ifdef JU_64BIT
case cJU_JPBRANCH_B4:
case cJU_JPBRANCH_B5:
case cJU_JPBRANCH_B6:
case cJU_JPBRANCH_B7:
#endif
case cJU_JPBRANCH_B:
{
Word_t subexp;
Word_t jpcount;
Pjbb_t Pjbb = P_JBB(Pjp->jp_Addr);
for (subexp = 0; subexp < cJU_NUMSUBEXPB; ++subexp)
{
jpcount = j__udyCountBitsB(JU_JBB_BITMAP(Pjbb, subexp));
Bytes += jpcount * sizeof(jp_t);
for (offset = 0; offset < jpcount; ++offset)
{
Bytes += j__udyGetMemActive(P_JP(JU_JBB_PJP(Pjbb, subexp))
+ offset);
}
}
return(Bytes + sizeof(jbb_t));
}
case cJU_JPBRANCH_U2:
case cJU_JPBRANCH_U3:
#ifdef JU_64BIT
case cJU_JPBRANCH_U4:
case cJU_JPBRANCH_U5:
case cJU_JPBRANCH_U6:
case cJU_JPBRANCH_U7:
#endif
case cJU_JPBRANCH_U:
{
Pjbu_t Pjbu = P_JBU(Pjp->jp_Addr);
for (offset = 0; offset < cJU_BRANCHUNUMJPS; ++offset)
{
if (((Pjbu->jbu_jp[offset].jp_Type) >= cJU_JPNULL1)
&& ((Pjbu->jbu_jp[offset].jp_Type) <= cJU_JPNULLMAX))
{
continue; // skip null JP to save time.
}
Bytes += j__udyGetMemActive(Pjbu->jbu_jp + offset);
}
return(Bytes + sizeof(jbu_t));
}
// -- Cases below here terminate and do not recurse. --
#if (defined(JUDYL) || (! defined(JU_64BIT)))
case cJU_JPLEAF1: IdxSz = 1; goto LeafWords;
#endif
case cJU_JPLEAF2: IdxSz = 2; goto LeafWords;
case cJU_JPLEAF3: IdxSz = 3; goto LeafWords;
#ifdef JU_64BIT
case cJU_JPLEAF4: IdxSz = 4; goto LeafWords;
case cJU_JPLEAF5: IdxSz = 5; goto LeafWords;
case cJU_JPLEAF6: IdxSz = 6; goto LeafWords;
case cJU_JPLEAF7: IdxSz = 7; goto LeafWords;
#endif
LeafWords:
#ifdef JUDY1
return(IdxSz * (JU_JPLEAF_POP0(Pjp) + 1));
#else
return((IdxSz + sizeof(Word_t))
* (JU_JPLEAF_POP0(Pjp) + 1));
#endif
case cJU_JPLEAF_B1:
{
#ifdef JUDY1
return(sizeof(jlb_t));
#else
Bytes = (JU_JPLEAF_POP0(Pjp) + 1) * sizeof(Word_t);
return(Bytes + sizeof(jlb_t));
#endif
}
JUDY1CODE(case cJ1_JPFULLPOPU1: return(0);)
#ifdef JUDY1
#define J__Mpy 0
#else
#define J__Mpy sizeof(Word_t)
#endif
case cJU_JPIMMED_1_01: return(0);
case cJU_JPIMMED_2_01: return(0);
case cJU_JPIMMED_3_01: return(0);
#ifdef JU_64BIT
case cJU_JPIMMED_4_01: return(0);
case cJU_JPIMMED_5_01: return(0);
case cJU_JPIMMED_6_01: return(0);
case cJU_JPIMMED_7_01: return(0);
#endif
case cJU_JPIMMED_1_02: return(J__Mpy * 2);
case cJU_JPIMMED_1_03: return(J__Mpy * 3);
#if (defined(JUDY1) || defined(JU_64BIT))
case cJU_JPIMMED_1_04: return(J__Mpy * 4);
case cJU_JPIMMED_1_05: return(J__Mpy * 5);
case cJU_JPIMMED_1_06: return(J__Mpy * 6);
case cJU_JPIMMED_1_07: return(J__Mpy * 7);
#endif
#if (defined(JUDY1) && defined(JU_64BIT))
case cJ1_JPIMMED_1_08: return(0);
case cJ1_JPIMMED_1_09: return(0);
case cJ1_JPIMMED_1_10: return(0);
case cJ1_JPIMMED_1_11: return(0);
case cJ1_JPIMMED_1_12: return(0);
case cJ1_JPIMMED_1_13: return(0);
case cJ1_JPIMMED_1_14: return(0);
case cJ1_JPIMMED_1_15: return(0);
#endif
#if (defined(JUDY1) || defined(JU_64BIT))
case cJU_JPIMMED_2_02: return(J__Mpy * 2);
case cJU_JPIMMED_2_03: return(J__Mpy * 3);
#endif
#if (defined(JUDY1) && defined(JU_64BIT))
case cJ1_JPIMMED_2_04: return(0);
case cJ1_JPIMMED_2_05: return(0);
case cJ1_JPIMMED_2_06: return(0);
case cJ1_JPIMMED_2_07: return(0);
#endif
#if (defined(JUDY1) || defined(JU_64BIT))
case cJU_JPIMMED_3_02: return(J__Mpy * 2);
#endif
#if (defined(JUDY1) && defined(JU_64BIT))
case cJ1_JPIMMED_3_03: return(0);
case cJ1_JPIMMED_3_04: return(0);
case cJ1_JPIMMED_3_05: return(0);
case cJ1_JPIMMED_4_02: return(0);
case cJ1_JPIMMED_4_03: return(0);
case cJ1_JPIMMED_5_02: return(0);
case cJ1_JPIMMED_5_03: return(0);
case cJ1_JPIMMED_6_02: return(0);
case cJ1_JPIMMED_7_02: return(0);
#endif
} // switch (JU_JPTYPE(Pjp))
FUNCTION static Word_t j__udy1LCountSM(
const Pjp_t Pjp, // top of Judy (sub)SM.
const Word_t Index, // count at or above this Index.
const Pjpm_t Pjpm) // for returning error info.
{
Pjbl_t Pjbl; // Pjp->jp_Addr masked and cast to types:
Pjbb_t Pjbb;
Pjbu_t Pjbu;
Pjll_t Pjll; // a Judy lower-level linear leaf.
Word_t digit; // next digit to decode from Index.
long jpnum; // JP number in a branch (base 0).
int offset; // index ordinal within a leaf, base 0.
Word_t pop1; // total population of an expanse.
Word_t pop1above; // to return.
// Common code to check Decode bits in a JP against the equivalent portion of
// Index; XOR together, then mask bits of interest; must be all 0:
//
// Note: Why does this code only assert() compliance rather than actively
// checking for outliers? Its because Index is supposed to be valid, hence
// always match any Dcd bits traversed.
//
// Note: This assertion turns out to be always true for cState = 3 on 32-bit
// and 7 on 64-bit, but its harmless, probably removed by the compiler.
#define CHECKDCD(Pjp,cState) \
assert(! JU_DCDNOTMATCHINDEX(Index, Pjp, cState))
// Common code to prepare to handle a root-level or lower-level branch:
// Extract a state-dependent digit from Index in a "constant" way, obtain the
// total population for the branch in a state-dependent way, and then branch to
// common code for multiple cases:
//
// For root-level branches, the state is always cJU_ROOTSTATE, and the
// population is received in Pjpm->jpm_Pop0.
//
// Note: The total population is only needed in cases where the common code
// "counts up" instead of down to minimize cache line fills. However, its
// available cheaply, and its better to do it with a constant shift (constant
// state value) instead of a variable shift later "when needed".
#define PREPB_ROOT(Pjp,Next) \
digit = JU_DIGITATSTATE(Index, cJU_ROOTSTATE); \
pop1 = (Pjpm->jpm_Pop0) + 1; \
goto Next
#define PREPB(Pjp,cState,Next) \
digit = JU_DIGITATSTATE(Index, cState); \
pop1 = JU_JPBRANCH_POP0(Pjp, (cState)) + 1; \
goto Next
// SWITCH ON JP TYPE:
//
// WARNING: For run-time efficiency the following cases replicate code with
// varying constants, rather than using common code with variable values!
switch (JU_JPTYPE(Pjp))
{
// ----------------------------------------------------------------------------
// ROOT-STATE LEAF that starts with a Pop0 word; just count within the leaf:
case cJU_LEAFW:
{
Pjlw_t Pjlw = P_JLW(Pjp->jp_Addr); // first word of leaf.
assert((Pjpm->jpm_Pop0) + 1 == Pjlw[0] + 1); // sent correctly.
offset = j__udySearchLeafW(Pjlw + 1, Pjpm->jpm_Pop0 + 1, Index);
assert(offset >= 0); // Index must exist.
assert(offset < (Pjpm->jpm_Pop0) + 1); // Index be in range.
return((Pjpm->jpm_Pop0) + 1 - offset); // INCLUSIVE of Index.
}
// ----------------------------------------------------------------------------
// LINEAR BRANCH; count populations in JPs in the JBL ABOVE the next digit in
// Index, and recurse for the next digit in Index:
//
// Note: There are no null JPs in a JBL; watch out for pop1 == 0.
//
// Note: A JBL should always fit in one cache line => no need to count up
// versus down to save cache line fills. (PREPB() sets pop1 for no reason.)
case cJU_JPBRANCH_L2: CHECKDCD(Pjp, 2); PREPB(Pjp, 2, BranchL);
case cJU_JPBRANCH_L3: CHECKDCD(Pjp, 3); PREPB(Pjp, 3, BranchL);
#ifdef JU_64BIT
case cJU_JPBRANCH_L4: CHECKDCD(Pjp, 4); PREPB(Pjp, 4, BranchL);
case cJU_JPBRANCH_L5: CHECKDCD(Pjp, 5); PREPB(Pjp, 5, BranchL);
case cJU_JPBRANCH_L6: CHECKDCD(Pjp, 6); PREPB(Pjp, 6, BranchL);
case cJU_JPBRANCH_L7: CHECKDCD(Pjp, 7); PREPB(Pjp, 7, BranchL);
#endif
case cJU_JPBRANCH_L: PREPB_ROOT(Pjp, BranchL);
// Common code (state-independent) for all cases of linear branches:
BranchL:
Pjbl = P_JBL(Pjp->jp_Addr);
jpnum = Pjbl->jbl_NumJPs; // above last JP.
pop1above = 0;
while (digit < (Pjbl->jbl_Expanse[--jpnum])) // still ABOVE digit.
{
if ((pop1 = j__udyJPPop1((Pjbl->jbl_jp) + jpnum)) == cJU_ALLONES)
{
JU_SET_ERRNO_NONNULL(Pjpm, JU_ERRNO_CORRUPT);
return(C_JERR);
}
pop1above += pop1;
assert(jpnum > 0); // should find digit.
}
assert(digit == (Pjbl->jbl_Expanse[jpnum])); // should find digit.
pop1 = j__udy1LCountSM((Pjbl->jbl_jp) + jpnum, Index, Pjpm);
if (pop1 == C_JERR) return(C_JERR); // pass error up.
assert(pop1above + pop1);
return(pop1above + pop1);
// ----------------------------------------------------------------------------
// BITMAP BRANCH; count populations in JPs in the JBB ABOVE the next digit in
// Index, and recurse for the next digit in Index:
//
// Note: There are no null JPs in a JBB; watch out for pop1 == 0.
case cJU_JPBRANCH_B2: CHECKDCD(Pjp, 2); PREPB(Pjp, 2, BranchB);
case cJU_JPBRANCH_B3: CHECKDCD(Pjp, 3); PREPB(Pjp, 3, BranchB);
#ifdef JU_64BIT
case cJU_JPBRANCH_B4: CHECKDCD(Pjp, 4); PREPB(Pjp, 4, BranchB);
case cJU_JPBRANCH_B5: CHECKDCD(Pjp, 5); PREPB(Pjp, 5, BranchB);
case cJU_JPBRANCH_B6: CHECKDCD(Pjp, 6); PREPB(Pjp, 6, BranchB);
case cJU_JPBRANCH_B7: CHECKDCD(Pjp, 7); PREPB(Pjp, 7, BranchB);
#endif
case cJU_JPBRANCH_B: PREPB_ROOT(Pjp, BranchB);
// Common code (state-independent) for all cases of bitmap branches:
BranchB:
{
intptr_t subexp; // for stepping through layer 1 (subexpanses).
intptr_t findsub; // subexpanse containing Index (digit).
Word_t findbit; // bit representing Index (digit).
Word_t lowermask; // bits for indexes at or below Index.
Word_t jpcount; // JPs in a subexpanse.
Word_t clbelow; // cache lines below digits cache line.
Word_t clabove; // cache lines above digits cache line.
Pjbb = P_JBB(Pjp->jp_Addr);
findsub = digit / cJU_BITSPERSUBEXPB;
findbit = digit % cJU_BITSPERSUBEXPB;
lowermask = JU_MASKLOWERINC(JU_BITPOSMASKB(findbit));
clbelow = clabove = 0; // initial/default => always downward.
assert(JU_BITMAPTESTB(Pjbb, digit)); // digit must have a JP.
assert(findsub < cJU_NUMSUBEXPB); // falls in expected range.
// Shorthand for one subexpanse in a bitmap and for one JP in a bitmap branch:
//
// Note: BMPJP0 exists separately to support assertions.
#define BMPJP0(Subexp) (P_JP(JU_JBB_PJP(Pjbb, Subexp)))
#define BMPJP(Subexp,JPnum) (BMPJP0(Subexp) + (JPnum))
#ifndef NOSMARTJBB // enable to turn off smart code for comparison purposes.
// FIGURE OUT WHICH DIRECTION CAUSES FEWER CACHE LINE FILLS; adding the pop1s
// in JPs above Indexs JP, or subtracting the pop1s in JPs below Indexs JP.
//
// This is tricky because, while each set bit in the bitmap represents a JP,
// the JPs are scattered over cJU_NUMSUBEXPB subexpanses, each of which can
// contain JPs packed into multiple cache lines, and this code must visit every
// JP either BELOW or ABOVE the JP for Index.
//
// Number of cache lines required to hold a linear list of the given number of
// JPs, assuming the first JP is at the start of a cache line or the JPs in
// jpcount fit wholly within a single cache line, which is ensured by
// JudyMalloc():
#define CLPERJPS(jpcount) \
((((jpcount) * cJU_WORDSPERJP) + cJU_WORDSPERCL - 1) / cJU_WORDSPERCL)
// Count cache lines below/above for each subexpanse:
for (subexp = 0; subexp < cJU_NUMSUBEXPB; ++subexp)
{
jpcount = j__udyCountBitsB(JU_JBB_BITMAP(Pjbb, subexp));
// When at the subexpanse containing Index (digit), add cache lines
// below/above appropriately, excluding the cache line containing the JP for
// Index itself:
if (subexp < findsub) clbelow += CLPERJPS(jpcount);
else if (subexp > findsub) clabove += CLPERJPS(jpcount);
else // (subexp == findsub)
{
Word_t clfind; // cache line containing Index (digit).
clfind = CLPERJPS(j__udyCountBitsB(
JU_JBB_BITMAP(Pjbb, subexp) & lowermask));
assert(clfind > 0); // digit itself should have 1 CL.
clbelow += clfind - 1;
clabove += CLPERJPS(jpcount) - clfind;
}
}
#endif // ! NOSMARTJBB
// Note: Its impossible to get through the following "if" without setting
// jpnum -- see some of the assertions below -- but gcc -Wall doesnt know
// this, so preset jpnum to make it happy:
jpnum = 0;
// COUNT POPULATION FOR A BITMAP BRANCH, in whichever direction should result
// in fewer cache line fills:
//
// Note: If the remainder of Index is zero, pop1above is the pop1 of the
// entire expanse and theres no point in recursing to lower levels; but this
// should be so rare that its not worth checking for;
// Judy1Count()/JudyLCount() never even calls the motor for Index == 0 (all
// bytes).
// COUNT UPWARD, subtracting each "below or at" JPs pop1 from the whole
// expanses pop1:
//
// Note: If this causes clbelow + 1 cache line fills including JPs cache
// line, thats OK; at worst this is the same as clabove.
if (clbelow < clabove)
{
#ifdef SMARTMETRICS
++jbb_upward;
#endif
pop1above = pop1; // subtract JPs at/below Index.
// Count JPs for which to accrue pop1s in this subexpanse:
//
// TBD: If JU_JBB_BITMAP is cJU_FULLBITMAPB, dont bother counting.
for (subexp = 0; subexp <= findsub; ++subexp)
{
jpcount = j__udyCountBitsB((subexp < findsub) ?
JU_JBB_BITMAP(Pjbb, subexp) :
JU_JBB_BITMAP(Pjbb, subexp) & lowermask);
// should always find findbit:
assert((subexp < findsub) || jpcount);
// Subtract pop1s from JPs BELOW OR AT Index (digit):
//
// Note: The pop1 for Indexs JP itself is partially added back later at a
// lower state.
//
// Note: An empty subexpanse (jpcount == 0) is handled "for free".
//
// Note: Must be null JP subexp pointer in empty subexpanse and non-empty in
// non-empty subexpanse:
assert( jpcount || (BMPJP0(subexp) == (Pjp_t) NULL));
assert((! jpcount) || (BMPJP0(subexp) != (Pjp_t) NULL));
for (jpnum = 0; jpnum < jpcount; ++jpnum)
{
if ((pop1 = j__udyJPPop1(BMPJP(subexp, jpnum)))
== cJU_ALLONES)
{
JU_SET_ERRNO_NONNULL(Pjpm, JU_ERRNO_CORRUPT);
return(C_JERR);
}
pop1above -= pop1;
}
jpnum = jpcount - 1; // make correct for digit.
}
}
// COUNT DOWNWARD, adding each "above" JPs pop1:
else
{
long jpcountbf; // below findbit, inclusive.
#ifdef SMARTMETRICS
++jbb_downward;
#endif
pop1above = 0; // add JPs above Index.
jpcountbf = 0; // until subexp == findsub.
// Count JPs for which to accrue pop1s in this subexpanse:
//
// This is more complicated than counting upward because the scan of digits
// subexpanse must count ALL JPs, to know where to START counting down, and
// ALSO note the offset of digits JP to know where to STOP counting down.
for (subexp = cJU_NUMSUBEXPB - 1; subexp >= findsub; --subexp)
{
jpcount = j__udyCountBitsB(JU_JBB_BITMAP(Pjbb, subexp));
// should always find findbit:
assert((subexp > findsub) || jpcount);
if (! jpcount) continue; // empty subexpanse, save time.
// Count JPs below digit, inclusive:
if (subexp == findsub)
{
jpcountbf = j__udyCountBitsB(JU_JBB_BITMAP(Pjbb, subexp)
& lowermask);
}
// should always find findbit:
assert((subexp > findsub) || jpcountbf);
assert(jpcount >= jpcountbf); // proper relationship.
// Add pop1s from JPs ABOVE Index (digit):
// no null JP subexp pointers:
assert(BMPJP0(subexp) != (Pjp_t) NULL);
for (jpnum = jpcount - 1; jpnum >= jpcountbf; --jpnum)
{
if ((pop1 = j__udyJPPop1(BMPJP(subexp, jpnum)))
== cJU_ALLONES)
{
JU_SET_ERRNO_NONNULL(Pjpm, JU_ERRNO_CORRUPT);
return(C_JERR);
}
pop1above += pop1;
}
// jpnum is now correct for digit.
}
} // else.
// Return the net population ABOVE the digits JP at this state (in this JBB)
// plus the population AT OR ABOVE Index in the SM under the digits JP:
pop1 = j__udy1LCountSM(BMPJP(findsub, jpnum), Index, Pjpm);
if (pop1 == C_JERR) return(C_JERR); // pass error up.
assert(pop1above + pop1);
return(pop1above + pop1);
} // case.
// ----------------------------------------------------------------------------
// UNCOMPRESSED BRANCH; count populations in JPs in the JBU ABOVE the next
// digit in Index, and recurse for the next digit in Index:
//
// Note: If the remainder of Index is zero, pop1above is the pop1 of the
// entire expanse and theres no point in recursing to lower levels; but this
// should be so rare that its not worth checking for;
// Judy1Count()/JudyLCount() never even calls the motor for Index == 0 (all
// bytes).
case cJU_JPBRANCH_U2: CHECKDCD(Pjp, 2); PREPB(Pjp, 2, BranchU);
case cJU_JPBRANCH_U3: CHECKDCD(Pjp, 3); PREPB(Pjp, 3, BranchU);
#ifdef JU_64BIT
case cJU_JPBRANCH_U4: CHECKDCD(Pjp, 4); PREPB(Pjp, 4, BranchU);
case cJU_JPBRANCH_U5: CHECKDCD(Pjp, 5); PREPB(Pjp, 5, BranchU);
case cJU_JPBRANCH_U6: CHECKDCD(Pjp, 6); PREPB(Pjp, 6, BranchU);
case cJU_JPBRANCH_U7: CHECKDCD(Pjp, 7); PREPB(Pjp, 7, BranchU);
#endif
case cJU_JPBRANCH_U: PREPB_ROOT(Pjp, BranchU);
// Common code (state-independent) for all cases of uncompressed branches:
BranchU:
Pjbu = P_JBU(Pjp->jp_Addr);
#ifndef NOSMARTJBU // enable to turn off smart code for comparison purposes.
// FIGURE OUT WHICH WAY CAUSES FEWER CACHE LINE FILLS; adding the JPs above
// Indexs JP, or subtracting the JPs below Indexs JP.
//
// COUNT UPWARD, subtracting the pop1 of each JP BELOW OR AT Index, from the
// whole expanses pop1:
if (digit < (cJU_BRANCHUNUMJPS / 2))
{
pop1above = pop1; // subtract JPs below Index.
#ifdef SMARTMETRICS
++jbu_upward;
#endif
for (jpnum = 0; jpnum <= digit; ++jpnum)
{
if ((Pjbu->jbu_jp[jpnum].jp_Type) <= cJU_JPNULLMAX)
continue; // shortcut, save a function call.
if ((pop1 = j__udyJPPop1(Pjbu->jbu_jp + jpnum))
== cJU_ALLONES)
{
JU_SET_ERRNO_NONNULL(Pjpm, JU_ERRNO_CORRUPT);
return(C_JERR);
}
pop1above -= pop1;
}
}
// COUNT DOWNWARD, simply adding the pop1 of each JP ABOVE Index:
else
#endif // NOSMARTJBU
{
assert(digit < cJU_BRANCHUNUMJPS);
#ifdef SMARTMETRICS
++jbu_downward;
#endif
pop1above = 0; // add JPs above Index.
for (jpnum = cJU_BRANCHUNUMJPS - 1; jpnum > digit; --jpnum)
{
if ((Pjbu->jbu_jp[jpnum].jp_Type) <= cJU_JPNULLMAX)
continue; // shortcut, save a function call.
if ((pop1 = j__udyJPPop1(Pjbu->jbu_jp + jpnum))
== cJU_ALLONES)
{
JU_SET_ERRNO_NONNULL(Pjpm, JU_ERRNO_CORRUPT);
return(C_JERR);
}
pop1above += pop1;
}
}
if ((pop1 = j__udy1LCountSM(Pjbu->jbu_jp + digit, Index, Pjpm))
== C_JERR) return(C_JERR); // pass error up.
assert(pop1above + pop1);
return(pop1above + pop1);
// ----------------------------------------------------------------------------
// LEAF COUNT MACROS:
//
// LEAF*ABOVE() are common code for different JP types (linear leaves, bitmap
// leaves, and immediates) and different leaf Index Sizes, which result in
// calling different leaf search functions. Linear leaves get the leaf address
// from jp_Addr and the Population from jp_DcdPopO, while immediates use Pjp
// itself as the leaf address and get Population from jp_Type.
#define LEAFLABOVE(Func) \
Pjll = P_JLL(Pjp->jp_Addr); \
pop1 = JU_JPLEAF_POP0(Pjp) + 1; \
LEAFABOVE(Func, Pjll, pop1)
#define LEAFB1ABOVE(Func) LEAFLABOVE(Func) // different Func, otherwise same.
#ifdef JUDY1
#define IMMABOVE(Func,Pop1) \
Pjll = (Pjll_t) Pjp; \
LEAFABOVE(Func, Pjll, Pop1)
#else
// Note: For JudyL immediates with >= 2 Indexes, the index bytes are in a
// different place than for Judy1:
#define IMMABOVE(Func,Pop1) \
LEAFABOVE(Func, (Pjll_t) (Pjp->jp_LIndex), Pop1)
#endif
// For all leaf types, the population AT OR ABOVE is the total pop1 less the
// offset of Index; and Index should always be found:
#define LEAFABOVE(Func,Pjll,Pop1) \
offset = Func(Pjll, Pop1, Index); \
assert(offset >= 0); \
assert(offset < (Pop1)); \
return((Pop1) - offset)
// IMMABOVE_01 handles the special case of an immediate JP with 1 index, which
// the search functions arent used for anyway:
//
// The target Index should be the one in this Immediate, in which case the
// count above (inclusive) is always 1.
#define IMMABOVE_01 \
assert((JU_JPDCDPOP0(Pjp)) == JU_TRIMTODCDSIZE(Index)); \
return(1)
// ----------------------------------------------------------------------------
// LINEAR LEAF; search the leaf for Index; size is computed from jp_Type:
#if (defined(JUDYL) || (! defined(JU_64BIT)))
case cJU_JPLEAF1: LEAFLABOVE(j__udySearchLeaf1);
#endif
case cJU_JPLEAF2: LEAFLABOVE(j__udySearchLeaf2);
case cJU_JPLEAF3: LEAFLABOVE(j__udySearchLeaf3);
#ifdef JU_64BIT
case cJU_JPLEAF4: LEAFLABOVE(j__udySearchLeaf4);
case cJU_JPLEAF5: LEAFLABOVE(j__udySearchLeaf5);
case cJU_JPLEAF6: LEAFLABOVE(j__udySearchLeaf6);
case cJU_JPLEAF7: LEAFLABOVE(j__udySearchLeaf7);
#endif
// ----------------------------------------------------------------------------
// BITMAP LEAF; search the leaf for Index:
//
// Since the bitmap describes Indexes digitally rather than linearly, this is
// not really a search, but just a count.
case cJU_JPLEAF_B1: LEAFB1ABOVE(j__udyCountLeafB1);
#ifdef JUDY1
// ----------------------------------------------------------------------------
// FULL POPULATION:
//
// Return the count of Indexes AT OR ABOVE Index, which is the total population
// of the expanse (a constant) less the value of the undecoded digit remaining
// in Index (its base-0 offset in the expanse), which yields an inclusive count
// above.
//
// TBD: This only supports a 1-byte full expanse. Should this extract a
// stored value for pop0 and possibly more LSBs of Index, to handle larger full
// expanses?
case cJ1_JPFULLPOPU1:
return(cJU_JPFULLPOPU1_POP0 + 1 - JU_DIGITATSTATE(Index, 1));
#endif
// ----------------------------------------------------------------------------
// IMMEDIATE:
case cJU_JPIMMED_1_01: IMMABOVE_01;
case cJU_JPIMMED_2_01: IMMABOVE_01;
case cJU_JPIMMED_3_01: IMMABOVE_01;
#ifdef JU_64BIT
case cJU_JPIMMED_4_01: IMMABOVE_01;
case cJU_JPIMMED_5_01: IMMABOVE_01;
case cJU_JPIMMED_6_01: IMMABOVE_01;
case cJU_JPIMMED_7_01: IMMABOVE_01;
#endif
case cJU_JPIMMED_1_02: IMMABOVE(j__udySearchLeaf1, 2);
case cJU_JPIMMED_1_03: IMMABOVE(j__udySearchLeaf1, 3);
#if (defined(JUDY1) || defined(JU_64BIT))
case cJU_JPIMMED_1_04: IMMABOVE(j__udySearchLeaf1, 4);
case cJU_JPIMMED_1_05: IMMABOVE(j__udySearchLeaf1, 5);
case cJU_JPIMMED_1_06: IMMABOVE(j__udySearchLeaf1, 6);
case cJU_JPIMMED_1_07: IMMABOVE(j__udySearchLeaf1, 7);
#endif
#if (defined(JUDY1) && defined(JU_64BIT))
case cJ1_JPIMMED_1_08: IMMABOVE(j__udySearchLeaf1, 8);
case cJ1_JPIMMED_1_09: IMMABOVE(j__udySearchLeaf1, 9);
case cJ1_JPIMMED_1_10: IMMABOVE(j__udySearchLeaf1, 10);
case cJ1_JPIMMED_1_11: IMMABOVE(j__udySearchLeaf1, 11);
case cJ1_JPIMMED_1_12: IMMABOVE(j__udySearchLeaf1, 12);
case cJ1_JPIMMED_1_13: IMMABOVE(j__udySearchLeaf1, 13);
case cJ1_JPIMMED_1_14: IMMABOVE(j__udySearchLeaf1, 14);
case cJ1_JPIMMED_1_15: IMMABOVE(j__udySearchLeaf1, 15);
#endif
#if (defined(JUDY1) || defined(JU_64BIT))
case cJU_JPIMMED_2_02: IMMABOVE(j__udySearchLeaf2, 2);
case cJU_JPIMMED_2_03: IMMABOVE(j__udySearchLeaf2, 3);
#endif
#if (defined(JUDY1) && defined(JU_64BIT))
case cJ1_JPIMMED_2_04: IMMABOVE(j__udySearchLeaf2, 4);
case cJ1_JPIMMED_2_05: IMMABOVE(j__udySearchLeaf2, 5);
case cJ1_JPIMMED_2_06: IMMABOVE(j__udySearchLeaf2, 6);
case cJ1_JPIMMED_2_07: IMMABOVE(j__udySearchLeaf2, 7);
#endif
#if (defined(JUDY1) || defined(JU_64BIT))
case cJU_JPIMMED_3_02: IMMABOVE(j__udySearchLeaf3, 2);
#endif
#if (defined(JUDY1) && defined(JU_64BIT))
case cJ1_JPIMMED_3_03: IMMABOVE(j__udySearchLeaf3, 3);
case cJ1_JPIMMED_3_04: IMMABOVE(j__udySearchLeaf3, 4);
case cJ1_JPIMMED_3_05: IMMABOVE(j__udySearchLeaf3, 5);
case cJ1_JPIMMED_4_02: IMMABOVE(j__udySearchLeaf4, 2);
case cJ1_JPIMMED_4_03: IMMABOVE(j__udySearchLeaf4, 3);
case cJ1_JPIMMED_5_02: IMMABOVE(j__udySearchLeaf5, 2);
case cJ1_JPIMMED_5_03: IMMABOVE(j__udySearchLeaf5, 3);
case cJ1_JPIMMED_6_02: IMMABOVE(j__udySearchLeaf6, 2);
case cJ1_JPIMMED_7_02: IMMABOVE(j__udySearchLeaf7, 2);
#endif
// ----------------------------------------------------------------------------
// OTHER CASES:
default: JU_SET_ERRNO_NONNULL(Pjpm, JU_ERRNO_CORRUPT); return(C_JERR);
} // switch on JP type
/*NOTREACHED*/
} // j__udy1LCountSM()
FUNCTION int j__udyCreateBranchB(
Pjp_t Pjp, // Build JPs from this place
Pjp_t PJPs, // Array of JPs to put into Bitmap branch
uint8_t Exp[], // Array of expanses to put into bitmap
Word_t ExpCnt, // Number of above JPs and Expanses
Pvoid_t Pjpm)
{
Pjbb_t PjbbRaw; // pointer to bitmap branch.
Pjbb_t Pjbb;
Word_t ii, jj; // Temps
uint8_t CurrSubExp; // Current sub expanse for BM
// This assertion says the number of populated subexpanses is not too large.
// This function is only called when a BranchL overflows to a BranchB or when a
// cascade occurs, meaning a leaf overflows. Either way ExpCnt cant be very
// large, in fact a lot smaller than cJU_BRANCHBMAXJPS. (Otherwise a BranchU
// would be used.) Popping this assertion means something (unspecified) has
// gone very wrong, or else Judys design criteria have changed, although in
// fact there should be no HARM in creating a BranchB with higher actual
// fanout.
assert(ExpCnt <= cJU_BRANCHBMAXJPS);
// Get memory for a Bitmap branch
PjbbRaw = j__udyAllocJBB(Pjpm);
if (PjbbRaw == (Pjbb_t) NULL) return(-1);
Pjbb = P_JBB(PjbbRaw);
// Get 1st "sub" expanse (0..7) of bitmap branch
CurrSubExp = Exp[0] / cJU_BITSPERSUBEXPB;
// Index thru all 1 byte sized expanses:
for (jj = ii = 0; ii <= ExpCnt; ii++)
{
Word_t SubExp; // Cannot be a uint8_t
// Make sure we cover the last one
if (ii == ExpCnt)
{
SubExp = cJU_ALLONES; // Force last one
}
else
{
// Calculate the "sub" expanse of the byte expanse
SubExp = Exp[ii] / cJU_BITSPERSUBEXPB; // Bits 5..7.
// Set the bit that represents the expanse in Exp[]
JU_JBB_BITMAP(Pjbb, SubExp) |= JU_BITPOSMASKB(Exp[ii]);
}
// Check if a new "sub" expanse range needed
if (SubExp != CurrSubExp)
{
// Get number of JPs in this sub expanse
Word_t NumJP = ii - jj;
Pjp_t PjpRaw;
Pjp_t Pjp;
PjpRaw = j__udyAllocJBBJP(NumJP, Pjpm);
Pjp = P_JP(PjpRaw);
if (PjpRaw == (Pjp_t) NULL) // out of memory.
{
// Free any previous allocations:
while(CurrSubExp--)
{
NumJP = j__udyCountBitsB(JU_JBB_BITMAP(Pjbb,
CurrSubExp));
if (NumJP)
{
j__udyFreeJBBJP(JU_JBB_PJP(Pjbb,
CurrSubExp), NumJP, Pjpm);
}
}
j__udyFreeJBB(PjbbRaw, Pjpm);
return(-1);
}
// Place the array of JPs in bitmap branch:
JU_JBB_PJP(Pjbb, CurrSubExp) = PjpRaw;
// Copy the JPs to new leaf:
JU_COPYMEM(Pjp, PJPs + jj, NumJP);
// On to the next bitmap branch "sub" expanse:
jj = ii;
CurrSubExp = SubExp;
}
} // for each 1-byte expanse
// Pass back some of the JP to the new Bitmap branch:
Pjp->jp_Addr = (Word_t) PjbbRaw;
return(1);
} // j__udyCreateBranchB()
