fast hashing for CreditIndex

2 files changed, 35 insertions, 2 deletions

diff --git a/pyisda/credit_index.pyx b/pyisda/credit_index.pyx
index 3238f2f..163f2f1 100644
--- a/pyisda/credit_index.pyx
+++ b/pyisda/credit_index.pyx
@@ -1,7 +1,7 @@
 #cython: cdivision=True, boundscheck=False
 from libc.stdlib cimport malloc, free
 from libc.math cimport nan
-from libc.string cimport memcpy
+from libc.string cimport memcpy, memset
 from libcpp.pair cimport pair
 from cython.operator cimport dereference as deref
 from cpython cimport PyObject, Py_INCREF
@@ -10,7 +10,8 @@ cimport cython
 from legs cimport (JpmcdsCdsContingentLegMake, JpmcdsCdsFeeLegMake,
                    JpmcdsContingentLegPV, JpmcdsFeeLegPV, FeeLegAI, JpmcdsFeeLegFree)
 from curve cimport (SpreadCurve, JpmcdsCopyCurve, tweak_curve, YieldCurve,
-                    JpmcdsFreeTCurve, JpmcdsForwardZeroPrice)
+                    JpmcdsFreeTCurve, JpmcdsForwardZeroPrice, Hash64WithSeed,
+                    Hash64, uint64_t, TCurve_size, serialize)
 from date cimport (pydate_to_TDate, TDate_to_pydate, ACT_365F, JpmcdsDtFwdAny,
                    TDateInterval, JpmcdsMakeDateInterval)
 from cdsone cimport JpmcdsStringToStubMethod, TStubMethod
@@ -279,6 +280,37 @@ cdef class CreditIndex(CurveList):
         return (self.__class__,
                 (TDate_to_pydate(self.start_date), self.maturities, self.curves, self.weights))
 
+    def __hash__(self):
+        cdef:
+            TCurve* curve = self._curves[0].get()
+            size_t size = TCurve_size(curve.fNumItems)
+            size_t buf_size = size + sizeof(size_t) + curve.fNumItems * sizeof(double) + 8
+            unsigned char* buf = <unsigned char*>malloc(buf_size)
+            unsigned char* cursor
+            unsigned char* start = buf + size
+            size_t i
+            uint64_t h = 0
+            pair[string, size_t] p
+
+        for p in self.tickers:
+            cursor = start
+            curve = self._curves[p.second].get()
+            serialize(curve, buf)
+            p.first.copy(<char*>cursor, p.first.length(), 0)
+            #0 padding
+            if p.first.length() < 8:
+                memset(cursor + p.first.length(), 0, 8 - p.first.length())
+            cursor += 8
+            memcpy(cursor, self.recovery_rates[p.second].get(), curve.fNumItems * sizeof(double))
+            h ^= Hash64(<char*>buf, buf_size)
+
+        free(buf)
+        h = Hash64WithSeed(<char*>&self.start_date, sizeof(TDate), h)
+        h = Hash64WithSeed(<char*>self._maturities.data(), sizeof(TDate) * self._maturities.size(), h)
+        h =  Hash64WithSeed(<char*>self._weights.data(), sizeof(double) * self._weights.size(), h)
+        return h
+
+
     #@cython.initializedcheck(False)
     def pv_vec(self, step_in_date, value_date, YieldCurve yc, double recovery_rate):
         cdef:
diff --git a/pyisda/curve.pxd b/pyisda/curve.pxd
index 269199c..8d62d65 100644
--- a/pyisda/curve.pxd
+++ b/pyisda/curve.pxd
@@ -41,6 +41,7 @@ cdef extern from "isda/bastypes.h":
 
 cdef extern from "farmhash.h" namespace 'util' nogil:
     uint64_t Hash64(const char *buff, size_t len)
+    uint64_t Hash64WithSeed(const char *buff, size_t len, uint64_t len)
 
 cdef inline size_t TCurve_size(int num_items) nogil:
     return sizeof(int) + sizeof(TDate) + sizeof(double) + \