Merge branch 'main' into fix-pickle-scalar-99

Author: SwayamInSync

.github/workflows/test_old_cpu.yml

@@ -39,7 +39,10 @@ jobs:
 
       - name: Install Intel SDE
         run: |
-          curl -o /tmp/sde.tar.xz https://downloadmirror.intel.com/859732/sde-external-9.58.0-2025-06-16-lin.tar.xz
+          SDE_URL="https://github.com/SwayamInSync/numpy-quaddtype/releases/download/sde-toolchain/sde-external-10.8.0-2026-03-15-lin.tar.xz"
+          SDE_SHA256="50b320cd226acef7a491f5b321fc1be3c3c7984f9e27a456e64894b5b0979dd3"
+          curl -fSL -o /tmp/sde.tar.xz "$SDE_URL"
+          echo "$SDE_SHA256  /tmp/sde.tar.xz" | sha256sum -c -
           mkdir /tmp/sde && tar -xvf /tmp/sde.tar.xz -C /tmp/sde/
           sudo mv /tmp/sde/* /opt/sde && sudo ln -s /opt/sde/sde64 /usr/bin/sde
 

src/csrc/quadblas_interface.cpp

@@ -25,7 +25,14 @@ int
 qblas_dot(size_t n, Sleef_quad *x, size_t incx,
           Sleef_quad *y, size_t incy, Sleef_quad *result)
 {
-    if (!x || !y || !result || n == 0) {
+    if (!result) {
+        return -1;
+    }
+    if (n == 0) {
+        *result = Sleef_cast_from_doubleq1(0.0);
+        return 0;
+    }
+    if (!x || !y) {
         return -1;
     }
     *result = cblas_qdot((int)n, x, (int)incx, y, (int)incy);
@@ -38,7 +45,10 @@ qblas_gemv(char layout, char trans, size_t m, size_t n,
            Sleef_quad *x, size_t incx,
            Sleef_quad *beta, Sleef_quad *y, size_t incy)
 {
-    if (!alpha || !A || !x || !beta || !y || m == 0 || n == 0) {
+    if (m == 0 || n == 0) {
+        return 0;
+    }
+    if (!alpha || !A || !x || !beta || !y) {
         return -1;
     }
     cblas_qgemv(to_layout(layout), to_trans(trans),
@@ -56,7 +66,10 @@ qblas_gemm(char layout, char transa, char transb,
            Sleef_quad *B, size_t ldb,
            Sleef_quad *beta, Sleef_quad *C, size_t ldc)
 {
-    if (!alpha || !A || !B || !beta || !C || m == 0 || n == 0 || k == 0) {
+    if (m == 0 || n == 0 || k == 0) {
+        return 0;
+    }
+    if (!alpha || !A || !B || !beta || !C) {
         return -1;
     }
     cblas_qgemm(to_layout(layout), to_trans(transa), to_trans(transb),

src/csrc/scalar.c

@@ -2,6 +2,7 @@
 #include <sleef.h>
 #include <sleefquad.h>
 #include <stdlib.h>
+#include <math.h>
 
 #define PY_ARRAY_UNIQUE_SYMBOL QuadPrecType_ARRAY_API
 #define NPY_NO_DEPRECATED_API NPY_2_0_API_VERSION
@@ -21,6 +22,11 @@
 #include "pythoncapi_compat.h"
 
 
+// forward declaration
+static Sleef_quad
+longdouble_to_quad(long double value);
+
+
 QuadPrecisionObject *
 QuadPrecision_raw_new(QuadBackendType backend)
 {
@@ -430,8 +436,7 @@ QuadPrecision_is_integer(QuadPrecisionObject *self, PyObject *Py_UNUSED(ignored)
         value = self->value.sleef_value;
     }
     else {
-        // lets also tackle ld from sleef functions as well
-        value = Sleef_cast_from_doubleq1((double)self->value.longdouble_value);
+        value = longdouble_to_quad(self->value.longdouble_value); 
     }
 
     if(Sleef_iunordq1(value, value)) {
@@ -449,9 +454,7 @@ QuadPrecision_is_integer(QuadPrecisionObject *self, PyObject *Py_UNUSED(ignored)
 
     // Check if value equals its truncated version
     Sleef_quad truncated = Sleef_truncq1(value);
-    int32_t is_equal = Sleef_icmpeqq1(value, truncated);
-    
-    if (is_equal) {
+              if (Sleef_icmpeqq1(value, truncated)) {
         Py_RETURN_TRUE;
     }
     else {
@@ -461,7 +464,7 @@ QuadPrecision_is_integer(QuadPrecisionObject *self, PyObject *Py_UNUSED(ignored)
 
 PyObject* quad_to_pylong(Sleef_quad value)
 {
-    char buffer[128];
+    char buffer[4936]; // 4933 + sign + null terminator, enough for 128-bit integer in decimal
 
     // Sleef_snprintf call is thread-unsafe
     LOCK_SLEEF;
@@ -479,6 +482,37 @@ PyObject* quad_to_pylong(Sleef_quad value)
     return result;
 }
 
+PyObject* longdouble_to_pylong(long double value)
+{
+  char buffer[4936]; // 4933 + sign + null terminator, enough for 128-bit integer in decimal
+
+  // POSIX guarantees thread-safety of snprintf
+  int written = snprintf(buffer, sizeof(buffer), "%.0Lf", value);
+  if (written < 0 || written >= sizeof(buffer)) {
+    PyErr_SetString(PyExc_RuntimeError, "Failed to convert long double to string");
+    return NULL;
+  }
+
+  // Already raises ValueError and returns NULL on failure
+  return PyLong_FromString(buffer, NULL, 10);
+}
+
+static Sleef_quad
+longdouble_to_quad(long double value)
+{
+  if (isnan(value) || isinf(value) || value == 0.0L)
+    return Sleef_cast_from_doubleq1((double)value);
+  
+  int exp;
+  long double mantissa = frexpl(value, &exp);
+  long double scaled = ldexpl(mantissa, 64);
+  exp -= 64;
+  Sleef_quad q = (scaled < 0) 
+    ? Sleef_negq1(Sleef_cast_from_uint64q1((uint64_t)(-scaled))) 
+    : Sleef_cast_from_uint64q1((uint64_t)scaled);
+  return Sleef_ldexpq1(q, exp);
+}
+
 // inspired by the CPython implementation
 // https://github.com/python/cpython/blob/ac1ffd77858b62d169a08040c08aa5de26e145ac/Objects/floatobject.c#L1503C1-L1572C2
 static PyObject *
@@ -491,10 +525,8 @@ QuadPrecision_as_integer_ratio(QuadPrecisionObject *self, PyObject *Py_UNUSED(ig
 
     if (self->backend == BACKEND_SLEEF) {
         value = self->value.sleef_value;
-    }
-    else {
-        // lets also tackle ld from sleef functions as well
-        value = Sleef_cast_from_doubleq1((double)self->value.longdouble_value);
+    } else {
+        value = longdouble_to_quad(self->value.longdouble_value);
     }
 
     if(Sleef_iunordq1(value, value)) {
@@ -697,7 +729,7 @@ QuadPrecision_hash(QuadPrecisionObject *self)
         value = self->value.sleef_value;
     }
     else {
-        value = Sleef_cast_from_doubleq1((double)self->value.longdouble_value);
+        value = longdouble_to_quad(self->value.longdouble_value);
     }
 
     // Check for NaN - use pointer hash (each NaN instance gets unique hash)

src/csrc/scalar_ops.cpp

@@ -3,6 +3,8 @@
 #define NPY_TARGET_VERSION NPY_2_4_API_VERSION
 #define NO_IMPORT_ARRAY
 
+#include <cmath>
+
 extern "C" {
 #include <Python.h>
 
@@ -224,12 +226,35 @@ QuadPrecision_float(QuadPrecisionObject *self)
 static PyObject *
 QuadPrecision_int(QuadPrecisionObject *self)
 {
-    if (self->backend == BACKEND_SLEEF) {
-        return PyLong_FromLongLong(Sleef_cast_to_int64q1(self->value.sleef_value));
+  if (self->backend == BACKEND_SLEEF) 
+  {
+      Sleef_quad value = self->value.sleef_value;
+      if (quad_isnan(&value)) {
+        PyErr_SetString(PyExc_ValueError, "cannot convert float NaN to integer");
+        return NULL;
     }
-    else {
-        return PyLong_FromLongLong((long long)self->value.longdouble_value);
+    if (quad_isinf(&value)) 
+    {
+        PyErr_SetString(PyExc_OverflowError,
+                        "cannot convert float infinity to integer");
+        return NULL;
+    }
+    return quad_to_pylong(Sleef_truncq1(value));
+
+    }
+
+    long double value = self->value.longdouble_value;
+    if(std::isnan(value))
+    {
+      PyErr_SetString(PyExc_ValueError, "cannot convert float NaN to integer");
+      return NULL;
+    }
+    if(std::isinf(value))
+    {
+      PyErr_SetString(PyExc_OverflowError, "cannot convert float infinity to integer");
+      return NULL;
     }
+    return longdouble_to_pylong(truncl(value));
 }
 
 template <binary_op_quad_def sleef_op, binary_op_longdouble_def longdouble_op>