BLD: 4.0.1 fails on i386 systems: call to 'countOnes' is ambiguous #3477
Comments
|
@yurivict thank you for this! I appreciate the bug reports. I'm not sure I can build it here without installing FreeBSD, which is probably what I ought to do. I will note that we've also run into issues with some strangeness in how the EWAH code compiles with MSVC++, too, and I will check to see if there's any overlap in the issues. |
neutrinoceros
changed the title
|
This issue is not specific to FreeBSD but to the i386 platform. I can reproduce it with the following steps on Linux (x86-64!): export CONDA_FORCE_32BIT=1
export CFLAGS=-m32
conda create -n py37_32 python=3.7
conda activate py37_32
pip install cython numpy
python setup.py developThis fails with the following traceback (abridged) OS: Arch Linux |
|
Seems to be fixed with this patch diff --git a/yt/utilities/lib/ewahboolarray/boolarray.h b/yt/utilities/lib/ewahboolarray/boolarray.h
index 822d576b4..e67f91bae 100644
--- a/yt/utilities/lib/ewahboolarray/boolarray.h
+++ b/yt/utilities/lib/ewahboolarray/boolarray.h
@@ -17,6 +17,13 @@
#include "ewahutil.h"
+
+#if ((ULONG_MAX) == (UINT_MAX))
+#define UWORD uint32_t
+#else
+#define UWORD uint64_t
+#endif
+
namespace ewah {
/**
* A dynamic bitset implementation. (without compression).
@@ -344,7 +351,7 @@ public:
size_t numberOfOnes() const {
size_t count = 0;
for (size_t i = 0; i < buffer.size(); ++i) {
- count += countOnes(buffer[i]);
+ count += countOnes((UWORD)buffer[i]);
}
return count;
}
diff --git a/yt/utilities/lib/ewahboolarray/ewah-inl.h b/yt/utilities/lib/ewahboolarray/ewah-inl.h
index 6176c5e15..5b6e0f281 100644
--- a/yt/utilities/lib/ewahboolarray/ewah-inl.h
+++ b/yt/utilities/lib/ewahboolarray/ewah-inl.h
@@ -3,6 +3,12 @@
#include "ewah.h"
+#if ((ULONG_MAX) == (UINT_MAX))
+#define UWORD uint32_t
+#else
+#define UWORD uint64_t
+#endif
+
namespace ewah {
/**
@@ -213,7 +219,7 @@ public:
return;
} else {
uword t = static_cast<uword>(word & (~word + 1));
- answer = literalPosition + countOnes((uword)(t - 1));
+ answer = literalPosition + countOnes((UWORD)(t - 1));
word ^= t;
}
hasNext = moveToNext();
@@ -371,7 +377,7 @@ template <class uword> size_t EWAHBoolArray<uword>::numberOfOnes() const {
}
++pointer;
for (size_t k = 0; k < rlw.getNumberOfLiteralWords(); ++k) {
- tot += countOnes((uword)buffer[pointer]);
+ tot += countOnes((UWORD)buffer[pointer]);
++pointer;
}
}Can anyone confirm? |
|
@cphyc I assume the answer-store being included in your patch is a mistake ? |
|
I'm adding this to the 4.0.2 milestone to make sure this is discussed before the bug fix release |
Correct. I'm amending my patch to remove the answer-store update. |
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Log: http://beefy17.nyi.freebsd.org/data/main-i386-default/pa8720dde236b_seba8e643b1/logs/py38-yt-4.0.1.log (IPv6 URL)
OS: FreeBSD 13
The text was updated successfully, but these errors were encountered: