Skip to content

Commit 1fe6c81

Browse files
committed
wrapping up
1 parent cec5ace commit 1fe6c81

File tree

3 files changed

+6
-27
lines changed

3 files changed

+6
-27
lines changed

quaddtype/README.md

Lines changed: 6 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -27,7 +27,7 @@ np.array([1,2,3], dtype=QuadPrecDType("longdouble"))
2727

2828
The code needs the quad precision pieces of the sleef library, which
2929
is not available on most systems by default, so we have to generate
30-
that first. The below assumes one has the required pieces to build
30+
that first. The below assumes one has the required pieces to build
3131
sleef (cmake and libmpfr-dev), and that one is in the package
3232
directory locally.
3333

@@ -40,6 +40,7 @@ cd ..
4040
```
4141

4242
Building the `numpy-quaddtype` package from locally installed sleef:
43+
4344
```bash
4445
export SLEEF_DIR=$PWD/sleef/build
4546
export LIBRARY_PATH=$SLEEF_DIR/lib
@@ -57,10 +58,13 @@ export LDFLAGS="-Wl,-rpath,$SLEEF_DIR/lib -fopenmp -latomic -lpthread"
5758
export CFLAGS="-fPIC"
5859
export CXXFLAGS="-fPIC"
5960

61+
# To build without QBLAS (default for MSVC)
62+
# export CFLAGS="-fPIC -DDISABLE_QUADBLAS"
63+
# export CXXFLAGS="-fPIC -DDISABLE_QUADBLAS"
64+
6065
python -m pip install . -v --no-build-isolation -Cbuilddir=build -C'compile-args=-v'
6166

6267
# Run the tests
6368
cd ..
6469
python -m pytest
6570
```
66-

quaddtype/numpy_quaddtype/src/umath/matmul.h

Lines changed: 0 additions & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -1,35 +1,12 @@
11
#ifndef _QUADDTYPE_MATMUL_H
22
#define _QUADDTYPE_MATMUL_H
33

4-
/**
5-
* Quad Precision Matrix Multiplication for NumPy
6-
*
7-
* This module implements matrix multiplication functionality for the QuadPrecDType
8-
* by registering custom loops with numpy's matmul generalized ufunc.
9-
*
10-
* Supports all matmul operation types:
11-
* - Vector-vector (dot product): (n,) @ (n,) -> scalar
12-
* - Matrix-vector: (m,n) @ (n,) -> (m,)
13-
* - Vector-matrix: (n,) @ (n,p) -> (p,)
14-
* - Matrix-matrix: (m,n) @ (n,p) -> (m,p)
15-
*
16-
* Uses naive algorithms optimized for correctness rather than performance.
17-
* For production use, consider integration with QBLAS optimized routines.
18-
*/
19-
204
#include <Python.h>
215

226
#ifdef __cplusplus
237
extern "C" {
248
#endif
259

26-
/**
27-
* Initialize the matmul operations for the quad precision dtype.
28-
* This function registers the matmul generalized ufunc with numpy.
29-
*
30-
* @param numpy The numpy module object
31-
* @return 0 on success, -1 on failure
32-
*/
3310
int
3411
init_matmul_ops(PyObject *numpy);
3512

quaddtype/release_tracker.md

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -91,5 +91,3 @@
9191
| floor ||
9292
| ceil ||
9393
| trunc ||
94-
95-
- Fixing QBLAS integration to work unaligned arrays without or recovering from bad allocation fallback

0 commit comments

Comments
 (0)