- Add patch to fix aarch64 support: * numpy-fix-cpu_asimdfhm.patch OBS-URL: https://build.opensuse.org/request/show/935455 OBS-URL: https://build.opensuse.org/package/show/devel:languages:python:numeric/python-numpy?expand=0&rev=100
30 lines
1.2 KiB
Diff
30 lines
1.2 KiB
Diff
From 2113cad8cbb4c4bc9469e0c94025bd7cbbe6f2d6 Mon Sep 17 00:00:00 2001
|
|
From: Thomas Green <tomgreen66@hotmail.com>
|
|
Date: Tue, 16 Nov 2021 22:36:18 +0000
|
|
Subject: [PATCH] Update cpu_asimdfhm.c
|
|
|
|
Updated `vfmlal_low_u32` and `vfmlslq_high_u32` to their `f16` new names. Described here:
|
|
|
|
https://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg664008.html
|
|
|
|
Many of the intrinsics had names updated. Supposedly previous specifications were not published so old names not required.
|
|
---
|
|
numpy/distutils/checks/cpu_asimdfhm.c | 4 ++--
|
|
1 file changed, 2 insertions(+), 2 deletions(-)
|
|
|
|
diff --git a/numpy/distutils/checks/cpu_asimdfhm.c b/numpy/distutils/checks/cpu_asimdfhm.c
|
|
index bb437aa4035..cb49751c4c2 100644
|
|
--- a/numpy/distutils/checks/cpu_asimdfhm.c
|
|
+++ b/numpy/distutils/checks/cpu_asimdfhm.c
|
|
@@ -10,8 +10,8 @@ int main(void)
|
|
float32x4_t vf = vdupq_n_f32(1.0f);
|
|
float32x2_t vlf = vdup_n_f32(1.0f);
|
|
|
|
- int ret = (int)vget_lane_f32(vfmlal_low_u32(vlf, vlhp, vlhp), 0);
|
|
- ret += (int)vgetq_lane_f32(vfmlslq_high_u32(vf, vhp, vhp), 0);
|
|
+ int ret = (int)vget_lane_f32(vfmlal_low_f16(vlf, vlhp, vlhp), 0);
|
|
+ ret += (int)vgetq_lane_f32(vfmlslq_high_f16(vf, vhp, vhp), 0);
|
|
|
|
return ret;
|
|
}
|