Re: Should we revert float16 loops (and what are the precision requirements)?
Sebastian Berg <[email protected]> Wed, 11 Mar 2026 12:07:27 +0100
| Newsgroups | gmane.comp.python.numeric.general |
|---|---|
| Message-ID | <[email protected]> |
T24gV2VkLCAyMDI2LTAzLTExIGF0IDExOjU5ICswMTAwLCBSYWxmIEdvbW1lcnMgdmlhIE51bVB5 LURpc2N1c3Npb24NCndyb3RlOg0KPiBPbiBXZWQsIE1hciAxMSwgMjAyNiBhdCAxMDo1OOKAr0FN IG1hdHRpIHBpY3VzIHZpYSBOdW1QeS1EaXNjdXNzaW9uIDwNCj4gbnVtcHktZGlzY3Vzc2lvbkBw eXRob24ub3JnPiB3cm90ZToNCj4gDQo+ID4gT24gVHVlLCBNYXIgMTAsIDIwMjYgYXQgMToyOOKA r1BNIFNlYmFzdGlhbiBCZXJnDQo+ID4gPHNlYmFzdGlhbkBzaXBzb2x1dGlvbnMubmV0PiB3cm90 ZToNCj4gPiA+IA0KPiA+ID4gSGkgYWxsLA0KPiA+ID4gDQo+ID4gPiBJbiB0aGUgTnVtUHkgMi40 IGN5Y2xlLCB0aGVyZSB3ZXJlIHNvbWUgbmF0aXZlIGZsb2F0MTYNCj4gPiA+IGltcGxlbWVudGF0 aW9ucw0KPiA+ID4gbWVyZ2VkIHdpdGggcmF0aGVyIGxvdyBwcmVjaXNpb24gbGVhZGluZyB0byB0 aGUgZm9sbG93aW5nIGlzc3VlOg0KPiA+ID4gaHR0cHM6Ly9naXRodWIuY29tL251bXB5L251bXB5 L2lzc3Vlcy8zMDgyMQ0KPiA+ID4gDQo+ID4gPiBUaGF0IGlzLCBwcmV2aW91c2x5LCBpdCB1c2Vk IGZsb2F0IGxvb3BzIHNvIH4wLjUgVUxQIGVycm9yLCBub3cNCj4gPiA+IGlzIGlzDQo+ID4gPiAy K1VMUCBmb3IgbWFueSBhbGdvcml0aG1zLCBvbiBfc29tZV8gaGFyZHdhcmU6DQo+ID4gPiBodHRw czovL2dpdGh1Yi5jb20vbnVtcHkvbnVtcHkvcHVsbC8yMzM1MQ0KPiA+ID4gDQo+ID4gPiBUaGVy ZSBpcyBhbHdheXMgYW4gYXJndW1lbnQgYXJvdW5kIHRoYXQgdXNlcnMgb2YgZmxvYXQxNiBwcm9i YWJseQ0KPiA+ID4gZG9uJ3QNCj4gPiA+IGNhcmUgYWJvdXQgbWFueSBVTFAsIGJ1dCBJIGd1ZXNz IHRoZXkgYWxzbyBoYXZlIHZlcnkgZmV3IGJpdHMgb2YNCj4gPiA+IHByZWNpc2lvbiB0byBiZWdp biB3aXRoPw0KPiA+ID4gSSBkb24ndCBoYXZlIGEgaHVnZSBvcGluaW9uIG9uIGl0LCBidXQgd2Ug YXJlIG1vcmUgYW5kIG1vcmUgaW4NCj4gPiA+IHRoZQ0KPiA+ID4gcG9zaXRpb24gd2hlcmUgaXQg aXMgdW5jbGVhciBpZiBzYWNyaWZpY2luZyBhIGJpdCBvZiBwcmVjaXNpb24gaXMNCj4gPiA+IHRo ZQ0KPiA+ID4gcmlnaHQgdGhpbmcgb3Igbm90Li4uDQo+ID4gPiANCj4gPiA+IFNpbWlsYXIgcXVl c3Rpb25zIGFjdHVhbGx5IGFyaXNlIGZvciBmbG9hdDMyIG1hdGgsIGlzIGl0IE9LIHRvDQo+ID4g PiB0cmFkZS0NCj4gPiA+IG9mZiBwcmVjaXNpb24gZm9yIHBlcmZvcm1hbmNlIChvciB0byB3aGF0 IGRlZ3JlZSwgZXZlcnl0aGluZw0KPiA+ID4gdHJhZGVzIGENCj4gPiA+IGJpdCk/DQo+ID4gPiBX ZSBoYXZlIGhhZCBkaXNjdXNzaW9ucyBhcm91bmQgdGhpcyBiZWZvcmUgYnV0IGl0IGlzIHN0aWxs IGENCj4gPiA+IGRpZmZpY3VsdA0KPiA+ID4gdHJhZGUtb2ZmIHRvIG1ha2UgYW5kIHRoZXJlIGlz IG5vIGNob2ljZSB0aGF0IG1ha2VzIGV2ZXJ5b25lDQo+ID4gPiBoYXBweS4gWzFdDQo+ID4gPiAN Cj4gPiA+IC0gU2ViYXN0aWFuDQo+ID4gPiANCj4gPiA+IFsxXSBXZSBjYW4gd29yayB0b3dhcmRz IHNvbWV0aGluZyBsaWtlIGBucC5vcHRzKHByZWNpc2lvbj0ibG93IilgDQo+ID4gPiBvcg0KPiA+ ID4gc28sIGJ1dCB0aGF0IGRvZXNuJ3QgY2hhbmdlIHRoZSBxdWVzdGlvbiBvZiBkZWZhdWx0cyBt dWNoLi4uDQo+ID4gDQo+ID4gSSBkbyBsaWtlIHRoZSBpZGVhIG9mIGhhdmluZyBhIHByZWNpc2Uv ZmFzdCB0b2dnbGUuIFVudGlsIHdlIGNhbg0KPiA+IGRldmVsb3Agb25lLCBJIHRoaW5rIHdlIHNo b3VsZCBwcmVmZXIgcHJlY2lzZS4gU28gd2Ugc2hvdWxkIHJldmVydA0KPiA+IGFuZA0KPiA+IGRv Y3VtZW50IHNvbWV3aGVyZSB0aGF0IGZsb2F0MTYgKGFuZCB0aGUgc29vbi10by1iZS1pbmNvbWlu Zw0KPiA+IGJmbG9hdDE2KQ0KPiA+IGFyZSwgaW4gTnVtUHksIGNvbnRhaW5lciB0eXBlcywgYW5k IHRoYXQgYWxsIHRoZSBtYXRoIGZvciB0aGVtIGlzDQo+ID4gZG9uZQ0KPiA+IGFzIGZsb2F0MTYu DQo+ID4gDQo+IA0KPiBZb3UgbWVhbnQgYGZsb2F0MzJgIGhlcmUuIEFuZCB5ZXMsIEkgYWdyZWUu IEhhdmluZyBhIGZldyBjb2RlIHBhdGhzDQo+IHVzZQ0KDQoNCk5vLCBJIG1lYW50IGZsb2F0MTYs IEkgZG9uJ3QgdGhpbmsgd2UgaGF2ZSBhIGJhZCB2YXJpYWJpbGl0eSBmb3INCmZsb2F0MzIgcmln aHQgbm93IGFuZCB3aGlsZSB0aGVyZSBpcyBhIGRpZmZlcmVudCBkaXNjdXNzaW9uIHRvIGJlIGhh ZA0KYWJvdXQgZmxvYXQzMiwgSSB0aGluayB0aG9zZSBwYXRocyB3b3VsZCBhdCBsZWFzdCBiZSBj b25zaXN0ZW50IGFjcm9zcw0KYXJjaGl0ZWN0dXJlcyAoYXMgaXQgd291bGQgYmUgY3VzdG9tIGlt cGxlbWVudGF0aW9ucykuDQoNCkJ1dCBpdCBzb3VuZHMgbGlrZSB5b3UgYWdyZWUgd2l0aCAicmV2 ZXJ0IiBoZXJlLCB3aGljaCB3b3VsZCBpcyBteQ0KdGVuZGVuY3ksIGV2ZW4gaWYgSSBkb24ndCBo YXZlIGEgY2xlYXIgcGljdHVyZSB3aGVyZSB0byBkcmF3IHRoZSBsaW5lLA0Kc2luY2UgaGFyZHdh cmUvcGxhdGZvcm0gZGlmZmVyZW5jZXMgYWx3YXlzIGV4aXN0IHRvIHNvbWUgZGVncmVlLg0KDQot IFNlYmFzdGlhbg0KDQoNCj4gcGxhdGZvcm0vQ1BVLWRlcGVuZGVudCBpbnN0cnVjdGlvbnMgbGlr ZSBBVlg1MTIteHh4IG9uZXMsIGFuZCBhcyBhDQo+IHJlc3VsdA0KPiBoYXZpbmcgYSBzbWFsbCBz dWJzZXQgb2YgdGhlIE51bVB5IEFQSSBoYXZlIGRpZmZlcmVudCBhY2N1cmFjeS9zcGVlZA0KPiB0 cmFkZS1vZmZzIHNlZW1zIG5vdCBhbGwgdGhhdCB1c2VmdWwgdG8gYWxtb3N0IGFsbCB1c2Vycy4g QW5kIG1ha2VzDQo+IGl0DQo+IGhhcmRlciB0byBidWlsZCB1cCBhIG1lbnRhbCBtb2RlbCBvZiB3 aGF0IE51bVB5IGlzIGFjdHVhbGx5IGRvaW5nLg0KPiANCj4gQ2hlZXJzLA0KPiBSYWxmDQo+IF9f X19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fDQo+IE51bVB5LURp c2N1c3Npb24gbWFpbGluZyBsaXN0IC0tIG51bXB5LWRpc2N1c3Npb25AcHl0aG9uLm9yZw0KPiBU byB1bnN1YnNjcmliZSBzZW5kIGFuIGVtYWlsIHRvIG51bXB5LWRpc2N1c3Npb24tbGVhdmVAcHl0 aG9uLm9yZw0KPiBodHRwczovL21haWwucHl0aG9uLm9yZy9tYWlsbWFuMy8vbGlzdHMvbnVtcHkt ZGlzY3Vzc2lvbi5weXRob24ub3JnDQo+IE1lbWJlciBhZGRyZXNzOiBzZWJhc3RpYW5Ac2lwc29s dXRpb25zLm5ldA0KX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19f X18KTnVtUHktRGlzY3Vzc2lvbiBtYWlsaW5nIGxpc3QgLS0gbnVtcHktZGlzY3Vzc2lvbkBweXRo b24ub3JnClRvIHVuc3Vic2NyaWJlIHNlbmQgYW4gZW1haWwgdG8gbnVtcHktZGlzY3Vzc2lvbi1s ZWF2ZUBweXRob24ub3JnCmh0dHBzOi8vbWFpbC5weXRob24ub3JnL21haWxtYW4zLy9saXN0cy9u dW1weS1kaXNjdXNzaW9uLnB5dGhvbi5vcmcKTWVtYmVyIGFkZHJlc3M6IGdjcG5nLW51bXB5LWRp c2N1c3Npb25AbS5nbWFuZS1teC5vcmcK