[Bug target/62178] [AArch64] Performance regression on matrix matrix multiply due to r211211

2014-08-19 Thread amker.cheng at gmail dot com
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62178 bin.cheng changed: What|Removed |Added CC||amker.cheng at gmail dot com --- Comment #3

[Bug target/62178] [AArch64] Performance regression on matrix matrix multiply due to r211211

2014-08-18 Thread amker at gcc dot gnu.org
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62178 --- Comment #2 from amker at gcc dot gnu.org --- Yes, the patch changes addressing modes choosing and further changes ivopts's decision. I shall take this one if you are ok. Thanks, bin

[Bug target/62178] [AArch64] Performance regression on matrix matrix multiply due to r211211

2014-08-18 Thread pinskia at gcc dot gnu.org
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62178 --- Comment #1 from Andrew Pinski --- Even -fno-ivopts produces better code: .L3: addx0, x5, x1, sxtw addw1, w1, 1 ldrd2, [x3], 124 ldrw0, [x4, x0, lsl 2] dupv0.2s, w0 mlav1.2s, v0.2s, v2.2s sub