@Kevin,

The compiler I built a few years back was optimized for pentium4; it
wouldn't accept the -mavx2 option.  That's why I had to hack libvpx to
disable avx completely.


Anyway, Here's the patch to libvpx if anyone is running into the same issue
at link time:

--- source/config/linux/ia32/vpx_dsp_rtcd.h.vanilla    2018-01-03
13:13:06.000000000 -0500
+++ source/config/linux/ia32/vpx_dsp_rtcd.h    2018-01-03
12:04:56.000000000 -0500
@@ -1969,6 +1969,8 @@
 #include "vpx_ports/x86.h"
 static void setup_rtcd_internal(void)
 {
+    // pham
+    return;
     int flags = x86_simd_caps();

     (void)flags;
debug-linux26:[libvpx]# diff -urN
source/config/linux/ia32/vp9_rtcd.h.vanilla
source/config/linux/ia32/vp9_rtcd.h
--- source/config/linux/ia32/vp9_rtcd.h.vanilla    2018-01-03
13:12:24.000000000 -0500
+++ source/config/linux/ia32/vp9_rtcd.h    2018-01-03 12:14:11.000000000
-0500
@@ -153,6 +153,9 @@
 {
     int flags = x86_simd_caps();

+    // pham
+    return;
+
     (void)flags;

     vp9_block_error = vp9_block_error_c;



On Wed, Jan 3, 2018 at 12:48 PM, Kevin Kofler <kevin.kof...@chello.at>
wrote:

> Toan Pham wrote:
> > I over came the memory limitation by forcing ninja to run in single
> thread
> > (pass -j 1 to ninja).  However, at linking stage using a 32bit linker,
> the
> > linker failed because it could not find references to avx2 in libvpx, a
> > third-party library that I hacked to get AVX2 disabled.  Since you've
> been
> > recommending that I should use 64bit linker, this error is completely
> > unrelated to the memory address limit of a 32bit linker.  While the 32bit
> > linker was doing its job linking libQt5WebEngine.so, its maximum memory
> > usage (RSS) was 1.8GB, not anywhere close to the maximum addressable
> limit
> > of 32bit executables.
>
> libvpx detects vector instructions at runtime. Therefore, it is not
> necessary to disable AVX2 support even if your machine does not support
> AVX2.
>
>         Kevin Kofler
>
> _______________________________________________
> Development mailing list
> Development@qt-project.org
> http://lists.qt-project.org/mailman/listinfo/development
>
_______________________________________________
Development mailing list
Development@qt-project.org
http://lists.qt-project.org/mailman/listinfo/development

Reply via email to