[Mesa-dev] [PATCH] utils/u_math: break dependency on gallium/utils

[Dylan Baker]

Sigh. Let me see if I can reproduce the failure with mingw, if we can't I'll
revert patch.

Dylan

Quoting Roland Scheidegger (2018-09-20 07:13:11)
> Looks like this broke the windows build, however.
> 
> Roland
> 
> Am 09.09.2018 um 08:39 schrieb Dylan Baker:
> > Currently u_math needs gallium utils for cpu detection.  Most of what
> > u_math uses out of u_cpu_detection is duplicated in src/mesa/x86
> > (surprise!), so I've just reworked it as much as possible to use the
> > x86/common_x86_features macros instead of the gallium ones. There is
> > one small function that was copied over, as promoting u_cpu_detection is
> > itself a fairly hefty undertaking, as it depends on u_debug, and this
> > fixes the bug for now.
> > 
> > bugzilla: https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fbugs.freedesktop.org%2Fshow_bug.cgi%3Fid%3D107870&data=02%7C01%7Csroland%40vmware.com%7C0c9a5dd4cb8946788a2308d6161f376f%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C1%7C0%7C636720720673702384&sdata=Y0wBuxQgoSiTbf8SPFfGD2qP2jzlbDu26f8QC5ksflg%3D&reserved=0
> > cc: vlee at freedesktop.org
> > ---
> > 
> > I have no idea if this fixes the build on mac, I had this in a branch where I
> > started replacing more of imports.{h,c} with utils stuff, and this fixes some
> > build problems on Linux. I don't have a mac to test on and I wont until Monday.
> > If this works let me know.
> > 
> >  src/util/u_math.c | 43 ++++++++++++++++++++++++++++++++++++++-----
> >  1 file changed, 38 insertions(+), 5 deletions(-)
> > 
> > diff --git a/src/util/u_math.c b/src/util/u_math.c
> > index c58af911be7..bf0c398eeec 100644
> > --- a/src/util/u_math.c
> > +++ b/src/util/u_math.c
> > @@ -29,7 +29,7 @@
> >  
> >  #include "pipe/p_config.h"
> >  #include "util/u_math.h"
> > -#include "util/u_cpu_detect.h"
> > +#include "x86/common_x86_features.h"
> >  
> >  #if defined(PIPE_ARCH_SSE)
> >  #include <xmmintrin.h>
> > @@ -90,7 +90,7 @@ util_fpstate_get(void)
> >     unsigned mxcsr = 0;
> >  
> >  #if defined(PIPE_ARCH_SSE)
> > -   if (util_cpu_caps.has_sse) {
> > +   if (cpu_has_xmm) {
> >        mxcsr = _mm_getcsr();
> >     }
> >  #endif
> > @@ -98,6 +98,31 @@ util_fpstate_get(void)
> >     return mxcsr;
> >  }
> >  
> > +/* TODO: this was copied from u_cpu_detection. It's another case of duplication
> > + * between gallium and core mesa, and it would be nice to get rid of that
> > + * duplication as well.
> > + */
> > +#if defined(PIPE_ARCH_X86)
> > +PIPE_ALIGN_STACK static inline bool sse2_has_daz(void)
> > +{
> > +   struct {
> > +      uint32_t pad1[7];
> > +      uint32_t mxcsr_mask;
> > +      uint32_t pad2[128-8];
> > +   } PIPE_ALIGN_VAR(16) fxarea;
> > +
> > +   fxarea.mxcsr_mask = 0;
> > +#if defined(PIPE_CC_GCC)
> > +   __asm __volatile ("fxsave %0" : "+m" (fxarea));
> > +#elif defined(PIPE_CC_MSVC) || defined(PIPE_CC_ICL)
> > +   _fxsave(&fxarea);
> > +#else
> > +   fxarea.mxcsr_mask = 0;
> > +#endif
> > +   return !!(fxarea.mxcsr_mask & (1 << 6));
> > +}
> > +#endif
> > +
> >  /**
> >   * Make sure that the fp treats the denormalized floating
> >   * point numbers as zero.
> > @@ -108,13 +133,21 @@ unsigned
> >  util_fpstate_set_denorms_to_zero(unsigned current_mxcsr)
> >  {
> >  #if defined(PIPE_ARCH_SSE)
> > -   if (util_cpu_caps.has_sse) {
> > +   if (cpu_has_xmm) {
> >        /* Enable flush to zero mode */
> >        current_mxcsr |= _MM_FLUSH_ZERO_MASK;
> > -      if (util_cpu_caps.has_daz) {
> > +      /* x86_64 cpus always have daz, as do cpus with sse3 in fact, there's
> > +       * basically only a handful of very early pentium 4's that have sse2 but
> > +       * not daz.
> > +       */
> > +# if !defined(PIPE_ARCH_x86_64) && !defined(PIPE_ARCH_SSSE3)
> > +      if (sse2_has_daz()) {
> > +# endif
> >           /* Enable denormals are zero mode */
> >           current_mxcsr |= _MM_DENORMALS_ZERO_MASK;
> > +# if !defined(PIPE_ARCH_x86_64) && !defined(PIPE_ARCH_SSSE3)
> >        }
> > +#endif
> >        util_fpstate_set(current_mxcsr);
> >     }
> >  #endif
> > @@ -130,7 +163,7 @@ void
> >  util_fpstate_set(unsigned mxcsr)
> >  {
> >  #if defined(PIPE_ARCH_SSE)
> > -   if (util_cpu_caps.has_sse) {
> > +   if (cpu_has_xmm) {
> >        _mm_setcsr(mxcsr);
> >     }
> >  #endif
> > 
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: signature
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20180920/e7a957ad/attachment.sig>