On Don, 2013-08-08 at 08:00 -0700, Tom Stellard wrote: > On Thu, Aug 08, 2013 at 02:20:54AM +0200, Marek Olšák wrote: > > > > diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c > > b/src/gallium/drivers/radeonsi/si_state_draw.c > > index 746ace6..4208fa7 100644 > > --- a/src/gallium/drivers/radeonsi/si_state_draw.c > > +++ b/src/gallium/drivers/radeonsi/si_state_draw.c > > @@ -241,6 +241,7 @@ static void si_pipe_shader_ps(struct pipe_context *ctx, > > struct si_pipe_shader *s > > /* Last 2 reserved SGPRs are used for VCC */ > > num_sgprs = num_user_sgprs + 2; > > } > > + num_sgprs += 1; /* XXX this fixes VM faults */ > > One problem is that the compiler is under reporting the number of SGPRs, > when there are unused USER_SGPRs in the shader. It should always be > reporting a number greater than or equal to the number of USER_SGPRs.
That itself shouldn't be a problem, as the radeonsi code ensures this. > I think Michel mentioned this earlier, but there may also be a problem > with the way we determine usage of the VCC register in the shader, maybe > it is being used for more instructions than we realize. I was just worrying that AMDGPUAsmPrinter::EmitProgramInfoSI() might not take implicit access to VCC into account, but Christian confirmed it should. AFAICT we should be encoding any such implicit access correctly (except for V_DIV_SCALE_F* and V_DIV_FMAS_F*, which aren't used yet). -- Earthling Michel Dänzer | http://www.amd.com Libre software enthusiast | Debian, X and DRI developer _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
