[Mesa-dev] [PATCH] radeon/llvm: improve select_cc lowering to generate CND* more often

[Tom Stellard]

> > diff --git a/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c b/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c
> > index 04469e2..bb37154 100644
> > --- a/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c
> > +++ b/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c
> > @@ -934,6 +934,20 @@ static void emit_u2f(
> >  			emit_data->args[0], bld_base->base.elem_type, "");
> >  }
> >  
> > +static void emit_cndlt(
> > +		const struct lp_build_tgsi_action * action,
> > +		struct lp_build_tgsi_context * bld_base,
> > +		struct lp_build_emit_data * emit_data)
> > +{
> > +	LLVMBuilderRef builder = bld_base->base.gallivm->builder;
> > +	LLVMValueRef float_zero = lp_build_const_float(
> > +		bld_base->base.gallivm, 0.0f);
> > +	LLVMValueRef cmp = LLVMBuildFCmp(
> > +		builder, LLVMRealULT, emit_data->args[0], float_zero, "");
> > +	emit_data->output[emit_data->chan] = LLVMBuildSelect(builder,
> > +		cmp, emit_data->args[1], emit_data->args[2], "");
> > +}
> > +
> >  static void emit_immediate(struct lp_build_tgsi_context * bld_base,
> >  		const struct tgsi_full_immediate *imm)
> >  {
> > @@ -1115,8 +1129,7 @@ void radeon_llvm_context_init(struct radeon_llvm_context * ctx)
> >  	bld_base->op_actions[TGSI_OPCODE_CONT].emit = cont_emit;
> >  	bld_base->op_actions[TGSI_OPCODE_CLAMP].emit = build_tgsi_intrinsic_nomem;
> >  	bld_base->op_actions[TGSI_OPCODE_CLAMP].intr_name = "llvm.AMDIL.clamp.";
> > -	bld_base->op_actions[TGSI_OPCODE_CMP].emit = build_tgsi_intrinsic_nomem;
> > -	bld_base->op_actions[TGSI_OPCODE_CMP].intr_name = "llvm.AMDGPU.cndlt";
> > +	bld_base->op_actions[TGSI_OPCODE_CMP].emit = emit_cndlt;
> >  	bld_base->op_actions[TGSI_OPCODE_COS].emit = build_tgsi_intrinsic_nomem;
> >  	bld_base->op_actions[TGSI_OPCODE_COS].intr_name = "llvm.AMDGPU.cos";
> >  	bld_base->op_actions[TGSI_OPCODE_DIV].emit = build_tgsi_intrinsic_nomem;
> 
> It would be nice if you could also remove the llvm.AMDGPU.cndlt
> intrinsic and also delete the SI pseudo instruction that uses it.  I can
> test on SI for you.
>

Unfortunately, removing this intrinsic from SI uncovers a bug that may
be difficult to solve, so for now just leave the intrinsic, so SI can
use it.

-Tom
> 
> > -- 
> > 1.7.11.4
> > 
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/mesa-dev
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev