I started out with predicated moves, but since this is about to go
away again anyway when I do ClipDistance support, I went with the
clearer option. No real perf impact.
-- Chris
On Thu, Jun 6, 2013 at 6:05 PM, Kenneth Graunke <kenn...@whitecape.org> wrote:
> On 06/05/2013 03:50 PM, Chris Forbes wrote:
>>
>> When clipping against a user clip plane, and gl_ClipVertex is provided
>> in the vertex, use it instead of hpos.
>>
>> Signed-off-by: Chris Forbes <chr...@ijw.co.nz>
>> ---
>> src/mesa/drivers/dri/i965/brw_clip.h | 1 +
>> src/mesa/drivers/dri/i965/brw_clip_tri.c | 44
>> +++++++++++++++++++++++++++++---
>> 2 files changed, 41 insertions(+), 4 deletions(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_clip.h
>> b/src/mesa/drivers/dri/i965/brw_clip.h
>> index c6581ad..a0d5d93 100644
>> --- a/src/mesa/drivers/dri/i965/brw_clip.h
>> +++ b/src/mesa/drivers/dri/i965/brw_clip.h
>> @@ -100,6 +100,7 @@ struct brw_clip_compile {
>> struct brw_reg plane_equation;
>>
>> struct brw_reg ff_sync;
>> + struct brw_reg vertex_src_mask;
>> } reg;
>>
>> /* Number of registers storing VUE data */
>> diff --git a/src/mesa/drivers/dri/i965/brw_clip_tri.c
>> b/src/mesa/drivers/dri/i965/brw_clip_tri.c
>> index 05959f6..5491ca8 100644
>> --- a/src/mesa/drivers/dri/i965/brw_clip_tri.c
>> +++ b/src/mesa/drivers/dri/i965/brw_clip_tri.c
>> @@ -119,6 +119,9 @@ void brw_clip_tri_alloc_regs( struct brw_clip_compile
>> *c,
>> i++;
>> }
>>
>> + c->reg.vertex_src_mask = retype(brw_vec1_grf(i, 0),
>> BRW_REGISTER_TYPE_UD);
>> + i++;
>> +
>> if (intel->needs_ff_sync) {
>> c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD);
>> i++;
>> @@ -219,6 +222,31 @@ void brw_clip_tri_flat_shade( struct brw_clip_compile
>> *c )
>> }
>>
>>
>> +static inline void
>> +load_vertex_pos(struct brw_clip_compile *c, struct brw_indirect vtx,
>> + struct brw_reg dst,
>> + GLuint hpos_offset, GLuint clip_offset)
>> +{
>> + struct brw_compile *p = &c->func;
>> +
>> + /*
>> + * Roughly:
>> + * dst = (vertex_src_mask & 1) ? src.hpos : src.clipvertex;
>> + */
>> +
>> + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
>> + brw_AND(p, vec1(brw_null_reg()), c->reg.vertex_src_mask,
>> brw_imm_ud(1));
>> + brw_IF(p, BRW_EXECUTE_1);
>> + {
>> + brw_MOV(p, dst, deref_4f(vtx, clip_offset));
>> + }
>> + brw_ELSE(p);
>> + {
>> + brw_MOV(p, dst, deref_4f(vtx, hpos_offset));
>> + }
>> + brw_ENDIF(p);
>
>
> I wonder, is it possible to use SEL here? Or failing that, predicated
> moves? If not, no worries...the rest of the code is pretty well littered
> with IFs already.
>
> I second Paul's request for additional comments. Other than that, this
> looks like a nice fix! Somehow I imagined it would be harder.
>
> Both patches are:
> Reviewed-by: Kenneth Graunke <kenn...@whitecape.org>
>
> Thanks for doing this, Chris!
>
>
>> +}
>> +
>>
>> /* Use mesa's clipping algorithms, translated to GEN4 assembly.
>> */
>> @@ -233,13 +261,17 @@ void brw_clip_tri( struct brw_clip_compile *c )
>> struct brw_indirect outlist_ptr = brw_indirect(5, 0);
>> struct brw_indirect freelist_ptr = brw_indirect(6, 0);
>> GLuint hpos_offset = brw_varying_to_offset(&c->vue_map,
>> VARYING_SLOT_POS);
>> -
>> + GLuint clipvert_offset = brw_clip_have_varying(c,
>> VARYING_SLOT_CLIP_VERTEX)
>> + ? brw_varying_to_offset(&c->vue_map, VARYING_SLOT_CLIP_VERTEX)
>> + : hpos_offset;
>> +
>> brw_MOV(p, get_addr_reg(vtxPrev), brw_address(c->reg.vertex[2])
>> );
>> brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
>> brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
>> brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
>>
>> brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3])
>> );
>> + brw_MOV(p, c->reg.vertex_src_mask, brw_imm_ud(0xfc0));
>>
>> brw_DO(p, BRW_EXECUTE_1);
>> {
>> @@ -269,15 +301,17 @@ void brw_clip_tri( struct brw_clip_compile *c )
>> */
>> brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0));
>>
>> + load_vertex_pos(c, vtxPrev, vec4(c->reg.dpPrev), hpos_offset,
>> clipvert_offset);
>> /* IS_NEGATIVE(prev) */
>> brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
>> - brw_DP4(p, vec4(c->reg.dpPrev), deref_4f(vtxPrev,
>> hpos_offset), c->reg.plane_equation);
>> + brw_DP4(p, vec4(c->reg.dpPrev), vec4(c->reg.dpPrev),
>> c->reg.plane_equation);
>> brw_IF(p, BRW_EXECUTE_1);
>> {
>> + load_vertex_pos(c, vtx, vec4(c->reg.dp), hpos_offset,
>> clipvert_offset);
>> /* IS_POSITIVE(next)
>> */
>> brw_set_conditionalmod(p, BRW_CONDITIONAL_GE);
>> - brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, hpos_offset),
>> c->reg.plane_equation);
>> + brw_DP4(p, vec4(c->reg.dp), vec4(c->reg.dp),
>> c->reg.plane_equation);
>> brw_IF(p, BRW_EXECUTE_1);
>> {
>>
>> @@ -316,10 +350,11 @@ void brw_clip_tri( struct brw_clip_compile *c )
>> brw_ADD(p, get_addr_reg(outlist_ptr),
>> get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
>> brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts,
>> brw_imm_ud(1));
>>
>> + load_vertex_pos(c, vtx, vec4(c->reg.dp), hpos_offset,
>> clipvert_offset);
>> /* IS_NEGATIVE(next)
>> */
>> brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
>> - brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, hpos_offset),
>> c->reg.plane_equation);
>> + brw_DP4(p, vec4(c->reg.dp), vec4(c->reg.dp),
>> c->reg.plane_equation);
>> brw_IF(p, BRW_EXECUTE_1);
>> {
>> /* Going out of bounds. Avoid division by zero as we
>> @@ -392,6 +427,7 @@ void brw_clip_tri( struct brw_clip_compile *c )
>> */
>> brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
>> brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
>> + brw_SHR(p, c->reg.vertex_src_mask, c->reg.vertex_src_mask,
>> brw_imm_ud(1));
>> }
>> brw_WHILE(p);
>> }
>>
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
