On Nov 25, 2015 4:18 PM, "Kenneth Graunke" <<a href="mailto:kenneth@whitecape.org">kenneth@whitecape.org</a>> wrote: > > On Tuesday, November 03, 2015 01:26:04 PM Jason Ekstrand wrote: > > --- > > src/mesa/drivers/dri/i965/brw_vec4.h | 4 +++- > > src/mesa/drivers/dri/i965/brw_vec4_nir.cpp | 27 ++++++++++++++++++-------- > > src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp | 12 ++++++++---- > > 3 files changed, 30 insertions(+), 13 deletions(-) > > > > diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h > > index ec8abf4..52d68c5 100644 > > --- a/src/mesa/drivers/dri/i965/brw_vec4.h > > +++ b/src/mesa/drivers/dri/i965/brw_vec4.h > > @@ -273,9 +273,11 @@ public: > > src_reg offset_value, > > src_reg mcs, > > bool is_cube_array, > > + uint32_t surface, src_reg surface_reg, > > uint32_t sampler, src_reg sampler_reg); > > > > - uint32_t gather_channel(unsigned gather_component, uint32_t sampler); > > + uint32_t gather_channel(unsigned gather_component, > > + uint32_t surface, uint32_t sampler); > > src_reg emit_mcs_fetch(const glsl_type *coordinate_type, src_reg coordinate, > > src_reg sampler); > > void emit_gen6_gather_wa(uint8_t wa, dst_reg dst); > > diff --git a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp > > index 7a6372b..86e3dec 100644 > > --- a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp > > +++ b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp > > @@ -1585,7 +1585,9 @@ glsl_type_for_nir_alu_type(nir_alu_type alu_type, > > void > > vec4_visitor::nir_emit_texture(nir_tex_instr *instr) > > { > > + unsigned texture = instr->texture_index; > > Here you call it texture, later in gather_channel() you call it surface. > > I kind of prefer surface, though I suppose it's not the actual binding > table index, but rather a texture unit number. *shrug* We should > probably be consistent though? (I don't feel too strongly though.) Yeah. Texture seemed better from a GL point of view but surface seemed better in the back end. I can rename it all texture if you'd like. > > unsigned sampler = instr->sampler_index; > > + src_reg texture_reg = src_reg(texture); > > src_reg sampler_reg = src_reg(sampler); > > src_reg coordinate; > > const glsl_type *coord_type = NULL; > > @@ -1666,8 +1668,8 @@ vec4_visitor::nir_emit_texture(nir_tex_instr *instr) > > sample_index = get_nir_src(instr->src[i].src, BRW_REGISTER_TYPE_D, 1); > > assert(coord_type != NULL); > > if (devinfo->gen >= 7 && > > - key_tex->compressed_multisample_layout_mask & (1 << sampler)) { > > - mcs = emit_mcs_fetch(coord_type, coordinate, sampler_reg); > > + key_tex->compressed_multisample_layout_mask & (1 << texture)) { > > + mcs = emit_mcs_fetch(coord_type, coordinate, texture_reg); > > } else { > > mcs = src_reg(0u); > > } > > @@ -1679,13 +1681,12 @@ vec4_visitor::nir_emit_texture(nir_tex_instr *instr) > > offset_value = get_nir_src(instr->src[i].src, BRW_REGISTER_TYPE_D, 2); > > break; > > > > - case nir_tex_src_sampler_offset: { > > - /* The highest sampler which may be used by this operation is > > + case nir_tex_src_texture_offset: { > > + /* The highest texture which may be used by this operation is > > * the last element of the array. Mark it here, because the generator > > * doesn't have enough information to determine the bound. > > */ > > - uint32_t array_size = instr->texture_array_size; > > - uint32_t max_used = sampler + array_size - 1; > > + uint32_t max_used = texture + instr->texture_array_size - 1; > > if (instr->op == nir_texop_tg4) { > > max_used += prog_data->base.binding_table.gather_texture_start; > > } else { > > @@ -1697,6 +1698,15 @@ vec4_visitor::nir_emit_texture(nir_tex_instr *instr) > > /* Emit code to evaluate the actual indexing expression */ > > src_reg src = get_nir_src(instr->src[i].src, 1); > > src_reg temp(this, glsl_type::uint_type); > > + emit(ADD(dst_reg(temp), src, src_reg(texture))); > > + texture_reg = emit_uniformize(temp); > > + break; > > + } > > + > > + case nir_tex_src_sampler_offset: { > > + /* Emit code to evaluate the actual indexing expression */ > > + src_reg src = get_nir_src(instr->src[i].src, 1); > > + src_reg temp(this, glsl_type::uint_type); > > emit(ADD(dst_reg(temp), src, src_reg(sampler))); > > sampler_reg = emit_uniformize(temp); > > break; > > @@ -1723,7 +1733,7 @@ vec4_visitor::nir_emit_texture(nir_tex_instr *instr) > > > > /* Stuff the channel select bits in the top of the texture offset */ > > if (instr->op == nir_texop_tg4) > > - constant_offset |= gather_channel(instr->component, sampler) << 16; > > + constant_offset |= gather_channel(instr->component, texture, sampler) << 16; > > > > ir_texture_opcode op = ir_texture_opcode_for_nir_texop(instr->op); > > > > @@ -1736,7 +1746,8 @@ vec4_visitor::nir_emit_texture(nir_tex_instr *instr) > > shadow_comparitor, > > lod, lod2, sample_index, > > constant_offset, offset_value, > > - mcs, is_cube_array, sampler, sampler_reg); > > + mcs, is_cube_array, > > + texture, texture_reg, sampler, sampler_reg); > > } > > > > void > > diff --git a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp > > index b8f90f2..5eb2f7b 100644 > > --- a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp > > +++ b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp > > @@ -879,6 +879,8 @@ vec4_visitor::emit_texture(ir_texture_opcode op, > > src_reg offset_value, > > src_reg mcs, > > bool is_cube_array, > > + uint32_t surface, > > + src_reg surface_reg, > > uint32_t sampler, > > src_reg sampler_reg) > > { > > @@ -937,7 +939,8 @@ vec4_visitor::emit_texture(ir_texture_opcode op, > > inst->dst.writemask = WRITEMASK_XYZW; > > inst->shadow_compare = shadow_comparitor.file != BAD_FILE; > > > > - inst->src[1] = sampler_reg; > > + inst->src[1] = surface_reg; > > + inst->src[2] = sampler_reg; > > > > /* MRF for the first parameter */ > > int param_base = inst->base_mrf + inst->header_size; > > @@ -1054,7 +1057,7 @@ vec4_visitor::emit_texture(ir_texture_opcode op, > > } > > > > if (devinfo->gen == 6 && op == ir_tg4) { > > - emit_gen6_gather_wa(key_tex->gen6_gather_wa[sampler], inst->dst); > > + emit_gen6_gather_wa(key_tex->gen6_gather_wa[surface], inst->dst); > > } > > > > swizzle_result(op, dest, > > @@ -1092,7 +1095,8 @@ vec4_visitor::emit_gen6_gather_wa(uint8_t wa, dst_reg dst) > > * Set up the gather channel based on the swizzle, for gather4. > > */ > > uint32_t > > -vec4_visitor::gather_channel(unsigned gather_component, uint32_t sampler) > > +vec4_visitor::gather_channel(unsigned gather_component, > > + uint32_t surface, uint32_t sampler) > > { > > int swiz = GET_SWZ(key_tex->swizzles[sampler], gather_component); > > switch (swiz) { > > @@ -1101,7 +1105,7 @@ vec4_visitor::gather_channel(unsigned gather_component, uint32_t sampler) > > /* gather4 sampler is broken for green channel on RG32F -- > > * we must ask for blue instead. > > */ > > - if (key_tex->gather_channel_quirk_mask & (1 << sampler)) > > + if (key_tex->gather_channel_quirk_mask & (1 << surface)) > > return 2; > > return 1; > > case SWIZZLE_Z: return 2; > > > > It doesn't look like gather_channel actually needs both parameters, > so maybe only pass in the surface/texture? Well, gather_channel is gone now, so that's not as big a deal.