[Mesa-dev] [PATCH 61/78] i965/nir/vec4: Add skeleton implementation of nir_emit_texture()

Jason Ekstrand jason at jlekstrand.net
Mon Jul 13 05:05:34 PDT 2015


On Mon, Jul 13, 2015 at 3:00 AM, Alejandro Piñeiro <apinheiro at igalia.com> wrote:
> On 01/07/15 01:37, Jason Ekstrand wrote:
>> If we can avoid duplication in the texturing code, that would be
>> really nice.  Could we do this as a refactor of the old code and then
>> a much smaller NIR function that calls some shared code?  That's what
>> we did for FS and it worked ok.  I looked at the layout of your code
>> and, after you finish reading instruction inputs, very little of it is
>> actually NIR-specific.
>
> Our wip second batch of commits has now a version that refactors the old
> code. But I have a doubt. As part of this refactor, I created a new
> method to be used by brw_fs and brw_vec4, that returns the ir texture
> opcode based on the nir texture opcode. You can take a look to that
> method here [1]. The issue with this method is that ir_texture_opcode is
> defined at glsl/ir.h inside a #ifdef __cplusplus. So I needed to rename
> brw_nir.c to brw_nir.cpp. On a previous early review [2] you mentioned
> that it would be better to keep it as brw_nir.c unless we have a good
> reason. Is using this enum good reason to do the renaming? If not, the
> other option I see is modify glsl/ir.h and move the enum definitions to
> the top, outside the ifdef __cplusplus, as other headers (like
> glsl/glsl_types.h) have.

Go ahead and make it a cpp file if you have to.  Another option would
be to do the map the other way: ir_opcode -> nir_tex_op and refactor
the code to use the nir_tex_op.  That's probably what we should be
doing in the FS now that the old visitor is gone.
--Jason

> BR
>
> [1]
> https://github.com/Igalia/mesa/commit/e81ce150ef931e50b6cb1aae0b42a79f448863f5
> [2] https://bugs.freedesktop.org/show_bug.cgi?id=89580#c7
>> --Jason
>>
>> On Fri, Jun 26, 2015 at 1:07 AM, Eduardo Lima Mitev <elima at igalia.com> wrote:
>>> From: Alejandro Piñeiro <apinheiro at igalia.com>
>>>
>>> This sets up the basic structure and placeholders for the implementation of
>>> texture operations, which will be added incrementally in subsequent patches.
>>>
>>> Apart from helping split the texture support into smaller patches, this patch
>>> is useful to get a simplified picture of the steps involved in the emission of
>>> texture operations.
>>>
>>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89580
>>> ---
>>>  src/mesa/drivers/dri/i965/brw_vec4_nir.cpp | 120 ++++++++++++++++++++++++++++-
>>>  1 file changed, 119 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
>>> index 08a70b0..c3638a0 100644
>>> --- a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
>>> +++ b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
>>> @@ -1412,7 +1412,125 @@ vec4_visitor::nir_swizzle_result(nir_tex_instr *instr, dst_reg dest,
>>>  void
>>>  vec4_visitor::nir_emit_texture(nir_tex_instr *instr)
>>>  {
>>> -   /* @TODO: Not yet implemented */
>>> +   unsigned sampler = instr->sampler_index;
>>> +   src_reg sampler_reg = src_reg(sampler);
>>> +
>>> +   const glsl_type *dest_type;
>>> +
>>> +   /* Load the texture operation sources */
>>> +   for (unsigned i = 0; i < instr->num_srcs; i++) {
>>> +      src_reg src = get_nir_src(instr->src[i].src);
>>> +
>>> +      switch (instr->src[i].src_type) {
>>> +      case nir_tex_src_comparitor:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_coord:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_ddx:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_ddy:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_lod:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_ms_index:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_offset:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_sampler_offset:
>>> +         /* @TODO: not yet implemented */
>>> +         break;
>>> +
>>> +      case nir_tex_src_projector:
>>> +         unreachable("Should be lowered by do_lower_texture_projection");
>>> +
>>> +      case nir_tex_src_bias:
>>> +         unreachable("LOD bias is not valid for vertex shaders.\n");
>>> +
>>> +      default:
>>> +         unreachable("unknown texture source");
>>> +      }
>>> +   }
>>> +
>>> +   /* Get the texture operation opcode */
>>> +   enum opcode opcode = shader_opcode_for_nir_opcode(instr->op);
>>> +
>>> +   /* Build the instruction */
>>> +   enum glsl_base_type dest_base_type =
>>> +      brw_glsl_base_type_for_nir_type(instr->dest_type);
>>> +
>>> +   dest_type =
>>> +      glsl_type::get_instance(dest_base_type, nir_tex_instr_dest_size(instr), 1);
>>> +
>>> +   vec4_instruction *inst = new(mem_ctx)
>>> +      vec4_instruction(opcode, dst_reg(this, dest_type));
>>> +
>>> +   for (unsigned i = 0; i < 3; i++) {
>>> +      if (instr->const_offset[i] != 0) {
>>> +         inst->offset = brw_texture_offset(instr->const_offset, 3);
>>> +         break;
>>> +      }
>>> +   }
>>> +
>>> +   /* The message header is necessary for:
>>> +    * - Texel offsets
>>> +    * - Gather channel selection
>>> +    * - Sampler indices too large to fit in a 4-bit value.
>>> +    */
>>> +   inst->header_size =
>>> +      (inst->offset != 0 ||
>>> +       instr->op == nir_texop_tg4 ||
>>> +       is_high_sampler(sampler_reg)) ? 1 : 0;
>>> +   inst->base_mrf = 2;
>>> +   inst->mlen = inst->header_size + 1;
>>> +   inst->dst.writemask = WRITEMASK_XYZW;
>>> +
>>> +   inst->src[1] = sampler_reg;
>>> +
>>> +   /* Calculate the MRF for the first parameter */
>>> +   int param_base = inst->base_mrf + inst->header_size;
>>> +
>>> +   if (instr->op == nir_texop_txs || instr->op == nir_texop_query_levels) {
>>> +      /* @TODO: not yet implemented */
>>> +   } else {
>>> +      /* @TODO: Load the coordinate */
>>> +
>>> +      /* @TODO: Load the shadow comparitor */
>>> +
>>> +      /* Load the LOD info */
>>> +      if (instr->op == nir_texop_tex || instr->op == nir_texop_txl) {
>>> +         /* @TODO: not yet implemented */
>>> +      } else if (instr->op == nir_texop_txf) {
>>> +         /* @TODO: not yet implemented */
>>> +      } else if (instr->op == nir_texop_txf_ms) {
>>> +         /* @TODO: not yet implemented */
>>> +      } else if (instr->op == nir_texop_txd) {
>>> +         /* @TODO: not yet implemented */
>>> +      } else if (instr->op == nir_texop_tg4) {
>>> +         /* @TODO: not yet implemented */
>>> +      }
>>> +   }
>>> +
>>> +   /* Emit the instruction */
>>> +   emit(inst);
>>> +
>>> +   /* Move the result to the destination registry, applying swizzle */
>>> +   dst_reg dest = get_nir_dest(instr->dest, brw_type_for_base_type(dest_type));
>>> +
>>> +   nir_swizzle_result(instr, dest, src_reg(inst->dst), sampler, dest_type);
>>>  }
>>>
>>>  }
>>> --
>>> 2.1.4
>>>
>>> _______________________________________________
>>> 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
>
> --
> Alejandro Piñeiro (apinheiro at igalia.com)
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev


More information about the mesa-dev mailing list