[Mesa-dev] [PATCH 07/14] i965/vs: Implement vec4_visitor::visit(ir_texture *).

[Kenneth Graunke]

On 12/13/2011 01:12 PM, Eric Anholt wrote:
> On Thu,  8 Dec 2011 17:07:58 -0800, Kenneth Graunke <kenneth at whitecape.org> wrote:
>> This translates the GLSL compiler's IR into vec4_instruction IR,
>> generating code to load coordinates, LOD info, shadow comparitors, and
>> so on into the appropriate message registers.
>>
>> It turns out that the SIMD4x2 parameters are identical on Gen 5-7, and
>> the Gen4 code is similar enough that, unlike in the FS, it's easy enough
>> to support all generations in a single function.
>>
>> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
>> ---
>>  src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp |  114 ++++++++++++++++++++++--
>>  1 files changed, 107 insertions(+), 7 deletions(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
>> index 853c3ee..85490bb 100644
>> --- a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
>> +++ b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
> 
>> +   vec4_instruction *inst;
>> +   switch (ir->op) {
>> +   case ir_tex:
> 
> Isn't this one invalid for vertex shaders, too?

Nope.  From the GLSL 1.30 spec, page 80 (86 in the PDF):
"The implicit level of detail is selected as follows: For a texture that
is not mip-mapped, the texture is used directly.  If it is mip-mapped
and running in a fragment shader, the LOD computed by the implementation
is used to do the texture lookup.  If it is mip-mapped and running on
the vertex shader, then the base texture is used."

So ir_tex is equivalent to ir_txl with lod 0.

>> +   case ir_txl:
>> +      inst = new(mem_ctx) vec4_instruction(this, SHADER_OPCODE_TXL);
>> +      break;
>> +   case ir_txd:
>> +      inst = new(mem_ctx) vec4_instruction(this, SHADER_OPCODE_TXD);
>> +      break;
>> +   case ir_txf:
>> +      inst = new(mem_ctx) vec4_instruction(this, SHADER_OPCODE_TXF);
>> +      break;
>> +   case ir_txs:
>> +      inst = new(mem_ctx) vec4_instruction(this, SHADER_OPCODE_TXS);
>> +      break;
>> +   case ir_txb:
>> +      assert(!"TXB is not valid for vertex shaders.");
>> +   }
> 
>> +      /* Load the LOD info */
>> +      if (ir->op == ir_txl) {
>> +	 ir->lod_info.lod->accept(this);
>> +	 int mrf, writemask;
>> +	 if (intel->gen >= 5) {
>> +	    mrf = param_base + 1;
>> +	    writemask = ir->shadow_comparitor ? WRITEMASK_Y : WRITEMASK_X;
>> +	    inst->mlen++;
>> +	 } else /* intel->gen == 4 */ {
>> +	    mrf = param_base;
>> +	    writemask = WRITEMASK_Z;
>> +	 }
>> +	 emit(MOV(dst_reg(MRF, mrf, ir->lod_info.lod->type, writemask),
>> +		  this->result));
> 
> I'd move the accept down next to the MOV.  I'm always afraid of the
> hidden return data from accept getting trashed before it gets consumed.
> I wish the visitor stuff let us do the sensible thing and produce a
> return value.

I totally agree - they used to be right next to each other before I
added the Gen4 code.  I'll move it.