[Mesa-dev] [PATCH 1/4] gallium/ttn: add support for texture offsets

[Ilia Mirkin]

On Mon, Apr 13, 2015 at 1:34 PM, Eric Anholt <eric at anholt.net> wrote:
> Rob Clark <robdclark at gmail.com> writes:
>
>> From: Rob Clark <robclark at freedesktop.org>
>>
>> Signed-off-by: Rob Clark <robclark at freedesktop.org>
>> ---
>>  src/gallium/auxiliary/nir/tgsi_to_nir.c | 30 +++++++++++++++++++++++++++++-
>>  1 file changed, 29 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/gallium/auxiliary/nir/tgsi_to_nir.c b/src/gallium/auxiliary/nir/tgsi_to_nir.c
>> index 9d988b06..c9f9e03 100644
>> --- a/src/gallium/auxiliary/nir/tgsi_to_nir.c
>> +++ b/src/gallium/auxiliary/nir/tgsi_to_nir.c
>> @@ -982,7 +982,7 @@ ttn_tex(struct ttn_compile *c, nir_alu_dest dest, nir_ssa_def **src)
>>     struct tgsi_full_instruction *tgsi_inst = &c->token->FullInstruction;
>>     nir_tex_instr *instr;
>>     nir_texop op;
>> -   unsigned num_srcs, samp = 1;
>> +   unsigned num_srcs, samp = 1, i;
>>
>>     switch (tgsi_inst->Instruction.Opcode) {
>>     case TGSI_OPCODE_TEX:
>> @@ -1026,6 +1026,8 @@ ttn_tex(struct ttn_compile *c, nir_alu_dest dest, nir_ssa_def **src)
>>        num_srcs++;
>>     }
>>
>> +   num_srcs += tgsi_inst->Texture.NumOffsets;
>> +
>>     instr = nir_tex_instr_create(b->shader, num_srcs);
>>     instr->op = op;
>>
>> @@ -1103,6 +1105,32 @@ ttn_tex(struct ttn_compile *c, nir_alu_dest dest, nir_ssa_def **src)
>>        src_number++;
>>     }
>>
>> +   for (i = 0; i < tgsi_inst->Texture.NumOffsets; i++) {
>> +      struct tgsi_texture_offset *tex_offset = &tgsi_inst->TexOffsets[i];
>> +      /* since TexOffset ins't using tgsi_full_src_register we get to
>> +       * do some extra gymnastics:
>> +       */
>> +      nir_alu_src src;
>> +
>> +      memset(&src, 0, sizeof(src));
>> +
>> +      src.src = ttn_src_for_file_and_index(c,
>> +                                           tex_offset->File,
>> +                                           tex_offset->Index,
>> +                                           NULL);
>> +
>> +      src.swizzle[0] = tex_offset->SwizzleX;
>> +      src.swizzle[1] = tex_offset->SwizzleY;
>> +      src.swizzle[2] = tex_offset->SwizzleZ;
>> +      src.swizzle[3] = TGSI_SWIZZLE_W;
>> +
>> +      nir_ssa_def *off = nir_fmov_alu(b, src, instr->coord_components);
>
> For a cubemap array, won't coord_components be 4, but you only want a
> 3-component value as the offset?  I'm suspicious of this fixed SWIZZLE_W
> here.  If you move the .src_type setting above this, you could use
> nir_tex_instr_src_size(instr, src_number) to get the right number of
> components.

When are offsets on cubemaps supported? The real question is what will
coord_components be for a 2darray -- if it's 2, then this is fine, if
it's 3, then it needs fixing.

  -ilia