Re: [Freedreno] [PATCH] freedreno/a4xx: add stencil texturing support
On Sun, Nov 19, 2017 at 3:14 PM, Ilia Mirkin wrote:
> Copied from a5xx, should be identical.
>
> Signed-off-by: Ilia Mirkin
Reviewed-by: Rob Clark
> ---
> docs/features.txt| 6 ++---
> src/gallium/drivers/freedreno/a4xx/fd4_emit.c| 2 ++
> src/gallium/drivers/freedreno/a4xx/fd4_format.c | 11 +---
> src/gallium/drivers/freedreno/a4xx/fd4_texture.c | 34
> ++--
> 4 files changed, 38 insertions(+), 15 deletions(-)
>
> diff --git a/docs/features.txt b/docs/features.txt
> index 99fb1715e0b..2d6e0b20fb5 100644
> --- a/docs/features.txt
> +++ b/docs/features.txt
> @@ -180,7 +180,7 @@ GL 4.3, GLSL 4.30 -- all DONE: i965/gen8+, nvc0, radeonsi
>GL_ARB_robust_buffer_access_behavior DONE (i965)
>GL_ARB_shader_image_size DONE
> (freedreno/a5xx, i965, r600, softpipe)
>GL_ARB_shader_storage_buffer_object DONE
> (freedreno/a5xx, i965, softpipe)
> - GL_ARB_stencil_texturing DONE
> (freedreno/a5xx, i965/hsw+, nv50, r600, llvmpipe, softpipe, swr)
> + GL_ARB_stencil_texturing DONE (freedreno,
> i965/hsw+, nv50, r600, llvmpipe, softpipe, swr)
>GL_ARB_texture_buffer_range DONE (freedreno,
> nv50, i965, r600, llvmpipe)
>GL_ARB_texture_query_levels DONE (all drivers
> that support GLSL 1.30)
>GL_ARB_texture_storage_multisampleDONE (all drivers
> that support GL_ARB_texture_multisample)
> @@ -203,7 +203,7 @@ GL 4.4, GLSL 4.40 -- all DONE: i965/gen8+, nvc0, radeonsi
>GL_ARB_multi_bind DONE (all drivers)
>GL_ARB_query_buffer_objectDONE (i965/hsw+)
>GL_ARB_texture_mirror_clamp_to_edge DONE (i965, nv50,
> r600, llvmpipe, softpipe, swr)
> - GL_ARB_texture_stencil8 DONE
> (freedreno/a5xx, i965/hsw+, nv50, r600, llvmpipe, softpipe, swr)
> + GL_ARB_texture_stencil8 DONE (freedreno,
> i965/hsw+, nv50, r600, llvmpipe, softpipe, swr)
>GL_ARB_vertex_type_10f_11f_11f_revDONE (i965, nv50,
> r600, llvmpipe, softpipe, swr)
>
> GL 4.5, GLSL 4.50 -- all DONE: nvc0, radeonsi
> @@ -252,7 +252,7 @@ GLES3.1, GLSL ES 3.1 -- all DONE: i965/hsw+, nvc0,
> radeonsi
>GL_ARB_shader_storage_buffer_object DONE
> (freedreno/a5xx, i965/gen7+, softpipe)
>GL_ARB_shading_language_packing DONE (all drivers)
>GL_ARB_separate_shader_objectsDONE (all drivers)
> - GL_ARB_stencil_texturing DONE
> (freedreno/a5xx, nv50, r600, llvmpipe, softpipe, swr)
> + GL_ARB_stencil_texturing DONE (freedreno,
> nv50, r600, llvmpipe, softpipe, swr)
>GL_ARB_texture_multisample (Multisample textures) DONE (i965/gen7+,
> nv50, r600, llvmpipe, softpipe)
>GL_ARB_texture_storage_multisampleDONE (all drivers
> that support GL_ARB_texture_multisample)
>GL_ARB_vertex_attrib_binding DONE (all drivers)
> diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
> b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
> index 0f7c6470330..8262b45daad 100644
> --- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
> +++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
> @@ -190,6 +190,8 @@ emit_textures(struct fd_context *ctx, struct
> fd_ringbuffer *ring,
> OUT_RING(ring, view->texconst3);
> if (view->base.texture) {
> struct fd_resource *rsc =
> fd_resource(view->base.texture);
> + if (view->base.format ==
> PIPE_FORMAT_X32_S8X24_UINT)
> + rsc = rsc->stencil;
> OUT_RELOC(ring, rsc->bo, view->offset,
> view->texconst4, 0);
> } else {
> OUT_RING(ring, 0x);
> diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_format.c
> b/src/gallium/drivers/freedreno/a4xx/fd4_format.c
> index 3e1dc277850..75d24126149 100644
> --- a/src/gallium/drivers/freedreno/a4xx/fd4_format.c
> +++ b/src/gallium/drivers/freedreno/a4xx/fd4_format.c
> @@ -211,10 +211,13 @@ static struct fd4_format formats[PIPE_FORMAT_COUNT] = {
> VT(R11G11B10_FLOAT, 11_11_10_FLOAT, R11G11B10_FLOAT, WZYX),
> _T(R9G9B9E5_FLOAT, 9_9_9_E5_FLOAT, NONE,WZYX),
>
> - _T(Z24X8_UNORM, X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
> - _T(Z24_UNORM_S8_UINT, X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
> - _T(Z32_FLOAT, 32_FLOAT, R8G8B8A8_UNORM, WZYX),
> - _T(Z32_FLOAT_S8X24_UINT, 32_FLOAT,R8G8B8A8_UNORM, WZYX),
> + _T(Z16_UNORM,16_UNORM, R16_UNORM, WZYX),
> + _T(Z24X8_UN
Re: [Freedreno] [PATCH 2/2] freedreno/ir3: add a pass to lower tg4 to txl, enable gather on a4xx
On Sun, Nov 19, 2017 at 2:54 PM, Ilia Mirkin wrote: > Unfortunately Adreno A4xx hardware returns incorrect results with the > GATHER4 opcodes. As a result, we have to lower to 4 individual texture > calls (txl since we have to force lod to 0). We achieve this using > offsets, including on cube maps which normally never have offsets. > > Signed-off-by: Ilia Mirkin > --- > > This pass relies on the hw doing the "right thing", working with nonconst > offsets, and not having the usual limits (since the gather offset will in > effect get offset by another 1). > > It fails two tests out of all the gather ones: > > bin/zero-tex-coord textureGather > tests/spec/arb_gpu_shader5/execution/built-in-functions/fs-textureGatherOffset-uniform-array-offset.shader_test > > We haven't fully investigated why yet, but this is a good start. > > Note that the blob does this differently - they modify the source coordinate. > However this seems unnecessary given that the hw can be made to use the > offsets. > > Also please note that my knowledge of nir is minimal. Please carefully check > that I used the right helpers/etc. This was largely a result of seeing what > doesn't result in assertions. > > docs/features.txt | 4 +- > src/gallium/drivers/freedreno/Makefile.sources | 1 + > src/gallium/drivers/freedreno/freedreno_screen.c | 2 +- > .../drivers/freedreno/ir3/ir3_compiler_nir.c | 7 +- > src/gallium/drivers/freedreno/ir3/ir3_nir.c| 2 + > src/gallium/drivers/freedreno/ir3/ir3_nir.h| 1 + > .../freedreno/ir3/ir3_nir_lower_tg4_to_tex.c | 139 > + > src/gallium/drivers/freedreno/meson.build | 1 + > 8 files changed, 152 insertions(+), 5 deletions(-) > create mode 100644 > src/gallium/drivers/freedreno/ir3/ir3_nir_lower_tg4_to_tex.c > > diff --git a/docs/features.txt b/docs/features.txt > index 633d2593738..99fb1715e0b 100644 > --- a/docs/features.txt > +++ b/docs/features.txt > @@ -130,7 +130,7 @@ GL 4.0, GLSL 4.00 --- all DONE: i965/gen7+, nvc0, r600, > radeonsi >GL_ARB_tessellation_shaderDONE (i965/gen7+) >GL_ARB_texture_buffer_object_rgb32DONE (freedreno, > i965/gen6+, llvmpipe, softpipe, swr) >GL_ARB_texture_cube_map_array DONE (i965/gen6+, > nv50, llvmpipe, softpipe) > - GL_ARB_texture_gather DONE > (freedreno/a5xx, i965/gen6+, nv50, llvmpipe, softpipe, swr) > + GL_ARB_texture_gather DONE (freedreno, > i965/gen6+, nv50, llvmpipe, softpipe, swr) >GL_ARB_texture_query_lod DONE (freedreno, > i965, nv50, llvmpipe, softpipe) >GL_ARB_transform_feedback2DONE (i965/gen6+, > nv50, llvmpipe, softpipe, swr) >GL_ARB_transform_feedback3DONE (i965/gen7+, > llvmpipe, softpipe, swr) > @@ -256,7 +256,7 @@ GLES3.1, GLSL ES 3.1 -- all DONE: i965/hsw+, nvc0, > radeonsi >GL_ARB_texture_multisample (Multisample textures) DONE (i965/gen7+, > nv50, r600, llvmpipe, softpipe) >GL_ARB_texture_storage_multisampleDONE (all drivers > that support GL_ARB_texture_multisample) >GL_ARB_vertex_attrib_binding DONE (all drivers) > - GS5 Enhanced textureGatherDONE (i965/gen7+, > r600) > + GS5 Enhanced textureGatherDONE (freedreno, > i965/gen7+, r600) >GS5 Packing/bitfield/conversion functions DONE (i965/gen6+, > r600) >GL_EXT_shader_integer_mix DONE (all drivers > that support GLSL) > > diff --git a/src/gallium/drivers/freedreno/Makefile.sources > b/src/gallium/drivers/freedreno/Makefile.sources > index b109a5a7a21..40c2eff0455 100644 > --- a/src/gallium/drivers/freedreno/Makefile.sources > +++ b/src/gallium/drivers/freedreno/Makefile.sources > @@ -168,6 +168,7 @@ ir3_SOURCES := \ > ir3/ir3_nir.c \ > ir3/ir3_nir.h \ > ir3/ir3_nir_lower_if_else.c \ > + ir3/ir3_nir_lower_tg4_to_tex.c \ > ir3/ir3_print.c \ > ir3/ir3_ra.c \ > ir3/ir3_sched.c \ > diff --git a/src/gallium/drivers/freedreno/freedreno_screen.c > b/src/gallium/drivers/freedreno/freedreno_screen.c > index e61344fd104..62e4a574b90 100644 > --- a/src/gallium/drivers/freedreno/freedreno_screen.c > +++ b/src/gallium/drivers/freedreno/freedreno_screen.c > @@ -264,7 +264,7 @@ fd_screen_get_param(struct pipe_screen *pscreen, enum > pipe_cap param) > return 0; > > case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS: > - if (is_a5xx(screen)) > + if (is_a4xx(screen) || is_a5xx(screen)) > return 4; > return 0; > > diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c > b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c > ind
Re: [Freedreno] [Mesa-dev] [PATCH 1/2] nir: allow texture offsets with cube maps
On Sun, Nov 19, 2017 at 11:54 AM, Ilia Mirkin wrote:
> GL doesn't have this, but some hardware supports it. This is convenient
> for lowering tg4 to plain texture calls, which is necessary on Adreno
> A4xx hardware.
>
> Signed-off-by: Ilia Mirkin
> ---
> src/compiler/nir/nir.h | 15 +--
> 1 file changed, 13 insertions(+), 2 deletions(-)
>
> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
> index f46f6147110..64965ae16d6 100644
> --- a/src/compiler/nir/nir.h
> +++ b/src/compiler/nir/nir.h
> @@ -1364,8 +1364,7 @@ nir_tex_instr_src_size(const nir_tex_instr *instr,
> unsigned src)
> if (instr->src[src].src_type == nir_tex_src_ms_mcs)
>return 4;
>
> - if (instr->src[src].src_type == nir_tex_src_offset ||
> - instr->src[src].src_type == nir_tex_src_ddx ||
> + if (instr->src[src].src_type == nir_tex_src_ddx ||
> instr->src[src].src_type == nir_tex_src_ddy) {
>if (instr->is_array)
> return instr->coord_components - 1;
> @@ -1373,6 +1372,18 @@ nir_tex_instr_src_size(const nir_tex_instr *instr,
> unsigned src)
> return instr->coord_components;
> }
>
> + /* Usual APIs don't allow cube + offset, but we allow it, with 2
> coords for
> +* the offset, since a cube maps to a single face.
> +*/
> + if (instr->src[src].src_type == nir_tex_src_offset) {
> + unsigned ret = instr->coord_components;
> + if (instr->is_array)
> + ret--;
> + if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
> + ret--;
> + return ret;
>
I think I'd rather this look more like the one above:
if (instr->is_array)
return instr->coord_components;
else if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
return 2;
else
return instr->coord_components - 1;
It seems a bit cleaner and/or more explicit to me.
Also, bonus points to anyone who converts this function to use a switch. :-P
--Jason
> + }
> +
> return 1;
> }
>
> --
> 2.13.6
>
> ___
> mesa-dev mailing list
> mesa-...@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
___
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno
Re: [Freedreno] [Mesa-dev] [PATCH 1/2] nir: allow texture offsets with cube maps
On Mon, Nov 20, 2017 at 5:16 PM, Jason Ekstrand wrote:
> On Sun, Nov 19, 2017 at 11:54 AM, Ilia Mirkin wrote:
>>
>> GL doesn't have this, but some hardware supports it. This is convenient
>> for lowering tg4 to plain texture calls, which is necessary on Adreno
>> A4xx hardware.
>>
>> Signed-off-by: Ilia Mirkin
>> ---
>> src/compiler/nir/nir.h | 15 +--
>> 1 file changed, 13 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
>> index f46f6147110..64965ae16d6 100644
>> --- a/src/compiler/nir/nir.h
>> +++ b/src/compiler/nir/nir.h
>> @@ -1364,8 +1364,7 @@ nir_tex_instr_src_size(const nir_tex_instr *instr,
>> unsigned src)
>> if (instr->src[src].src_type == nir_tex_src_ms_mcs)
>>return 4;
>>
>> - if (instr->src[src].src_type == nir_tex_src_offset ||
>> - instr->src[src].src_type == nir_tex_src_ddx ||
>> + if (instr->src[src].src_type == nir_tex_src_ddx ||
>> instr->src[src].src_type == nir_tex_src_ddy) {
>>if (instr->is_array)
>> return instr->coord_components - 1;
>> @@ -1373,6 +1372,18 @@ nir_tex_instr_src_size(const nir_tex_instr *instr,
>> unsigned src)
>> return instr->coord_components;
>> }
>>
>> + /* Usual APIs don't allow cube + offset, but we allow it, with 2
>> coords for
>> +* the offset, since a cube maps to a single face.
>> +*/
>> + if (instr->src[src].src_type == nir_tex_src_offset) {
>> + unsigned ret = instr->coord_components;
>> + if (instr->is_array)
>> + ret--;
>> + if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
>> + ret--;
>> + return ret;
>
>
> I think I'd rather this look more like the one above:
>
> if (instr->is_array)
>return instr->coord_components;
> else if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
>return 2;
> else
>return instr->coord_components - 1;
>
> It seems a bit cleaner and/or more explicit to me.
OK. Although your version is slightly wrong, but I get the idea. Will
do that in a v2. (array should get -1, and cube should always get 2
even if it's an array)
-ilia
>
> Also, bonus points to anyone who converts this function to use a switch. :-P
>
> --Jason
>
>>
>> + }
>> +
>> return 1;
>> }
>>
>> --
>> 2.13.6
>>
>> ___
>> mesa-dev mailing list
>> mesa-...@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
>
>
> ___
> Freedreno mailing list
> Freedreno@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/freedreno
>
___
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno
Re: [Freedreno] [Mesa-dev] [PATCH 1/2] nir: allow texture offsets with cube maps
On Mon, Nov 20, 2017 at 3:11 PM, Ilia Mirkin wrote:
> On Mon, Nov 20, 2017 at 5:16 PM, Jason Ekstrand
> wrote:
> > On Sun, Nov 19, 2017 at 11:54 AM, Ilia Mirkin
> wrote:
> >>
> >> GL doesn't have this, but some hardware supports it. This is convenient
> >> for lowering tg4 to plain texture calls, which is necessary on Adreno
> >> A4xx hardware.
> >>
> >> Signed-off-by: Ilia Mirkin
> >> ---
> >> src/compiler/nir/nir.h | 15 +--
> >> 1 file changed, 13 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
> >> index f46f6147110..64965ae16d6 100644
> >> --- a/src/compiler/nir/nir.h
> >> +++ b/src/compiler/nir/nir.h
> >> @@ -1364,8 +1364,7 @@ nir_tex_instr_src_size(const nir_tex_instr *instr,
> >> unsigned src)
> >> if (instr->src[src].src_type == nir_tex_src_ms_mcs)
> >>return 4;
> >>
> >> - if (instr->src[src].src_type == nir_tex_src_offset ||
> >> - instr->src[src].src_type == nir_tex_src_ddx ||
> >> + if (instr->src[src].src_type == nir_tex_src_ddx ||
> >> instr->src[src].src_type == nir_tex_src_ddy) {
> >>if (instr->is_array)
> >> return instr->coord_components - 1;
> >> @@ -1373,6 +1372,18 @@ nir_tex_instr_src_size(const nir_tex_instr
> *instr,
> >> unsigned src)
> >> return instr->coord_components;
> >> }
> >>
> >> + /* Usual APIs don't allow cube + offset, but we allow it, with 2
> >> coords for
> >> +* the offset, since a cube maps to a single face.
> >> +*/
> >> + if (instr->src[src].src_type == nir_tex_src_offset) {
> >> + unsigned ret = instr->coord_components;
> >> + if (instr->is_array)
> >> + ret--;
> >> + if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
> >> + ret--;
> >> + return ret;
> >
> >
> > I think I'd rather this look more like the one above:
> >
> > if (instr->is_array)
> >return instr->coord_components;
> > else if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
> >return 2;
> > else
> >return instr->coord_components - 1;
> >
> > It seems a bit cleaner and/or more explicit to me.
>
> OK. Although your version is slightly wrong, but I get the idea. Will
> do that in a v2. (array should get -1, and cube should always get 2
> even if it's an array)
>
I'd forgotten about cube arrays, yes, those would naturally be -2. In that
case, maybe -- for each subtraction is a good idea...
> -ilia
>
> >
> > Also, bonus points to anyone who converts this function to use a switch.
> :-P
> >
> > --Jason
> >
> >>
> >> + }
> >> +
> >> return 1;
> >> }
> >>
> >> --
> >> 2.13.6
> >>
> >> ___
> >> mesa-dev mailing list
> >> mesa-...@lists.freedesktop.org
> >> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> >
> >
> >
> > ___
> > Freedreno mailing list
> > Freedreno@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/freedreno
> >
>
___
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno
Re: [Freedreno] [Mesa-dev] [PATCH 1/2] nir: allow texture offsets with cube maps
On Mon, Nov 20, 2017 at 7:08 PM, Jason Ekstrand wrote:
> On Mon, Nov 20, 2017 at 3:11 PM, Ilia Mirkin wrote:
>>
>> On Mon, Nov 20, 2017 at 5:16 PM, Jason Ekstrand
>> wrote:
>> > On Sun, Nov 19, 2017 at 11:54 AM, Ilia Mirkin
>> > wrote:
>> >>
>> >> GL doesn't have this, but some hardware supports it. This is convenient
>> >> for lowering tg4 to plain texture calls, which is necessary on Adreno
>> >> A4xx hardware.
>> >>
>> >> Signed-off-by: Ilia Mirkin
>> >> ---
>> >> src/compiler/nir/nir.h | 15 +--
>> >> 1 file changed, 13 insertions(+), 2 deletions(-)
>> >>
>> >> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
>> >> index f46f6147110..64965ae16d6 100644
>> >> --- a/src/compiler/nir/nir.h
>> >> +++ b/src/compiler/nir/nir.h
>> >> @@ -1364,8 +1364,7 @@ nir_tex_instr_src_size(const nir_tex_instr
>> >> *instr,
>> >> unsigned src)
>> >> if (instr->src[src].src_type == nir_tex_src_ms_mcs)
>> >>return 4;
>> >>
>> >> - if (instr->src[src].src_type == nir_tex_src_offset ||
>> >> - instr->src[src].src_type == nir_tex_src_ddx ||
>> >> + if (instr->src[src].src_type == nir_tex_src_ddx ||
>> >> instr->src[src].src_type == nir_tex_src_ddy) {
>> >>if (instr->is_array)
>> >> return instr->coord_components - 1;
>> >> @@ -1373,6 +1372,18 @@ nir_tex_instr_src_size(const nir_tex_instr
>> >> *instr,
>> >> unsigned src)
>> >> return instr->coord_components;
>> >> }
>> >>
>> >> + /* Usual APIs don't allow cube + offset, but we allow it, with 2
>> >> coords for
>> >> +* the offset, since a cube maps to a single face.
>> >> +*/
>> >> + if (instr->src[src].src_type == nir_tex_src_offset) {
>> >> + unsigned ret = instr->coord_components;
>> >> + if (instr->is_array)
>> >> + ret--;
>> >> + if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
>> >> + ret--;
>> >> + return ret;
>> >
>> >
>> > I think I'd rather this look more like the one above:
>> >
>> > if (instr->is_array)
>> >return instr->coord_components;
>> > else if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
>> >return 2;
>> > else
>> >return instr->coord_components - 1;
>> >
>> > It seems a bit cleaner and/or more explicit to me.
>>
>> OK. Although your version is slightly wrong, but I get the idea. Will
>> do that in a v2. (array should get -1, and cube should always get 2
>> even if it's an array)
>
>
> I'd forgotten about cube arrays, yes, those would naturally be -2. In that
> case, maybe -- for each subtraction is a good idea...
Well, rearranging it, we get
if (instr->sampler_dim == CUBE)
return 2;
else if (instr->is_array)
return comp - 1;
else
return comp;
Happy to leave it alone too though.
___
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno
Re: [Freedreno] [Mesa-dev] [PATCH 1/2] nir: allow texture offsets with cube maps
On Mon, Nov 20, 2017 at 4:10 PM, Ilia Mirkin wrote:
> On Mon, Nov 20, 2017 at 7:08 PM, Jason Ekstrand
> wrote:
> > On Mon, Nov 20, 2017 at 3:11 PM, Ilia Mirkin
> wrote:
> >>
> >> On Mon, Nov 20, 2017 at 5:16 PM, Jason Ekstrand
> >> wrote:
> >> > On Sun, Nov 19, 2017 at 11:54 AM, Ilia Mirkin
> >> > wrote:
> >> >>
> >> >> GL doesn't have this, but some hardware supports it. This is
> convenient
> >> >> for lowering tg4 to plain texture calls, which is necessary on Adreno
> >> >> A4xx hardware.
> >> >>
> >> >> Signed-off-by: Ilia Mirkin
> >> >> ---
> >> >> src/compiler/nir/nir.h | 15 +--
> >> >> 1 file changed, 13 insertions(+), 2 deletions(-)
> >> >>
> >> >> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
> >> >> index f46f6147110..64965ae16d6 100644
> >> >> --- a/src/compiler/nir/nir.h
> >> >> +++ b/src/compiler/nir/nir.h
> >> >> @@ -1364,8 +1364,7 @@ nir_tex_instr_src_size(const nir_tex_instr
> >> >> *instr,
> >> >> unsigned src)
> >> >> if (instr->src[src].src_type == nir_tex_src_ms_mcs)
> >> >>return 4;
> >> >>
> >> >> - if (instr->src[src].src_type == nir_tex_src_offset ||
> >> >> - instr->src[src].src_type == nir_tex_src_ddx ||
> >> >> + if (instr->src[src].src_type == nir_tex_src_ddx ||
> >> >> instr->src[src].src_type == nir_tex_src_ddy) {
> >> >>if (instr->is_array)
> >> >> return instr->coord_components - 1;
> >> >> @@ -1373,6 +1372,18 @@ nir_tex_instr_src_size(const nir_tex_instr
> >> >> *instr,
> >> >> unsigned src)
> >> >> return instr->coord_components;
> >> >> }
> >> >>
> >> >> + /* Usual APIs don't allow cube + offset, but we allow it, with 2
> >> >> coords for
> >> >> +* the offset, since a cube maps to a single face.
> >> >> +*/
> >> >> + if (instr->src[src].src_type == nir_tex_src_offset) {
> >> >> + unsigned ret = instr->coord_components;
> >> >> + if (instr->is_array)
> >> >> + ret--;
> >> >> + if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
> >> >> + ret--;
> >> >> + return ret;
> >> >
> >> >
> >> > I think I'd rather this look more like the one above:
> >> >
> >> > if (instr->is_array)
> >> >return instr->coord_components;
> >> > else if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE)
> >> >return 2;
> >> > else
> >> >return instr->coord_components - 1;
> >> >
> >> > It seems a bit cleaner and/or more explicit to me.
> >>
> >> OK. Although your version is slightly wrong, but I get the idea. Will
> >> do that in a v2. (array should get -1, and cube should always get 2
> >> even if it's an array)
> >
> >
> > I'd forgotten about cube arrays, yes, those would naturally be -2. In
> that
> > case, maybe -- for each subtraction is a good idea...
>
> Well, rearranging it, we get
>
> if (instr->sampler_dim == CUBE)
> return 2;
> else if (instr->is_array)
> return comp - 1;
> else
> return comp;
>
Right. With that if-ladder,
Reviewed-by: Jason Ekstrand
___
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno
