[Mesa-dev] mesa : ARB_texture_gather implementation - WIP[PATCH1/1] but need advice !
Kenneth Graunke
kenneth at whitecape.org
Tue Nov 13 14:53:50 PST 2012
On 11/13/2012 01:44 PM, Maxence Le Doré wrote:
> Oh sorry ! I simply forgot to add
> src/glsl/builtins/profiles/ARB_texture_gather.glsl to tracked files in
> my patch !
>
> Here comes the new one. I also add the forgotten xml api file.
>
> Thanks a lot Kenneth
Cool, thanks!
Could you please split this into several patches? I would break it up
like this:
1. mesa: Add basic plumbing for the ARB_texture_gather extension.
This would have the src/mesa/main and src/mapi/glapi changes (adding the
extension enable bit, the XML, the glGet() changes, etc.)
2. glsl: Add support for the ARB_texture_gather extension.
This would have your src/glsl changes.
3. The Gallium changes (TGSI and st/mesa).
Also, when you send out those patches, could you please use
git-send-email? I wasn't able to apply your current patch cleanly.
Using git-send-email allows us to easily save the email and apply it
using 'git am'. You can do that via:
git send-email --to mesa-dev at lists.freedesktop.org --annotate HEAD~3
(that sends the top three patches on your tree. you can also do
<sha>..<sha>, similar to git log.)
A few comments below:
> From 9673f66ede72bf872824fd3430ae01b104b3d69f Mon Sep 17 00:00:00 2001
> From: Maxence Le Dore <maxence.ledore at gmail.com
> <mailto:maxence.ledore at gmail.com>>
> Date: Sat, 20 Oct 2012 02:31:41 +0200
> Subject: [PATCH] texture_gather
>
> ---
> src/gallium/auxiliary/tgsi/tgsi_exec.c | 74
> +++++++++++++++++++-
> src/gallium/drivers/llvmpipe/lp_screen.c | 3 +
> src/gallium/drivers/softpipe/sp_screen.c | 6 ++
> src/gallium/drivers/softpipe/sp_tex_sample.c | 8 +++
> src/gallium/include/pipe/p_defines.h | 5 +-
> src/glsl/builtins/profiles/ARB_texture_gather.glsl | 26 +++++++
> src/glsl/builtins/tools/generate_builtins.py | 1 +
> src/glsl/builtins/tools/texture_builtins.py | 14 +++-
> src/glsl/glcpp/glcpp-parse.y | 3 +
> src/glsl/glsl_parser_extras.cpp | 1 +
> src/glsl/glsl_parser_extras.h | 2 +
> src/glsl/ir.cpp | 2 +-
> src/glsl/ir.h | 4 +-
> src/glsl/ir_clone.cpp | 1 +
> src/glsl/ir_hv_accept.cpp | 1 +
> src/glsl/ir_print_visitor.cpp | 3 +-
> src/glsl/ir_rvalue_visitor.cpp | 1 +
> src/glsl/opt_tree_grafting.cpp | 1 +
> src/glsl/standalone_scaffolding.cpp | 1 +
> src/mapi/glapi/gen/ARB_texture_gather.xml | 14 ++++
> src/mapi/glapi/gen/gl_API.xml | 2 +-
> src/mesa/main/config.h | 6 ++
> src/mesa/main/extensions.c | 1 +
> src/mesa/main/get_hash_params.py | 5 ++
> src/mesa/main/mtypes.h | 6 ++
> src/mesa/main/tests/enum_strings.cpp | 3 +
> src/mesa/program/ir_to_mesa.cpp | 2 +
> src/mesa/program/prog_instruction.h | 1 +
> src/mesa/state_tracker/st_extensions.c | 7 +-
> src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 7 ++
> 30 files changed, 203 insertions(+), 8 deletions(-)
> create mode 100644 src/glsl/builtins/profiles/ARB_texture_gather.glsl
> create mode 100644 src/mapi/glapi/gen/ARB_texture_gather.xml
>
> diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c
> b/src/gallium/auxiliary/tgsi/tgsi_exec.c
> index 1b7150b..0af1458 100644
> --- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
> +++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
> @@ -2119,6 +2119,78 @@ exec_txq(struct tgsi_exec_machine *mach,
> }
>
> static void
> +exec_txg(struct tgsi_exec_machine *mach,
> + const struct tgsi_full_instruction *inst)
> +{
> + struct tgsi_sampler *sampler;
> + const uint unit = inst->Src[2].Register.Index;
> + union tgsi_exec_channel r[4];
> + union tgsi_exec_channel offset[3];
> + uint chan;
> + float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
> + int j;
> + int8_t offsets[3];
> +
> + if (inst->Texture.NumOffsets == 1) {
> + union tgsi_exec_channel index;
> + index.i[0] = index.i[1] = index.i[2] = index.i[3] =
> inst->TexOffsets[0].Index;
> + fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File,
> + inst->TexOffsets[0].SwizzleX, &index,
> &ZeroVec, &offset[0]);
> + fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File,
> + inst->TexOffsets[0].SwizzleY, &index,
> &ZeroVec, &offset[1]);
> + fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File,
> + inst->TexOffsets[0].SwizzleZ, &index,
> &ZeroVec, &offset[2]);
> + offsets[0] = offset[0].i[0];
> + offsets[1] = offset[1].i[0];
> + offsets[2] = offset[2].i[0];
> + } else
> + offsets[0] = offsets[1] = offsets[2] = 0;
> +
> + IFETCH(&r[3], 0, TGSI_CHAN_W);
> +
> + switch(inst->Texture.Texture) {
> + case TGSI_TEXTURE_2D_ARRAY:
> + case TGSI_TEXTURE_CUBE_ARRAY:
> + IFETCH(&r[2], 0, TGSI_CHAN_Z);
> + /* fallthrough */
> + case TGSI_TEXTURE_2D:
> + case TGSI_TEXTURE_CUBE:
> + IFETCH(&r[1], 0, TGSI_CHAN_Y);
> + break;
> + case TGSI_TEXTURE_1D:
> + case TGSI_TEXTURE_3D:
> + case TGSI_TEXTURE_SHADOW1D:
> + case TGSI_TEXTURE_1D_ARRAY:
> + case TGSI_TEXTURE_SHADOW2D:
> + case TGSI_TEXTURE_SHADOWRECT:
> + case TGSI_TEXTURE_SHADOW1D_ARRAY:
> + case TGSI_TEXTURE_SHADOW2D_ARRAY:
> + assert(0);
> + /* fallthrough */
> + default:
> + assert(0);
> + break;
> + }
> +
> + sampler = mach->Samplers[unit];
> + sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i,
> + offsets, rgba);
> +
> + for (j = 0; j < TGSI_QUAD_SIZE; j++) {
> + r[0].f[j] = rgba[0][j];
> + r[1].f[j] = rgba[1][j];
> + r[2].f[j] = rgba[2][j];
> + r[3].f[j] = rgba[3][j];
> + }
> +
> + for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
> + if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
> + store_dest(mach, &r[chan], &inst->Dst[0], inst, chan,
> TGSI_EXEC_DATA_FLOAT);
> + }
> + }
> +}
> +
> +static void
> exec_sample(struct tgsi_exec_machine *mach,
> const struct tgsi_full_instruction *inst,
> uint modifier)
> @@ -4215,7 +4287,7 @@ exec_instruction(
> break;
>
> case TGSI_OPCODE_GATHER4:
> - assert(0);
> + exec_txg(mach,inst);
> break;
>
> case TGSI_OPCODE_SVIEWINFO:
> diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c
> b/src/gallium/drivers/llvmpipe/lp_screen.c
> index e81c441..157e043 100644
> --- a/src/gallium/drivers/llvmpipe/lp_screen.c
> +++ b/src/gallium/drivers/llvmpipe/lp_screen.c
> @@ -214,6 +214,9 @@ llvmpipe_get_param(struct pipe_screen *screen, enum
> pipe_cap param)
> case PIPE_CAP_TEXTURE_MULTISAMPLE:
> case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
> case PIPE_CAP_CUBE_MAP_ARRAY:
> + case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
> + case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
> + case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
> return 0;
> }
> /* should only get here on unhandled cases */
> diff --git a/src/gallium/drivers/softpipe/sp_screen.c
> b/src/gallium/drivers/softpipe/sp_screen.c
> index 3a38182..99f3344 100644
> --- a/src/gallium/drivers/softpipe/sp_screen.c
> +++ b/src/gallium/drivers/softpipe/sp_screen.c
> @@ -171,6 +171,12 @@ softpipe_get_param(struct pipe_screen *screen, enum
> pipe_cap param)
> case PIPE_CAP_QUERY_TIMESTAMP:
> case PIPE_CAP_CUBE_MAP_ARRAY:
> return 1;
> + case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
> + return -16;
> + case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
> + return 15;
> + case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
> + return 4;
> }
> /* should only get here on unhandled cases */
> debug_printf("Unexpected PIPE_CAP %d query\n", param);
> diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c
> b/src/gallium/drivers/softpipe/sp_tex_sample.c
> index 31eab39..3d53c10 100644
> --- a/src/gallium/drivers/softpipe/sp_tex_sample.c
> +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
> @@ -1528,6 +1528,14 @@ img_filter_3d_linear(struct tgsi_sampler
> *tgsi_sampler,
> tx12[c], tx13[c]);
> }
>
> +img_gather_2d(struct tgsi_sampler *tgsi_sampler,
> + float s,
> + float t,
> + unsigned level,
> + enum tgsi_sampler_control control,
> + float *rgba)
> +{
> +}
>
> /* Calculate level of detail for every fragment.
> * Note that lambda has already been biased by global LOD bias.
> diff --git a/src/gallium/include/pipe/p_defines.h
> b/src/gallium/include/pipe/p_defines.h
> index fa2fb07..4e99158 100644
> --- a/src/gallium/include/pipe/p_defines.h
> +++ b/src/gallium/include/pipe/p_defines.h
> @@ -491,7 +491,10 @@ enum pipe_cap {
> PIPE_CAP_QUERY_TIMESTAMP = 73,
> PIPE_CAP_TEXTURE_MULTISAMPLE = 74,
> PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT = 75,
> - PIPE_CAP_CUBE_MAP_ARRAY = 76
> + PIPE_CAP_CUBE_MAP_ARRAY = 76,
> + PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET = 77,
> + PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET = 78,
> + PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS = 79
> };
>
> /**
> diff --git a/src/glsl/builtins/profiles/ARB_texture_gather.glsl
> b/src/glsl/builtins/profiles/ARB_texture_gather.glsl
> new file mode 100644
> index 0000000..c75a9bf
> --- /dev/null
> +++ b/src/glsl/builtins/profiles/ARB_texture_gather.glsl
> @@ -0,0 +1,26 @@
> +#version 130
> +#extension GL_ARB_texture_gather: enable
> +
> + vec4 textureGather( sampler2D sampler, vec2 coord);
> +ivec4 textureGather(isampler2D sampler, vec2 coord);
> +uvec4 textureGather(usampler2D sampler, vec2 coord);
> +
> + vec4 textureGather( sampler2DArray sampler, vec3 coord);
> +ivec4 textureGather(isampler2DArray sampler, vec3 coord);
> +uvec4 textureGather(usampler2DArray sampler, vec3 coord);
> +
> + vec4 textureGather( sampler2DCube sampler, vec3 coord);
> +ivec4 textureGather(isampler2DCube sampler, vec3 coord);
> +uvec4 textureGather(usampler2DCube sampler, vec3 coord);
This fails to compile because sampler2DCube is not a type :) You want
"samplerCube" here.
> + vec4 textureGather( samplerCubeArray sampler, vec4 coord);
> +ivec4 textureGather(isamplerCubeArray sampler, vec4 coord);
> +uvec4 textureGather(usamplerCubeArray sampler, vec4 coord);
This also fails to compile since the samplerCubeArray types aren't
available unless you do:
#extension GL_ARB_texture_cube_map_array : enable
With both of those changes, your code compiles.
> + vec4 textureGatherOffset( sampler2D sampler, vec2 coord, ivec2 offset);
> +ivec4 textureGatherOffset(isampler2D sampler, vec2 coord, ivec2 offset);
> +uvec4 textureGatherOffset(usampler2D sampler, vec2 coord, ivec2 offset);
> +
> + vec4 textureGatherOffset( sampler2DArray sampler, vec3 coord, ivec2
> offset);
> +ivec4 textureGatherOffset(isampler2DArray sampler, vec3 coord, ivec2
> offset);
> +uvec4 textureGatherOffset(usampler2DArray sampler, vec3 coord, ivec2
> offset);
> \ No newline at end of file
> diff --git a/src/glsl/builtins/tools/generate_builtins.py
> b/src/glsl/builtins/tools/generate_builtins.py
> index 7eccb7d..2f024c2 100755
> --- a/src/glsl/builtins/tools/generate_builtins.py
> +++ b/src/glsl/builtins/tools/generate_builtins.py
> @@ -188,6 +188,7 @@ read_builtins(GLenum target, const char *protos,
> const char **functions, unsigne
> st->OES_EGL_image_external_enable = true;
> st->ARB_shader_bit_encoding_enable = true;
> st->ARB_texture_cube_map_array_enable = true;
> + st->ARB_texture_gather_enable = true;
> _mesa_glsl_initialize_types(st);
>
> sh->ir = new(sh) exec_list;
> diff --git a/src/glsl/builtins/tools/texture_builtins.py
> b/src/glsl/builtins/tools/texture_builtins.py
> index 654eb06..42b20d4 100755
> --- a/src/glsl/builtins/tools/texture_builtins.py
> +++ b/src/glsl/builtins/tools/texture_builtins.py
> @@ -105,7 +105,7 @@ def generate_sigs(g, tex_inst, sampler_type, variant
> = 0, unused_fields = 0):
> else:
> print "0",
>
> - if tex_inst != "txf" and tex_inst != "txs":
> + if tex_inst != "txf" and tex_inst != "txs" and tex_inst != "txg":
> # Projective divisor
> if variant & Proj:
> print "(swiz " + "xyzw"[coord_dim + extra_dim-1] + "
> (var_ref P))",
> @@ -619,6 +619,18 @@ def generate_texture_functions(fs):
> generate_sigs("", "txl", "2DShadow", Proj)
> end_function(fs, "shadow2DProjLod")
>
> + start_function("textureGather")
> + generate_fiu_sigs("txg", "2D")
> + generate_fiu_sigs("txg", "2DArray")
> + generate_fiu_sigs("txg", "Cube")
> + generate_fiu_sigs("txg", "CubeArray")
> + end_function(fs, "textureGather")
> +
> + start_function("textureGatherOffset")
> + generate_fiu_sigs("txg", "2D", Offset)
> + generate_fiu_sigs("txg", "2DArray", Offset)
> + end_function(fs, "textureGatherOffset")
> +
> sys.stdout = sys.__stdout__
> return fs
>
> diff --git a/src/glsl/glcpp/glcpp-parse.y b/src/glsl/glcpp/glcpp-parse.y
> index 9628637..9eae568 100644
> --- a/src/glsl/glcpp/glcpp-parse.y
> +++ b/src/glsl/glcpp/glcpp-parse.y
> @@ -1190,6 +1190,9 @@ glcpp_parser_create (const struct gl_extensions
> *extensions, int api)
>
> if (extensions->ARB_texture_cube_map_array)
> add_builtin_define(parser, "GL_ARB_texture_cube_map_array", 1);
> +
> + if (extensions->ARB_texture_cube_map_array)
> + add_builtin_define(parser, "GL_ARB_texture_gather", 1);
Your check here is wrong: you want extensions->ARB_texture_gather.
> }
>
> language_version = 110;
> diff --git a/src/glsl/glsl_parser_extras.cpp
> b/src/glsl/glsl_parser_extras.cpp
> index f1fdd3a..ad99cdc 100644
> --- a/src/glsl/glsl_parser_extras.cpp
> +++ b/src/glsl/glsl_parser_extras.cpp
> @@ -290,6 +290,7 @@ static const _mesa_glsl_extension
> _mesa_glsl_supported_extensions[] = {
> EXT(ARB_uniform_buffer_object, true, false, true, true,
> false, ARB_uniform_buffer_object),
> EXT(OES_standard_derivatives, false, false, true, false,
> true, OES_standard_derivatives),
> EXT(ARB_texture_cube_map_array, true, false, true, true,
> false, ARB_texture_cube_map_array),
> + EXT(ARB_texture_gather, true, false, true, true,
> false, ARB_texture_gather),
> };
>
> #undef EXT
> diff --git a/src/glsl/glsl_parser_extras.h b/src/glsl/glsl_parser_extras.h
> index 0b208f6..c058049 100644
> --- a/src/glsl/glsl_parser_extras.h
> +++ b/src/glsl/glsl_parser_extras.h
> @@ -183,6 +183,8 @@ struct _mesa_glsl_parse_state {
> bool ARB_fragment_coord_conventions_warn;
> bool ARB_texture_rectangle_enable;
> bool ARB_texture_rectangle_warn;
> + bool ARB_texture_gather_enable;
> + bool ARB_texture_gather_warn;
> bool EXT_texture_array_enable;
> bool EXT_texture_array_warn;
> bool ARB_shader_texture_lod_enable;
> diff --git a/src/glsl/ir.cpp b/src/glsl/ir.cpp
> index 7b0a487..a826395 100644
> --- a/src/glsl/ir.cpp
> +++ b/src/glsl/ir.cpp
> @@ -1282,7 +1282,7 @@ ir_dereference::is_lvalue() const
> }
>
>
> -static const char *tex_opcode_strs[] = { "tex", "txb", "txl", "txd",
> "txf", "txs" };
> +static const char *tex_opcode_strs[] = { "tex", "txb", "txl", "txd",
> "txf", "txs", "txg" };
Hmm. I'm not a fan of the "txg" name, since at least in our hardware
docs, sample_g means textureGrad() (an alias for
sample_d...gradients/derivatives). textureGather() is called "gather4"
in our hardware.
I asked on IRC if people had ideas for a different name, and Dave
suggested "ir_tg4". Could we use that instead?
> const char *ir_texture::opcode_string()
> {
> diff --git a/src/glsl/ir.h b/src/glsl/ir.h
> index 89c516c..5d5b56d 100644
> --- a/src/glsl/ir.h
> +++ b/src/glsl/ir.h
> @@ -1334,7 +1334,8 @@ enum ir_texture_opcode {
> ir_txl, /**< Texture look-up with explicit LOD */
> ir_txd, /**< Texture look-up with partial derivatvies */
> ir_txf, /**< Texel fetch with explicit LOD */
> - ir_txs /**< Texture size */
> + ir_txs, /**< Texture size */
> + ir_txg /**< Texture gathering */
> };
>
>
> @@ -1356,6 +1357,7 @@ enum ir_texture_opcode {
> * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
> * (txf <type> <sampler> <coordinate> 0 <lod>)
> * (txs <type> <sampler> <lod>)
> + * (txg <type> <sampler> <coordinate>)
> */
> class ir_texture : public ir_rvalue {
> public:
> diff --git a/src/glsl/ir_clone.cpp b/src/glsl/ir_clone.cpp
> index 4314efa..a6d8207 100644
> --- a/src/glsl/ir_clone.cpp
> +++ b/src/glsl/ir_clone.cpp
> @@ -246,6 +246,7 @@ ir_texture::clone(void *mem_ctx, struct hash_table
> *ht) const
>
> switch (this->op) {
> case ir_tex:
> + case ir_txg:
> break;
> case ir_txb:
> new_tex->lod_info.bias = this->lod_info.bias->clone(mem_ctx, ht);
> diff --git a/src/glsl/ir_hv_accept.cpp b/src/glsl/ir_hv_accept.cpp
> index 3ce8959..155ed08 100644
> --- a/src/glsl/ir_hv_accept.cpp
> +++ b/src/glsl/ir_hv_accept.cpp
> @@ -213,6 +213,7 @@ ir_texture::accept(ir_hierarchical_visitor *v)
>
> switch (this->op) {
> case ir_tex:
> + case ir_txg:
> break;
> case ir_txb:
> s = this->lod_info.bias->accept(v);
> diff --git a/src/glsl/ir_print_visitor.cpp b/src/glsl/ir_print_visitor.cpp
> index 8aa26e5..e54fef0 100644
> --- a/src/glsl/ir_print_visitor.cpp
> +++ b/src/glsl/ir_print_visitor.cpp
> @@ -259,7 +259,7 @@ void ir_print_visitor::visit(ir_texture *ir)
> printf(" ");
> }
>
> - if (ir->op != ir_txf && ir->op != ir_txs) {
> + if (ir->op != ir_txf && ir->op != ir_txs && ir->op != ir_txg) {
> if (ir->projector)
> ir->projector->accept(this);
> else
> @@ -277,6 +277,7 @@ void ir_print_visitor::visit(ir_texture *ir)
> switch (ir->op)
> {
> case ir_tex:
> + case ir_txg:
> break;
> case ir_txb:
> ir->lod_info.bias->accept(this);
> diff --git a/src/glsl/ir_rvalue_visitor.cpp b/src/glsl/ir_rvalue_visitor.cpp
> index b34a419..280ca74 100644
> --- a/src/glsl/ir_rvalue_visitor.cpp
> +++ b/src/glsl/ir_rvalue_visitor.cpp
> @@ -57,6 +57,7 @@ ir_rvalue_base_visitor::rvalue_visit(ir_texture *ir)
>
> switch (ir->op) {
> case ir_tex:
> + case ir_txg:
> break;
> case ir_txb:
> handle_rvalue(&ir->lod_info.bias);
> diff --git a/src/glsl/opt_tree_grafting.cpp b/src/glsl/opt_tree_grafting.cpp
> index 25b18ea..c7f9e5b 100644
> --- a/src/glsl/opt_tree_grafting.cpp
> +++ b/src/glsl/opt_tree_grafting.cpp
> @@ -273,6 +273,7 @@ ir_tree_grafting_visitor::visit_enter(ir_texture *ir)
>
> switch (ir->op) {
> case ir_tex:
> + case ir_txg:
> break;
> case ir_txb:
> if (do_graft(&ir->lod_info.bias))
> diff --git a/src/glsl/standalone_scaffolding.cpp
> b/src/glsl/standalone_scaffolding.cpp
> index 120ee95..e92dfd8 100644
> --- a/src/glsl/standalone_scaffolding.cpp
> +++ b/src/glsl/standalone_scaffolding.cpp
> @@ -82,6 +82,7 @@ void initialize_context_to_defaults(struct gl_context
> *ctx, gl_api api)
> ctx->Extensions.ARB_shader_bit_encoding = true;
> ctx->Extensions.OES_standard_derivatives = true;
> ctx->Extensions.ARB_texture_cube_map_array = true;
> + ctx->Extensions.ARB_texture_gather = true;
>
> ctx->Const.GLSLVersion = 120;
>
> diff --git a/src/mapi/glapi/gen/ARB_texture_gather.xml
> b/src/mapi/glapi/gen/ARB_texture_gather.xml
> new file mode 100644
> index 0000000..cd331ac
> --- /dev/null
> +++ b/src/mapi/glapi/gen/ARB_texture_gather.xml
> @@ -0,0 +1,14 @@
> +<?xml version="1.0"?>
> +<!DOCTYPE OpenGLAPI SYSTEM "gl_API.dtd">
> +
> +<OpenGLAPI>
> +
> +<category name="GL_ARB_texture_gather" number="72">
> +
> + <enum name="MIN_PROGRAM_TEXTURE_GATHER_OFFSET_ARB" value="0x8E5E"/>
> + <enum name="MAX_PROGRAM_TEXTURE_GATHER_OFFSET_ARB" value="0x8E5F"/>
> + <enum name="MAX_PROGRAM_TEXTURE_GATHER_COMPONENTS_ARB" value="0x8F9F"/>
> +
> +</category>
> +
> +</OpenGLAPI>
> \ No newline at end of file
> diff --git a/src/mapi/glapi/gen/gl_API.xml b/src/mapi/glapi/gen/gl_API.xml
> index a47316a..a130b2c 100644
> --- a/src/mapi/glapi/gen/gl_API.xml
> +++ b/src/mapi/glapi/gen/gl_API.xml
> @@ -8187,7 +8187,7 @@
>
> <!-- 70. GL_ARB_sample_shading -->
> <xi:include href="ARB_texture_cube_map_array.xml"
> xmlns:xi="http://www.w3.org/2001/XInclude"/>
> -<!-- 72. GL_ARB_texture_gather -->
> +<xi:include href="ARB_texture_gather.xml"
> xmlns:xi="http://www.w3.org/2001/XInclude"/>
> <!-- 73. GL_ARB_texture_query_lod -->
>
> <!-- ARB extension number 74 is a WGL extension. -->
> diff --git a/src/mesa/main/config.h b/src/mesa/main/config.h
> index 99910d7..2a99514 100644
> --- a/src/mesa/main/config.h
> +++ b/src/mesa/main/config.h
> @@ -259,6 +259,12 @@
> #define MAX_DEBUG_MESSAGE_LENGTH 4096
> /*@}*/
>
> +/** For GL_ARB_texture_gather */
> +/*@{*/
> +#define MIN_PROGRAM_TEXTURE_GATHER_OFFSET -16
> +#define MAX_PROGRAM_TEXTURE_GATHER_OFFSET 15
> +#define MAX_PROGRAM_TEXTURE_GATHER_COMPONENTS 1
> +/*@{*/
These #defines should not exist. There are no users of them, and these
values don't match anything that I can find.
> /*
> * Color channel component order
> diff --git a/src/mesa/main/extensions.c b/src/mesa/main/extensions.c
> index d5a7886..8da6359 100644
> --- a/src/mesa/main/extensions.c
> +++ b/src/mesa/main/extensions.c
> @@ -135,6 +135,7 @@ static const struct extension extension_table[] = {
> { "GL_ARB_texture_env_crossbar",
> o(ARB_texture_env_crossbar), GLL, 2001 },
> { "GL_ARB_texture_env_dot3",
> o(ARB_texture_env_dot3), GLL, 2001 },
> { "GL_ARB_texture_float",
> o(ARB_texture_float), GL, 2004 },
> + { "GL_ARB_texture_gather",
> o(ARB_texture_gather), GL, 2009 },
> { "GL_ARB_texture_mirrored_repeat",
> o(dummy_true), GLL, 2001 },
> { "GL_ARB_texture_multisample",
> o(ARB_texture_multisample), GL, 2009 },
> { "GL_ARB_texture_non_power_of_two",
> o(ARB_texture_non_power_of_two), GL, 2003 },
> diff --git a/src/mesa/main/get_hash_params.py
> b/src/mesa/main/get_hash_params.py
> index ac9c79c..9f33b83 100644
> --- a/src/mesa/main/get_hash_params.py
> +++ b/src/mesa/main/get_hash_params.py
> @@ -691,6 +691,11 @@ descriptor=[
>
> # GL_ARB_texture_cube_map_array
> [ "TEXTURE_BINDING_CUBE_MAP_ARRAY_ARB", "LOC_CUSTOM, TYPE_INT,
> TEXTURE_CUBE_ARRAY_INDEX, extra_ARB_texture_cube_map_array" ],
> +
> +# GL_ARB_texture_gather
> +# [ "MIN_PROGRAM_TEXTURE_GATHER_OFFSET",
> "CONTEXT_INT(Const.MinProgramTextureGatherOffset),
> extra_ARB_texture_gather"],
> +# [ "MAX_PROGRAM_TEXTURE_GATHER_OFFSET",
> "CONTEXT_INT(Const.MaxProgramTextureGatherOffset),
> extra_ARB_texture_gather"],
> +# [ "MAX_PROGRAM_TEXTURE_GATHER_COMPONENTS",
> "CONTEXT_INT(Const.MaxProgramTextureGatherComponents),
> extra_ARB_texture_gather"],
> ]}
>
> ]
> diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
> index da72da9..758cd18 100644
> --- a/src/mesa/main/mtypes.h
> +++ b/src/mesa/main/mtypes.h
> @@ -2889,6 +2889,11 @@ struct gl_constants
> /** GL_EXT_gpu_shader4 */
> GLint MinProgramTexelOffset, MaxProgramTexelOffset;
>
> + /* GL_ARB_texture_gather */
> + GLuint MinProgramTextureGatherOffest;
> + GLuint MaxProgramTextureGatherOffest;
> + GLuint MaxProgramTextureGatherComponents;
> +
You'll want to add code to initialize these to sane default values. In
context.c's _mesa_init_constants() function, look for
ctx->Const.MinProgramTexelOffset and add similar code there.
According to the GL 4.2 spec, the minimum/maximum required values are -8
and 7. I would set the defaults to those---the minimum
requirements---and let drivers that can provide more functionality raise
the limits explicitly.
> /* GL_ARB_robustness */
> GLenum ResetStrategy;
>
> @@ -2981,6 +2986,7 @@ struct gl_extensions
> GLboolean ARB_texture_env_crossbar;
> GLboolean ARB_texture_env_dot3;
> GLboolean ARB_texture_float;
> + GLboolean ARB_texture_gather;
> GLboolean ARB_texture_multisample;
> GLboolean ARB_texture_non_power_of_two;
> GLboolean ARB_texture_rg;
> diff --git a/src/mesa/main/tests/enum_strings.cpp
> b/src/mesa/main/tests/enum_strings.cpp
> index d473c30..18202a3 100644
> --- a/src/mesa/main/tests/enum_strings.cpp
> +++ b/src/mesa/main/tests/enum_strings.cpp
> @@ -1782,6 +1782,8 @@ const struct enum_info everything[] = {
> { 0x8E4D, "GL_FIRST_VERTEX_CONVENTION" },
> { 0x8E4E, "GL_LAST_VERTEX_CONVENTION" },
> { 0x8E4F, "GL_PROVOKING_VERTEX" },
> + { 0x8E5E, "GL_MIN_PROGRAM_TEXTURE_GATHER_OFFSET" },
> + { 0x8E5F, "GL_MAX_PROGRAM_TEXTURE_GATHER_OFFSET" },
> { 0x8F36, "GL_COPY_READ_BUFFER" },
> { 0x8F37, "GL_COPY_WRITE_BUFFER" },
> { 0x8F90, "GL_RED_SNORM" },
> @@ -1799,6 +1801,7 @@ const struct enum_info everything[] = {
> { 0x8F9C, "GL_SIGNED_NORMALIZED" },
> { 0x8F9D, "GL_PRIMITIVE_RESTART" },
> { 0x8F9E, "GL_PRIMITIVE_RESTART_INDEX" },
> + { 0x8F9F, "GL_MAX_PROGRAM_TEXTURE_GATHER_COMPONENTS" },
> { 0x906F, "GL_RGB10_A2UI" },
> { 0x9111, "GL_MAX_SERVER_WAIT_TIMEOUT" },
> { 0x9112, "GL_OBJECT_TYPE" },
> diff --git a/src/mesa/program/ir_to_mesa.cpp
> b/src/mesa/program/ir_to_mesa.cpp
> index d5b9683..3cdc625 100644
> --- a/src/mesa/program/ir_to_mesa.cpp
> +++ b/src/mesa/program/ir_to_mesa.cpp
> @@ -2023,6 +2023,8 @@ ir_to_mesa_visitor::visit(ir_texture *ir)
> ir->lod_info.grad.dPdy->accept(this);
> dy = this->result;
> break;
> + case ir_txg:
> + opcode = OPCODE_TXG;
> }
ir_to_mesa is only used for old GPUs like i915, which will never support
textureGather(). So I wouldn't add an opcode here. Instead, I'd just:
assert(!"Invalid texture opcode.");
> const glsl_type *sampler_type = ir->sampler->type;
> diff --git a/src/mesa/program/prog_instruction.h
> b/src/mesa/program/prog_instruction.h
> index a7bba75..3bbe756 100644
> --- a/src/mesa/program/prog_instruction.h
> +++ b/src/mesa/program/prog_instruction.h
> @@ -222,6 +222,7 @@ typedef enum prog_opcode {
> OPCODE_TXL, /* 3 2 X */
> OPCODE_TXP, /* X X */
> OPCODE_TXP_NV, /* 3 X */
> + OPCODE_TXG, /* X */
> OPCODE_TRUNC, /* X */
> OPCODE_UP2H, /* X */
> OPCODE_UP2US, /* X */
You don't need to add this opcode. GLSL doesn't go through this code.
> diff --git a/src/mesa/state_tracker/st_extensions.c
> b/src/mesa/state_tracker/st_extensions.c
> index 7570d68..aa7ec5b 100644
> --- a/src/mesa/state_tracker/st_extensions.c
> +++ b/src/mesa/state_tracker/st_extensions.c
> @@ -234,6 +234,10 @@ void st_init_limits(struct st_context *st)
> c->MinProgramTexelOffset = screen->get_param(screen,
> PIPE_CAP_MIN_TEXEL_OFFSET);
> c->MaxProgramTexelOffset = screen->get_param(screen,
> PIPE_CAP_MAX_TEXEL_OFFSET);
>
> + c->MinProgramTextureGatherOffest = screen->get_param(screen,
> PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET);
> + c->MaxProgramTextureGatherOffest = screen->get_param(screen,
> PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET);
> + c->MaxProgramTextureGatherComponents = screen->get_param(screen,
> PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS);
> +
> c->UniformBooleanTrue = ~0;
>
> c->MaxTransformFeedbackBuffers =
> @@ -379,7 +383,8 @@ void st_init_extensions(struct st_context *st)
> { o(MESA_texture_array),
> PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS },
>
> { o(OES_standard_derivatives),
> PIPE_CAP_SM3 },
> - { o(ARB_texture_cube_map_array),
> PIPE_CAP_CUBE_MAP_ARRAY }
> + { o(ARB_texture_cube_map_array),
> PIPE_CAP_CUBE_MAP_ARRAY },
> + { o(ARB_texture_gather),
> PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS }
> };
>
> /* Required: render target and sampler support */
> diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> index c030a6b..3aee6a5 100644
> --- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> +++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> @@ -2637,6 +2637,13 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
> offset = this->result;
> }
> break;
> + case ir_txg:
> + opcode = TGSI_OPCODE_GATHER4;
> + if (ir->offset) {
> + ir->offset->accept(this);
> + offset = this->result;
> + }
> + break;
> }
>
> if (ir->projector) {
> --
> 1.7.9.5
Thanks again for doing this! I'm looking forward to seeing your next
revision of the patch series.
More information about the mesa-dev
mailing list