r600/sfn: lower VS IO and drop old deref code

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9874>

src/gallium/drivers/r600/sfn/sfn_nir.cpp

@@ -303,13 +303,9 @@ bool ShaderFromNir::emit_instruction(nir_instr *instr)
 
 bool ShaderFromNir::process_declaration()
 {
-   // scan declarations
+
-   nir_foreach_shader_in_variable(variable, sh) {
+   if (!impl->scan_inputs_read(sh))
-      if (!impl->process_inputs(variable)) {
+      return false;
-         fprintf(stderr, "R600: error parsing input variable %s\n", variable->name);
-         return false;
-      }
-   }
 
    // scan declarations
    nir_foreach_shader_out_variable(variable, sh) {
@@ -899,9 +895,7 @@ int r600_shader_from_nir(struct r600_context *rctx,
 
    NIR_PASS_V(sel->nir, r600_nir_lower_pack_unpack_2x16);
 
-   nir_variable_mode io_modes = nir_var_uniform;
+   nir_variable_mode io_modes = nir_var_uniform | nir_var_shader_in;
-   if (sel->nir->info.stage != MESA_SHADER_VERTEX)
-      io_modes |= nir_var_shader_in;
 
    if (sel->nir->info.stage != MESA_SHADER_FRAGMENT)
       io_modes |= nir_var_shader_out;

src/gallium/drivers/r600/sfn/sfn_nir_lower_tess_io.cpp

@@ -8,6 +8,7 @@ bool r600_lower_tess_io_filter(const nir_instr *instr, gl_shader_stage stage)
    nir_intrinsic_instr *op = nir_instr_as_intrinsic(instr);
    switch (op->intrinsic) {
    case nir_intrinsic_load_input:
+      return stage == MESA_SHADER_TESS_CTRL || stage == MESA_SHADER_TESS_EVAL;
    case nir_intrinsic_load_output:
    case nir_intrinsic_load_per_vertex_input:
    case nir_intrinsic_load_per_vertex_output:

src/gallium/drivers/r600/sfn/sfn_shader_base.cpp

@@ -292,9 +292,9 @@ bool ShaderFromNirProcessor::process_uniforms(nir_variable *uniform)
    return true;
 }
 
-bool ShaderFromNirProcessor::process_inputs(nir_variable *input)
+bool ShaderFromNirProcessor::scan_inputs_read(const nir_shader *sh)
 {
-   return do_process_inputs(input);
+   return true;
 }
 
 bool ShaderFromNirProcessor::process_outputs(nir_variable *output)

src/gallium/drivers/r600/sfn/sfn_shader_base.h

@@ -92,6 +92,8 @@ public:
 
    void get_array_info(r600_shader& shader) const;
 
+   virtual bool scan_inputs_read(const nir_shader *sh);
+
 protected:
 
    void set_var_address(nir_deref_instr *instr);
@@ -176,7 +178,6 @@ private:
 
    virtual void emit_shader_start();
    virtual bool emit_deref_instruction_override(nir_deref_instr* instr);
-   virtual bool do_process_inputs(nir_variable *input) = 0;
    virtual bool do_process_outputs(nir_variable *output) = 0;
    virtual bool do_emit_load_deref(const nir_variable *in_var, nir_intrinsic_instr* instr) = 0;
    virtual bool do_emit_store_deref(const nir_variable *out_var, nir_intrinsic_instr* instr) = 0;

src/gallium/drivers/r600/sfn/sfn_shader_compute.cpp

@@ -104,11 +104,6 @@ bool ComputeShaderFromNir::emit_load_num_work_groups(nir_intrinsic_instr* instr)
    return true;
 }
 
-bool ComputeShaderFromNir::do_process_inputs(UNUSED nir_variable *input)
-{
-   return true;
-}
-
 bool ComputeShaderFromNir::do_process_outputs(UNUSED nir_variable *output)
 {
    return true;

src/gallium/drivers/r600/sfn/sfn_shader_compute.h

@@ -47,7 +47,6 @@ private:
    bool emit_intrinsic_instruction_override(nir_intrinsic_instr* instr) override;
 
    bool do_allocate_reserved_registers() override;
-   bool do_process_inputs(nir_variable *input) override;
    bool do_process_outputs(nir_variable *output) override;
    bool do_emit_load_deref(const nir_variable *in_var, nir_intrinsic_instr* instr) override;
    bool do_emit_store_deref(const nir_variable *out_var, nir_intrinsic_instr* instr) override;

src/gallium/drivers/r600/sfn/sfn_shader_fragment.cpp

@@ -60,12 +60,6 @@ FragmentShaderFromNir::FragmentShaderFromNir(const nir_shader& nir,
    sh_info().atomic_base = key.ps.first_atomic_counter;
 }
 
-bool FragmentShaderFromNir::do_process_inputs(nir_variable *input)
-{
-   /*  inputs have been lowered */
-   return true;
-}
-
 bool FragmentShaderFromNir::do_emit_load_deref(const nir_variable *in_var, nir_intrinsic_instr* instr)
 {
    assert(0 && "all input derefs should have been lowered");

src/gallium/drivers/r600/sfn/sfn_shader_fragment.h

@@ -49,7 +49,6 @@ private:
    };
 
    void emit_shader_start() override;
-   bool do_process_inputs(nir_variable *input) override;
    bool do_allocate_reserved_registers() override;
    bool do_process_outputs(nir_variable *output) override;
    bool do_emit_load_deref(const nir_variable *in_var, nir_intrinsic_instr* instr) override;

src/gallium/drivers/r600/sfn/sfn_shader_geometry.cpp

@@ -193,39 +193,6 @@ bool GeometryShaderFromNir::process_load_input(nir_intrinsic_instr* instr)
    return false;
 }
 
-bool GeometryShaderFromNir::do_process_inputs(nir_variable *input)
-{
-
-   if (input->data.location == VARYING_SLOT_POS ||
-       input->data.location == VARYING_SLOT_PSIZ ||
-       input->data.location == VARYING_SLOT_FOGC ||
-       input->data.location == VARYING_SLOT_CLIP_VERTEX ||
-       input->data.location == VARYING_SLOT_CLIP_DIST0 ||
-       input->data.location == VARYING_SLOT_CLIP_DIST1 ||
-       input->data.location == VARYING_SLOT_COL0 ||
-       input->data.location == VARYING_SLOT_COL1 ||
-       input->data.location == VARYING_SLOT_BFC0 ||
-       input->data.location == VARYING_SLOT_BFC1 ||
-       input->data.location == VARYING_SLOT_PNTC ||
-       (input->data.location >= VARYING_SLOT_VAR0 &&
-       input->data.location <= VARYING_SLOT_VAR31) ||
-       (input->data.location >= VARYING_SLOT_TEX0 &&
-       input->data.location <= VARYING_SLOT_TEX7)) {
-
-      r600_shader_io& io = sh_info().input[input->data.driver_location];
-      auto semantic = r600_get_varying_semantic(input->data.location);
-      io.name = semantic.first;
-      io.sid = semantic.second;
-
-      io.ring_offset = 16 * input->data.driver_location;
-      ++sh_info().ninput;
-      m_next_input_ring_offset += 16;
-      return true;
-   }
-
-   return false;
-}
-
 bool GeometryShaderFromNir::do_process_outputs(nir_variable *output)
 {
    if (output->data.location == VARYING_SLOT_COL0 ||

src/gallium/drivers/r600/sfn/sfn_shader_geometry.h

@@ -44,7 +44,6 @@ private:
    bool do_emit_load_deref(const nir_variable *in_var, nir_intrinsic_instr* instr) override;
    bool do_emit_store_deref(const nir_variable *out_var, nir_intrinsic_instr* instr) override;
 
-   bool do_process_inputs(nir_variable *input) override;
    bool do_allocate_reserved_registers() override;
    bool do_process_outputs(nir_variable *output) override;
    bool emit_intrinsic_instruction_override(nir_intrinsic_instr* instr) override;

src/gallium/drivers/r600/sfn/sfn_shader_tcs.h

@@ -16,7 +16,6 @@ private:
    bool emit_intrinsic_instruction_override(nir_intrinsic_instr* instr) override;
    bool store_tess_factor(nir_intrinsic_instr* instr);
 
-   bool do_process_inputs(nir_variable *input) override { return true;}
    bool do_process_outputs(nir_variable *output) override;
    bool do_emit_load_deref(const nir_variable *in_var, nir_intrinsic_instr* instr) override { return true;}
    bool do_emit_store_deref(const nir_variable *out_var, nir_intrinsic_instr* instr) override { return true;}

src/gallium/drivers/r600/sfn/sfn_shader_tess_eval.cpp

@@ -19,30 +19,6 @@ TEvalShaderFromNir::TEvalShaderFromNir(r600_pipe_shader *sh, r600_pipe_shader_se
       m_export_processor.reset(new VertexStageExportForFS(*this, &sel.so, sh, key));
 }
 
-bool TEvalShaderFromNir::do_process_inputs(nir_variable *input)
-{
-   if (input->data.location == VARYING_SLOT_POS ||
-       input->data.location == VARYING_SLOT_PSIZ ||
-       input->data.location == VARYING_SLOT_CLIP_DIST0 ||
-       input->data.location == VARYING_SLOT_CLIP_DIST1 ||
-       (input->data.location >= VARYING_SLOT_VAR0 &&
-       input->data.location <= VARYING_SLOT_VAR31) ||
-       (input->data.location >= VARYING_SLOT_TEX0 &&
-       input->data.location <= VARYING_SLOT_TEX7) ||
-        (input->data.location >= VARYING_SLOT_PATCH0 &&
-         input->data.location <= VARYING_SLOT_TESS_MAX)) {
-
-      r600_shader_io& io = sh_info().input[input->data.driver_location];
-      tgsi_get_gl_varying_semantic(static_cast<gl_varying_slot>( input->data.location),
-                                   true, &io.name, &io.sid);
-      ++sh_info().ninput;
-      return true;
-   }
-
-   return false;
-
-}
-
 bool TEvalShaderFromNir::scan_sysvalue_access(nir_instr *instr)
 {
    if (instr->type != nir_instr_type_intrinsic)

src/gallium/drivers/r600/sfn/sfn_shader_tess_eval.h

@@ -21,7 +21,6 @@ public:
         bool emit_load_tess_coord(nir_intrinsic_instr* instr);
         bool load_tess_z_coord(nir_intrinsic_instr* instr);
 
-        bool do_process_inputs(nir_variable *input) override;
         bool do_process_outputs(nir_variable *output) override;
         bool do_emit_load_deref(const nir_variable *in_var, nir_intrinsic_instr* instr) override { return true;}
         bool do_emit_store_deref(const nir_variable *out_var, nir_intrinsic_instr* instr) override;

src/gallium/drivers/r600/sfn/sfn_shader_vertex.cpp

@@ -72,19 +72,18 @@ VertexShaderFromNir::VertexShaderFromNir(r600_pipe_shader *sh,
    }
 }
 
-bool VertexShaderFromNir::do_process_inputs(nir_variable *input)
+bool VertexShaderFromNir::scan_inputs_read(const nir_shader *sh)
 {
-   ++sh_info().ninput;
+   uint64_t inputs = sh->info.inputs_read;
 
-   if (input->data.location < VERT_ATTRIB_MAX) {
+   while (inputs) {
-      increment_reserved_registers();
+      unsigned i = u_bit_scan64(&inputs);
-      if (m_max_attrib < input->data.driver_location)
+      if (i < VERT_ATTRIB_MAX) {
-         m_max_attrib = input->data.driver_location;
+         ++sh_info().ninput;
-
+      }
-      return true;
    }
-   fprintf(stderr, "r600-NIR-VS: Unimplemented process_inputs for %d\n", input->data.location);
+   m_max_attrib = sh_info().ninput;
-   return false;
+   return true;
 }
 
 bool VertexShaderFromNir::do_allocate_reserved_registers()
@@ -177,11 +176,33 @@ bool VertexShaderFromNir::emit_intrinsic_instruction_override(nir_intrinsic_inst
       return emit_store_local_shared(instr);
    case nir_intrinsic_store_output:
       return m_export_processor->store_output(instr);
+   case nir_intrinsic_load_input:
+      return load_input(instr);
+
    default:
       return false;
    }
 }
 
+bool VertexShaderFromNir::load_input(nir_intrinsic_instr* instr)
+{
+   unsigned location = nir_intrinsic_base(instr);
+
+   if (location < VERT_ATTRIB_MAX) {
+      for (unsigned i = 0; i < nir_dest_num_components(instr->dest); ++i) {
+         auto src = m_attribs[4 * location + i];
+
+         if (i == 0)
+            set_input(location, src);
+
+         load_preloaded_value(instr->dest, i, src, i == (unsigned)(instr->num_components - 1));
+      }
+      return true;
+   }
+   fprintf(stderr, "r600-NIR: Unimplemented load_deref for %d\n", location);
+   return false;
+}
+
 bool VertexShaderFromNir::emit_store_local_shared(nir_intrinsic_instr* instr)
 {
    unsigned write_mask = nir_intrinsic_write_mask(instr);

src/gallium/drivers/r600/sfn/sfn_shader_vertex.h

@@ -55,12 +55,16 @@ protected:
    void do_finalize() override;
 
    std::map<unsigned, unsigned> m_param_map;
+
+   bool scan_inputs_read(const nir_shader *sh) override;
+
 private:
+   bool load_input(nir_intrinsic_instr* instr);
+
    bool do_emit_store_deref(const nir_variable *out_var, nir_intrinsic_instr* instr) override;
    void finalize_exports();
 
    void emit_shader_start() override;
-   bool do_process_inputs(nir_variable *input) override;
    bool do_allocate_reserved_registers() override;
    bool do_process_outputs(nir_variable *output) override;
    bool emit_intrinsic_instruction_override(nir_intrinsic_instr* instr) override;