nir: move nir_shader_info to a common compiler header
[mesa.git] / src / mesa / drivers / dri / i965 / brw_nir.c
index 9274f2e1616d6d240a8b327fed6907674a7bcbed..04dbf01f6d356229b8c0fda9bcff2bfa4dab02f2 100644 (file)
 #include "brw_nir.h"
 #include "brw_shader.h"
 #include "compiler/glsl_types.h"
-#include "compiler/nir/glsl_to_nir.h"
 #include "compiler/nir/nir_builder.h"
-#include "program/prog_to_nir.h"
-#include "program/prog_parameter.h"
 
 static bool
 is_input(nir_intrinsic_instr *intrin)
 {
    return intrin->intrinsic == nir_intrinsic_load_input ||
-          intrin->intrinsic == nir_intrinsic_load_per_vertex_input;
+          intrin->intrinsic == nir_intrinsic_load_per_vertex_input ||
+          intrin->intrinsic == nir_intrinsic_load_interpolated_input;
 }
 
 static bool
@@ -98,7 +96,7 @@ add_const_offset_to_base(nir_shader *nir, nir_variable_mode mode)
 }
 
 static bool
-remap_vs_attrs(nir_block *block, struct nir_shader_info *nir_info)
+remap_vs_attrs(nir_block *block, shader_info *nir_info)
 {
    nir_foreach_instr(instr, block) {
       if (instr->type != nir_instr_type_intrinsic)
@@ -193,7 +191,6 @@ remap_patch_urb_offsets(nir_block *block, nir_builder *b,
 
 void
 brw_nir_lower_vs_inputs(nir_shader *nir,
-                        const struct brw_device_info *devinfo,
                         bool is_scalar,
                         bool use_legacy_snorm_formula,
                         const uint8_t *vs_attrib_wa_flags)
@@ -207,7 +204,7 @@ brw_nir_lower_vs_inputs(nir_shader *nir,
     * loaded as one vec4 or dvec4 per element (or matrix column), depending on
     * whether it is a double-precision type or not.
     */
-   nir_lower_io(nir, nir_var_shader_in, type_size_vs_input);
+   nir_lower_io(nir, nir_var_shader_in, type_size_vs_input, 0);
 
    /* This pass needs actual constants */
    nir_opt_constant_folding(nir);
@@ -239,7 +236,7 @@ brw_nir_lower_vue_inputs(nir_shader *nir, bool is_scalar,
    }
 
    /* Inputs are stored in vec4 slots, so use type_size_vec4(). */
-   nir_lower_io(nir, nir_var_shader_in, type_size_vec4);
+   nir_lower_io(nir, nir_var_shader_in, type_size_vec4, 0);
 
    if (is_scalar || nir->stage != MESA_SHADER_GEOMETRY) {
       /* This pass needs actual constants */
@@ -264,7 +261,7 @@ brw_nir_lower_tes_inputs(nir_shader *nir, const struct brw_vue_map *vue_map)
       var->data.driver_location = var->data.location;
    }
 
-   nir_lower_io(nir, nir_var_shader_in, type_size_vec4);
+   nir_lower_io(nir, nir_var_shader_in, type_size_vec4, 0);
 
    /* This pass needs actual constants */
    nir_opt_constant_folding(nir);
@@ -283,25 +280,58 @@ brw_nir_lower_tes_inputs(nir_shader *nir, const struct brw_vue_map *vue_map)
 }
 
 void
-brw_nir_lower_fs_inputs(nir_shader *nir)
+brw_nir_lower_fs_inputs(nir_shader *nir,
+                        const struct gen_device_info *devinfo,
+                        const struct brw_wm_prog_key *key)
 {
-   nir_assign_var_locations(&nir->inputs, &nir->num_inputs, type_size_scalar);
-   nir_lower_io(nir, nir_var_shader_in, type_size_scalar);
+   foreach_list_typed(nir_variable, var, node, &nir->inputs) {
+      var->data.driver_location = var->data.location;
+
+      /* Apply default interpolation mode.
+       *
+       * Everything defaults to smooth except for the legacy GL color
+       * built-in variables, which might be flat depending on API state.
+       */
+      if (var->data.interpolation == INTERP_MODE_NONE) {
+         const bool flat = key->flat_shade &&
+            (var->data.location == VARYING_SLOT_COL0 ||
+             var->data.location == VARYING_SLOT_COL1);
+
+         var->data.interpolation = flat ? INTERP_MODE_FLAT
+                                        : INTERP_MODE_SMOOTH;
+      }
+
+      /* On Ironlake and below, there is only one interpolation mode.
+       * Centroid interpolation doesn't mean anything on this hardware --
+       * there is no multisampling.
+       */
+      if (devinfo->gen < 6) {
+         var->data.centroid = false;
+         var->data.sample = false;
+      }
+   }
+
+   nir_lower_io_options lower_io_options = 0;
+   if (key->persample_interp)
+      lower_io_options |= nir_lower_io_force_sample_interpolation;
+
+   nir_lower_io(nir, nir_var_shader_in, type_size_vec4, lower_io_options);
+
+   /* This pass needs actual constants */
+   nir_opt_constant_folding(nir);
+
+   add_const_offset_to_base(nir, nir_var_shader_in);
 }
 
 void
 brw_nir_lower_vue_outputs(nir_shader *nir,
                           bool is_scalar)
 {
-   if (is_scalar) {
-      nir_assign_var_locations(&nir->outputs, &nir->num_outputs,
-                               type_size_vec4_times_4);
-      nir_lower_io(nir, nir_var_shader_out, type_size_vec4_times_4);
-   } else {
-      nir_foreach_variable(var, &nir->outputs)
-         var->data.driver_location = var->data.location;
-      nir_lower_io(nir, nir_var_shader_out, type_size_vec4);
+   nir_foreach_variable(var, &nir->outputs) {
+      var->data.driver_location = var->data.location;
    }
+
+   nir_lower_io(nir, nir_var_shader_out, type_size_vec4, 0);
 }
 
 void
@@ -311,7 +341,7 @@ brw_nir_lower_tcs_outputs(nir_shader *nir, const struct brw_vue_map *vue_map)
       var->data.driver_location = var->data.location;
    }
 
-   nir_lower_io(nir, nir_var_shader_out, type_size_vec4);
+   nir_lower_io(nir, nir_var_shader_out, type_size_vec4, 0);
 
    /* This pass needs actual constants */
    nir_opt_constant_folding(nir);
@@ -332,43 +362,21 @@ brw_nir_lower_tcs_outputs(nir_shader *nir, const struct brw_vue_map *vue_map)
 void
 brw_nir_lower_fs_outputs(nir_shader *nir)
 {
-   nir_assign_var_locations(&nir->outputs, &nir->num_outputs,
-                            type_size_scalar);
-   nir_lower_io(nir, nir_var_shader_out, type_size_scalar);
-}
-
-static int
-type_size_scalar_bytes(const struct glsl_type *type)
-{
-   return type_size_scalar(type) * 4;
-}
-
-static int
-type_size_vec4_bytes(const struct glsl_type *type)
-{
-   return type_size_vec4(type) * 16;
-}
-
-static void
-brw_nir_lower_uniforms(nir_shader *nir, bool is_scalar)
-{
-   if (is_scalar) {
-      nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
-                               type_size_scalar_bytes);
-      nir_lower_io(nir, nir_var_uniform, type_size_scalar_bytes);
-   } else {
-      nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
-                               type_size_vec4_bytes);
-      nir_lower_io(nir, nir_var_uniform, type_size_vec4_bytes);
+   nir_foreach_variable(var, &nir->outputs) {
+      var->data.driver_location =
+         SET_FIELD(var->data.index, BRW_NIR_FRAG_OUTPUT_INDEX) |
+         SET_FIELD(var->data.location, BRW_NIR_FRAG_OUTPUT_LOCATION);
    }
+
+   nir_lower_io(nir, nir_var_shader_out, type_size_dvec4, 0);
 }
 
 void
 brw_nir_lower_cs_shared(nir_shader *nir)
 {
-   nir_assign_var_locations(&nir->shared, &nir->num_shared,
+   nir_assign_var_locations(&nir->shared, &nir->num_shared, 0,
                             type_size_scalar_bytes);
-   nir_lower_io(nir, nir_var_shared, type_size_scalar_bytes);
+   nir_lower_io(nir, nir_var_shared, type_size_scalar_bytes, 0);
 }
 
 #define OPT(pass, ...) ({                                  \
@@ -390,19 +398,19 @@ nir_optimize(nir_shader *nir, bool is_scalar)
       OPT_V(nir_lower_vars_to_ssa);
 
       if (is_scalar) {
-         OPT_V(nir_lower_alu_to_scalar);
+         OPT(nir_lower_alu_to_scalar);
       }
 
       OPT(nir_copy_prop);
 
       if (is_scalar) {
-         OPT_V(nir_lower_phis_to_scalar);
+         OPT(nir_lower_phis_to_scalar);
       }
 
       OPT(nir_copy_prop);
       OPT(nir_opt_dce);
       OPT(nir_opt_cse);
-      OPT(nir_opt_peephole_select);
+      OPT(nir_opt_peephole_select, 0);
       OPT(nir_opt_algebraic);
       OPT(nir_opt_constant_folding);
       OPT(nir_opt_dead_cf);
@@ -448,6 +456,8 @@ brw_preprocess_nir(const struct brw_compiler *compiler, nir_shader *nir)
 
    static const nir_lower_tex_options tex_options = {
       .lower_txp = ~0,
+      .lower_txf_offset = true,
+      .lower_rect_offset = true,
    };
 
    OPT(nir_lower_tex, &tex_options);
@@ -483,7 +493,7 @@ brw_preprocess_nir(const struct brw_compiler *compiler, nir_shader *nir)
  */
 nir_shader *
 brw_postprocess_nir(nir_shader *nir,
-                    const struct brw_device_info *devinfo,
+                    const struct gen_device_info *devinfo,
                     bool is_scalar)
 {
    bool debug_enabled =
@@ -545,62 +555,9 @@ brw_postprocess_nir(nir_shader *nir,
    return nir;
 }
 
-nir_shader *
-brw_create_nir(struct brw_context *brw,
-               const struct gl_shader_program *shader_prog,
-               const struct gl_program *prog,
-               gl_shader_stage stage,
-               bool is_scalar)
-{
-   struct gl_context *ctx = &brw->ctx;
-   const nir_shader_compiler_options *options =
-      ctx->Const.ShaderCompilerOptions[stage].NirOptions;
-   bool progress;
-   nir_shader *nir;
-
-   /* First, lower the GLSL IR or Mesa IR to NIR */
-   if (shader_prog) {
-      nir = glsl_to_nir(shader_prog, stage, options);
-      nir_remove_dead_variables(nir, nir_var_shader_in | nir_var_shader_out);
-      OPT_V(nir_lower_io_to_temporaries,
-            nir_shader_get_entrypoint(nir),
-            true, false);
-   } else {
-      nir = prog_to_nir(prog, options);
-      OPT_V(nir_convert_to_ssa); /* turn registers into SSA */
-   }
-   nir_validate_shader(nir);
-
-   (void)progress;
-
-   nir = brw_preprocess_nir(brw->intelScreen->compiler, nir);
-
-   if (stage == MESA_SHADER_FRAGMENT) {
-      static const struct nir_lower_wpos_ytransform_options wpos_options = {
-         .state_tokens = {STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0},
-         .fs_coord_pixel_center_integer = 1,
-         .fs_coord_origin_upper_left = 1,
-      };
-      _mesa_add_state_reference(prog->Parameters,
-                                (gl_state_index *) wpos_options.state_tokens);
-
-      OPT(nir_lower_wpos_ytransform, &wpos_options);
-   }
-
-   OPT(nir_lower_system_values);
-   OPT_V(brw_nir_lower_uniforms, is_scalar);
-
-   if (shader_prog) {
-      OPT_V(nir_lower_samplers, shader_prog);
-      OPT_V(nir_lower_atomics, shader_prog);
-   }
-
-   return nir;
-}
-
 nir_shader *
 brw_nir_apply_sampler_key(nir_shader *nir,
-                          const struct brw_device_info *devinfo,
+                          const struct gen_device_info *devinfo,
                           const struct brw_sampler_prog_key_data *key_tex,
                           bool is_scalar)
 {
@@ -627,6 +584,10 @@ brw_nir_apply_sampler_key(nir_shader *nir,
          tex_options.swizzles[s][c] = GET_SWZ(key_tex->swizzles[s], c);
    }
 
+   tex_options.lower_y_uv_external = key_tex->y_uv_image_mask;
+   tex_options.lower_y_u_v_external = key_tex->y_u_v_image_mask;
+   tex_options.lower_yx_xuxv_external = key_tex->yx_xuxv_image_mask;
+
    if (nir_lower_tex(nir, &tex_options)) {
       nir_validate_shader(nir);
       nir = nir_optimize(nir, is_scalar);