v3d: Refactor compiler entrypoints.

Before, I had per-stage entryoints with some helpers shared between them.
As I extended for compute shaders and shader-db, it turned out that the
other common code in the middle wanted to be shared too.

3 files changed, 170 insertions, 189 deletions

diff --git a/src/broadcom/compiler/v3d_compiler.h b/src/broadcom/compiler/v3d_compiler.h
index 717d85890b0..7068b9029be 100644
--- a/src/broadcom/compiler/v3d_compiler.h
+++ b/src/broadcom/compiler/v3d_compiler.h
@@ -706,25 +706,15 @@ const struct v3d_compiler *v3d_compiler_init(const struct v3d_device_info *devin
 void v3d_compiler_free(const struct v3d_compiler *compiler);
 void v3d_optimize_nir(struct nir_shader *s);
 
-uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
-                         struct v3d_vs_key *key,
-                         struct v3d_vs_prog_data *prog_data,
-                         nir_shader *s,
-                         void (*debug_output)(const char *msg,
-                                              void *debug_output_data),
-                         void *debug_output_data,
-                         int program_id, int variant_id,
-                         uint32_t *final_assembly_size);
-
-uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,
-                         struct v3d_fs_key *key,
-                         struct v3d_fs_prog_data *prog_data,
-                         nir_shader *s,
-                         void (*debug_output)(const char *msg,
-                                              void *debug_output_data),
-                         void *debug_output_data,
-                         int program_id, int variant_id,
-                         uint32_t *final_assembly_size);
+uint64_t *v3d_compile(const struct v3d_compiler *compiler,
+                      struct v3d_key *key,
+                      struct v3d_prog_data **prog_data,
+                      nir_shader *s,
+                      void (*debug_output)(const char *msg,
+                                           void *debug_output_data),
+                      void *debug_output_data,
+                      int program_id, int variant_id,
+                      uint32_t *final_assembly_size);
 
 void v3d_nir_to_vir(struct v3d_compile *c);
 
diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index c553e724616..8cf45630c4b 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -626,14 +626,6 @@ v3d_lower_nir(struct v3d_compile *c)
 }
 
 static void
-v3d_lower_nir_late(struct v3d_compile *c)
-{
-        NIR_PASS_V(c->s, v3d_nir_lower_io, c);
-        NIR_PASS_V(c->s, v3d_nir_lower_txf_ms, c);
-        NIR_PASS_V(c->s, nir_lower_idiv);
-}
-
-static void
 v3d_set_prog_data_uniforms(struct v3d_compile *c,
                            struct v3d_prog_data *prog_data)
 {
@@ -685,109 +677,9 @@ v3d_set_prog_data_ubo(struct v3d_compile *c,
 }
 
 static void
-v3d_set_prog_data(struct v3d_compile *c,
-                  struct v3d_prog_data *prog_data)
-{
-        prog_data->threads = c->threads;
-        prog_data->single_seg = !c->last_thrsw;
-        prog_data->spill_size = c->spill_size;
-
-        v3d_set_prog_data_uniforms(c, prog_data);
-        v3d_set_prog_data_ubo(c, prog_data);
-}
-
-static uint64_t *
-v3d_return_qpu_insts(struct v3d_compile *c, uint32_t *final_assembly_size)
-{
-        *final_assembly_size = c->qpu_inst_count * sizeof(uint64_t);
-
-        uint64_t *qpu_insts = malloc(*final_assembly_size);
-        if (!qpu_insts)
-                return NULL;
-
-        memcpy(qpu_insts, c->qpu_insts, *final_assembly_size);
-
-        char *shaderdb;
-        int ret = asprintf(&shaderdb,
-                           "%s shader: %d inst, %d threads, %d loops, "
-                           "%d uniforms, %d:%d spills:fills",
-                           vir_get_stage_name(c),
-                           c->qpu_inst_count,
-                           c->threads,
-                           c->loops,
-                           c->num_uniforms,
-                           c->spills,
-                           c->fills);
-        if (ret >= 0) {
-                c->debug_output(shaderdb, c->debug_output_data);
-                free(shaderdb);
-        }
-
-        vir_compile_destroy(c);
-
-        return qpu_insts;
-}
-
-uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
-                         struct v3d_vs_key *key,
-                         struct v3d_vs_prog_data *prog_data,
-                         nir_shader *s,
-                         void (*debug_output)(const char *msg,
-                                              void *debug_output_data),
-                         void *debug_output_data,
-                         int program_id, int variant_id,
-                         uint32_t *final_assembly_size)
+v3d_vs_set_prog_data(struct v3d_compile *c,
+                     struct v3d_vs_prog_data *prog_data)
 {
-        struct v3d_compile *c = vir_compile_init(compiler, &key->base, s,
-                                                 debug_output, debug_output_data,
-                                                 program_id, variant_id);
-
-        c->vs_key = key;
-
-        /* Split our I/O vars and dead code eliminate the unused
-         * components.
-         */
-        NIR_PASS_V(c->s, nir_lower_io_to_scalar_early,
-                   nir_var_shader_in | nir_var_shader_out);
-        uint64_t used_outputs[4] = {0};
-        for (int i = 0; i < c->vs_key->num_fs_inputs; i++) {
-                int slot = v3d_slot_get_slot(c->vs_key->fs_inputs[i]);
-                int comp = v3d_slot_get_component(c->vs_key->fs_inputs[i]);
-                used_outputs[comp] |= 1ull << slot;
-        }
-        NIR_PASS_V(c->s, nir_remove_unused_io_vars,
-                   &c->s->outputs, used_outputs, NULL); /* demotes to globals */
-        NIR_PASS_V(c->s, nir_lower_global_vars_to_local);
-        v3d_optimize_nir(c->s);
-        NIR_PASS_V(c->s, nir_remove_dead_variables, nir_var_shader_in);
-        NIR_PASS_V(c->s, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
-                   type_size_vec4,
-                   (nir_lower_io_options)0);
-
-        v3d_lower_nir(c);
-
-        if (key->clamp_color)
-                NIR_PASS_V(c->s, nir_lower_clamp_color_outputs);
-
-        if (key->base.ucp_enables) {
-                NIR_PASS_V(c->s, nir_lower_clip_vs, key->base.ucp_enables,
-                           false);
-                NIR_PASS_V(c->s, nir_lower_io_to_scalar,
-                           nir_var_shader_out);
-        }
-
-        /* Note: VS output scalarizing must happen after nir_lower_clip_vs. */
-        NIR_PASS_V(c->s, nir_lower_io_to_scalar, nir_var_shader_out);
-
-        v3d_lower_nir_late(c);
-        v3d_optimize_nir(c->s);
-        NIR_PASS_V(c->s, nir_lower_bool_to_int32);
-        NIR_PASS_V(c->s, nir_convert_from_ssa, true);
-
-        v3d_nir_to_vir(c);
-
-        v3d_set_prog_data(c, &prog_data->base);
-
         prog_data->base.num_inputs = c->num_inputs;
 
         /* The vertex data gets format converted by the VPM so that
@@ -799,9 +691,9 @@ uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
                 prog_data->vpm_input_size += c->vattr_sizes[i];
         }
 
-        prog_data->uses_vid = (s->info.system_values_read &
+        prog_data->uses_vid = (c->s->info.system_values_read &
                                (1ull << SYSTEM_VALUE_VERTEX_ID));
-        prog_data->uses_iid = (s->info.system_values_read &
+        prog_data->uses_iid = (c->s->info.system_values_read &
                                (1ull << SYSTEM_VALUE_INSTANCE_ID));
 
         if (prog_data->uses_vid)
@@ -839,8 +731,6 @@ uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
         int vpm_output_batches = vpm_output_sectors / prog_data->vpm_output_size;
         assert(vpm_output_batches >= 2);
         prog_data->vcm_cache_size = CLAMP(vpm_output_batches - 1, 2, 4);
-
-        return v3d_return_qpu_insts(c, final_assembly_size);
 }
 
 static void
@@ -866,6 +756,77 @@ v3d_set_fs_prog_data_inputs(struct v3d_compile *c,
 }
 
 static void
+v3d_fs_set_prog_data(struct v3d_compile *c,
+                     struct v3d_fs_prog_data *prog_data)
+{
+        v3d_set_fs_prog_data_inputs(c, prog_data);
+        prog_data->writes_z = (c->s->info.outputs_written &
+                               (1 << FRAG_RESULT_DEPTH));
+        prog_data->discard = (c->s->info.fs.uses_discard ||
+                              c->fs_key->sample_alpha_to_coverage);
+        prog_data->uses_center_w = c->uses_center_w;
+}
+
+static void
+v3d_set_prog_data(struct v3d_compile *c,
+                  struct v3d_prog_data *prog_data)
+{
+        prog_data->threads = c->threads;
+        prog_data->single_seg = !c->last_thrsw;
+        prog_data->spill_size = c->spill_size;
+
+        v3d_set_prog_data_uniforms(c, prog_data);
+        v3d_set_prog_data_ubo(c, prog_data);
+
+        if (c->s->info.stage == MESA_SHADER_VERTEX) {
+                v3d_vs_set_prog_data(c, (struct v3d_vs_prog_data *)prog_data);
+        } else {
+                assert(c->s->info.stage == MESA_SHADER_FRAGMENT);
+                v3d_fs_set_prog_data(c, (struct v3d_fs_prog_data *)prog_data);
+        }
+}
+
+static uint64_t *
+v3d_return_qpu_insts(struct v3d_compile *c, uint32_t *final_assembly_size)
+{
+        *final_assembly_size = c->qpu_inst_count * sizeof(uint64_t);
+
+        uint64_t *qpu_insts = malloc(*final_assembly_size);
+        if (!qpu_insts)
+                return NULL;
+
+        memcpy(qpu_insts, c->qpu_insts, *final_assembly_size);
+
+        vir_compile_destroy(c);
+
+        return qpu_insts;
+}
+
+static void
+v3d_nir_lower_vs_early(struct v3d_compile *c)
+{
+        /* Split our I/O vars and dead code eliminate the unused
+         * components.
+         */
+        NIR_PASS_V(c->s, nir_lower_io_to_scalar_early,
+                   nir_var_shader_in | nir_var_shader_out);
+        uint64_t used_outputs[4] = {0};
+        for (int i = 0; i < c->vs_key->num_fs_inputs; i++) {
+                int slot = v3d_slot_get_slot(c->vs_key->fs_inputs[i]);
+                int comp = v3d_slot_get_component(c->vs_key->fs_inputs[i]);
+                used_outputs[comp] |= 1ull << slot;
+        }
+        NIR_PASS_V(c->s, nir_remove_unused_io_vars,
+                   &c->s->outputs, used_outputs, NULL); /* demotes to globals */
+        NIR_PASS_V(c->s, nir_lower_global_vars_to_local);
+        v3d_optimize_nir(c->s);
+        NIR_PASS_V(c->s, nir_remove_dead_variables, nir_var_shader_in);
+        NIR_PASS_V(c->s, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
+                   type_size_vec4,
+                   (nir_lower_io_options)0);
+}
+
+static void
 v3d_fixup_fs_output_types(struct v3d_compile *c)
 {
         nir_foreach_variable(var, &c->s->outputs) {
@@ -895,61 +856,111 @@ v3d_fixup_fs_output_types(struct v3d_compile *c)
         }
 }
 
-uint64_t *v3d_compile_fs(const struct v3d_compiler *compiler,
-                         struct v3d_fs_key *key,
-                         struct v3d_fs_prog_data *prog_data,
-                         nir_shader *s,
-                         void (*debug_output)(const char *msg,
-                                              void *debug_output_data),
-                         void *debug_output_data,
-                         int program_id, int variant_id,
-                         uint32_t *final_assembly_size)
+static void
+v3d_nir_lower_fs_early(struct v3d_compile *c)
 {
-        struct v3d_compile *c = vir_compile_init(compiler, &key->base, s,
-                                                 debug_output, debug_output_data,
-                                                 program_id, variant_id);
+        if (c->fs_key->int_color_rb || c->fs_key->uint_color_rb)
+                v3d_fixup_fs_output_types(c);
+}
 
-        c->fs_key = key;
+static void
+v3d_nir_lower_vs_late(struct v3d_compile *c)
+{
+        if (c->vs_key->clamp_color)
+                NIR_PASS_V(c->s, nir_lower_clamp_color_outputs);
 
-        if (key->int_color_rb || key->uint_color_rb)
-                v3d_fixup_fs_output_types(c);
+        if (c->key->ucp_enables) {
+                NIR_PASS_V(c->s, nir_lower_clip_vs, c->key->ucp_enables,
+                           false);
+                NIR_PASS_V(c->s, nir_lower_io_to_scalar,
+                           nir_var_shader_out);
+        }
 
-        v3d_lower_nir(c);
+        /* Note: VS output scalarizing must happen after nir_lower_clip_vs. */
+        NIR_PASS_V(c->s, nir_lower_io_to_scalar, nir_var_shader_out);
+}
 
-        if (key->light_twoside)
+static void
+v3d_nir_lower_fs_late(struct v3d_compile *c)
+{
+        if (c->fs_key->light_twoside)
                 NIR_PASS_V(c->s, nir_lower_two_sided_color);
 
-        if (key->clamp_color)
+        if (c->fs_key->clamp_color)
                 NIR_PASS_V(c->s, nir_lower_clamp_color_outputs);
 
-        if (key->alpha_test) {
-                NIR_PASS_V(c->s, nir_lower_alpha_test, key->alpha_test_func,
+        if (c->fs_key->alpha_test) {
+                NIR_PASS_V(c->s, nir_lower_alpha_test,
+                           c->fs_key->alpha_test_func,
                            false);
         }
 
-        if (key->base.ucp_enables)
-                NIR_PASS_V(c->s, nir_lower_clip_fs, key->base.ucp_enables);
+        if (c->key->ucp_enables)
+                NIR_PASS_V(c->s, nir_lower_clip_fs, c->key->ucp_enables);
 
         /* Note: FS input scalarizing must happen after
          * nir_lower_two_sided_color, which only handles a vec4 at a time.
          */
         NIR_PASS_V(c->s, nir_lower_io_to_scalar, nir_var_shader_in);
+}
+
+uint64_t *v3d_compile(const struct v3d_compiler *compiler,
+                      struct v3d_key *key,
+                      struct v3d_prog_data **out_prog_data,
+                      nir_shader *s,
+                      void (*debug_output)(const char *msg,
+                                           void *debug_output_data),
+                      void *debug_output_data,
+                      int program_id, int variant_id,
+                      uint32_t *final_assembly_size)
+{
+        struct v3d_prog_data *prog_data;
+        struct v3d_compile *c = vir_compile_init(compiler, key, s,
+                                                 debug_output, debug_output_data,
+                                                 program_id, variant_id);
+
+        switch (c->s->info.stage) {
+        case MESA_SHADER_VERTEX:
+                c->vs_key = (struct v3d_vs_key *)key;
+                prog_data = rzalloc_size(NULL, sizeof(struct v3d_vs_prog_data));
+                break;
+        case MESA_SHADER_FRAGMENT:
+                c->fs_key = (struct v3d_fs_key *)key;
+                prog_data = rzalloc_size(NULL, sizeof(struct v3d_fs_prog_data));
+                break;
+        default:
+                unreachable("unsupported shader stage");
+        }
+
+        if (c->s->info.stage == MESA_SHADER_VERTEX) {
+                v3d_nir_lower_vs_early(c);
+        } else {
+                assert(c->s->info.stage == MESA_SHADER_FRAGMENT);
+                v3d_nir_lower_fs_early(c);
+        }
+
+        v3d_lower_nir(c);
+
+        if (c->s->info.stage == MESA_SHADER_VERTEX) {
+                v3d_nir_lower_vs_late(c);
+        } else {
+                assert(c->s->info.stage == MESA_SHADER_FRAGMENT);
+                v3d_nir_lower_fs_late(c);
+        }
+
+        NIR_PASS_V(c->s, v3d_nir_lower_io, c);
+        NIR_PASS_V(c->s, v3d_nir_lower_txf_ms, c);
+        NIR_PASS_V(c->s, nir_lower_idiv);
 
-        v3d_lower_nir_late(c);
         v3d_optimize_nir(c->s);
         NIR_PASS_V(c->s, nir_lower_bool_to_int32);
         NIR_PASS_V(c->s, nir_convert_from_ssa, true);
 
         v3d_nir_to_vir(c);
 
-        v3d_set_prog_data(c, &prog_data->base);
-        v3d_set_fs_prog_data_inputs(c, prog_data);
-        prog_data->writes_z = (c->s->info.outputs_written &
-                               (1 << FRAG_RESULT_DEPTH));
-        prog_data->discard = (c->s->info.fs.uses_discard ||
-                              c->fs_key->sample_alpha_to_coverage);
-        prog_data->uses_center_w = c->uses_center_w;
+        v3d_set_prog_data(c, prog_data);
 
+        *out_prog_data = prog_data;
         return v3d_return_qpu_insts(c, final_assembly_size);
 }
 
diff --git a/src/gallium/drivers/v3d/v3d_program.c b/src/gallium/drivers/v3d/v3d_program.c
index ec9d0339528..e0a77847880 100644
--- a/src/gallium/drivers/v3d/v3d_program.c
+++ b/src/gallium/drivers/v3d/v3d_program.c
@@ -354,32 +354,12 @@ v3d_get_compiled_shader(struct v3d_context *v3d, struct v3d_key *key)
         uint64_t *qpu_insts;
         uint32_t shader_size;
 
-        switch (s->info.stage) {
-        case MESA_SHADER_VERTEX:
-                shader->prog_data.vs = rzalloc(shader, struct v3d_vs_prog_data);
-
-                qpu_insts = v3d_compile_vs(v3d->screen->compiler,
-                                           (struct v3d_vs_key *)key,
-                                           shader->prog_data.vs, s,
-                                           v3d_shader_debug_output,
-                                           v3d,
-                                           program_id, variant_id,
-                                           &shader_size);
-                break;
-        case MESA_SHADER_FRAGMENT:
-                shader->prog_data.fs = rzalloc(shader, struct v3d_fs_prog_data);
-
-                qpu_insts = v3d_compile_fs(v3d->screen->compiler,
-                                           (struct v3d_fs_key *)key,
-                                           shader->prog_data.fs, s,
-                                           v3d_shader_debug_output,
-                                           v3d,
-                                           program_id, variant_id,
-                                           &shader_size);
-                break;
-        default:
-                unreachable("bad stage");
-        }
+        qpu_insts = v3d_compile(v3d->screen->compiler, key,
+                                &shader->prog_data.base, s,
+                                v3d_shader_debug_output,
+                                v3d,
+                                program_id, variant_id, &shader_size);
+        ralloc_steal(shader, shader->prog_data.base);
 
         v3d_set_shader_uniform_dirty_flags(shader);