xref: /third_party/mesa3d/src/amd/vulkan/radv_shader.c

/*
 * Copyright © 2016 Red Hat.
 * Copyright © 2016 Bas Nieuwenhuizen
 *
 * based in part on anv driver which is:
 * Copyright © 2015 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include "radv_shader.h"
#include "nir/nir.h"
#include "nir/nir_builder.h"
#include "spirv/nir_spirv.h"
#include "util/memstream.h"
#include "util/mesa-sha1.h"
#include "util/u_atomic.h"
#include "radv_debug.h"
#include "radv_private.h"
#include "radv_shader_args.h"

#include "util/debug.h"
#include "ac_binary.h"
#include "ac_exp_param.h"
#include "ac_nir.h"
#include "ac_rtld.h"
#include "aco_interface.h"
#include "sid.h"
#include "vk_format.h"

#ifdef LLVM_AVAILABLE
#include "ac_llvm_util.h"
#endif

void
radv_get_nir_options(struct radv_physical_device *device)
{
   device->nir_options = (nir_shader_compiler_options){
      .vertex_id_zero_based = true,
      .lower_scmp = true,
      .lower_flrp16 = true,
      .lower_flrp32 = true,
      .lower_flrp64 = true,
      .lower_device_index_to_zero = true,
      .lower_fdiv = true,
      .lower_fmod = true,
      .lower_ineg = true,
      .lower_bitfield_insert_to_bitfield_select = true,
      .lower_bitfield_extract = true,
      .lower_pack_snorm_2x16 = true,
      .lower_pack_snorm_4x8 = true,
      .lower_pack_unorm_2x16 = true,
      .lower_pack_unorm_4x8 = true,
      .lower_pack_half_2x16 = true,
      .lower_pack_64_2x32 = true,
      .lower_pack_64_4x16 = true,
      .lower_pack_32_2x16 = true,
      .lower_unpack_snorm_2x16 = true,
      .lower_unpack_snorm_4x8 = true,
      .lower_unpack_unorm_2x16 = true,
      .lower_unpack_unorm_4x8 = true,
      .lower_unpack_half_2x16 = true,
      .lower_ffma16 = true,
      .lower_ffma32 = true,
      .lower_ffma64 = true,
      .lower_fpow = true,
      .lower_mul_2x32_64 = true,
      .lower_rotate = true,
      .lower_iadd_sat = device->rad_info.chip_class <= GFX8,
      .has_fsub = true,
      .has_isub = true,
      .has_dot_4x8 = device->rad_info.has_accelerated_dot_product,
      .has_dot_2x16 = device->rad_info.has_accelerated_dot_product,
      .use_scoped_barrier = true,
      .max_unroll_iterations = 32,
      .max_unroll_iterations_aggressive = 128,
      .use_interpolated_input_intrinsics = true,
      .vectorize_vec2_16bit = true,
      /* nir_lower_int64() isn't actually called for the LLVM backend,
       * but this helps the loop unrolling heuristics. */
      .lower_int64_options = nir_lower_imul64 | nir_lower_imul_high64 | nir_lower_imul_2x32_64 |
                             nir_lower_divmod64 | nir_lower_minmax64 | nir_lower_iabs64,
      .lower_doubles_options = nir_lower_drcp | nir_lower_dsqrt | nir_lower_drsq | nir_lower_ddiv,
      .divergence_analysis_options = nir_divergence_view_index_uniform,
   };
}

bool
radv_can_dump_shader(struct radv_device *device, struct vk_shader_module *module,
                     bool meta_shader)
{
   if (!(device->instance->debug_flags & RADV_DEBUG_DUMP_SHADERS))
      return false;
   if (module)
      return !module->nir || (device->instance->debug_flags & RADV_DEBUG_DUMP_META_SHADERS);

   return meta_shader;
}

bool
radv_can_dump_shader_stats(struct radv_device *device, struct vk_shader_module *module)
{
   /* Only dump non-meta shader stats. */
   return device->instance->debug_flags & RADV_DEBUG_DUMP_SHADER_STATS && module && !module->nir;
}

void
radv_optimize_nir(const struct radv_device *device, struct nir_shader *shader,
                  bool optimize_conservatively, bool allow_copies)
{
   bool progress;

   do {
      progress = false;

      NIR_PASS(progress, shader, nir_split_array_vars, nir_var_function_temp);
      NIR_PASS(progress, shader, nir_shrink_vec_array_vars, nir_var_function_temp);

      NIR_PASS_V(shader, nir_lower_vars_to_ssa);

      if (allow_copies) {
         /* Only run this pass in the first call to
          * radv_optimize_nir.  Later calls assume that we've
          * lowered away any copy_deref instructions and we
          *  don't want to introduce any more.
          */
         NIR_PASS(progress, shader, nir_opt_find_array_copies);
      }

      NIR_PASS(progress, shader, nir_opt_copy_prop_vars);
      NIR_PASS(progress, shader, nir_opt_dead_write_vars);
      NIR_PASS(progress, shader, nir_remove_dead_variables,
               nir_var_function_temp | nir_var_shader_in | nir_var_shader_out, NULL);

      NIR_PASS_V(shader, nir_lower_alu_to_scalar, NULL, NULL);
      NIR_PASS_V(shader, nir_lower_phis_to_scalar, true);

      NIR_PASS(progress, shader, nir_copy_prop);
      NIR_PASS(progress, shader, nir_opt_remove_phis);
      NIR_PASS(progress, shader, nir_opt_dce);
      if (nir_opt_trivial_continues(shader)) {
         progress = true;
         NIR_PASS(progress, shader, nir_copy_prop);
         NIR_PASS(progress, shader, nir_opt_remove_phis);
         NIR_PASS(progress, shader, nir_opt_dce);
      }
      NIR_PASS(progress, shader, nir_opt_if, true);
      NIR_PASS(progress, shader, nir_opt_dead_cf);
      NIR_PASS(progress, shader, nir_opt_cse);
      NIR_PASS(progress, shader, nir_opt_peephole_select, 8, true, true);
      NIR_PASS(progress, shader, nir_opt_constant_folding);
      NIR_PASS(progress, shader, nir_opt_algebraic);

      NIR_PASS(progress, shader, nir_opt_undef);
      NIR_PASS(progress, shader, nir_opt_shrink_vectors,
               !device->instance->disable_shrink_image_store);
      if (shader->options->max_unroll_iterations) {
         NIR_PASS(progress, shader, nir_opt_loop_unroll);
      }
   } while (progress && !optimize_conservatively);

   NIR_PASS(progress, shader, nir_opt_conditional_discard);
   NIR_PASS(progress, shader, nir_opt_move, nir_move_load_ubo);
}

void
radv_optimize_nir_algebraic(nir_shader *nir, bool opt_offsets)
{
   bool more_algebraic = true;
   while (more_algebraic) {
      more_algebraic = false;
      NIR_PASS_V(nir, nir_copy_prop);
      NIR_PASS_V(nir, nir_opt_dce);
      NIR_PASS_V(nir, nir_opt_constant_folding);
      NIR_PASS_V(nir, nir_opt_cse);
      NIR_PASS(more_algebraic, nir, nir_opt_algebraic);
   }

   if (opt_offsets)
      NIR_PASS_V(nir, nir_opt_offsets);

   /* Do late algebraic optimization to turn add(a,
    * neg(b)) back into subs, then the mandatory cleanup
    * after algebraic.  Note that it may produce fnegs,
    * and if so then we need to keep running to squash
    * fneg(fneg(a)).
    */
   bool more_late_algebraic = true;
   while (more_late_algebraic) {
      more_late_algebraic = false;
      NIR_PASS(more_late_algebraic, nir, nir_opt_algebraic_late);
      NIR_PASS_V(nir, nir_opt_constant_folding);
      NIR_PASS_V(nir, nir_copy_prop);
      NIR_PASS_V(nir, nir_opt_dce);
      NIR_PASS_V(nir, nir_opt_cse);
   }
}

static void
shared_var_info(const struct glsl_type *type, unsigned *size, unsigned *align)
{
   assert(glsl_type_is_vector_or_scalar(type));

   uint32_t comp_size = glsl_type_is_boolean(type) ? 4 : glsl_get_bit_size(type) / 8;
   unsigned length = glsl_get_vector_elements(type);
   *size = comp_size * length, *align = comp_size;
}

struct radv_shader_debug_data {
   struct radv_device *device;
   const struct vk_shader_module *module;
};

static void
radv_spirv_nir_debug(void *private_data, enum nir_spirv_debug_level level, size_t spirv_offset,
                     const char *message)
{
   struct radv_shader_debug_data *debug_data = private_data;
   struct radv_instance *instance = debug_data->device->instance;

   static const VkDebugReportFlagsEXT vk_flags[] = {
      [NIR_SPIRV_DEBUG_LEVEL_INFO] = VK_DEBUG_REPORT_INFORMATION_BIT_EXT,
      [NIR_SPIRV_DEBUG_LEVEL_WARNING] = VK_DEBUG_REPORT_WARNING_BIT_EXT,
      [NIR_SPIRV_DEBUG_LEVEL_ERROR] = VK_DEBUG_REPORT_ERROR_BIT_EXT,
   };
   char buffer[256];

   snprintf(buffer, sizeof(buffer), "SPIR-V offset %lu: %s", (unsigned long)spirv_offset, message);

   vk_debug_report(&instance->vk, vk_flags[level], &debug_data->module->base, 0, 0, "radv", buffer);
}

static void
radv_compiler_debug(void *private_data, enum radv_compiler_debug_level level, const char *message)
{
   struct radv_shader_debug_data *debug_data = private_data;
   struct radv_instance *instance = debug_data->device->instance;

   static const VkDebugReportFlagsEXT vk_flags[] = {
      [RADV_COMPILER_DEBUG_LEVEL_PERFWARN] = VK_DEBUG_REPORT_PERFORMANCE_WARNING_BIT_EXT,
      [RADV_COMPILER_DEBUG_LEVEL_ERROR] = VK_DEBUG_REPORT_ERROR_BIT_EXT,
   };

   /* VK_DEBUG_REPORT_DEBUG_BIT_EXT specifies diagnostic information
    * from the implementation and layers.
    */
   vk_debug_report(&instance->vk, vk_flags[level] | VK_DEBUG_REPORT_DEBUG_BIT_EXT,
                   &debug_data->module->base, 0, 0, "radv", message);
}

static nir_ssa_def *
convert_pointer_to_64(nir_builder *b, const struct radv_physical_device *pdev, nir_ssa_def *ptr)
{
   nir_ssa_def *comp[] = {ptr, nir_imm_int(b, pdev->rad_info.address32_hi)};
   return nir_pack_64_2x32(b, nir_vec(b, comp, 2));
}

static bool
lower_intrinsics(nir_shader *nir, const struct radv_pipeline_key *key,
                 const struct radv_pipeline_layout *layout, const struct radv_physical_device *pdev)
{
   nir_function_impl *entry = nir_shader_get_entrypoint(nir);
   bool progress = false;
   nir_builder b;

   nir_builder_init(&b, entry);

   nir_foreach_block (block, entry) {
      nir_foreach_instr_safe (instr, block) {
         if (instr->type != nir_instr_type_intrinsic)
            continue;

         nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
         b.cursor = nir_before_instr(&intrin->instr);

         nir_ssa_def *def = NULL;
         switch (intrin->intrinsic) {
         case nir_intrinsic_load_vulkan_descriptor:
            if (nir_intrinsic_desc_type(intrin) == VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR) {
               nir_ssa_def *addr =
                  convert_pointer_to_64(&b, pdev,
                                        nir_iadd(&b, nir_channel(&b, intrin->src[0].ssa, 0),
                                                 nir_channel(&b, intrin->src[0].ssa, 1)));

               def = nir_build_load_global(&b, 1, 64, addr, .access = ACCESS_NON_WRITEABLE,
                                           .align_mul = 8, .align_offset = 0);
            } else {
               def = nir_vector_insert_imm(&b, intrin->src[0].ssa, nir_imm_int(&b, 0), 2);
            }
            break;
         case nir_intrinsic_vulkan_resource_index: {
            unsigned desc_set = nir_intrinsic_desc_set(intrin);
            unsigned binding = nir_intrinsic_binding(intrin);
            struct radv_descriptor_set_layout *desc_layout = layout->set[desc_set].layout;

            nir_ssa_def *new_res = nir_vulkan_resource_index(
               &b, 3, 32, intrin->src[0].ssa, .desc_set = desc_set, .binding = binding,
               .desc_type = nir_intrinsic_desc_type(intrin));

            nir_ssa_def *stride;
            if (desc_layout->binding[binding].type == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC ||
                desc_layout->binding[binding].type == VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC) {
               stride = nir_imm_int(&b, 16);
            } else {
               stride = nir_imm_int(&b, desc_layout->binding[binding].size);
            }
            def = nir_vector_insert_imm(&b, new_res, stride, 2);
            break;
         }
         case nir_intrinsic_vulkan_resource_reindex: {
            nir_ssa_def *binding_ptr = nir_channel(&b, intrin->src[0].ssa, 1);
            nir_ssa_def *stride = nir_channel(&b, intrin->src[0].ssa, 2);
            binding_ptr = nir_iadd(&b, binding_ptr, nir_imul(&b, intrin->src[1].ssa, stride));
            def = nir_vector_insert_imm(&b, intrin->src[0].ssa, binding_ptr, 1);
            break;
         }
         case nir_intrinsic_is_sparse_texels_resident:
            def = nir_ieq_imm(&b, intrin->src[0].ssa, 0);
            break;
         case nir_intrinsic_sparse_residency_code_and:
            def = nir_ior(&b, intrin->src[0].ssa, intrin->src[1].ssa);
            break;
         case nir_intrinsic_load_view_index:
            if (key->has_multiview_view_index)
               continue;
            def = nir_imm_zero(&b, 1, 32);
            break;
         default:
            continue;
         }

         nir_ssa_def_rewrite_uses(&intrin->dest.ssa, def);

         nir_instr_remove(instr);
         progress = true;
      }
   }

   return progress;
}

static bool
radv_lower_primitive_shading_rate(nir_shader *nir)
{
   nir_function_impl *impl = nir_shader_get_entrypoint(nir);
   bool progress = false;

   nir_builder b;
   nir_builder_init(&b, impl);

   /* Iterate in reverse order since there should be only one deref store to PRIMITIVE_SHADING_RATE
    * after lower_io_to_temporaries for vertex shaders.
    */
   nir_foreach_block_reverse(block, impl) {
      nir_foreach_instr_reverse(instr, block) {
         if (instr->type != nir_instr_type_intrinsic)
            continue;

         nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
         if (intr->intrinsic != nir_intrinsic_store_deref)
            continue;

         nir_variable *var = nir_intrinsic_get_var(intr, 0);
         if (var->data.mode != nir_var_shader_out ||
             var->data.location != VARYING_SLOT_PRIMITIVE_SHADING_RATE)
            continue;

         b.cursor = nir_before_instr(instr);

         nir_ssa_def *val = nir_ssa_for_src(&b, intr->src[1], 1);

         /* x_rate = (shadingRate & (Horizontal2Pixels | Horizontal4Pixels)) ? 0x1 : 0x0; */
         nir_ssa_def *x_rate = nir_iand(&b, val, nir_imm_int(&b, 12));
         x_rate = nir_b2i32(&b, nir_ine(&b, x_rate, nir_imm_int(&b, 0)));

         /* y_rate = (shadingRate & (Vertical2Pixels | Vertical4Pixels)) ? 0x1 : 0x0; */
         nir_ssa_def *y_rate = nir_iand(&b, val, nir_imm_int(&b, 3));
         y_rate = nir_b2i32(&b, nir_ine(&b, y_rate, nir_imm_int(&b, 0)));

         /* Bits [2:3] = VRS rate X
          * Bits [4:5] = VRS rate Y
          * HW shading rate = (xRate << 2) | (yRate << 4)
          */
         nir_ssa_def *out = nir_ior(&b, nir_ishl(&b, x_rate, nir_imm_int(&b, 2)),
                                        nir_ishl(&b, y_rate, nir_imm_int(&b, 4)));

         nir_instr_rewrite_src(&intr->instr, &intr->src[1], nir_src_for_ssa(out));

         progress = true;
         if (nir->info.stage == MESA_SHADER_VERTEX)
            return progress;
      }
   }

   return progress;
}

nir_shader *
radv_shader_compile_to_nir(struct radv_device *device, struct vk_shader_module *module,
                           const char *entrypoint_name, gl_shader_stage stage,
                           const VkSpecializationInfo *spec_info,
                           const struct radv_pipeline_layout *layout,
                           const struct radv_pipeline_key *key)
{
   unsigned subgroup_size = 64, ballot_bit_size = 64;
   if (key->cs.compute_subgroup_size) {
      /* Only compute shaders currently support requiring a
       * specific subgroup size.
       */
      assert(stage == MESA_SHADER_COMPUTE);
      subgroup_size = key->cs.compute_subgroup_size;
      ballot_bit_size = key->cs.compute_subgroup_size;
   }

   nir_shader *nir;

   if (module->nir) {
      /* Some things such as our meta clear/blit code will give us a NIR
       * shader directly.  In that case, we just ignore the SPIR-V entirely
       * and just use the NIR shader */
      nir = module->nir;
      nir->options = &device->physical_device->nir_options;
      nir_validate_shader(nir, "in internal shader");

      assert(exec_list_length(&nir->functions) == 1);
   } else {
      uint32_t *spirv = (uint32_t *)module->data;
      assert(module->size % 4 == 0);

      if (device->instance->debug_flags & RADV_DEBUG_DUMP_SPIRV)
         radv_print_spirv(module->data, module->size, stderr);

      uint32_t num_spec_entries = 0;
      struct nir_spirv_specialization *spec_entries =
         vk_spec_info_to_nir_spirv(spec_info, &num_spec_entries);
      struct radv_shader_debug_data spirv_debug_data = {
         .device = device,
         .module = module,
      };
      const struct spirv_to_nir_options spirv_options = {
         .caps =
            {
               .amd_fragment_mask = true,
               .amd_gcn_shader = true,
               .amd_image_gather_bias_lod = true,
               .amd_image_read_write_lod = true,
               .amd_shader_ballot = true,
               .amd_shader_explicit_vertex_parameter = true,
               .amd_trinary_minmax = true,
               .demote_to_helper_invocation = true,
               .derivative_group = true,
               .descriptor_array_dynamic_indexing = true,
               .descriptor_array_non_uniform_indexing = true,
               .descriptor_indexing = true,
               .device_group = true,
               .draw_parameters = true,
               .float_controls = true,
               .float16 = device->physical_device->rad_info.has_packed_math_16bit,
               .float32_atomic_add = true,
               .float32_atomic_min_max = true,
               .float64 = true,
               .float64_atomic_min_max = true,
               .geometry_streams = true,
               .groups = true,
               .image_atomic_int64 = true,
               .image_ms_array = true,
               .image_read_without_format = true,
               .image_write_without_format = true,
               .int8 = true,
               .int16 = true,
               .int64 = true,
               .int64_atomics = true,
               .min_lod = true,
               .multiview = true,
               .physical_storage_buffer_address = true,
               .post_depth_coverage = true,
               .ray_tracing = true,
               .runtime_descriptor_array = true,
               .shader_clock = true,
               .shader_viewport_index_layer = true,
               .sparse_residency = true,
               .stencil_export = true,
               .storage_8bit = true,
               .storage_16bit = true,
               .storage_image_ms = true,
               .subgroup_arithmetic = true,
               .subgroup_ballot = true,
               .subgroup_basic = true,
               .subgroup_quad = true,
               .subgroup_shuffle = true,
               .subgroup_uniform_control_flow = true,
               .subgroup_vote = true,
               .tessellation = true,
               .transform_feedback = true,
               .variable_pointers = true,
               .vk_memory_model = true,
               .vk_memory_model_device_scope = true,
               .fragment_shading_rate = device->physical_device->rad_info.chip_class >= GFX10_3,
               .workgroup_memory_explicit_layout = true,
            },
         .ubo_addr_format = nir_address_format_vec2_index_32bit_offset,
         .ssbo_addr_format = nir_address_format_vec2_index_32bit_offset,
         .phys_ssbo_addr_format = nir_address_format_64bit_global,
         .push_const_addr_format = nir_address_format_logical,
         .shared_addr_format = nir_address_format_32bit_offset,
         .constant_addr_format = nir_address_format_64bit_global,
         .use_deref_buffer_array_length = true,
         .debug =
            {
               .func = radv_spirv_nir_debug,
               .private_data = &spirv_debug_data,
            },
      };
      nir = spirv_to_nir(spirv, module->size / 4, spec_entries, num_spec_entries, stage,
                         entrypoint_name, &spirv_options, &device->physical_device->nir_options);
      assert(nir->info.stage == stage);
      nir_validate_shader(nir, "after spirv_to_nir");

      free(spec_entries);

      const struct nir_lower_sysvals_to_varyings_options sysvals_to_varyings = {
         .point_coord = true,
      };
      NIR_PASS_V(nir, nir_lower_sysvals_to_varyings, &sysvals_to_varyings);

      /* We have to lower away local constant initializers right before we
       * inline functions.  That way they get properly initialized at the top
       * of the function and not at the top of its caller.
       */
      NIR_PASS_V(nir, nir_lower_variable_initializers, nir_var_function_temp);
      NIR_PASS_V(nir, nir_lower_returns);
      NIR_PASS_V(nir, nir_inline_functions);
      NIR_PASS_V(nir, nir_copy_prop);
      NIR_PASS_V(nir, nir_opt_deref);

      /* Pick off the single entrypoint that we want */
      foreach_list_typed_safe(nir_function, func, node, &nir->functions)
      {
         if (func->is_entrypoint)
            func->name = ralloc_strdup(func, "main");
         else
            exec_node_remove(&func->node);
      }
      assert(exec_list_length(&nir->functions) == 1);

      /* Make sure we lower constant initializers on output variables so that
       * nir_remove_dead_variables below sees the corresponding stores
       */
      NIR_PASS_V(nir, nir_lower_variable_initializers, nir_var_shader_out);

      /* Now that we've deleted all but the main function, we can go ahead and
       * lower the rest of the constant initializers.
       */
      NIR_PASS_V(nir, nir_lower_variable_initializers, ~0);

      /* Split member structs.  We do this before lower_io_to_temporaries so that
       * it doesn't lower system values to temporaries by accident.
       */
      NIR_PASS_V(nir, nir_split_var_copies);
      NIR_PASS_V(nir, nir_split_per_member_structs);

      if (nir->info.stage == MESA_SHADER_FRAGMENT)
         NIR_PASS_V(nir, nir_lower_io_to_vector, nir_var_shader_out);
      if (nir->info.stage == MESA_SHADER_FRAGMENT)
         NIR_PASS_V(nir, nir_lower_input_attachments,
                    &(nir_input_attachment_options){
                       .use_fragcoord_sysval = true,
                       .use_layer_id_sysval = false,
                    });

      NIR_PASS_V(nir, nir_remove_dead_variables,
                 nir_var_shader_in | nir_var_shader_out | nir_var_system_value | nir_var_mem_shared,
                 NULL);

      /* Variables can make nir_propagate_invariant more conservative
       * than it needs to be.
       */
      NIR_PASS_V(nir, nir_lower_global_vars_to_local);
      NIR_PASS_V(nir, nir_lower_vars_to_ssa);

      NIR_PASS_V(nir, nir_propagate_invariant, key->invariant_geom);

      NIR_PASS_V(nir, nir_lower_clip_cull_distance_arrays);

      NIR_PASS_V(nir, nir_lower_discard_or_demote, key->ps.lower_discard_to_demote);

      nir_lower_doubles_options lower_doubles = nir->options->lower_doubles_options;

      if (device->physical_device->rad_info.chip_class == GFX6) {
         /* GFX6 doesn't support v_floor_f64 and the precision
          * of v_fract_f64 which is used to implement 64-bit
          * floor is less than what Vulkan requires.
          */
         lower_doubles |= nir_lower_dfloor;
      }

      NIR_PASS_V(nir, nir_lower_doubles, NULL, lower_doubles);
   }

   NIR_PASS_V(nir, nir_lower_system_values);
   NIR_PASS_V(nir, nir_lower_compute_system_values, NULL);

   /* Vulkan uses the separate-shader linking model */
   nir->info.separate_shader = true;

   nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));

   if (nir->info.stage == MESA_SHADER_GEOMETRY) {
      unsigned nir_gs_flags = nir_lower_gs_intrinsics_per_stream;

      if (key->use_ngg && !radv_use_llvm_for_stage(device, stage)) {
         /* ACO needs NIR to do some of the hard lifting */
         nir_gs_flags |= nir_lower_gs_intrinsics_count_primitives |
                         nir_lower_gs_intrinsics_count_vertices_per_primitive |
                         nir_lower_gs_intrinsics_overwrite_incomplete;
      }

      nir_lower_gs_intrinsics(nir, nir_gs_flags);
   }

   static const nir_lower_tex_options tex_options = {
      .lower_txp = ~0,
      .lower_tg4_offsets = true,
      .lower_txs_cube_array = true,
      .lower_to_fragment_fetch_amd = true,
   };

   nir_lower_tex(nir, &tex_options);

   static const nir_lower_image_options image_options = {
      .lower_cube_size = true,
   };

   nir_lower_image(nir, &image_options);

   nir_lower_vars_to_ssa(nir);

   if (nir->info.stage == MESA_SHADER_VERTEX || nir->info.stage == MESA_SHADER_GEOMETRY ||
       nir->info.stage == MESA_SHADER_FRAGMENT) {
      NIR_PASS_V(nir, nir_lower_io_to_temporaries, nir_shader_get_entrypoint(nir), true, true);
   } else if (nir->info.stage == MESA_SHADER_TESS_EVAL) {
      NIR_PASS_V(nir, nir_lower_io_to_temporaries, nir_shader_get_entrypoint(nir), true, false);
   }

   nir_split_var_copies(nir);

   nir_lower_global_vars_to_local(nir);
   nir_remove_dead_variables(nir, nir_var_function_temp, NULL);
   bool gfx7minus = device->physical_device->rad_info.chip_class <= GFX7;
   nir_lower_subgroups(nir, &(struct nir_lower_subgroups_options){
                               .subgroup_size = subgroup_size,
                               .ballot_bit_size = ballot_bit_size,
                               .ballot_components = 1,
                               .lower_to_scalar = 1,
                               .lower_subgroup_masks = 1,
                               .lower_shuffle = 1,
                               .lower_shuffle_to_32bit = 1,
                               .lower_vote_eq = 1,
                               .lower_quad_broadcast_dynamic = 1,
                               .lower_quad_broadcast_dynamic_to_const = gfx7minus,
                               .lower_shuffle_to_swizzle_amd = 1,
                            });

   nir_lower_load_const_to_scalar(nir);

   if (!key->optimisations_disabled)
      radv_optimize_nir(device, nir, false, true);

   /* call radv_nir_lower_ycbcr_textures() late as there might still be
    * tex with undef texture/sampler before first optimization */
   NIR_PASS_V(nir, radv_nir_lower_ycbcr_textures, layout);

   /* We call nir_lower_var_copies() after the first radv_optimize_nir()
    * to remove any copies introduced by nir_opt_find_array_copies().
    */
   nir_lower_var_copies(nir);

   unsigned lower_flrp = (nir->options->lower_flrp16 ? 16 : 0) |
                         (nir->options->lower_flrp32 ? 32 : 0) |
                         (nir->options->lower_flrp64 ? 64 : 0);
   if (lower_flrp != 0) {
      if (nir_lower_flrp(nir, lower_flrp, false /* always_precise */))
         NIR_PASS_V(nir, nir_opt_constant_folding);
   }

   const nir_opt_access_options opt_access_options = {
      .is_vulkan = true,
      .infer_non_readable = true,
   };
   NIR_PASS_V(nir, nir_opt_access, &opt_access_options);

   NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_push_const, nir_address_format_32bit_offset);

   NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_ubo | nir_var_mem_ssbo,
              nir_address_format_vec2_index_32bit_offset);

   NIR_PASS_V(nir, lower_intrinsics, key, layout, device->physical_device);

   /* Lower deref operations for compute shared memory. */
   if (nir->info.stage == MESA_SHADER_COMPUTE) {
      if (!nir->info.shared_memory_explicit_layout) {
         NIR_PASS_V(nir, nir_lower_vars_to_explicit_types, nir_var_mem_shared, shared_var_info);
      }
      NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_shared, nir_address_format_32bit_offset);

      if (nir->info.zero_initialize_shared_memory && nir->info.shared_size > 0) {
         const unsigned chunk_size = 16; /* max single store size */
         const unsigned shared_size = ALIGN(nir->info.shared_size, chunk_size);
         NIR_PASS_V(nir, nir_zero_initialize_shared_memory, shared_size, chunk_size);
      }
   }

   nir_lower_explicit_io(nir, nir_var_mem_global | nir_var_mem_constant,
                         nir_address_format_64bit_global);

   /* Lower large variables that are always constant with load_constant
    * intrinsics, which get turned into PC-relative loads from a data
    * section next to the shader.
    */
   NIR_PASS_V(nir, nir_opt_large_constants, glsl_get_natural_size_align_bytes, 16);

   /* Lower primitive shading rate to match HW requirements. */
   if ((nir->info.stage == MESA_SHADER_VERTEX ||
        nir->info.stage == MESA_SHADER_GEOMETRY) &&
       nir->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_PRIMITIVE_SHADING_RATE)) {
      NIR_PASS_V(nir, radv_lower_primitive_shading_rate);
   }

   /* Indirect lowering must be called after the radv_optimize_nir() loop
    * has been called at least once. Otherwise indirect lowering can
    * bloat the instruction count of the loop and cause it to be
    * considered too large for unrolling.
    */
   if (ac_nir_lower_indirect_derefs(nir, device->physical_device->rad_info.chip_class) &&
       !key->optimisations_disabled && nir->info.stage != MESA_SHADER_COMPUTE) {
      /* Optimize the lowered code before the linking optimizations. */
      radv_optimize_nir(device, nir, false, false);
   }

   return nir;
}

static int
type_size_vec4(const struct glsl_type *type, bool bindless)
{
   return glsl_count_attribute_slots(type, false);
}

static nir_variable *
find_layer_in_var(nir_shader *nir)
{
   nir_variable *var = nir_find_variable_with_location(nir, nir_var_shader_in, VARYING_SLOT_LAYER);
   if (var != NULL)
      return var;

   var = nir_variable_create(nir, nir_var_shader_in, glsl_int_type(), "layer id");
   var->data.location = VARYING_SLOT_LAYER;
   var->data.interpolation = INTERP_MODE_FLAT;
   return var;
}

/* We use layered rendering to implement multiview, which means we need to map
 * view_index to gl_Layer. The code generates a load from the layer_id sysval,
 * but since we don't have a way to get at this information from the fragment
 * shader, we also need to lower this to the gl_Layer varying.  This pass
 * lowers both to a varying load from the LAYER slot, before lowering io, so
 * that nir_assign_var_locations() will give the LAYER varying the correct
 * driver_location.
 */

static bool
lower_view_index(nir_shader *nir)
{
   bool progress = false;
   nir_function_impl *entry = nir_shader_get_entrypoint(nir);
   nir_builder b;
   nir_builder_init(&b, entry);

   nir_variable *layer = NULL;
   nir_foreach_block (block, entry) {
      nir_foreach_instr_safe (instr, block) {
         if (instr->type != nir_instr_type_intrinsic)
            continue;

         nir_intrinsic_instr *load = nir_instr_as_intrinsic(instr);
         if (load->intrinsic != nir_intrinsic_load_view_index)
            continue;

         if (!layer)
            layer = find_layer_in_var(nir);

         b.cursor = nir_before_instr(instr);
         nir_ssa_def *def = nir_load_var(&b, layer);
         nir_ssa_def_rewrite_uses(&load->dest.ssa, def);

         nir_instr_remove(instr);
         progress = true;
      }
   }

   return progress;
}

void
radv_lower_io(struct radv_device *device, nir_shader *nir)
{
   if (nir->info.stage == MESA_SHADER_COMPUTE)
      return;

   if (nir->info.stage == MESA_SHADER_FRAGMENT) {
      NIR_PASS_V(nir, lower_view_index);
      nir_assign_io_var_locations(nir, nir_var_shader_in, &nir->num_inputs, MESA_SHADER_FRAGMENT);
   }

   /* The RADV/LLVM backend expects 64-bit IO to be lowered. */
   nir_lower_io_options options =
      radv_use_llvm_for_stage(device, nir->info.stage) ? nir_lower_io_lower_64bit_to_32 : 0;

   NIR_PASS_V(nir, nir_lower_io, nir_var_shader_in | nir_var_shader_out, type_size_vec4, options);

   /* This pass needs actual constants */
   nir_opt_constant_folding(nir);

   NIR_PASS_V(nir, nir_io_add_const_offset_to_base, nir_var_shader_in | nir_var_shader_out);
}

bool
radv_lower_io_to_mem(struct radv_device *device, struct nir_shader *nir,
                     const struct radv_shader_info *info, const struct radv_pipeline_key *pl_key)
{
   if (nir->info.stage == MESA_SHADER_VERTEX) {
      if (info->vs.as_ls) {
         ac_nir_lower_ls_outputs_to_mem(nir, info->vs.tcs_in_out_eq,
                                        info->vs.tcs_temp_only_input_mask,
                                        info->vs.num_linked_outputs);
         return true;
      } else if (info->vs.as_es) {
         ac_nir_lower_es_outputs_to_mem(nir, device->physical_device->rad_info.chip_class,
                                        info->vs.num_linked_outputs);
         return true;
      }
   } else if (nir->info.stage == MESA_SHADER_TESS_CTRL) {
      ac_nir_lower_hs_inputs_to_mem(nir, info->vs.tcs_in_out_eq, info->tcs.num_linked_inputs);
      ac_nir_lower_hs_outputs_to_mem(
         nir, device->physical_device->rad_info.chip_class, info->tcs.tes_reads_tess_factors,
         info->tcs.tes_inputs_read, info->tcs.tes_patch_inputs_read, info->tcs.num_linked_inputs,
         info->tcs.num_linked_outputs, info->tcs.num_linked_patch_outputs, true);
      ac_nir_lower_tess_to_const(nir, pl_key->tcs.tess_input_vertices, info->num_tess_patches,
                                 ac_nir_lower_patch_vtx_in | ac_nir_lower_num_patches);

      return true;
   } else if (nir->info.stage == MESA_SHADER_TESS_EVAL) {
      ac_nir_lower_tes_inputs_to_mem(nir, info->tes.num_linked_inputs,
                                     info->tes.num_linked_patch_inputs);
      ac_nir_lower_tess_to_const(nir, nir->info.tess.tcs_vertices_out, info->num_tess_patches,
                                 ac_nir_lower_patch_vtx_in | ac_nir_lower_num_patches);

      if (info->tes.as_es) {
         ac_nir_lower_es_outputs_to_mem(nir, device->physical_device->rad_info.chip_class,
                                        info->tes.num_linked_outputs);
      }

      return true;
   } else if (nir->info.stage == MESA_SHADER_GEOMETRY) {
      ac_nir_lower_gs_inputs_to_mem(nir, device->physical_device->rad_info.chip_class,
                                    info->gs.num_linked_inputs);
      return true;
   }

   return false;
}

bool
radv_consider_culling(struct radv_device *device, struct nir_shader *nir, uint64_t ps_inputs_read,
                      unsigned num_vertices_per_primitive, const struct radv_shader_info *info)
{
   /* Culling doesn't make sense for meta shaders. */
   if (!!nir->info.name)
      return false;

   /* We don't support culling with multiple viewports yet. */
   if (nir->info.outputs_written & (VARYING_BIT_VIEWPORT | VARYING_BIT_VIEWPORT_MASK))
      return false;

   /* We don't support culling with vertex shader prologs. */
   if (info->vs.has_prolog)
      return false;

   if (!device->physical_device->use_ngg_culling)
      return false;

   /* Shader based culling efficiency can depend on PS throughput.
    * Estimate an upper limit for PS input param count based on GPU info.
    */
   unsigned max_ps_params;
   unsigned max_render_backends = device->physical_device->rad_info.max_render_backends;
   unsigned max_se = device->physical_device->rad_info.max_se;

   if (max_render_backends / max_se == 4)
      max_ps_params = 6; /* Sienna Cichlid and other GFX10.3 dGPUs. */
   else
      max_ps_params = 4; /* Navi 1x. */

   /* TODO: consider other heuristics here, such as PS execution time */
   if (util_bitcount64(ps_inputs_read & ~VARYING_BIT_POS) > max_ps_params)
      return false;

   /* Only triangle culling is supported. */
   if (num_vertices_per_primitive != 3)
      return false;

   /* When the shader writes memory, it is difficult to guarantee correctness.
    * Future work:
    * - if only write-only SSBOs are used
    * - if we can prove that non-position outputs don't rely on memory stores
    * then may be okay to keep the memory stores in the 1st shader part, and delete them from the 2nd.
    */
   if (nir->info.writes_memory)
      return false;

   /* When the shader relies on the subgroup invocation ID, we'd break it, because the ID changes after the culling.
    * Future work: try to save this to LDS and reload, but it can still be broken in subtle ways.
    */
   if (BITSET_TEST(nir->info.system_values_read, SYSTEM_VALUE_SUBGROUP_INVOCATION))
      return false;

   return true;
}

void radv_lower_ngg(struct radv_device *device, struct nir_shader *nir,
                    const struct radv_shader_info *info,
                    const struct radv_pipeline_key *pl_key)
{
   /* TODO: support the LLVM backend with the NIR lowering */
   assert(!radv_use_llvm_for_stage(device, nir->info.stage));

   assert(nir->info.stage == MESA_SHADER_VERTEX ||
          nir->info.stage == MESA_SHADER_TESS_EVAL ||
          nir->info.stage == MESA_SHADER_GEOMETRY);

   const struct gfx10_ngg_info *ngg_info = &info->ngg_info;
   unsigned num_vertices_per_prim = 3;

   /* Get the number of vertices per input primitive */
   if (nir->info.stage == MESA_SHADER_TESS_EVAL) {
      if (nir->info.tess.point_mode)
         num_vertices_per_prim = 1;
      else if (nir->info.tess.primitive_mode == GL_ISOLINES)
         num_vertices_per_prim = 2;

      /* Manually mark the primitive ID used, so the shader can repack it. */
      if (info->tes.outinfo.export_prim_id)
         BITSET_SET(nir->info.system_values_read, SYSTEM_VALUE_PRIMITIVE_ID);

   } else if (nir->info.stage == MESA_SHADER_VERTEX) {
      /* Need to add 1, because: V_028A6C_POINTLIST=0, V_028A6C_LINESTRIP=1, V_028A6C_TRISTRIP=2, etc. */
      num_vertices_per_prim = si_conv_prim_to_gs_out(pl_key->vs.topology) + 1;

      /* Manually mark the instance ID used, so the shader can repack it. */
      if (pl_key->vs.instance_rate_inputs)
         BITSET_SET(nir->info.system_values_read, SYSTEM_VALUE_INSTANCE_ID);

   } else if (nir->info.stage == MESA_SHADER_GEOMETRY) {
      num_vertices_per_prim = nir->info.gs.vertices_in;
   } else {
      unreachable("NGG needs to be VS, TES or GS.");
   }

   /* Invocations that process an input vertex */
   unsigned max_vtx_in = MIN2(256, ngg_info->enable_vertex_grouping ? ngg_info->hw_max_esverts : num_vertices_per_prim * ngg_info->max_gsprims);

   if (nir->info.stage == MESA_SHADER_VERTEX ||
       nir->info.stage == MESA_SHADER_TESS_EVAL) {
      bool export_prim_id;

      assert(info->is_ngg);

      if (info->has_ngg_culling)
         radv_optimize_nir_algebraic(nir, false);

      if (nir->info.stage == MESA_SHADER_VERTEX) {
         export_prim_id = info->vs.outinfo.export_prim_id;
      } else {
         export_prim_id = info->tes.outinfo.export_prim_id;
      }

      ac_nir_lower_ngg_nogs(
         nir,
         max_vtx_in,
         num_vertices_per_prim,
         info->workgroup_size,
         info->wave_size,
         info->has_ngg_culling,
         info->has_ngg_early_prim_export,
         info->is_ngg_passthrough,
         export_prim_id,
         pl_key->vs.provoking_vtx_last,
         false,
         pl_key->vs.instance_rate_inputs);
   } else if (nir->info.stage == MESA_SHADER_GEOMETRY) {
      assert(info->is_ngg);
      ac_nir_lower_ngg_gs(
         nir, info->wave_size, info->workgroup_size,
         info->ngg_info.esgs_ring_size,
         info->gs.gsvs_vertex_size,
         info->ngg_info.ngg_emit_size * 4u,
         pl_key->vs.provoking_vtx_last);
   } else {
      unreachable("invalid SW stage passed to radv_lower_ngg");
   }
}

static unsigned
get_size_class(unsigned size, bool round_up)
{
   size = round_up ? util_logbase2_ceil(size) : util_logbase2(size);
   unsigned size_class =
      MAX2(size, RADV_SHADER_ALLOC_MIN_SIZE_CLASS) - RADV_SHADER_ALLOC_MIN_SIZE_CLASS;
   return MIN2(size_class, RADV_SHADER_ALLOC_NUM_FREE_LISTS - 1);
}

static void
remove_hole(struct radv_device *device, union radv_shader_arena_block *hole)
{
   unsigned size_class = get_size_class(hole->size, false);
   list_del(&hole->freelist);
   if (list_is_empty(&device->shader_free_lists[size_class]))
      device->shader_free_list_mask &= ~(1u << size_class);
}

static void
add_hole(struct radv_device *device, union radv_shader_arena_block *hole)
{
   unsigned size_class = get_size_class(hole->size, false);
   list_addtail(&hole->freelist, &device->shader_free_lists[size_class]);
   device->shader_free_list_mask |= 1u << size_class;
}

static union radv_shader_arena_block *
alloc_block_obj(struct radv_device *device)
{
   if (!list_is_empty(&device->shader_block_obj_pool)) {
      union radv_shader_arena_block *block =
         list_first_entry(&device->shader_block_obj_pool, union radv_shader_arena_block, pool);
      list_del(&block->pool);
      return block;
   }

   return malloc(sizeof(union radv_shader_arena_block));
}

static void
free_block_obj(struct radv_device *device, union radv_shader_arena_block *block)
{
   list_add(&block->pool, &device->shader_block_obj_pool);
}

/* Segregated fit allocator, implementing a good-fit allocation policy.
 *
 * This is an variation of sequential fit allocation with several lists of free blocks ("holes")
 * instead of one. Each list of holes only contains holes of a certain range of sizes, so holes that
 * are too small can easily be ignored while allocating. Because this also ignores holes that are
 * larger than necessary (approximating best-fit allocation), this could be described as a
 * "good-fit" allocator.
 *
 * Typically, shaders are allocated and only free'd when the device is destroyed. For this pattern,
 * this should allocate blocks for shaders fast and with no fragmentation, while still allowing
 * free'd memory to be re-used.
 */
static union radv_shader_arena_block *
alloc_shader_memory(struct radv_device *device, uint32_t size, void *ptr)
{
   size = align(size, RADV_SHADER_ALLOC_ALIGNMENT);

   mtx_lock(&device->shader_arena_mutex);

   /* Try to use an existing hole. Unless the shader is very large, this should only have to look
    * at the first one available.
    */
   unsigned free_list_mask = BITFIELD_MASK(RADV_SHADER_ALLOC_NUM_FREE_LISTS);
   unsigned size_class =
      ffs(device->shader_free_list_mask & (free_list_mask << get_size_class(size, true)));
   if (size_class) {
      size_class--;

      list_for_each_entry(union radv_shader_arena_block, hole,
                          &device->shader_free_lists[size_class], freelist)
      {
         if (hole->size < size)
            continue;

         assert(hole->offset % RADV_SHADER_ALLOC_ALIGNMENT == 0);

         if (size == hole->size) {
            remove_hole(device, hole);
            hole->freelist.next = ptr;
            mtx_unlock(&device->shader_arena_mutex);
            return hole;
         } else {
            union radv_shader_arena_block *alloc = alloc_block_obj(device);
            if (!alloc) {
               mtx_unlock(&device->shader_arena_mutex);
               return NULL;
            }
            list_addtail(&alloc->list, &hole->list);
            alloc->freelist.prev = NULL;
            alloc->freelist.next = ptr;
            alloc->arena = hole->arena;
            alloc->offset = hole->offset;
            alloc->size = size;

            remove_hole(device, hole);
            hole->offset += size;
            hole->size -= size;
            add_hole(device, hole);

            mtx_unlock(&device->shader_arena_mutex);
            return alloc;
         }
      }
   }

   /* Allocate a new shader arena. */
   struct radv_shader_arena *arena = calloc(1, sizeof(struct radv_shader_arena));
   union radv_shader_arena_block *alloc = NULL, *hole = NULL;
   if (!arena)
      goto fail;

   unsigned arena_size = MAX2(RADV_SHADER_ALLOC_MIN_ARENA_SIZE, size);
   VkResult result = device->ws->buffer_create(
      device->ws, arena_size, RADV_SHADER_ALLOC_ALIGNMENT, RADEON_DOMAIN_VRAM,
      RADEON_FLAG_NO_INTERPROCESS_SHARING | RADEON_FLAG_32BIT |
         (device->physical_device->rad_info.cpdma_prefetch_writes_memory ? 0
                                                                         : RADEON_FLAG_READ_ONLY),
      RADV_BO_PRIORITY_SHADER, 0, &arena->bo);
   if (result != VK_SUCCESS)
      goto fail;

   list_inithead(&arena->entries);

   arena->ptr = (char *)device->ws->buffer_map(arena->bo);
   if (!arena->ptr)
      goto fail;

   alloc = alloc_block_obj(device);
   hole = arena_size - size > 0 ? alloc_block_obj(device) : alloc;
   if (!alloc || !hole)
      goto fail;
   list_addtail(&alloc->list, &arena->entries);
   alloc->freelist.prev = NULL;
   alloc->freelist.next = ptr;
   alloc->arena = arena;
   alloc->offset = 0;
   alloc->size = size;

   if (hole != alloc) {
      hole->arena = arena;
      hole->offset = size;
      hole->size = arena_size - size;

      list_addtail(&hole->list, &arena->entries);
      add_hole(device, hole);
   }

   list_addtail(&arena->list, &device->shader_arenas);

   mtx_unlock(&device->shader_arena_mutex);
   return alloc;

fail:
   mtx_unlock(&device->shader_arena_mutex);
   free(alloc);
   free(hole);
   if (arena && arena->bo)
      device->ws->buffer_destroy(device->ws, arena->bo);
   free(arena);
   return NULL;
}

static union radv_shader_arena_block *
get_hole(struct radv_shader_arena *arena, struct list_head *head)
{
   if (head == &arena->entries)
      return NULL;

   union radv_shader_arena_block *hole = LIST_ENTRY(union radv_shader_arena_block, head, list);
   return hole->freelist.prev ? hole : NULL;
}

static void
free_shader_memory(struct radv_device *device, union radv_shader_arena_block *alloc)
{
   mtx_lock(&device->shader_arena_mutex);

   union radv_shader_arena_block *hole_prev = get_hole(alloc->arena, alloc->list.prev);
   union radv_shader_arena_block *hole_next = get_hole(alloc->arena, alloc->list.next);

   union radv_shader_arena_block *hole = alloc;

   /* merge with previous hole */
   if (hole_prev) {
      remove_hole(device, hole_prev);

      hole_prev->size += hole->size;
      list_del(&hole->list);
      free_block_obj(device, hole);

      hole = hole_prev;
   }

   /* merge with next hole */
   if (hole_next) {
      remove_hole(device, hole_next);

      hole_next->offset -= hole->size;
      hole_next->size += hole->size;
      list_del(&hole->list);
      free_block_obj(device, hole);

      hole = hole_next;
   }

   if (list_is_singular(&hole->list)) {
      struct radv_shader_arena *arena = hole->arena;
      free_block_obj(device, hole);

      device->ws->buffer_destroy(device->ws, arena->bo);
      list_del(&arena->list);
      free(arena);
   } else {
      add_hole(device, hole);
   }

   mtx_unlock(&device->shader_arena_mutex);
}

static void *
radv_alloc_shader_memory(struct radv_device *device, struct radv_shader_variant *shader)
{
   shader->alloc = alloc_shader_memory(device, shader->code_size, shader);
   if (!shader->alloc)
      return NULL;
   shader->bo = shader->alloc->arena->bo;
   return shader->alloc->arena->ptr + shader->alloc->offset;
}

void
radv_init_shader_arenas(struct radv_device *device)
{
   mtx_init(&device->shader_arena_mutex, mtx_plain);

   device->shader_free_list_mask = 0;

   list_inithead(&device->shader_arenas);
   list_inithead(&device->shader_block_obj_pool);
   for (unsigned i = 0; i < RADV_SHADER_ALLOC_NUM_FREE_LISTS; i++)
      list_inithead(&device->shader_free_lists[i]);
}

void
radv_destroy_shader_arenas(struct radv_device *device)
{
   list_for_each_entry_safe(union radv_shader_arena_block, block, &device->shader_block_obj_pool,
                            pool) free(block);

   list_for_each_entry_safe(struct radv_shader_arena, arena, &device->shader_arenas, list)
   {
      device->ws->buffer_destroy(device->ws, arena->bo);
      free(arena);
   }
   mtx_destroy(&device->shader_arena_mutex);
}

/* For the UMR disassembler. */
#define DEBUGGER_END_OF_CODE_MARKER 0xbf9f0000 /* invalid instruction */
#define DEBUGGER_NUM_MARKERS        5

static unsigned
radv_get_shader_binary_size(size_t code_size)
{
   return code_size + DEBUGGER_NUM_MARKERS * 4;
}

static bool
radv_should_use_wgp_mode(const struct radv_device *device, gl_shader_stage stage,
                         const struct radv_shader_info *info)
{
   enum chip_class chip = device->physical_device->rad_info.chip_class;
   switch (stage) {
   case MESA_SHADER_COMPUTE:
   case MESA_SHADER_TESS_CTRL:
      return chip >= GFX10;
   case MESA_SHADER_GEOMETRY:
      return chip == GFX10 || (chip >= GFX10_3 && !info->is_ngg);
   case MESA_SHADER_VERTEX:
   case MESA_SHADER_TESS_EVAL:
      return chip == GFX10 && info->is_ngg;
   default:
      return false;
   }
}

static void
radv_postprocess_config(const struct radv_device *device, const struct ac_shader_config *config_in,
                        const struct radv_shader_info *info, gl_shader_stage stage,
                        struct ac_shader_config *config_out)
{
   const struct radv_physical_device *pdevice = device->physical_device;
   bool scratch_enabled = config_in->scratch_bytes_per_wave > 0;
   bool trap_enabled = !!device->trap_handler_shader;
   unsigned vgpr_comp_cnt = 0;
   unsigned num_input_vgprs = info->num_input_vgprs;

   if (stage == MESA_SHADER_FRAGMENT) {
      num_input_vgprs = ac_get_fs_input_vgpr_cnt(config_in, NULL, NULL);
   }

   unsigned num_vgprs = MAX2(config_in->num_vgprs, num_input_vgprs);
   /* +3 for scratch wave offset and VCC */
   unsigned num_sgprs = MAX2(config_in->num_sgprs, info->num_input_sgprs + 3);
   unsigned num_shared_vgprs = config_in->num_shared_vgprs;
   /* shared VGPRs are introduced in Navi and are allocated in blocks of 8 (RDNA ref 3.6.5) */
   assert((pdevice->rad_info.chip_class >= GFX10 && num_shared_vgprs % 8 == 0) ||
          (pdevice->rad_info.chip_class < GFX10 && num_shared_vgprs == 0));
   unsigned num_shared_vgpr_blocks = num_shared_vgprs / 8;
   unsigned excp_en = 0;

   *config_out = *config_in;
   config_out->num_vgprs = num_vgprs;
   config_out->num_sgprs = num_sgprs;
   config_out->num_shared_vgprs = num_shared_vgprs;

   config_out->rsrc2 = S_00B12C_USER_SGPR(info->num_user_sgprs) |
                       S_00B12C_SCRATCH_EN(scratch_enabled) | S_00B12C_TRAP_PRESENT(trap_enabled);

   if (trap_enabled) {
      /* Configure the shader exceptions like memory violation, etc.
       * TODO: Enable (and validate) more exceptions.
       */
      excp_en = 1 << 8; /* mem_viol */
   }

   if (!pdevice->use_ngg_streamout) {
      config_out->rsrc2 |=
         S_00B12C_SO_BASE0_EN(!!info->so.strides[0]) | S_00B12C_SO_BASE1_EN(!!info->so.strides[1]) |
         S_00B12C_SO_BASE2_EN(!!info->so.strides[2]) | S_00B12C_SO_BASE3_EN(!!info->so.strides[3]) |
         S_00B12C_SO_EN(!!info->so.num_outputs);
   }

   config_out->rsrc1 = S_00B848_VGPRS((num_vgprs - 1) / (info->wave_size == 32 ? 8 : 4)) |
                       S_00B848_DX10_CLAMP(1) | S_00B848_FLOAT_MODE(config_out->float_mode);

   if (pdevice->rad_info.chip_class >= GFX10) {
      config_out->rsrc2 |= S_00B22C_USER_SGPR_MSB_GFX10(info->num_user_sgprs >> 5);
   } else {
      config_out->rsrc1 |= S_00B228_SGPRS((num_sgprs - 1) / 8);
      config_out->rsrc2 |= S_00B22C_USER_SGPR_MSB_GFX9(info->num_user_sgprs >> 5);
   }

   bool wgp_mode = radv_should_use_wgp_mode(device, stage, info);

   switch (stage) {
   case MESA_SHADER_TESS_EVAL:
      if (info->is_ngg) {
         config_out->rsrc1 |= S_00B228_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10);
         config_out->rsrc2 |= S_00B22C_OC_LDS_EN(1) | S_00B22C_EXCP_EN(excp_en);
      } else if (info->tes.as_es) {
         assert(pdevice->rad_info.chip_class <= GFX8);
         vgpr_comp_cnt = info->uses_prim_id ? 3 : 2;

         config_out->rsrc2 |= S_00B12C_OC_LDS_EN(1) | S_00B12C_EXCP_EN(excp_en);
      } else {
         bool enable_prim_id = info->tes.outinfo.export_prim_id || info->uses_prim_id;
         vgpr_comp_cnt = enable_prim_id ? 3 : 2;

         config_out->rsrc1 |= S_00B128_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10);
         config_out->rsrc2 |= S_00B12C_OC_LDS_EN(1) | S_00B12C_EXCP_EN(excp_en);
      }
      config_out->rsrc2 |= S_00B22C_SHARED_VGPR_CNT(num_shared_vgpr_blocks);
      break;
   case MESA_SHADER_TESS_CTRL:
      if (pdevice->rad_info.chip_class >= GFX9) {
         /* We need at least 2 components for LS.
          * VGPR0-3: (VertexID, RelAutoindex, InstanceID / StepRate0, InstanceID).
          * StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
          */
         if (pdevice->rad_info.chip_class >= GFX10) {
            vgpr_comp_cnt = info->vs.needs_instance_id ? 3 : 1;
            config_out->rsrc2 |=
               S_00B42C_LDS_SIZE_GFX10(info->tcs.num_lds_blocks) | S_00B42C_EXCP_EN_GFX6(excp_en);
         } else {
            vgpr_comp_cnt = info->vs.needs_instance_id ? 2 : 1;
            config_out->rsrc2 |=
               S_00B42C_LDS_SIZE_GFX9(info->tcs.num_lds_blocks) | S_00B42C_EXCP_EN_GFX9(excp_en);
         }
      } else {
         config_out->rsrc2 |= S_00B12C_OC_LDS_EN(1) | S_00B12C_EXCP_EN(excp_en);
      }
      config_out->rsrc1 |=
         S_00B428_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10) | S_00B428_WGP_MODE(wgp_mode);
      config_out->rsrc2 |= S_00B42C_SHARED_VGPR_CNT(num_shared_vgpr_blocks);
      break;
   case MESA_SHADER_VERTEX:
      if (info->is_ngg) {
         config_out->rsrc1 |= S_00B228_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10);
      } else if (info->vs.as_ls) {
         assert(pdevice->rad_info.chip_class <= GFX8);
         /* We need at least 2 components for LS.
          * VGPR0-3: (VertexID, RelAutoindex, InstanceID / StepRate0, InstanceID).
          * StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
          */
         vgpr_comp_cnt = info->vs.needs_instance_id ? 2 : 1;
      } else if (info->vs.as_es) {
         assert(pdevice->rad_info.chip_class <= GFX8);
         /* VGPR0-3: (VertexID, InstanceID / StepRate0, ...) */
         vgpr_comp_cnt = info->vs.needs_instance_id ? 1 : 0;
      } else {
         /* VGPR0-3: (VertexID, InstanceID / StepRate0, PrimID, InstanceID)
          * If PrimID is disabled. InstanceID / StepRate1 is loaded instead.
          * StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
          */
         if (info->vs.needs_instance_id && pdevice->rad_info.chip_class >= GFX10) {
            vgpr_comp_cnt = 3;
         } else if (info->vs.outinfo.export_prim_id) {
            vgpr_comp_cnt = 2;
         } else if (info->vs.needs_instance_id) {
            vgpr_comp_cnt = 1;
         } else {
            vgpr_comp_cnt = 0;
         }

         config_out->rsrc1 |= S_00B128_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10);
      }
      config_out->rsrc2 |=
         S_00B12C_SHARED_VGPR_CNT(num_shared_vgpr_blocks) | S_00B12C_EXCP_EN(excp_en);
      break;
   case MESA_SHADER_FRAGMENT:
      config_out->rsrc1 |= S_00B028_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10);
      config_out->rsrc2 |= S_00B02C_SHARED_VGPR_CNT(num_shared_vgpr_blocks) |
                           S_00B02C_EXCP_EN(excp_en);
      break;
   case MESA_SHADER_GEOMETRY:
      config_out->rsrc1 |= S_00B228_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10);
      config_out->rsrc2 |=
         S_00B22C_SHARED_VGPR_CNT(num_shared_vgpr_blocks) | S_00B22C_EXCP_EN(excp_en);
      break;
   case MESA_SHADER_COMPUTE:
      config_out->rsrc1 |=
         S_00B848_MEM_ORDERED(pdevice->rad_info.chip_class >= GFX10) | S_00B848_WGP_MODE(wgp_mode);
      config_out->rsrc2 |= S_00B84C_TGID_X_EN(info->cs.uses_block_id[0]) |
                           S_00B84C_TGID_Y_EN(info->cs.uses_block_id[1]) |
                           S_00B84C_TGID_Z_EN(info->cs.uses_block_id[2]) |
                           S_00B84C_TIDIG_COMP_CNT(info->cs.uses_thread_id[2]   ? 2
                                                   : info->cs.uses_thread_id[1] ? 1
                                                                                : 0) |
                           S_00B84C_TG_SIZE_EN(info->cs.uses_local_invocation_idx) |
                           S_00B84C_LDS_SIZE(config_in->lds_size) | S_00B84C_EXCP_EN(excp_en);
      config_out->rsrc3 |= S_00B8A0_SHARED_VGPR_CNT(num_shared_vgpr_blocks);

      break;
   default:
      unreachable("unsupported shader type");
      break;
   }

   if (pdevice->rad_info.chip_class >= GFX10 && info->is_ngg &&
       (stage == MESA_SHADER_VERTEX || stage == MESA_SHADER_TESS_EVAL ||
        stage == MESA_SHADER_GEOMETRY)) {
      unsigned gs_vgpr_comp_cnt, es_vgpr_comp_cnt;
      gl_shader_stage es_stage = stage;
      if (stage == MESA_SHADER_GEOMETRY)
         es_stage = info->gs.es_type;

      /* VGPR5-8: (VertexID, UserVGPR0, UserVGPR1, UserVGPR2 / InstanceID) */
      if (es_stage == MESA_SHADER_VERTEX) {
         es_vgpr_comp_cnt = info->vs.needs_instance_id ? 3 : 0;
      } else if (es_stage == MESA_SHADER_TESS_EVAL) {
         bool enable_prim_id = info->tes.outinfo.export_prim_id || info->uses_prim_id;
         es_vgpr_comp_cnt = enable_prim_id ? 3 : 2;
      } else
         unreachable("Unexpected ES shader stage");

      bool nggc = info->has_ngg_culling; /* Culling uses GS vertex offsets 0, 1, 2. */
      bool tes_triangles =
         stage == MESA_SHADER_TESS_EVAL && info->tes.primitive_mode >= 4; /* GL_TRIANGLES */
      if (info->uses_invocation_id) {
         gs_vgpr_comp_cnt = 3; /* VGPR3 contains InvocationID. */
      } else if (info->uses_prim_id || (es_stage == MESA_SHADER_VERTEX &&
                                        info->vs.outinfo.export_prim_id)) {
         gs_vgpr_comp_cnt = 2; /* VGPR2 contains PrimitiveID. */
      } else if (info->gs.vertices_in >= 3 || tes_triangles || nggc) {
         gs_vgpr_comp_cnt = 1; /* VGPR1 contains offsets 2, 3 */
      } else {
         gs_vgpr_comp_cnt = 0; /* VGPR0 contains offsets 0, 1 */
      }

      /* Disable the WGP mode on gfx10.3 because it can hang. (it
       * happened on VanGogh) Let's disable it on all chips that
       * disable exactly 1 CU per SA for GS.
       */
      config_out->rsrc1 |=
         S_00B228_GS_VGPR_COMP_CNT(gs_vgpr_comp_cnt) | S_00B228_WGP_MODE(wgp_mode);
      config_out->rsrc2 |= S_00B22C_ES_VGPR_COMP_CNT(es_vgpr_comp_cnt) |
                           S_00B22C_LDS_SIZE(config_in->lds_size) |
                           S_00B22C_OC_LDS_EN(es_stage == MESA_SHADER_TESS_EVAL);
   } else if (pdevice->rad_info.chip_class >= GFX9 && stage == MESA_SHADER_GEOMETRY) {
      unsigned es_type = info->gs.es_type;
      unsigned gs_vgpr_comp_cnt, es_vgpr_comp_cnt;

      if (es_type == MESA_SHADER_VERTEX) {
         /* VGPR0-3: (VertexID, InstanceID / StepRate0, ...) */
         if (info->vs.needs_instance_id) {
            es_vgpr_comp_cnt = pdevice->rad_info.chip_class >= GFX10 ? 3 : 1;
         } else {
            es_vgpr_comp_cnt = 0;
         }
      } else if (es_type == MESA_SHADER_TESS_EVAL) {
         es_vgpr_comp_cnt = info->uses_prim_id ? 3 : 2;
      } else {
         unreachable("invalid shader ES type");
      }

      /* If offsets 4, 5 are used, GS_VGPR_COMP_CNT is ignored and
       * VGPR[0:4] are always loaded.
       */
      if (info->uses_invocation_id) {
         gs_vgpr_comp_cnt = 3; /* VGPR3 contains InvocationID. */
      } else if (info->uses_prim_id) {
         gs_vgpr_comp_cnt = 2; /* VGPR2 contains PrimitiveID. */
      } else if (info->gs.vertices_in >= 3) {
         gs_vgpr_comp_cnt = 1; /* VGPR1 contains offsets 2, 3 */
      } else {
         gs_vgpr_comp_cnt = 0; /* VGPR0 contains offsets 0, 1 */
      }

      config_out->rsrc1 |=
         S_00B228_GS_VGPR_COMP_CNT(gs_vgpr_comp_cnt) | S_00B228_WGP_MODE(wgp_mode);
      config_out->rsrc2 |= S_00B22C_ES_VGPR_COMP_CNT(es_vgpr_comp_cnt) |
                           S_00B22C_OC_LDS_EN(es_type == MESA_SHADER_TESS_EVAL);
   } else if (pdevice->rad_info.chip_class >= GFX9 && stage == MESA_SHADER_TESS_CTRL) {
      config_out->rsrc1 |= S_00B428_LS_VGPR_COMP_CNT(vgpr_comp_cnt);
   } else {
      config_out->rsrc1 |= S_00B128_VGPR_COMP_CNT(vgpr_comp_cnt);
   }
}

struct radv_shader_variant *
radv_shader_variant_create(struct radv_device *device, const struct radv_shader_binary *binary,
                           bool keep_shader_info, bool from_cache)
{
   struct ac_shader_config config = {0};
   struct ac_rtld_binary rtld_binary = {0};
   struct radv_shader_variant *variant = calloc(1, sizeof(struct radv_shader_variant));
   if (!variant)
      return NULL;

   variant->ref_count = 1;

   if (binary->type == RADV_BINARY_TYPE_RTLD) {
      struct ac_rtld_symbol lds_symbols[2];
      unsigned num_lds_symbols = 0;
      const char *elf_data = (const char *)((struct radv_shader_binary_rtld *)binary)->data;
      size_t elf_size = ((struct radv_shader_binary_rtld *)binary)->elf_size;

      if (device->physical_device->rad_info.chip_class >= GFX9 &&
          (binary->stage == MESA_SHADER_GEOMETRY || binary->info.is_ngg) &&
          !binary->is_gs_copy_shader) {
         struct ac_rtld_symbol *sym = &lds_symbols[num_lds_symbols++];
         sym->name = "esgs_ring";
         sym->size = binary->info.ngg_info.esgs_ring_size;
         sym->align = 64 * 1024;
      }

      if (binary->info.is_ngg && binary->stage == MESA_SHADER_GEOMETRY) {
         struct ac_rtld_symbol *sym = &lds_symbols[num_lds_symbols++];
         sym->name = "ngg_emit";
         sym->size = binary->info.ngg_info.ngg_emit_size * 4;
         sym->align = 4;
      }

      struct ac_rtld_open_info open_info = {
         .info = &device->physical_device->rad_info,
         .shader_type = binary->stage,
         .wave_size = binary->info.wave_size,
         .num_parts = 1,
         .elf_ptrs = &elf_data,
         .elf_sizes = &elf_size,
         .num_shared_lds_symbols = num_lds_symbols,
         .shared_lds_symbols = lds_symbols,
      };

      if (!ac_rtld_open(&rtld_binary, open_info)) {
         free(variant);
         return NULL;
      }

      if (!ac_rtld_read_config(&device->physical_device->rad_info, &rtld_binary, &config)) {
         ac_rtld_close(&rtld_binary);
         free(variant);
         return NULL;
      }

      if (rtld_binary.lds_size > 0) {
         unsigned encode_granularity = device->physical_device->rad_info.lds_encode_granularity;
         config.lds_size = align(rtld_binary.lds_size, encode_granularity) / encode_granularity;
      }
      if (!config.lds_size && binary->stage == MESA_SHADER_TESS_CTRL) {
         /* This is used for reporting LDS statistics */
         config.lds_size = binary->info.tcs.num_lds_blocks;
      }

      variant->code_size = rtld_binary.rx_size;
      variant->exec_size = rtld_binary.exec_size;
   } else {
      assert(binary->type == RADV_BINARY_TYPE_LEGACY);
      config = ((struct radv_shader_binary_legacy *)binary)->base.config;
      variant->code_size =
         radv_get_shader_binary_size(((struct radv_shader_binary_legacy *)binary)->code_size);
      variant->exec_size = ((struct radv_shader_binary_legacy *)binary)->exec_size;
   }

   variant->info = binary->info;

   if (from_cache) {
      /* Copy the shader binary configuration from the cache. */
      memcpy(&variant->config, &binary->config, sizeof(variant->config));
   } else {
      radv_postprocess_config(device, &config, &binary->info, binary->stage, &variant->config);
   }

   void *dest_ptr = radv_alloc_shader_memory(device, variant);
   if (!dest_ptr) {
      if (binary->type == RADV_BINARY_TYPE_RTLD)
         ac_rtld_close(&rtld_binary);
      free(variant);
      return NULL;
   }

   if (binary->type == RADV_BINARY_TYPE_RTLD) {
      struct radv_shader_binary_rtld *bin = (struct radv_shader_binary_rtld *)binary;
      struct ac_rtld_upload_info info = {
         .binary = &rtld_binary,
         .rx_va = radv_shader_variant_get_va(variant),
         .rx_ptr = dest_ptr,
      };

      if (!ac_rtld_upload(&info)) {
         radv_shader_variant_destroy(device, variant);
         ac_rtld_close(&rtld_binary);
         return NULL;
      }

      if (keep_shader_info || (device->instance->debug_flags & RADV_DEBUG_DUMP_SHADERS)) {
         const char *disasm_data;
         size_t disasm_size;
         if (!ac_rtld_get_section_by_name(&rtld_binary, ".AMDGPU.disasm", &disasm_data,
                                          &disasm_size)) {
            radv_shader_variant_destroy(device, variant);
            ac_rtld_close(&rtld_binary);
            return NULL;
         }

         variant->ir_string =
            bin->llvm_ir_size ? strdup((const char *)(bin->data + bin->elf_size)) : NULL;
         variant->disasm_string = malloc(disasm_size + 1);
         memcpy(variant->disasm_string, disasm_data, disasm_size);
         variant->disasm_string[disasm_size] = 0;
      }

      variant->code_ptr = dest_ptr;
      ac_rtld_close(&rtld_binary);
   } else {
      struct radv_shader_binary_legacy *bin = (struct radv_shader_binary_legacy *)binary;
      memcpy(dest_ptr, bin->data + bin->stats_size, bin->code_size);

      /* Add end-of-code markers for the UMR disassembler. */
      uint32_t *ptr32 = (uint32_t *)dest_ptr + bin->code_size / 4;
      for (unsigned i = 0; i < DEBUGGER_NUM_MARKERS; i++)
         ptr32[i] = DEBUGGER_END_OF_CODE_MARKER;

      variant->code_ptr = dest_ptr;
      variant->ir_string =
         bin->ir_size ? strdup((const char *)(bin->data + bin->stats_size + bin->code_size)) : NULL;
      variant->disasm_string =
         bin->disasm_size
            ? strdup((const char *)(bin->data + bin->stats_size + bin->code_size + bin->ir_size))
            : NULL;

      if (bin->stats_size) {
         variant->statistics = calloc(bin->stats_size, 1);
         memcpy(variant->statistics, bin->data, bin->stats_size);
      }
   }
   return variant;
}

static char *
radv_dump_nir_shaders(struct nir_shader *const *shaders, int shader_count)
{
   char *data = NULL;
   char *ret = NULL;
   size_t size = 0;
   struct u_memstream mem;
   if (u_memstream_open(&mem, &data, &size)) {
      FILE *const memf = u_memstream_get(&mem);
      for (int i = 0; i < shader_count; ++i)
         nir_print_shader(shaders[i], memf);
      u_memstream_close(&mem);
   }

   ret = malloc(size + 1);
   if (ret) {
      memcpy(ret, data, size);
      ret[size] = 0;
   }
   free(data);
   return ret;
}

static struct radv_shader_variant *
shader_variant_compile(struct radv_device *device, struct vk_shader_module *module,
                       struct nir_shader *const *shaders, int shader_count, gl_shader_stage stage,
                       struct radv_shader_info *info, struct radv_nir_compiler_options *options,
                       bool gs_copy_shader, bool trap_handler_shader, bool keep_shader_info,
                       bool keep_statistic_info, struct radv_shader_binary **binary_out)
{
   enum radeon_family chip_family = device->physical_device->rad_info.family;
   struct radv_shader_binary *binary = NULL;

   struct radv_shader_debug_data debug_data = {
      .device = device,
      .module = module,
   };

   options->family = chip_family;
   options->chip_class = device->physical_device->rad_info.chip_class;
   options->info = &device->physical_device->rad_info;
   options->dump_shader = radv_can_dump_shader(device, module, gs_copy_shader || trap_handler_shader);
   options->dump_preoptir =
      options->dump_shader && device->instance->debug_flags & RADV_DEBUG_PREOPTIR;
   options->record_ir = keep_shader_info;
   options->record_stats = keep_statistic_info;
   options->check_ir = device->instance->debug_flags & RADV_DEBUG_CHECKIR;
   options->address32_hi = device->physical_device->rad_info.address32_hi;
   options->has_ls_vgpr_init_bug = device->physical_device->rad_info.has_ls_vgpr_init_bug;
   options->enable_mrt_output_nan_fixup =
      module && !module->nir && options->key.ps.enable_mrt_output_nan_fixup;
   options->adjust_frag_coord_z = device->adjust_frag_coord_z;
   options->has_image_load_dcc_bug = device->physical_device->rad_info.has_image_load_dcc_bug;
   options->debug.func = radv_compiler_debug;
   options->debug.private_data = &debug_data;

   switch (options->key.ps.force_vrs) {
   case RADV_FORCE_VRS_2x2:
      options->force_vrs_rates = (1u << 2) | (1u << 4);
      break;
   case RADV_FORCE_VRS_2x1:
      options->force_vrs_rates = (1u << 2) | (0u << 4);
      break;
   case RADV_FORCE_VRS_1x2:
      options->force_vrs_rates = (0u << 2) | (1u << 4);
      break;
   default:
      break;
   }

   struct radv_shader_args args = {0};
   args.options = options;
   args.shader_info = info;
   args.is_gs_copy_shader = gs_copy_shader;
   args.is_trap_handler_shader = trap_handler_shader;

   radv_declare_shader_args(
      &args, gs_copy_shader ? MESA_SHADER_VERTEX : shaders[shader_count - 1]->info.stage,
      shader_count >= 2,
      shader_count >= 2 ? shaders[shader_count - 2]->info.stage : MESA_SHADER_VERTEX);

#ifdef LLVM_AVAILABLE
   if (radv_use_llvm_for_stage(device, stage) || options->dump_shader || options->record_ir)
      ac_init_llvm_once();

   if (radv_use_llvm_for_stage(device, stage)) {
      llvm_compile_shader(device, shader_count, shaders, &binary, &args);
#else
   if (false) {
#endif
   } else {
      aco_compile_shader(shader_count, shaders, &binary, &args);
   }

   binary->info = *info;

   struct radv_shader_variant *variant =
      radv_shader_variant_create(device, binary, keep_shader_info, false);
   if (!variant) {
      free(binary);
      return NULL;
   }

   if (options->dump_shader) {
      fprintf(stderr, "%s", radv_get_shader_name(info, shaders[0]->info.stage));
      for (int i = 1; i < shader_count; ++i)
         fprintf(stderr, " + %s", radv_get_shader_name(info, shaders[i]->info.stage));

      fprintf(stderr, "\ndisasm:\n%s\n", variant->disasm_string);
   }

   if (keep_shader_info) {
      variant->nir_string = radv_dump_nir_shaders(shaders, shader_count);
      if (!gs_copy_shader && !trap_handler_shader && !module->nir) {
         variant->spirv = malloc(module->size);
         if (!variant->spirv) {
            free(variant);
            free(binary);
            return NULL;
         }

         memcpy(variant->spirv, module->data, module->size);
         variant->spirv_size = module->size;
      }
   }

   /* Copy the shader binary configuration to store it in the cache. */
   memcpy(&binary->config, &variant->config, sizeof(binary->config));

   if (binary_out)
      *binary_out = binary;
   else
      free(binary);

   return variant;
}

struct radv_shader_variant *
radv_shader_variant_compile(struct radv_device *device, struct vk_shader_module *module,
                            struct nir_shader *const *shaders, int shader_count,
                            struct radv_pipeline_layout *layout,
                            const struct radv_pipeline_key *key,
                            struct radv_shader_info *info, bool keep_shader_info,
                            bool keep_statistic_info,
                            struct radv_shader_binary **binary_out)
{
   gl_shader_stage stage = shaders[shader_count - 1]->info.stage;
   struct radv_nir_compiler_options options = {0};

   options.layout = layout;
   if (key)
      options.key = *key;

   options.explicit_scratch_args = !radv_use_llvm_for_stage(device, stage);
   options.remap_spi_ps_input = !radv_use_llvm_for_stage(device, stage);
   options.robust_buffer_access = device->robust_buffer_access;
   options.wgp_mode = radv_should_use_wgp_mode(device, stage, info);

   return shader_variant_compile(device, module, shaders, shader_count, stage, info, &options,
                                 false, false, keep_shader_info, keep_statistic_info, binary_out);
}

struct radv_shader_variant *
radv_create_gs_copy_shader(struct radv_device *device, struct nir_shader *shader,
                           struct radv_shader_info *info, struct radv_shader_binary **binary_out,
                           bool keep_shader_info, bool keep_statistic_info, bool multiview,
                           bool disable_optimizations)
{
   struct radv_nir_compiler_options options = {0};
   gl_shader_stage stage = MESA_SHADER_VERTEX;

   options.explicit_scratch_args = !radv_use_llvm_for_stage(device, stage);
   options.remap_spi_ps_input = !radv_use_llvm_for_stage(device, stage);
   options.key.has_multiview_view_index = multiview;
   options.key.optimisations_disabled = disable_optimizations;

   return shader_variant_compile(device, NULL, &shader, 1, stage, info, &options, true, false,
                                 keep_shader_info, keep_statistic_info, binary_out);
}

struct radv_shader_variant *
radv_create_trap_handler_shader(struct radv_device *device)
{
   struct radv_nir_compiler_options options = {0};
   struct radv_shader_variant *shader = NULL;
   struct radv_shader_binary *binary = NULL;
   struct radv_shader_info info = {0};

   nir_builder b = nir_builder_init_simple_shader(MESA_SHADER_COMPUTE, NULL, "meta_trap_handler");

   options.explicit_scratch_args = true;
   options.wgp_mode = radv_should_use_wgp_mode(device, MESA_SHADER_COMPUTE, &info);
   info.wave_size = 64;

   shader = shader_variant_compile(device, NULL, &b.shader, 1, MESA_SHADER_COMPUTE, &info, &options,
                                   false, true, true, false, &binary);

   ralloc_free(b.shader);
   free(binary);

   return shader;
}

static struct radv_shader_prolog *
upload_vs_prolog(struct radv_device *device, struct radv_prolog_binary *bin, unsigned wave_size)
{
   struct radv_shader_prolog *prolog = malloc(sizeof(struct radv_shader_prolog));
   if (!prolog)
      return NULL;

   prolog->alloc = alloc_shader_memory(device, bin->code_size, NULL);
   if (!prolog->alloc) {
      free(prolog);
      return NULL;
   }

   prolog->bo = prolog->alloc->arena->bo;
   char *dest_ptr = prolog->alloc->arena->ptr + prolog->alloc->offset;

   memcpy(dest_ptr, bin->data, bin->code_size);

   prolog->rsrc1 = S_00B848_VGPRS((bin->num_vgprs - 1) / (wave_size == 32 ? 8 : 4)) |
                   S_00B228_SGPRS((bin->num_sgprs - 1) / 8);
   prolog->num_preserved_sgprs = bin->num_preserved_sgprs;

   return prolog;
}

struct radv_shader_prolog *
radv_create_vs_prolog(struct radv_device *device, const struct radv_vs_prolog_key *key)
{
   struct radv_nir_compiler_options options = {0};
   options.explicit_scratch_args = true;
   options.family = device->physical_device->rad_info.family;
   options.chip_class = device->physical_device->rad_info.chip_class;
   options.info = &device->physical_device->rad_info;
   options.address32_hi = device->physical_device->rad_info.address32_hi;
   options.dump_shader = device->instance->debug_flags & RADV_DEBUG_DUMP_PROLOGS;

   struct radv_shader_info info = {0};
   info.wave_size = key->wave32 ? 32 : 64;
   info.vs.needs_instance_id = true;
   info.vs.needs_base_instance = true;
   info.vs.needs_draw_id = true;
   info.vs.use_per_attribute_vb_descs = true;
   info.vs.vb_desc_usage_mask = BITFIELD_MASK(key->num_attributes);
   info.vs.has_prolog = true;
   info.vs.as_ls = key->as_ls;
   info.is_ngg = key->is_ngg;

   struct radv_shader_args args = {0};
   args.options = &options;
   args.shader_info = &info;
   radv_declare_shader_args(&args, key->next_stage, key->next_stage != MESA_SHADER_VERTEX,
                            MESA_SHADER_VERTEX);

#ifdef LLVM_AVAILABLE
   if (options.dump_shader)
      ac_init_llvm_once();
#endif

   struct radv_prolog_binary *binary = NULL;
   aco_compile_vs_prolog(key, &binary, &args);
   struct radv_shader_prolog *prolog = upload_vs_prolog(device, binary, info.wave_size);
   if (prolog) {
      prolog->nontrivial_divisors = key->state->nontrivial_divisors;
   }
   free(binary);

   return prolog;
}

void
radv_shader_variant_destroy(struct radv_device *device, struct radv_shader_variant *variant)
{
   if (!p_atomic_dec_zero(&variant->ref_count))
      return;

   free_shader_memory(device, variant->alloc);

   free(variant->spirv);
   free(variant->nir_string);
   free(variant->disasm_string);
   free(variant->ir_string);
   free(variant->statistics);
   free(variant);
}

void
radv_prolog_destroy(struct radv_device *device, struct radv_shader_prolog *prolog)
{
   if (!prolog)
      return;

   free_shader_memory(device, prolog->alloc);
   free(prolog);
}

uint64_t
radv_shader_variant_get_va(const struct radv_shader_variant *variant)
{
   return radv_buffer_get_va(variant->bo) + variant->alloc->offset;
}

struct radv_shader_variant *
radv_find_shader_variant(struct radv_device *device, uint64_t pc)
{
   mtx_lock(&device->shader_arena_mutex);
   list_for_each_entry(struct radv_shader_arena, arena, &device->shader_arenas, list)
   {
#ifdef __GNUC__
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wshadow"
#endif
      list_for_each_entry(union radv_shader_arena_block, block, &arena->entries, list)
      {
#ifdef __GNUC__
#pragma GCC diagnostic pop
#endif
         uint64_t start = radv_buffer_get_va(block->arena->bo) + block->offset;
         if (!block->freelist.prev && pc >= start && pc < start + block->size) {
            mtx_unlock(&device->shader_arena_mutex);
            return (struct radv_shader_variant *)block->freelist.next;
         }
      }
   }

   mtx_unlock(&device->shader_arena_mutex);
   return NULL;
}

const char *
radv_get_shader_name(struct radv_shader_info *info, gl_shader_stage stage)
{
   switch (stage) {
   case MESA_SHADER_VERTEX:
      if (info->vs.as_ls)
         return "Vertex Shader as LS";
      else if (info->vs.as_es)
         return "Vertex Shader as ES";
      else if (info->is_ngg)
         return "Vertex Shader as ESGS";
      else
         return "Vertex Shader as VS";
   case MESA_SHADER_TESS_CTRL:
      return "Tessellation Control Shader";
   case MESA_SHADER_TESS_EVAL:
      if (info->tes.as_es)
         return "Tessellation Evaluation Shader as ES";
      else if (info->is_ngg)
         return "Tessellation Evaluation Shader as ESGS";
      else
         return "Tessellation Evaluation Shader as VS";
   case MESA_SHADER_GEOMETRY:
      return "Geometry Shader";
   case MESA_SHADER_FRAGMENT:
      return "Pixel Shader";
   case MESA_SHADER_COMPUTE:
      return "Compute Shader";
   default:
      return "Unknown shader";
   };
}

unsigned
radv_get_max_waves(const struct radv_device *device, struct radv_shader_variant *variant,
                   gl_shader_stage stage)
{
   struct radeon_info *info = &device->physical_device->rad_info;
   enum chip_class chip_class = info->chip_class;
   uint8_t wave_size = variant->info.wave_size;
   struct ac_shader_config *conf = &variant->config;
   unsigned max_simd_waves;
   unsigned lds_per_wave = 0;

   max_simd_waves = info->max_wave64_per_simd * (64 / wave_size);

   if (stage == MESA_SHADER_FRAGMENT) {
      lds_per_wave =
         conf->lds_size * info->lds_encode_granularity + variant->info.ps.num_interp * 48;
      lds_per_wave = align(lds_per_wave, info->lds_alloc_granularity);
   } else if (stage == MESA_SHADER_COMPUTE) {
      unsigned max_workgroup_size = variant->info.workgroup_size;
      lds_per_wave =
         align(conf->lds_size * info->lds_encode_granularity, info->lds_alloc_granularity);
      lds_per_wave /= DIV_ROUND_UP(max_workgroup_size, wave_size);
   }

   if (conf->num_sgprs && chip_class < GFX10) {
      unsigned sgprs = align(conf->num_sgprs, chip_class >= GFX8 ? 16 : 8);
      max_simd_waves = MIN2(max_simd_waves, info->num_physical_sgprs_per_simd / sgprs);
   }

   if (conf->num_vgprs) {
      unsigned physical_vgprs = info->num_physical_wave64_vgprs_per_simd * (64 / wave_size);
      unsigned vgprs = align(conf->num_vgprs, wave_size == 32 ? 8 : 4);
      if (chip_class >= GFX10_3)
         vgprs = align(vgprs, wave_size == 32 ? 16 : 8);
      max_simd_waves = MIN2(max_simd_waves, physical_vgprs / vgprs);
   }

   unsigned simd_per_workgroup = info->num_simd_per_compute_unit;
   if (chip_class >= GFX10)
      simd_per_workgroup *= 2; /* like lds_size_per_workgroup, assume WGP on GFX10+ */

   unsigned max_lds_per_simd = info->lds_size_per_workgroup / simd_per_workgroup;
   if (lds_per_wave)
      max_simd_waves = MIN2(max_simd_waves, DIV_ROUND_UP(max_lds_per_simd, lds_per_wave));

   return chip_class >= GFX10 ? max_simd_waves * (wave_size / 32) : max_simd_waves;
}

unsigned
radv_compute_spi_ps_input(const struct radv_device *device,
                          const struct radv_shader_info *info)
{
   unsigned spi_ps_input;

   spi_ps_input = S_0286CC_PERSP_CENTER_ENA(info->ps.reads_persp_center) |
                  S_0286CC_PERSP_CENTROID_ENA(info->ps.reads_persp_centroid) |
                  S_0286CC_PERSP_SAMPLE_ENA(info->ps.reads_persp_sample) |
                  S_0286CC_LINEAR_CENTER_ENA(info->ps.reads_linear_center) |
                  S_0286CC_LINEAR_CENTROID_ENA(info->ps.reads_linear_centroid) |
                  S_0286CC_LINEAR_SAMPLE_ENA(info->ps.reads_linear_sample)|
                  S_0286CC_PERSP_PULL_MODEL_ENA(info->ps.reads_barycentric_model) |
                  S_0286CC_FRONT_FACE_ENA(info->ps.reads_front_face);

   if (info->ps.reads_frag_coord_mask ||
       info->ps.reads_sample_pos_mask) {
      uint8_t mask = info->ps.reads_frag_coord_mask | info->ps.reads_sample_pos_mask;

      for (unsigned i = 0; i < 4; i++) {
         if (mask & (1 << i))
            spi_ps_input |= S_0286CC_POS_X_FLOAT_ENA(1) << i;
      }

      if (device->adjust_frag_coord_z && info->ps.reads_frag_coord_mask & (1 << 2)) {
         spi_ps_input |= S_0286CC_ANCILLARY_ENA(1);
      }
   }

   if (info->ps.reads_sample_id || info->ps.reads_frag_shading_rate || info->ps.reads_sample_mask_in) {
      spi_ps_input |= S_0286CC_ANCILLARY_ENA(1);
   }

   if (info->ps.reads_sample_mask_in) {
      spi_ps_input |= S_0286CC_SAMPLE_COVERAGE_ENA(1);
   }

   if (G_0286CC_POS_W_FLOAT_ENA(spi_ps_input)) {
      /* If POS_W_FLOAT (11) is enabled, at least one of PERSP_* must be enabled too */
      spi_ps_input |= S_0286CC_PERSP_CENTER_ENA(1);
   }

   if (!(spi_ps_input & 0x7F)) {
      /* At least one of PERSP_* (0xF) or LINEAR_* (0x70) must be enabled */
      spi_ps_input |= S_0286CC_PERSP_CENTER_ENA(1);
   }

   return spi_ps_input;
}

VkResult
radv_GetShaderInfoAMD(VkDevice _device, VkPipeline _pipeline, VkShaderStageFlagBits shaderStage,
                      VkShaderInfoTypeAMD infoType, size_t *pInfoSize, void *pInfo)
{
   RADV_FROM_HANDLE(radv_device, device, _device);
   RADV_FROM_HANDLE(radv_pipeline, pipeline, _pipeline);
   gl_shader_stage stage = vk_to_mesa_shader_stage(shaderStage);
   struct radv_shader_variant *variant = pipeline->shaders[stage];
   VkResult result = VK_SUCCESS;

   /* Spec doesn't indicate what to do if the stage is invalid, so just
    * return no info for this. */
   if (!variant)
      return vk_error(device, VK_ERROR_FEATURE_NOT_PRESENT);

   switch (infoType) {
   case VK_SHADER_INFO_TYPE_STATISTICS_AMD:
      if (!pInfo) {
         *pInfoSize = sizeof(VkShaderStatisticsInfoAMD);
      } else {
         unsigned lds_multiplier = device->physical_device->rad_info.lds_encode_granularity;
         struct ac_shader_config *conf = &variant->config;

         VkShaderStatisticsInfoAMD statistics = {0};
         statistics.shaderStageMask = shaderStage;
         statistics.numPhysicalVgprs =
            device->physical_device->rad_info.num_physical_wave64_vgprs_per_simd;
         statistics.numPhysicalSgprs =
            device->physical_device->rad_info.num_physical_sgprs_per_simd;
         statistics.numAvailableSgprs = statistics.numPhysicalSgprs;

         if (stage == MESA_SHADER_COMPUTE) {
            unsigned *local_size = variant->info.cs.block_size;
            unsigned workgroup_size = pipeline->shaders[MESA_SHADER_COMPUTE]->info.workgroup_size;

            statistics.numAvailableVgprs =
               statistics.numPhysicalVgprs /
               ceil((double)workgroup_size / statistics.numPhysicalVgprs);

            statistics.computeWorkGroupSize[0] = local_size[0];
            statistics.computeWorkGroupSize[1] = local_size[1];
            statistics.computeWorkGroupSize[2] = local_size[2];
         } else {
            statistics.numAvailableVgprs = statistics.numPhysicalVgprs;
         }

         statistics.resourceUsage.numUsedVgprs = conf->num_vgprs;
         statistics.resourceUsage.numUsedSgprs = conf->num_sgprs;
         statistics.resourceUsage.ldsSizePerLocalWorkGroup = 32768;
         statistics.resourceUsage.ldsUsageSizeInBytes = conf->lds_size * lds_multiplier;
         statistics.resourceUsage.scratchMemUsageInBytes = conf->scratch_bytes_per_wave;

         size_t size = *pInfoSize;
         *pInfoSize = sizeof(statistics);

         memcpy(pInfo, &statistics, MIN2(size, *pInfoSize));

         if (size < *pInfoSize)
            result = VK_INCOMPLETE;
      }

      break;
   case VK_SHADER_INFO_TYPE_DISASSEMBLY_AMD: {
      char *out;
      size_t outsize;
      struct u_memstream mem;
      u_memstream_open(&mem, &out, &outsize);
      FILE *const memf = u_memstream_get(&mem);

      fprintf(memf, "%s:\n", radv_get_shader_name(&variant->info, stage));
      fprintf(memf, "%s\n\n", variant->ir_string);
      if (variant->disasm_string) {
         fprintf(memf, "%s\n\n", variant->disasm_string);
      }
      radv_dump_shader_stats(device, pipeline, stage, memf);
      u_memstream_close(&mem);

      /* Need to include the null terminator. */
      size_t length = outsize + 1;

      if (!pInfo) {
         *pInfoSize = length;
      } else {
         size_t size = *pInfoSize;
         *pInfoSize = length;

         memcpy(pInfo, out, MIN2(size, length));

         if (size < length)
            result = VK_INCOMPLETE;
      }

      free(out);
      break;
   }
   default:
      /* VK_SHADER_INFO_TYPE_BINARY_AMD unimplemented for now. */
      result = VK_ERROR_FEATURE_NOT_PRESENT;
      break;
   }

   return result;
}

VkResult
radv_dump_shader_stats(struct radv_device *device, struct radv_pipeline *pipeline,
                       gl_shader_stage stage, FILE *output)
{
   struct radv_shader_variant *shader = pipeline->shaders[stage];
   VkPipelineExecutablePropertiesKHR *props = NULL;
   uint32_t prop_count = 0;
   VkResult result;

   VkPipelineInfoKHR pipeline_info = {0};
   pipeline_info.sType = VK_STRUCTURE_TYPE_PIPELINE_INFO_KHR;
   pipeline_info.pipeline = radv_pipeline_to_handle(pipeline);

   result = radv_GetPipelineExecutablePropertiesKHR(radv_device_to_handle(device), &pipeline_info,
                                                    &prop_count, NULL);
   if (result != VK_SUCCESS)
      return result;

   props = calloc(prop_count, sizeof(*props));
   if (!props)
      return VK_ERROR_OUT_OF_HOST_MEMORY;

   result = radv_GetPipelineExecutablePropertiesKHR(radv_device_to_handle(device), &pipeline_info,
                                                    &prop_count, props);
   if (result != VK_SUCCESS)
      goto fail;

   for (unsigned exec_idx = 0; exec_idx < prop_count; exec_idx++) {
      if (!(props[exec_idx].stages & mesa_to_vk_shader_stage(stage)))
         continue;

      VkPipelineExecutableStatisticKHR *stats = NULL;
      uint32_t stat_count = 0;

      VkPipelineExecutableInfoKHR exec_info = {0};
      exec_info.pipeline = radv_pipeline_to_handle(pipeline);
      exec_info.executableIndex = exec_idx;

      result = radv_GetPipelineExecutableStatisticsKHR(radv_device_to_handle(device), &exec_info,
                                                       &stat_count, NULL);
      if (result != VK_SUCCESS)
         goto fail;

      stats = calloc(stat_count, sizeof(*stats));
      if (!stats) {
         result = VK_ERROR_OUT_OF_HOST_MEMORY;
         goto fail;
      }

      result = radv_GetPipelineExecutableStatisticsKHR(radv_device_to_handle(device), &exec_info,
                                                       &stat_count, stats);
      if (result != VK_SUCCESS) {
         free(stats);
         goto fail;
      }

      fprintf(output, "\n%s:\n", radv_get_shader_name(&shader->info, stage));
      fprintf(output, "*** SHADER STATS ***\n");

      for (unsigned i = 0; i < stat_count; i++) {
         fprintf(output, "%s: ", stats[i].name);
         switch (stats[i].format) {
         case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_BOOL32_KHR:
            fprintf(output, "%s", stats[i].value.b32 == VK_TRUE ? "true" : "false");
            break;
         case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_INT64_KHR:
            fprintf(output, "%" PRIi64, stats[i].value.i64);
            break;
         case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_UINT64_KHR:
            fprintf(output, "%" PRIu64, stats[i].value.u64);
            break;
         case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_FLOAT64_KHR:
            fprintf(output, "%f", stats[i].value.f64);
            break;
         default:
            unreachable("Invalid pipeline statistic format");
         }
         fprintf(output, "\n");
      }

      fprintf(output, "********************\n\n\n");

      free(stats);
   }

fail:
   free(props);
   return result;
}