#include "radv_shader.h"
#include "nir/nir.h"
#include "nir/nir_deref.h"
+#include "nir/nir_xfb_info.h"
static void mark_sampler_desc(const nir_variable *var,
struct radv_shader_info *info)
info->tcs.outputs_written |= (mask << param);
}
-static void get_deref_offset(nir_deref_var *deref, unsigned *const_out)
-{
- nir_deref *tail = &deref->deref;
- unsigned const_offset = 0;
-
- if (deref->var->data.compact) {
- assert(tail->child->deref_type == nir_deref_type_array);
- assert(glsl_type_is_scalar(glsl_without_array(deref->var->type)));
-
- nir_deref_array *deref_array = nir_deref_as_array(tail->child);
- /* We always lower indirect dereferences for "compact" array vars. */
- assert(deref_array->deref_array_type == nir_deref_array_type_direct);
-
- *const_out = deref_array->base_offset;
- return;
- }
-
- while (tail->child != NULL) {
- const struct glsl_type *parent_type = tail->type;
- tail = tail->child;
-
- if (tail->deref_type == nir_deref_type_array) {
- nir_deref_array *deref_array = nir_deref_as_array(tail);
- unsigned size = glsl_count_attribute_slots(tail->type, false);
-
- const_offset += size * deref_array->base_offset;
- } else if (tail->deref_type == nir_deref_type_struct) {
- nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
-
- for (unsigned i = 0; i < deref_struct->index; i++) {
- const struct glsl_type *ft = glsl_get_struct_field(parent_type, i);
- const_offset += glsl_count_attribute_slots(ft, false);
- }
- } else
- unreachable("unsupported deref type");
- }
-
- *const_out = const_offset;
-}
-
static void
-get_deref_instr_offset(nir_deref_instr *instr,
- unsigned *const_out)
+get_deref_offset(nir_deref_instr *instr,
+ unsigned *const_out)
{
nir_variable *var = nir_deref_instr_get_variable(instr);
nir_deref_path path;
if (var->data.compact) {
assert(instr->deref_type == nir_deref_type_array);
- nir_const_value *v = nir_src_as_const_value(instr->arr.index);
- assert(v);
- *const_out = v->u32[0];
+ *const_out = nir_src_as_uint(instr->arr.index);
return;
}
}
} else if(path.path[idx_lvl]->deref_type == nir_deref_type_array) {
unsigned size = glsl_count_attribute_slots(path.path[idx_lvl]->type, false);
- nir_const_value *v = nir_src_as_const_value(path.path[idx_lvl]->arr.index);
- if (v)
- const_offset += v->u32[0] * size;
+ if (nir_src_is_const(path.path[idx_lvl]->arr.index))
+ const_offset += nir_src_as_uint(path.path[idx_lvl]->arr.index) * size;
} else
unreachable("Uhandled deref type in get_deref_instr_offset");
}
}
static void
-gather_intrinsic_load_var_info(const nir_shader *nir,
+gather_intrinsic_load_deref_info(const nir_shader *nir,
const nir_intrinsic_instr *instr,
struct radv_shader_info *info)
{
switch (nir->info.stage) {
case MESA_SHADER_VERTEX: {
- nir_variable *var = instr->intrinsic == nir_intrinsic_load_var ? instr->variables[0]->var :
- nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[0].ssa->parent_instr));
+ nir_variable *var = nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[0].ssa->parent_instr));
- if (var->data.mode == nir_var_shader_in) {
+ if (var && var->data.mode == nir_var_shader_in) {
unsigned idx = var->data.location;
uint8_t mask = nir_ssa_def_components_read(&instr->dest.ssa);
}
}
+static uint32_t
+widen_writemask(uint32_t wrmask)
+{
+ uint32_t new_wrmask = 0;
+ for(unsigned i = 0; i < 4; i++)
+ new_wrmask |= (wrmask & (1 << i) ? 0x3 : 0x0) << (i * 2);
+ return new_wrmask;
+}
+
static void
-gather_intrinsic_store_var_info(const nir_shader *nir,
+set_output_usage_mask(const nir_shader *nir, const nir_intrinsic_instr *instr,
+ uint8_t *output_usage_mask)
+{
+ nir_deref_instr *deref_instr =
+ nir_instr_as_deref(instr->src[0].ssa->parent_instr);
+ nir_variable *var = nir_deref_instr_get_variable(deref_instr);
+ unsigned attrib_count = glsl_count_attribute_slots(deref_instr->type, false);
+ unsigned idx = var->data.location;
+ unsigned comp = var->data.location_frac;
+ unsigned const_offset = 0;
+
+ get_deref_offset(deref_instr, &const_offset);
+
+ if (var->data.compact) {
+ assert(!glsl_type_is_64bit(deref_instr->type));
+ const_offset += comp;
+ output_usage_mask[idx + const_offset / 4] |= 1 << (const_offset % 4);
+ return;
+ }
+
+ uint32_t wrmask = nir_intrinsic_write_mask(instr);
+ if (glsl_type_is_64bit(deref_instr->type))
+ wrmask = widen_writemask(wrmask);
+
+ for (unsigned i = 0; i < attrib_count; i++)
+ output_usage_mask[idx + i + const_offset] |=
+ ((wrmask >> (i * 4)) & 0xf) << comp;
+}
+
+static void
+gather_intrinsic_store_deref_info(const nir_shader *nir,
const nir_intrinsic_instr *instr,
struct radv_shader_info *info)
{
- nir_variable *var = instr->intrinsic == nir_intrinsic_store_var ? instr->variables[0]->var :
- nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[0].ssa->parent_instr));
+ nir_variable *var = nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[0].ssa->parent_instr));
- if (var->data.mode == nir_var_shader_out) {
- unsigned attrib_count = glsl_count_attribute_slots(var->type, false);
+ if (var && var->data.mode == nir_var_shader_out) {
unsigned idx = var->data.location;
- unsigned comp = var->data.location_frac;
- unsigned const_offset = 0;
-
- if (instr->intrinsic == nir_intrinsic_store_var)
- get_deref_offset(instr->variables[0], &const_offset);
- else
- get_deref_instr_offset(nir_instr_as_deref(instr->src[0].ssa->parent_instr), &const_offset);
switch (nir->info.stage) {
case MESA_SHADER_VERTEX:
- for (unsigned i = 0; i < attrib_count; i++) {
- info->vs.output_usage_mask[idx + i + const_offset] |=
- instr->const_index[0] << comp;
- }
+ set_output_usage_mask(nir, instr,
+ info->vs.output_usage_mask);
break;
case MESA_SHADER_GEOMETRY:
- for (unsigned i = 0; i < attrib_count; i++) {
- info->gs.output_usage_mask[idx + i + const_offset] |=
- instr->const_index[0] << comp;
- }
+ set_output_usage_mask(nir, instr,
+ info->gs.output_usage_mask);
break;
case MESA_SHADER_TESS_EVAL:
- for (unsigned i = 0; i < attrib_count; i++) {
- info->tes.output_usage_mask[idx + i + const_offset] |=
- instr->const_index[0] << comp;
- }
+ set_output_usage_mask(nir, instr,
+ info->tes.output_usage_mask);
break;
case MESA_SHADER_TESS_CTRL: {
unsigned param = shader_io_get_unique_index(idx);
type = glsl_get_array_element(var->type);
unsigned slots =
- var->data.compact ? DIV_ROUND_UP(glsl_get_length(type), 4)
+ var->data.compact ? DIV_ROUND_UP(var->data.location_frac + glsl_get_length(type), 4)
: glsl_count_attribute_slots(type, false);
- if (idx == VARYING_SLOT_CLIP_DIST0)
- slots = (nir->info.clip_distance_array_size +
- nir->info.cull_distance_array_size > 4) ? 2 : 1;
-
mark_tess_output(info, var->data.patch, param, slots);
break;
}
}
}
+static void
+gather_push_constant_info(const nir_shader *nir,
+ const nir_intrinsic_instr *instr,
+ struct radv_shader_info *info)
+{
+ int base = nir_intrinsic_base(instr);
+
+ if (!nir_src_is_const(instr->src[0])) {
+ info->has_indirect_push_constants = true;
+ } else {
+ uint32_t min = base + nir_src_as_uint(instr->src[0]);
+ uint32_t max = min + instr->num_components * 4;
+
+ info->max_push_constant_used =
+ MAX2(max, info->max_push_constant_used);
+ info->min_push_constant_used =
+ MIN2(min, info->min_push_constant_used);
+ }
+
+ if (instr->dest.ssa.bit_size != 32)
+ info->has_only_32bit_push_constants = false;
+
+ info->loads_push_constants = true;
+}
+
static void
gather_intrinsic_info(const nir_shader *nir, const nir_intrinsic_instr *instr,
struct radv_shader_info *info)
{
switch (instr->intrinsic) {
- case nir_intrinsic_interp_var_at_sample:
case nir_intrinsic_interp_deref_at_sample:
info->ps.needs_sample_positions = true;
break;
info->uses_prim_id = true;
break;
case nir_intrinsic_load_push_constant:
- info->loads_push_constants = true;
+ gather_push_constant_info(nir, instr, info);
break;
case nir_intrinsic_vulkan_resource_index:
info->desc_set_used_mask |= (1 << nir_intrinsic_desc_set(instr));
break;
- case nir_intrinsic_image_var_load:
- case nir_intrinsic_image_var_store:
- case nir_intrinsic_image_var_atomic_add:
- case nir_intrinsic_image_var_atomic_min:
- case nir_intrinsic_image_var_atomic_max:
- case nir_intrinsic_image_var_atomic_and:
- case nir_intrinsic_image_var_atomic_or:
- case nir_intrinsic_image_var_atomic_xor:
- case nir_intrinsic_image_var_atomic_exchange:
- case nir_intrinsic_image_var_atomic_comp_swap:
- case nir_intrinsic_image_var_size: {
- const struct glsl_type *type = glsl_without_array(instr->variables[0]->var->type);
-
- enum glsl_sampler_dim dim = glsl_get_sampler_dim(type);
- if (dim == GLSL_SAMPLER_DIM_SUBPASS ||
- dim == GLSL_SAMPLER_DIM_SUBPASS_MS) {
- info->ps.layer_input = true;
- info->ps.uses_input_attachments = true;
- }
- mark_sampler_desc(instr->variables[0]->var, info);
-
- if (nir_intrinsic_image_var_store ||
- nir_intrinsic_image_var_atomic_add ||
- nir_intrinsic_image_var_atomic_min ||
- nir_intrinsic_image_var_atomic_max ||
- nir_intrinsic_image_var_atomic_and ||
- nir_intrinsic_image_var_atomic_or ||
- nir_intrinsic_image_var_atomic_xor ||
- nir_intrinsic_image_var_atomic_exchange ||
- nir_intrinsic_image_var_atomic_comp_swap) {
- if (nir->info.stage == MESA_SHADER_FRAGMENT)
- info->ps.writes_memory = true;
- }
- break;
- }
case nir_intrinsic_image_deref_load:
case nir_intrinsic_image_deref_store:
case nir_intrinsic_image_deref_atomic_add:
}
mark_sampler_desc(var, info);
- if (nir_intrinsic_image_deref_store ||
- nir_intrinsic_image_deref_atomic_add ||
- nir_intrinsic_image_deref_atomic_min ||
- nir_intrinsic_image_deref_atomic_max ||
- nir_intrinsic_image_deref_atomic_and ||
- nir_intrinsic_image_deref_atomic_or ||
- nir_intrinsic_image_deref_atomic_xor ||
- nir_intrinsic_image_deref_atomic_exchange ||
- nir_intrinsic_image_deref_atomic_comp_swap) {
+ if (instr->intrinsic == nir_intrinsic_image_deref_store ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_add ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_min ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_max ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_and ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_or ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_xor ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_exchange ||
+ instr->intrinsic == nir_intrinsic_image_deref_atomic_comp_swap) {
if (nir->info.stage == MESA_SHADER_FRAGMENT)
info->ps.writes_memory = true;
}
if (nir->info.stage == MESA_SHADER_FRAGMENT)
info->ps.writes_memory = true;
break;
- case nir_intrinsic_load_var:
case nir_intrinsic_load_deref:
- gather_intrinsic_load_var_info(nir, instr, info);
+ gather_intrinsic_load_deref_info(nir, instr, info);
break;
- case nir_intrinsic_store_var:
case nir_intrinsic_store_deref:
- gather_intrinsic_store_var_info(nir, instr, info);
+ gather_intrinsic_store_deref_info(nir, instr, info);
break;
default:
break;
break;
}
}
-
- if (instr->sampler)
- mark_sampler_desc(instr->sampler->var, info);
- if (instr->texture)
- mark_sampler_desc(instr->texture->var, info);
}
static void
gather_info_input_decl_ps(const nir_shader *nir, const nir_variable *var,
struct radv_shader_info *info)
{
+ unsigned attrib_count = glsl_count_attribute_slots(var->type, false);
const struct glsl_type *type = glsl_without_array(var->type);
int idx = var->data.location;
case VARYING_SLOT_LAYER:
info->ps.layer_input = true;
break;
+ case VARYING_SLOT_CLIP_DIST0:
+ case VARYING_SLOT_CLIP_DIST1:
+ info->ps.num_input_clips_culls += attrib_count;
+ break;
default:
break;
}
int idx = var->data.location;
unsigned param = shader_io_get_unique_index(idx);
int num_slots = glsl_count_attribute_slots(var->type, false);
- if (idx == VARYING_SLOT_CLIP_DIST0)
- num_slots = (nir->info.clip_distance_array_size + nir->info.cull_distance_array_size > 4) ? 2 : 1;
+ if (var->data.compact)
+ num_slots = DIV_ROUND_UP(var->data.location_frac + glsl_get_length(var->type), 4);
mark_ls_output(info, param, num_slots);
}
}
}
+static void
+gather_info_output_decl_gs(const nir_shader *nir, const nir_variable *var,
+ struct radv_shader_info *info)
+{
+ unsigned num_components = glsl_get_component_slots(var->type);
+ unsigned stream = var->data.stream;
+ unsigned idx = var->data.location;
+
+ assert(stream < 4);
+
+ info->gs.max_stream = MAX2(info->gs.max_stream, stream);
+ info->gs.num_stream_output_components[stream] += num_components;
+ info->gs.output_streams[idx] = stream;
+}
+
static void
gather_info_output_decl(const nir_shader *nir, const nir_variable *var,
struct radv_shader_info *info,
if (options->key.vs.as_ls)
gather_info_output_decl_ls(nir, var, info);
break;
+ case MESA_SHADER_GEOMETRY:
+ gather_info_output_decl_gs(nir, var, info);
+ break;
default:
break;
}
}
+static void
+gather_xfb_info(const nir_shader *nir, struct radv_shader_info *info)
+{
+ nir_xfb_info *xfb = nir_gather_xfb_info(nir, NULL);
+ struct radv_streamout_info *so = &info->so;
+
+ if (!xfb)
+ return;
+
+ assert(xfb->output_count < MAX_SO_OUTPUTS);
+ so->num_outputs = xfb->output_count;
+
+ for (unsigned i = 0; i < xfb->output_count; i++) {
+ struct radv_stream_output *output = &so->outputs[i];
+
+ output->buffer = xfb->outputs[i].buffer;
+ output->stream = xfb->buffer_to_stream[xfb->outputs[i].buffer];
+ output->offset = xfb->outputs[i].offset;
+ output->location = xfb->outputs[i].location;
+ output->component_mask = xfb->outputs[i].component_mask;
+
+ so->enabled_stream_buffers_mask |=
+ (1 << output->buffer) << (output->stream * 4);
+
+ }
+
+ for (unsigned i = 0; i < NIR_MAX_XFB_BUFFERS; i++) {
+ so->strides[i] = xfb->buffers[i].stride / 4;
+ }
+
+ ralloc_free(xfb);
+}
+
+void
+radv_nir_shader_info_init(struct radv_shader_info *info)
+{
+ /* Assume that shaders only have 32-bit push constants by default. */
+ info->min_push_constant_used = UINT8_MAX;
+ info->has_only_32bit_push_constants = true;
+}
+
void
radv_nir_shader_info_pass(const struct nir_shader *nir,
const struct radv_nir_compiler_options *options,
struct nir_function *func =
(struct nir_function *)exec_list_get_head_const(&nir->functions);
- if (options->layout && options->layout->dynamic_offset_count)
+ if (options->layout && options->layout->dynamic_offset_count &&
+ (options->layout->dynamic_shader_stages & mesa_to_vk_shader_stage(nir->info.stage))) {
info->loads_push_constants = true;
+ info->loads_dynamic_offsets = true;
+ }
nir_foreach_variable(variable, &nir->inputs)
gather_info_input_decl(nir, variable, info);
nir_foreach_variable(variable, &nir->outputs)
gather_info_output_decl(nir, variable, info, options);
+
+ if (nir->info.stage == MESA_SHADER_VERTEX ||
+ nir->info.stage == MESA_SHADER_TESS_EVAL ||
+ nir->info.stage == MESA_SHADER_GEOMETRY)
+ gather_xfb_info(nir, info);
}