#include "nir.h"
#include "nir_control_flow_private.h"
+#include "util/half_float.h"
+#include <limits.h>
#include <assert.h>
+#include <math.h>
+#include "util/u_math.h"
+
+#include "main/menums.h" /* BITFIELD64_MASK */
nir_shader *
nir_shader_create(void *mem_ctx,
assert(!"invalid mode");
break;
- case nir_var_local:
+ case nir_var_function_temp:
assert(!"nir_shader_add_variable cannot be used for local variables");
break;
- case nir_var_param:
- assert(!"nir_shader_add_variable cannot be used for function parameters");
- break;
-
- case nir_var_global:
+ case nir_var_shader_temp:
exec_list_push_tail(&shader->globals, &var->node);
break;
break;
case nir_var_uniform:
- case nir_var_shader_storage:
+ case nir_var_mem_ubo:
+ case nir_var_mem_ssbo:
exec_list_push_tail(&shader->uniforms, &var->node);
break;
- case nir_var_shared:
+ case nir_var_mem_shared:
assert(shader->info.stage == MESA_SHADER_COMPUTE);
exec_list_push_tail(&shader->shared, &var->node);
break;
var->name = ralloc_strdup(var, name);
var->type = type;
var->data.mode = mode;
+ var->data.how_declared = nir_var_declared_normally;
if ((mode == nir_var_shader_in &&
shader->info.stage != MESA_SHADER_VERTEX) ||
nir_variable *var = rzalloc(impl->function->shader, nir_variable);
var->name = ralloc_strdup(var, name);
var->type = type;
- var->data.mode = nir_var_local;
+ var->data.mode = nir_var_function_temp;
nir_function_impl_add_variable(impl, var);
func->shader = shader;
func->num_params = 0;
func->params = NULL;
- func->return_type = glsl_void_type();
func->impl = NULL;
+ func->is_entrypoint = false;
return func;
}
+/* NOTE: if the instruction you are copying a src to is already added
+ * to the IR, use nir_instr_rewrite_src() instead.
+ */
void nir_src_copy(nir_src *dest, const nir_src *src, void *mem_ctx)
{
dest->is_ssa = src->is_ssa;
nir_src_copy(&dest->src, &src->src, &instr->instr);
dest->abs = src->abs;
dest->negate = src->negate;
- for (unsigned i = 0; i < 4; i++)
+ for (unsigned i = 0; i < NIR_MAX_VEC_COMPONENTS; i++)
dest->swizzle[i] = src->swizzle[i];
}
exec_list_make_empty(&impl->body);
exec_list_make_empty(&impl->registers);
exec_list_make_empty(&impl->locals);
- impl->num_params = 0;
- impl->params = NULL;
- impl->return_var = NULL;
impl->reg_alloc = 0;
impl->ssa_alloc = 0;
impl->valid_metadata = nir_metadata_none;
function->impl = impl;
impl->function = function;
- impl->num_params = function->num_params;
- impl->params = ralloc_array(function->shader,
- nir_variable *, impl->num_params);
-
- for (unsigned i = 0; i < impl->num_params; i++) {
- impl->params[i] = rzalloc(function->shader, nir_variable);
- impl->params[i]->type = function->params[i].type;
- impl->params[i]->data.mode = nir_var_param;
- impl->params[i]->data.location = i;
- }
-
- if (!glsl_type_is_void(function->return_type)) {
- impl->return_var = rzalloc(function->shader, nir_variable);
- impl->return_var->type = function->return_type;
- impl->return_var->data.mode = nir_var_param;
- impl->return_var->data.location = -1;
- } else {
- impl->return_var = NULL;
- }
-
return impl;
}
cf_init(&block->cf_node, nir_cf_node_block);
block->successors[0] = block->successors[1] = NULL;
- block->predecessors = _mesa_set_create(block, _mesa_hash_pointer,
- _mesa_key_pointer_equal);
+ block->predecessors = _mesa_pointer_set_create(block);
block->imm_dom = NULL;
/* XXX maybe it would be worth it to defer allocation? This
* way it doesn't get allocated for shader refs that never run
* which is later used to do state specific lowering and futher
* opt. Do any of the references not need dominance metadata?
*/
- block->dom_frontier = _mesa_set_create(block, _mesa_hash_pointer,
- _mesa_key_pointer_equal);
+ block->dom_frontier = _mesa_pointer_set_create(block);
exec_list_make_empty(&block->instr_list);
{
src_init(&src->src);
src->abs = src->negate = false;
- src->swizzle[0] = 0;
- src->swizzle[1] = 1;
- src->swizzle[2] = 2;
- src->swizzle[3] = 3;
+ for (int i = 0; i < NIR_MAX_VEC_COMPONENTS; ++i)
+ src->swizzle[i] = i;
}
nir_alu_instr *
return instr;
}
+nir_deref_instr *
+nir_deref_instr_create(nir_shader *shader, nir_deref_type deref_type)
+{
+ nir_deref_instr *instr =
+ rzalloc_size(shader, sizeof(nir_deref_instr));
+
+ instr_init(&instr->instr, nir_instr_type_deref);
+
+ instr->deref_type = deref_type;
+ if (deref_type != nir_deref_type_var)
+ src_init(&instr->parent);
+
+ if (deref_type == nir_deref_type_array ||
+ deref_type == nir_deref_type_ptr_as_array)
+ src_init(&instr->arr.index);
+
+ dest_init(&instr->dest);
+
+ return instr;
+}
+
nir_jump_instr *
nir_jump_instr_create(nir_shader *shader, nir_jump_type type)
{
nir_call_instr *
nir_call_instr_create(nir_shader *shader, nir_function *callee)
{
- nir_call_instr *instr = ralloc(shader, nir_call_instr);
- instr_init(&instr->instr, nir_instr_type_call);
+ const unsigned num_params = callee->num_params;
+ nir_call_instr *instr =
+ rzalloc_size(shader, sizeof(*instr) +
+ num_params * sizeof(instr->params[0]));
+ instr_init(&instr->instr, nir_instr_type_call);
instr->callee = callee;
- instr->num_params = callee->num_params;
- instr->params = ralloc_array(instr, nir_deref_var *, instr->num_params);
- instr->return_deref = NULL;
+ instr->num_params = num_params;
+ for (unsigned i = 0; i < num_params; i++)
+ src_init(&instr->params[i]);
return instr;
}
instr->texture_index = 0;
instr->texture_array_size = 0;
- instr->texture = NULL;
instr->sampler_index = 0;
- instr->sampler = NULL;
return instr;
}
return instr;
}
-nir_deref_var *
-nir_deref_var_create(void *mem_ctx, nir_variable *var)
+static nir_const_value
+const_value_float(double d, unsigned bit_size)
{
- nir_deref_var *deref = ralloc(mem_ctx, nir_deref_var);
- deref->deref.deref_type = nir_deref_type_var;
- deref->deref.child = NULL;
- deref->deref.type = var->type;
- deref->var = var;
- return deref;
-}
-
-nir_deref_array *
-nir_deref_array_create(void *mem_ctx)
-{
- nir_deref_array *deref = ralloc(mem_ctx, nir_deref_array);
- deref->deref.deref_type = nir_deref_type_array;
- deref->deref.child = NULL;
- deref->deref_array_type = nir_deref_array_type_direct;
- src_init(&deref->indirect);
- deref->base_offset = 0;
- return deref;
-}
-
-nir_deref_struct *
-nir_deref_struct_create(void *mem_ctx, unsigned field_index)
-{
- nir_deref_struct *deref = ralloc(mem_ctx, nir_deref_struct);
- deref->deref.deref_type = nir_deref_type_struct;
- deref->deref.child = NULL;
- deref->index = field_index;
- return deref;
-}
-
-nir_deref_var *
-nir_deref_var_clone(const nir_deref_var *deref, void *mem_ctx)
-{
- if (deref == NULL)
- return NULL;
-
- nir_deref_var *ret = nir_deref_var_create(mem_ctx, deref->var);
- ret->deref.type = deref->deref.type;
- if (deref->deref.child)
- ret->deref.child = nir_deref_clone(deref->deref.child, ret);
- return ret;
-}
-
-static nir_deref_array *
-deref_array_clone(const nir_deref_array *deref, void *mem_ctx)
-{
- nir_deref_array *ret = nir_deref_array_create(mem_ctx);
- ret->base_offset = deref->base_offset;
- ret->deref_array_type = deref->deref_array_type;
- if (deref->deref_array_type == nir_deref_array_type_indirect) {
- nir_src_copy(&ret->indirect, &deref->indirect, mem_ctx);
- }
- ret->deref.type = deref->deref.type;
- if (deref->deref.child)
- ret->deref.child = nir_deref_clone(deref->deref.child, ret);
- return ret;
-}
-
-static nir_deref_struct *
-deref_struct_clone(const nir_deref_struct *deref, void *mem_ctx)
-{
- nir_deref_struct *ret = nir_deref_struct_create(mem_ctx, deref->index);
- ret->deref.type = deref->deref.type;
- if (deref->deref.child)
- ret->deref.child = nir_deref_clone(deref->deref.child, ret);
- return ret;
-}
-
-nir_deref *
-nir_deref_clone(const nir_deref *deref, void *mem_ctx)
-{
- if (deref == NULL)
- return NULL;
-
- switch (deref->deref_type) {
- case nir_deref_type_var:
- return &nir_deref_var_clone(nir_deref_as_var(deref), mem_ctx)->deref;
- case nir_deref_type_array:
- return &deref_array_clone(nir_deref_as_array(deref), mem_ctx)->deref;
- case nir_deref_type_struct:
- return &deref_struct_clone(nir_deref_as_struct(deref), mem_ctx)->deref;
+ nir_const_value v;
+ switch (bit_size) {
+ case 16: v.u16[0] = _mesa_float_to_half(d); break;
+ case 32: v.f32[0] = d; break;
+ case 64: v.f64[0] = d; break;
default:
- unreachable("Invalid dereference type");
+ unreachable("Invalid bit size");
}
-
- return NULL;
+ return v;
}
-/* This is the second step in the recursion. We've found the tail and made a
- * copy. Now we need to iterate over all possible leaves and call the
- * callback on each one.
- */
-static bool
-deref_foreach_leaf_build_recur(nir_deref_var *deref, nir_deref *tail,
- nir_deref_foreach_leaf_cb cb, void *state)
-{
- unsigned length;
- union {
- nir_deref_array arr;
- nir_deref_struct str;
- } tmp;
-
- assert(tail->child == NULL);
- switch (glsl_get_base_type(tail->type)) {
- case GLSL_TYPE_UINT:
- case GLSL_TYPE_UINT16:
- case GLSL_TYPE_UINT64:
- case GLSL_TYPE_INT:
- case GLSL_TYPE_INT16:
- case GLSL_TYPE_INT64:
- case GLSL_TYPE_FLOAT:
- case GLSL_TYPE_FLOAT16:
- case GLSL_TYPE_DOUBLE:
- case GLSL_TYPE_BOOL:
- if (glsl_type_is_vector_or_scalar(tail->type))
- return cb(deref, state);
- /* Fall Through */
-
- case GLSL_TYPE_ARRAY:
- tmp.arr.deref.deref_type = nir_deref_type_array;
- tmp.arr.deref.type = glsl_get_array_element(tail->type);
- tmp.arr.deref_array_type = nir_deref_array_type_direct;
- tmp.arr.indirect = NIR_SRC_INIT;
- tail->child = &tmp.arr.deref;
-
- length = glsl_get_length(tail->type);
- for (unsigned i = 0; i < length; i++) {
- tmp.arr.deref.child = NULL;
- tmp.arr.base_offset = i;
- if (!deref_foreach_leaf_build_recur(deref, &tmp.arr.deref, cb, state))
- return false;
- }
- return true;
-
- case GLSL_TYPE_STRUCT:
- tmp.str.deref.deref_type = nir_deref_type_struct;
- tail->child = &tmp.str.deref;
-
- length = glsl_get_length(tail->type);
- for (unsigned i = 0; i < length; i++) {
- tmp.arr.deref.child = NULL;
- tmp.str.deref.type = glsl_get_struct_field(tail->type, i);
- tmp.str.index = i;
- if (!deref_foreach_leaf_build_recur(deref, &tmp.arr.deref, cb, state))
- return false;
- }
- return true;
-
- default:
- unreachable("Invalid type for dereference");
- }
-}
-
-/* This is the first step of the foreach_leaf recursion. In this step we are
- * walking to the end of the deref chain and making a copy in the stack as we
- * go. This is because we don't want to mutate the deref chain that was
- * passed in by the caller. The downside is that this deref chain is on the
- * stack and , if the caller wants to do anything with it, they will have to
- * make their own copy because this one will go away.
- */
-static bool
-deref_foreach_leaf_copy_recur(nir_deref_var *deref, nir_deref *tail,
- nir_deref_foreach_leaf_cb cb, void *state)
-{
- union {
- nir_deref_array arr;
- nir_deref_struct str;
- } c;
-
- if (tail->child) {
- switch (tail->child->deref_type) {
- case nir_deref_type_array:
- c.arr = *nir_deref_as_array(tail->child);
- tail->child = &c.arr.deref;
- return deref_foreach_leaf_copy_recur(deref, &c.arr.deref, cb, state);
-
- case nir_deref_type_struct:
- c.str = *nir_deref_as_struct(tail->child);
- tail->child = &c.str.deref;
- return deref_foreach_leaf_copy_recur(deref, &c.str.deref, cb, state);
-
- case nir_deref_type_var:
- default:
- unreachable("Invalid deref type for a child");
- }
- } else {
- /* We've gotten to the end of the original deref. Time to start
- * building our own derefs.
- */
- return deref_foreach_leaf_build_recur(deref, tail, cb, state);
- }
-}
-
-/**
- * This function iterates over all of the possible derefs that can be created
- * with the given deref as the head. It then calls the provided callback with
- * a full deref for each one.
- *
- * The deref passed to the callback will be allocated on the stack. You will
- * need to make a copy if you want it to hang around.
- */
-bool
-nir_deref_foreach_leaf(nir_deref_var *deref,
- nir_deref_foreach_leaf_cb cb, void *state)
+static nir_const_value
+const_value_int(int64_t i, unsigned bit_size)
{
- nir_deref_var copy = *deref;
- return deref_foreach_leaf_copy_recur(©, ©.deref, cb, state);
-}
-
-/* Returns a load_const instruction that represents the constant
- * initializer for the given deref chain. The caller is responsible for
- * ensuring that there actually is a constant initializer.
- */
-nir_load_const_instr *
-nir_deref_get_const_initializer_load(nir_shader *shader, nir_deref_var *deref)
-{
- nir_constant *constant = deref->var->constant_initializer;
- assert(constant);
-
- const nir_deref *tail = &deref->deref;
- unsigned matrix_col = 0;
- while (tail->child) {
- switch (tail->child->deref_type) {
- case nir_deref_type_array: {
- nir_deref_array *arr = nir_deref_as_array(tail->child);
- assert(arr->deref_array_type == nir_deref_array_type_direct);
- if (glsl_type_is_matrix(tail->type)) {
- assert(arr->deref.child == NULL);
- matrix_col = arr->base_offset;
- } else {
- constant = constant->elements[arr->base_offset];
- }
- break;
- }
-
- case nir_deref_type_struct: {
- constant = constant->elements[nir_deref_as_struct(tail->child)->index];
- break;
- }
-
- default:
- unreachable("Invalid deref child type");
- }
-
- tail = tail->child;
- }
-
- unsigned bit_size = glsl_get_bit_size(tail->type);
- nir_load_const_instr *load =
- nir_load_const_instr_create(shader, glsl_get_vector_elements(tail->type),
- bit_size);
-
- switch (glsl_get_base_type(tail->type)) {
- case GLSL_TYPE_FLOAT:
- case GLSL_TYPE_INT:
- case GLSL_TYPE_UINT:
- case GLSL_TYPE_FLOAT16:
- case GLSL_TYPE_DOUBLE:
- case GLSL_TYPE_INT16:
- case GLSL_TYPE_UINT16:
- case GLSL_TYPE_UINT64:
- case GLSL_TYPE_INT64:
- case GLSL_TYPE_BOOL:
- load->value = constant->values[matrix_col];
- break;
+ nir_const_value v;
+ switch (bit_size) {
+ case 1: v.b[0] = i & 1; break;
+ case 8: v.i8[0] = i; break;
+ case 16: v.i16[0] = i; break;
+ case 32: v.i32[0] = i; break;
+ case 64: v.i64[0] = i; break;
+ default:
+ unreachable("Invalid bit size");
+ }
+ return v;
+}
+
+nir_const_value
+nir_alu_binop_identity(nir_op binop, unsigned bit_size)
+{
+ const int64_t max_int = (1ull << (bit_size - 1)) - 1;
+ const int64_t min_int = -max_int - 1;
+ switch (binop) {
+ case nir_op_iadd:
+ return const_value_int(0, bit_size);
+ case nir_op_fadd:
+ return const_value_float(0, bit_size);
+ case nir_op_imul:
+ return const_value_int(1, bit_size);
+ case nir_op_fmul:
+ return const_value_float(1, bit_size);
+ case nir_op_imin:
+ return const_value_int(max_int, bit_size);
+ case nir_op_umin:
+ return const_value_int(~0ull, bit_size);
+ case nir_op_fmin:
+ return const_value_float(INFINITY, bit_size);
+ case nir_op_imax:
+ return const_value_int(min_int, bit_size);
+ case nir_op_umax:
+ return const_value_int(0, bit_size);
+ case nir_op_fmax:
+ return const_value_float(-INFINITY, bit_size);
+ case nir_op_iand:
+ return const_value_int(~0ull, bit_size);
+ case nir_op_ior:
+ return const_value_int(0, bit_size);
+ case nir_op_ixor:
+ return const_value_int(0, bit_size);
default:
- unreachable("Invalid immediate type");
+ unreachable("Invalid reduction operation");
}
-
- return load;
}
nir_function_impl *
nir_foreach_src(instr, remove_use_cb, instr);
}
-void nir_instr_remove(nir_instr *instr)
+void nir_instr_remove_v(nir_instr *instr)
{
remove_defs_uses(instr);
exec_node_remove(&instr->node);
return cb(&instr->dest.dest, state);
}
+static bool
+visit_deref_dest(nir_deref_instr *instr, nir_foreach_dest_cb cb, void *state)
+{
+ return cb(&instr->dest, state);
+}
+
static bool
visit_intrinsic_dest(nir_intrinsic_instr *instr, nir_foreach_dest_cb cb,
void *state)
switch (instr->type) {
case nir_instr_type_alu:
return visit_alu_dest(nir_instr_as_alu(instr), cb, state);
+ case nir_instr_type_deref:
+ return visit_deref_dest(nir_instr_as_deref(instr), cb, state);
case nir_instr_type_intrinsic:
return visit_intrinsic_dest(nir_instr_as_intrinsic(instr), cb, state);
case nir_instr_type_tex:
{
switch (instr->type) {
case nir_instr_type_alu:
+ case nir_instr_type_deref:
case nir_instr_type_tex:
case nir_instr_type_intrinsic:
case nir_instr_type_phi:
}
static bool
-visit_deref_array_src(nir_deref_array *deref, nir_foreach_src_cb cb,
- void *state)
+visit_alu_src(nir_alu_instr *instr, nir_foreach_src_cb cb, void *state)
{
- if (deref->deref_array_type == nir_deref_array_type_indirect)
- return visit_src(&deref->indirect, cb, state);
+ for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++)
+ if (!visit_src(&instr->src[i].src, cb, state))
+ return false;
+
return true;
}
static bool
-visit_deref_src(nir_deref_var *deref, nir_foreach_src_cb cb, void *state)
+visit_deref_instr_src(nir_deref_instr *instr,
+ nir_foreach_src_cb cb, void *state)
{
- nir_deref *cur = &deref->deref;
- while (cur != NULL) {
- if (cur->deref_type == nir_deref_type_array) {
- if (!visit_deref_array_src(nir_deref_as_array(cur), cb, state))
- return false;
- }
-
- cur = cur->child;
+ if (instr->deref_type != nir_deref_type_var) {
+ if (!visit_src(&instr->parent, cb, state))
+ return false;
}
- return true;
-}
-
-static bool
-visit_alu_src(nir_alu_instr *instr, nir_foreach_src_cb cb, void *state)
-{
- for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++)
- if (!visit_src(&instr->src[i].src, cb, state))
+ if (instr->deref_type == nir_deref_type_array ||
+ instr->deref_type == nir_deref_type_ptr_as_array) {
+ if (!visit_src(&instr->arr.index, cb, state))
return false;
+ }
return true;
}
return false;
}
- if (instr->texture != NULL) {
- if (!visit_deref_src(instr->texture, cb, state))
- return false;
- }
-
- if (instr->sampler != NULL) {
- if (!visit_deref_src(instr->sampler, cb, state))
- return false;
- }
-
return true;
}
return false;
}
- unsigned num_vars =
- nir_intrinsic_infos[instr->intrinsic].num_variables;
- for (unsigned i = 0; i < num_vars; i++) {
- if (!visit_deref_src(instr->variables[i], cb, state))
+ return true;
+}
+
+static bool
+visit_call_src(nir_call_instr *instr, nir_foreach_src_cb cb, void *state)
+{
+ for (unsigned i = 0; i < instr->num_params; i++) {
+ if (!visit_src(&instr->params[i], cb, state))
return false;
}
if (!visit_alu_src(nir_instr_as_alu(instr), cb, state))
return false;
break;
+ case nir_instr_type_deref:
+ if (!visit_deref_instr_src(nir_instr_as_deref(instr), cb, state))
+ return false;
+ break;
case nir_instr_type_intrinsic:
if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr), cb, state))
return false;
return false;
break;
case nir_instr_type_call:
- /* Call instructions have no regular sources */
+ if (!visit_call_src(nir_instr_as_call(instr), cb, state))
+ return false;
break;
case nir_instr_type_load_const:
/* Constant load instructions have no regular sources */
return nir_foreach_dest(instr, visit_dest_indirect, &dest_state);
}
+int64_t
+nir_src_comp_as_int(nir_src src, unsigned comp)
+{
+ assert(nir_src_is_const(src));
+ nir_load_const_instr *load = nir_instr_as_load_const(src.ssa->parent_instr);
+
+ assert(comp < load->def.num_components);
+ switch (load->def.bit_size) {
+ /* int1_t uses 0/-1 convention */
+ case 1: return -(int)load->value.b[comp];
+ case 8: return load->value.i8[comp];
+ case 16: return load->value.i16[comp];
+ case 32: return load->value.i32[comp];
+ case 64: return load->value.i64[comp];
+ default:
+ unreachable("Invalid bit size");
+ }
+}
+
+uint64_t
+nir_src_comp_as_uint(nir_src src, unsigned comp)
+{
+ assert(nir_src_is_const(src));
+ nir_load_const_instr *load = nir_instr_as_load_const(src.ssa->parent_instr);
+
+ assert(comp < load->def.num_components);
+ switch (load->def.bit_size) {
+ case 1: return load->value.b[comp];
+ case 8: return load->value.u8[comp];
+ case 16: return load->value.u16[comp];
+ case 32: return load->value.u32[comp];
+ case 64: return load->value.u64[comp];
+ default:
+ unreachable("Invalid bit size");
+ }
+}
+
+bool
+nir_src_comp_as_bool(nir_src src, unsigned comp)
+{
+ int64_t i = nir_src_comp_as_int(src, comp);
+
+ /* Booleans of any size use 0/-1 convention */
+ assert(i == 0 || i == -1);
+
+ return i;
+}
+
+double
+nir_src_comp_as_float(nir_src src, unsigned comp)
+{
+ assert(nir_src_is_const(src));
+ nir_load_const_instr *load = nir_instr_as_load_const(src.ssa->parent_instr);
+
+ assert(comp < load->def.num_components);
+ switch (load->def.bit_size) {
+ case 16: return _mesa_half_to_float(load->value.u16[comp]);
+ case 32: return load->value.f32[comp];
+ case 64: return load->value.f64[comp];
+ default:
+ unreachable("Invalid bit size");
+ }
+}
+
+int64_t
+nir_src_as_int(nir_src src)
+{
+ assert(nir_src_num_components(src) == 1);
+ return nir_src_comp_as_int(src, 0);
+}
+
+uint64_t
+nir_src_as_uint(nir_src src)
+{
+ assert(nir_src_num_components(src) == 1);
+ return nir_src_comp_as_uint(src, 0);
+}
+
+bool
+nir_src_as_bool(nir_src src)
+{
+ assert(nir_src_num_components(src) == 1);
+ return nir_src_comp_as_bool(src, 0);
+}
+
+double
+nir_src_as_float(nir_src src)
+{
+ assert(nir_src_num_components(src) == 1);
+ return nir_src_comp_as_float(src, 0);
+}
+
nir_const_value *
nir_src_as_const_value(nir_src src)
{
src_add_all_uses(dest->reg.indirect, instr, NULL);
}
-void
-nir_instr_rewrite_deref(nir_instr *instr, nir_deref_var **deref,
- nir_deref_var *new_deref)
-{
- if (*deref)
- visit_deref_src(*deref, remove_use_cb, NULL);
-
- *deref = new_deref;
-
- if (*deref)
- visit_deref_src(*deref, add_use_cb, instr);
-}
-
/* note: does *not* take ownership of 'name' */
void
nir_ssa_def_init(nir_instr *instr, nir_ssa_def *def,
nir_if_rewrite_condition(use_src->parent_if, new_src);
}
-uint8_t
-nir_ssa_def_components_read(nir_ssa_def *def)
+nir_component_mask_t
+nir_ssa_def_components_read(const nir_ssa_def *def)
{
- uint8_t read_mask = 0;
+ nir_component_mask_t read_mask = 0;
nir_foreach_use(use, def) {
if (use->parent_instr->type == nir_instr_type_alu) {
nir_alu_instr *alu = nir_instr_as_alu(use->parent_instr);
nir_alu_src *alu_src = exec_node_data(nir_alu_src, use, src);
int src_idx = alu_src - &alu->src[0];
assert(src_idx >= 0 && src_idx < nir_op_infos[alu->op].num_inputs);
-
- for (unsigned c = 0; c < 4; c++) {
- if (!nir_alu_instr_channel_used(alu, src_idx, c))
- continue;
-
- read_mask |= (1 << alu_src->swizzle[c]);
- }
+ read_mask |= nir_alu_instr_src_read_mask(alu, src_idx);
} else {
return (1 << def->num_components) - 1;
}
}
+ if (!list_empty(&def->if_uses))
+ read_mask |= 1;
+
return read_mask;
}
block->index = index++;
}
- impl->num_blocks = index;
+ /* The end_block isn't really part of the program, which is why its index
+ * is >= num_blocks.
+ */
+ impl->num_blocks = impl->end_block->index = index;
}
static bool
return nir_intrinsic_load_base_instance;
case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
return nir_intrinsic_load_vertex_id_zero_base;
+ case SYSTEM_VALUE_IS_INDEXED_DRAW:
+ return nir_intrinsic_load_is_indexed_draw;
+ case SYSTEM_VALUE_FIRST_VERTEX:
+ return nir_intrinsic_load_first_vertex;
case SYSTEM_VALUE_BASE_VERTEX:
return nir_intrinsic_load_base_vertex;
case SYSTEM_VALUE_INVOCATION_ID:
return nir_intrinsic_load_subgroup_le_mask;
case SYSTEM_VALUE_SUBGROUP_LT_MASK:
return nir_intrinsic_load_subgroup_lt_mask;
+ case SYSTEM_VALUE_NUM_SUBGROUPS:
+ return nir_intrinsic_load_num_subgroups;
+ case SYSTEM_VALUE_SUBGROUP_ID:
+ return nir_intrinsic_load_subgroup_id;
case SYSTEM_VALUE_LOCAL_GROUP_SIZE:
return nir_intrinsic_load_local_group_size;
+ case SYSTEM_VALUE_GLOBAL_INVOCATION_ID:
+ return nir_intrinsic_load_global_invocation_id;
+ case SYSTEM_VALUE_WORK_DIM:
+ return nir_intrinsic_load_work_dim;
default:
unreachable("system value does not directly correspond to intrinsic");
}
return SYSTEM_VALUE_BASE_INSTANCE;
case nir_intrinsic_load_vertex_id_zero_base:
return SYSTEM_VALUE_VERTEX_ID_ZERO_BASE;
+ case nir_intrinsic_load_first_vertex:
+ return SYSTEM_VALUE_FIRST_VERTEX;
+ case nir_intrinsic_load_is_indexed_draw:
+ return SYSTEM_VALUE_IS_INDEXED_DRAW;
case nir_intrinsic_load_base_vertex:
return SYSTEM_VALUE_BASE_VERTEX;
case nir_intrinsic_load_invocation_id:
return SYSTEM_VALUE_SUBGROUP_LE_MASK;
case nir_intrinsic_load_subgroup_lt_mask:
return SYSTEM_VALUE_SUBGROUP_LT_MASK;
+ case nir_intrinsic_load_num_subgroups:
+ return SYSTEM_VALUE_NUM_SUBGROUPS;
+ case nir_intrinsic_load_subgroup_id:
+ return SYSTEM_VALUE_SUBGROUP_ID;
case nir_intrinsic_load_local_group_size:
return SYSTEM_VALUE_LOCAL_GROUP_SIZE;
+ case nir_intrinsic_load_global_invocation_id:
+ return SYSTEM_VALUE_GLOBAL_INVOCATION_ID;
default:
unreachable("intrinsic doesn't produce a system value");
}
}
+
+/* OpenGL utility method that remaps the location attributes if they are
+ * doubles. Not needed for vulkan due the differences on the input location
+ * count for doubles on vulkan vs OpenGL
+ *
+ * The bitfield returned in dual_slot is one bit for each double input slot in
+ * the original OpenGL single-slot input numbering. The mapping from old
+ * locations to new locations is as follows:
+ *
+ * new_loc = loc + util_bitcount(dual_slot & BITFIELD64_MASK(loc))
+ */
+void
+nir_remap_dual_slot_attributes(nir_shader *shader, uint64_t *dual_slot)
+{
+ assert(shader->info.stage == MESA_SHADER_VERTEX);
+
+ *dual_slot = 0;
+ nir_foreach_variable(var, &shader->inputs) {
+ if (glsl_type_is_dual_slot(glsl_without_array(var->type))) {
+ unsigned slots = glsl_count_attribute_slots(var->type, true);
+ *dual_slot |= BITFIELD64_MASK(slots) << var->data.location;
+ }
+ }
+
+ nir_foreach_variable(var, &shader->inputs) {
+ var->data.location +=
+ util_bitcount64(*dual_slot & BITFIELD64_MASK(var->data.location));
+ }
+}
+
+/* Returns an attribute mask that has been re-compacted using the given
+ * dual_slot mask.
+ */
+uint64_t
+nir_get_single_slot_attribs_mask(uint64_t attribs, uint64_t dual_slot)
+{
+ while (dual_slot) {
+ unsigned loc = u_bit_scan64(&dual_slot);
+ /* mask of all bits up to and including loc */
+ uint64_t mask = BITFIELD64_MASK(loc + 1);
+ attribs = (attribs & mask) | ((attribs & ~mask) >> 1);
+ }
+ return attribs;
+}