ppir_dest *dest = ppir_node_get_dest(node);
dest->type = ppir_target_ssa;
dest->ssa.num_components = ssa->num_components;
- dest->ssa.live_in = INT_MAX;
- dest->ssa.live_out = 0;
dest->write_mask = u_bit_consecutive(0, ssa->num_components);
if (node->type == ppir_node_type_load ||
if (ns->is_ssa) {
child = comp->var_nodes[ns->ssa->index];
- /* Clone consts for each successor */
switch (child->op) {
- case ppir_op_const:
- child = ppir_node_clone(node->block, child);
- break;
- /* Clone uniforms and load textures for each block */
- case ppir_op_load_texture:
- case ppir_op_load_uniform:
case ppir_op_load_varying:
- if (child->block != node->block) {
- child = ppir_node_clone(node->block, child);
- comp->var_nodes[ns->ssa->index] = child;
+ /* If at least one successor is load_texture, promote it to
+ * load_coords to ensure that is has exactly one successor */
+ if (node->op == ppir_op_load_texture) {
+ nir_tex_src *nts = (nir_tex_src *)ns;
+ if (nts->src_type == nir_tex_src_coord)
+ child->op = ppir_op_load_coords;
}
break;
default:
break;
}
- ppir_node_add_dep(node, child);
+ if (child->op != ppir_op_undef)
+ ppir_node_add_dep(node, child, ppir_dep_src);
}
else {
nir_register *reg = ns->reg.reg;
}
/* Don't add dummies or recursive deps for ops like r1 = r1 + ssa1 */
if (child && node != child && child->op != ppir_op_dummy)
- ppir_node_add_dep(node, child);
+ ppir_node_add_dep(node, child, ppir_dep_src);
}
}
[nir_op_fceil] = ppir_op_ceil,
[nir_op_ffract] = ppir_op_fract,
[nir_op_sge] = ppir_op_ge,
- [nir_op_fge] = ppir_op_ge,
[nir_op_slt] = ppir_op_lt,
- [nir_op_flt] = ppir_op_lt,
[nir_op_seq] = ppir_op_eq,
- [nir_op_feq] = ppir_op_eq,
[nir_op_sne] = ppir_op_ne,
- [nir_op_fne] = ppir_op_ne,
[nir_op_fcsel] = ppir_op_select,
[nir_op_inot] = ppir_op_not,
[nir_op_ftrunc] = ppir_op_trunc,
lnode->num_components = instr->num_components;
lnode->index = nir_intrinsic_base(instr) * 4 + nir_intrinsic_component(instr);
+ if (nir_src_is_const(instr->src[0]))
+ lnode->index += (uint32_t)(nir_src_as_float(instr->src[0]) * 4);
+ else {
+ lnode->num_src = 1;
+ ppir_node_add_src(block->comp, &lnode->node, &lnode->src, instr->src, 1);
+ }
return &lnode->node;
case nir_intrinsic_load_frag_coord:
lnode->num_components = instr->num_components;
lnode->index = nir_intrinsic_base(instr);
- lnode->index += (uint32_t)nir_src_as_float(instr->src[0]);
+ if (nir_src_is_const(instr->src[0]))
+ lnode->index += (uint32_t)nir_src_as_float(instr->src[0]);
+ else {
+ lnode->num_src = 1;
+ ppir_node_add_src(block->comp, &lnode->node, &lnode->src, instr->src, 1);
+ }
return &lnode->node;
ppir_dest *dest = ppir_node_get_dest(&alu_node->node);
dest->type = ppir_target_ssa;
dest->ssa.num_components = instr->num_components;
- dest->ssa.live_in = INT_MAX;
- dest->ssa.live_out = 0;
dest->ssa.index = 0;
dest->write_mask = u_bit_consecutive(0, instr->num_components);
static ppir_node *ppir_emit_ssa_undef(ppir_block *block, nir_instr *ni)
{
- ppir_error("nir_ssa_undef_instr not support\n");
- return NULL;
+ nir_ssa_undef_instr *undef = nir_instr_as_ssa_undef(ni);
+ ppir_node *node = ppir_node_create_ssa(block, ppir_op_undef, &undef->def);
+ if (!node)
+ return NULL;
+ ppir_alu_node *alu = ppir_node_to_alu(node);
+
+ ppir_dest *dest = &alu->dest;
+ dest->ssa.undef = true;
+
+ return node;
}
static ppir_node *ppir_emit_tex(ppir_block *block, nir_instr *ni)
nir_tex_instr *instr = nir_instr_as_tex(ni);
ppir_load_texture_node *node;
- if (instr->op != nir_texop_tex) {
+ switch (instr->op) {
+ case nir_texop_tex:
+ case nir_texop_txb:
+ case nir_texop_txl:
+ break;
+ default:
ppir_error("unsupported texop %d\n", instr->op);
return NULL;
}
switch (instr->sampler_dim) {
case GLSL_SAMPLER_DIM_2D:
+ case GLSL_SAMPLER_DIM_CUBE:
case GLSL_SAMPLER_DIM_RECT:
case GLSL_SAMPLER_DIM_EXTERNAL:
break;
node->sampler_dim = instr->sampler_dim;
for (int i = 0; i < instr->coord_components; i++)
- node->src_coords.swizzle[i] = i;
+ node->src[0].swizzle[i] = i;
for (int i = 0; i < instr->num_srcs; i++) {
switch (instr->src[i].src_type) {
case nir_tex_src_coord:
- ppir_node_add_src(block->comp, &node->node, &node->src_coords, &instr->src[i].src,
+ ppir_node_add_src(block->comp, &node->node, &node->src[0], &instr->src[i].src,
u_bit_consecutive(0, instr->coord_components));
+ node->num_src++;
+ break;
+ case nir_tex_src_bias:
+ case nir_tex_src_lod:
+ node->lod_bias_en = true;
+ node->explicit_lod = (instr->src[i].src_type == nir_tex_src_lod);
+ ppir_node_add_src(block->comp, &node->node, &node->src[1], &instr->src[i].src, 1);
+ node->num_src++;
break;
default:
ppir_error("unsupported texture source type\n");
- assert(0);
return NULL;
}
}
static ppir_node *ppir_emit_jump(ppir_block *block, nir_instr *ni)
{
- ppir_error("nir_jump_instr not support\n");
- return NULL;
+ ppir_node *node;
+ ppir_compiler *comp = block->comp;
+ ppir_branch_node *branch;
+ ppir_block *jump_block;
+ nir_jump_instr *jump = nir_instr_as_jump(ni);
+
+ switch (jump->type) {
+ case nir_jump_break: {
+ assert(comp->current_block->successors[0]);
+ assert(!comp->current_block->successors[1]);
+ jump_block = comp->current_block->successors[0];
+ }
+ break;
+ case nir_jump_continue:
+ jump_block = comp->loop_cont_block;
+ break;
+ default:
+ ppir_error("nir_jump_instr not support\n");
+ return NULL;
+ }
+
+ assert(jump_block != NULL);
+
+ node = ppir_node_create(block, ppir_op_branch, -1, 0);
+ if (!node)
+ return NULL;
+ branch = ppir_node_to_branch(node);
+
+ /* Unconditional */
+ branch->num_src = 0;
+ branch->target = jump_block;
+
+ return node;
}
static ppir_node *(*ppir_emit_instr[nir_instr_type_phi])(ppir_block *, nir_instr *) = {
{
ppir_block *block = ppir_get_block(comp, nblock);
+ comp->current_block = block;
+
list_addtail(&block->list, &comp->block_list);
nir_foreach_instr(instr, nblock) {
return true;
}
-static bool ppir_emit_if(ppir_compiler *comp, nir_if *nif)
+static bool ppir_emit_cf_list(ppir_compiler *comp, struct exec_list *list);
+
+static bool ppir_emit_if(ppir_compiler *comp, nir_if *if_stmt)
{
- ppir_error("if nir_cf_node not support\n");
- return false;
+ ppir_node *node;
+ ppir_branch_node *else_branch, *after_branch;
+ nir_block *nir_else_block = nir_if_first_else_block(if_stmt);
+ bool empty_else_block =
+ (nir_else_block == nir_if_last_else_block(if_stmt) &&
+ exec_list_is_empty(&nir_else_block->instr_list));
+ ppir_block *block = comp->current_block;
+
+ node = ppir_node_create(block, ppir_op_branch, -1, 0);
+ if (!node)
+ return false;
+ else_branch = ppir_node_to_branch(node);
+ ppir_node_add_src(block->comp, node, &else_branch->src[0],
+ &if_stmt->condition, 1);
+ else_branch->num_src = 1;
+ /* Negate condition to minimize branching. We're generating following:
+ * current_block: { ...; if (!statement) branch else_block; }
+ * then_block: { ...; branch after_block; }
+ * else_block: { ... }
+ * after_block: { ... }
+ *
+ * or if else list is empty:
+ * block: { if (!statement) branch else_block; }
+ * then_block: { ... }
+ * else_block: after_block: { ... }
+ */
+ else_branch->negate = true;
+ list_addtail(&else_branch->node.list, &block->node_list);
+
+ ppir_emit_cf_list(comp, &if_stmt->then_list);
+ if (empty_else_block) {
+ nir_block *nblock = nir_if_last_else_block(if_stmt);
+ assert(nblock->successors[0]);
+ assert(!nblock->successors[1]);
+ else_branch->target = ppir_get_block(comp, nblock->successors[0]);
+ /* Add empty else block to the list */
+ list_addtail(&block->successors[1]->list, &comp->block_list);
+ return true;
+ }
+
+ else_branch->target = ppir_get_block(comp, nir_if_first_else_block(if_stmt));
+
+ nir_block *last_then_block = nir_if_last_then_block(if_stmt);
+ assert(last_then_block->successors[0]);
+ assert(!last_then_block->successors[1]);
+ block = ppir_get_block(comp, last_then_block);
+ node = ppir_node_create(block, ppir_op_branch, -1, 0);
+ if (!node)
+ return false;
+ after_branch = ppir_node_to_branch(node);
+ /* Unconditional */
+ after_branch->num_src = 0;
+ after_branch->target = ppir_get_block(comp, last_then_block->successors[0]);
+ /* Target should be after_block, will fixup later */
+ list_addtail(&after_branch->node.list, &block->node_list);
+
+ ppir_emit_cf_list(comp, &if_stmt->else_list);
+
+ return true;
}
static bool ppir_emit_loop(ppir_compiler *comp, nir_loop *nloop)
{
- ppir_error("loop nir_cf_node not support\n");
- return false;
+ ppir_block *save_loop_cont_block = comp->loop_cont_block;
+ ppir_block *block;
+ ppir_branch_node *loop_branch;
+ nir_block *loop_last_block;
+ ppir_node *node;
+
+ comp->loop_cont_block = ppir_get_block(comp, nir_loop_first_block(nloop));
+
+ ppir_emit_cf_list(comp, &nloop->body);
+
+ loop_last_block = nir_loop_last_block(nloop);
+ block = ppir_get_block(comp, loop_last_block);
+ node = ppir_node_create(block, ppir_op_branch, -1, 0);
+ if (!node)
+ return false;
+ loop_branch = ppir_node_to_branch(node);
+ /* Unconditional */
+ loop_branch->num_src = 0;
+ loop_branch->target = comp->loop_cont_block;
+ list_addtail(&loop_branch->node.list, &block->node_list);
+
+ comp->loop_cont_block = save_loop_cont_block;
+
+ comp->num_loops++;
+
+ return true;
}
static bool ppir_emit_function(ppir_compiler *comp, nir_function_impl *nfunc)
ppir_node *prev_node = NULL;
list_for_each_entry_rev(ppir_node, node, &block->node_list, list) {
if (prev_node && ppir_node_is_root(node) && node->op != ppir_op_const) {
- ppir_node_add_dep(prev_node, node);
+ ppir_node_add_dep(prev_node, node, ppir_dep_sequence);
}
if (node->op == ppir_op_discard ||
node->op == ppir_op_store_color ||
ppir_src *src = ppir_node_get_src(node, i);
if (src && src->type == ppir_target_register &&
src->reg == reg &&
- write)
- ppir_node_add_dep(write, node);
+ write) {
+ ppir_debug("Adding dep %d for write %d\n", node->index, write->index);
+ ppir_node_add_dep(write, node, ppir_dep_write_after_read);
+ }
}
ppir_dest *dest = ppir_node_get_dest(node);
if (dest && dest->type == ppir_target_register &&
}
}
+ /* Validate outputs, we support only gl_FragColor */
+ nir_foreach_variable(var, &nir->outputs) {
+ switch (var->data.location) {
+ case FRAG_RESULT_COLOR:
+ case FRAG_RESULT_DATA0:
+ break;
+ default:
+ ppir_error("unsupported output type\n");
+ goto err_out0;
+ break;
+ }
+ }
+
foreach_list_typed(nir_register, reg, node, &func->registers) {
ppir_reg *r = rzalloc(comp, ppir_reg);
if (!r)
r->index = reg->index;
r->num_components = reg->num_components;
- r->live_in = INT_MAX;
- r->live_out = 0;
r->is_head = false;
list_addtail(&r->list, &comp->reg_list);
}
projects / mesa.git / blobdiff