From: Erico Nunes Date: Sun, 21 Jul 2019 11:30:34 +0000 (+0200) Subject: lima/ppir: support pipeline registers in scheduler X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=eb82637c2f9cad0c84d09502b7e65166448c0475;p=mesa.git lima/ppir: support pipeline registers in scheduler The ppir scheduler grew to be rather complicated and containing many exceptions as it also has to take care of inserting additional nodes when it is mandatory for nodes to be in the same instruction. As such, the lima lowering and scheduling process can be difficult to understand and maintain. The ppir lowering step created nodes hoping that the scheduler would notice the exception and do the right thing. This proposal adds a simple refactor to the scheduler so that it places nodes with pipeline registers in the same instruction. With the scheduler handling this in a general way, it is possible to create same-instruction dependencies by using pipeline registers during the lowering stage. This is simpler to maintain because now we can make these dependencies explicit in a single place (lowering), and we can drop exceptions from scheduling. Reducing the complexity of the scheduler is also useful as preparatory work to support control flow in ppir. Signed-off-by: Erico Nunes Reviewed-by: Vasily Khoruzhick Reviewed-by: Qiang Yu --- diff --git a/src/gallium/drivers/lima/ir/pp/lower.c b/src/gallium/drivers/lima/ir/pp/lower.c index 2b4da2c2992..c3be6acfe8f 100644 --- a/src/gallium/drivers/lima/ir/pp/lower.c +++ b/src/gallium/drivers/lima/ir/pp/lower.c @@ -112,6 +112,30 @@ static bool ppir_lower_texture(ppir_block *block, ppir_node *node) } ppir_node_add_dep(node, &load->node); + + /* Create move node */ + ppir_node *move = ppir_node_create(block, ppir_op_mov, -1 , 0); + if (unlikely(!move)) + return false; + + ppir_alu_node *alu = ppir_node_to_alu(move); + + ppir_dest *dest = ppir_node_get_dest(node); + alu->dest = *dest; + + ppir_node_replace_all_succ(move, node); + + dest->type = ppir_target_pipeline; + dest->pipeline = ppir_pipeline_reg_sampler; + + alu->num_src = 1; + ppir_node_target_assign(&alu->src[0], dest); + for (int i = 0; i < 4; i++) + alu->src->swizzle[i] = i; + + ppir_node_add_dep(move, node); + list_addtail(&move->list, &node->list); + return true; } diff --git a/src/gallium/drivers/lima/ir/pp/node_to_instr.c b/src/gallium/drivers/lima/ir/pp/node_to_instr.c index baf5b0d815a..dbc1c9c32ce 100644 --- a/src/gallium/drivers/lima/ir/pp/node_to_instr.c +++ b/src/gallium/drivers/lima/ir/pp/node_to_instr.c @@ -37,45 +37,6 @@ static bool create_new_instr(ppir_block *block, ppir_node *node) return true; } -static bool insert_to_load_tex(ppir_block *block, ppir_node *load_coords, ppir_node *ldtex) -{ - ppir_dest *dest = ppir_node_get_dest(ldtex); - ppir_node *move = NULL; - - /* Insert load_coords to ldtex instruction */ - if (!ppir_instr_insert_node(ldtex->instr, load_coords)) - return false; - - /* Create move node */ - move = ppir_node_create(block, ppir_op_mov, -1 , 0); - if (unlikely(!move)) - return false; - - ppir_debug("insert_load_tex: create move %d for %d\n", - move->index, ldtex->index); - - ppir_alu_node *alu = ppir_node_to_alu(move); - alu->dest = *dest; - - ppir_node_replace_all_succ(move, ldtex); - - dest->type = ppir_target_pipeline; - dest->pipeline = ppir_pipeline_reg_sampler; - - alu->num_src = 1; - ppir_node_target_assign(&alu->src[0], dest); - for (int i = 0; i < 4; i++) - alu->src->swizzle[i] = i; - - ppir_node_add_dep(move, ldtex); - list_addtail(&move->list, &ldtex->list); - - if (!ppir_instr_insert_node(ldtex->instr, move)) - return false; - - return true; -} - static bool insert_to_each_succ_instr(ppir_block *block, ppir_node *node) { ppir_dest *dest = ppir_node_get_dest(node); @@ -176,7 +137,31 @@ static bool insert_to_each_succ_instr(ppir_block *block, ppir_node *node) return true; } -static bool ppir_do_node_to_instr(ppir_block *block, ppir_node *node) +/* + * If a node has a pipeline dest, schedule it in the same instruction as its + * successor. + * Since it has a pipeline dest, it must have only one successor and since we + * schedule nodes backwards, its successor must have already been scheduled. + */ +static bool ppir_do_node_to_instr_pipeline(ppir_block *block, ppir_node *node) +{ + ppir_dest *dest = ppir_node_get_dest(node); + + if (!dest || dest->type != ppir_target_pipeline) + return false; + + assert(ppir_node_has_single_succ(node)); + ppir_node *succ = ppir_node_first_succ(node); + assert(succ); + assert(succ->instr); + + if (!ppir_instr_insert_node(succ->instr, node)) + return false; + + return true; +} + +static bool ppir_do_one_node_to_instr(ppir_block *block, ppir_node *node, ppir_node **next) { switch (node->type) { case ppir_node_type_alu: @@ -239,11 +224,6 @@ static bool ppir_do_node_to_instr(ppir_block *block, ppir_node *node) if (!create_new_instr(block, node)) return false; } - else if (node->op == ppir_op_load_coords) { - ppir_node *ldtex = ppir_node_first_succ(node); - if (!insert_to_load_tex(block, node, ldtex)) - return false; - } else { /* not supported yet */ assert(0); @@ -315,7 +295,7 @@ static bool ppir_do_node_to_instr(ppir_block *block, ppir_node *node) node->instr = move->instr; /* use move for the following recursion */ - node = move; + *next = move; break; } case ppir_node_type_discard: @@ -331,6 +311,21 @@ static bool ppir_do_node_to_instr(ppir_block *block, ppir_node *node) return false; } + return true; +} + +static bool ppir_do_node_to_instr(ppir_block *block, ppir_node *node) +{ + ppir_node *next = node; + + /* first try pipeline sched, if that didn't succeed try normal scheduling */ + if (!ppir_do_node_to_instr_pipeline(block, node)) + if (!ppir_do_one_node_to_instr(block, node, &next)) + return false; + + /* next may have been updated in ppir_do_one_node_to_instr */ + node = next; + /* we have to make sure the dep not be destroyed (due to * succ change) in ppir_do_node_to_instr, otherwise we can't * do recursion like this */