d/numam-dpdk
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

pipeline: enable pipeline compilation

Browse Source 
Commit the pipeline changes when the compilation process is
successful: change the table lookup instructions to execute the action
function for each action, replace the regular pipeline instructions
with the custom instructions.

Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
This commit is contained in:
Cristian Dumitrescu 2021-09-13 17:44:43 +01:00 committed by Thomas Monjalon
parent f898a475c3
commit 1c6571c837
1 changed files with 94 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

94
lib/pipeline/rte_swx_pipeline.c
View File

@ -12177,6 +12177,26 @@ instruction_group_list_codegen(struct instruction_group_list *igl,
}
}
static uint32_t
instruction_group_list_custom_instructions_count(struct instruction_group_list *igl)
{
struct instruction_group *g;
uint32_t n_custom_instr = 0;
/* Groups with a single instruction: no function is generated for this group, the group
* keeps its current instruction. Groups with more than two instructions: one function and
* the associated custom instruction get generated for each such group.
*/
TAILQ_FOREACH(g, igl, node) {
if (g->first_instr_id == g->last_instr_id)
continue;
n_custom_instr++;
}
return n_custom_instr;
}
static int
pipeline_codegen(struct rte_swx_pipeline *p, struct instruction_group_list *igl)
{
@ -12331,6 +12351,73 @@ free:
return status;
}
static int
pipeline_adjust_check(struct rte_swx_pipeline *p __rte_unused,
struct instruction_group_list *igl)
{
uint32_t n_custom_instr = instruction_group_list_custom_instructions_count(igl);
/* Check that enough space is available within the pipeline instruction table to store all
* the custom instructions.
*/
if (INSTR_CUSTOM_0 + n_custom_instr > RTE_SWX_PIPELINE_INSTRUCTION_TABLE_SIZE_MAX)
return -ENOSPC;
return 0;
}
static void
pipeline_adjust(struct rte_swx_pipeline *p, struct instruction_group_list *igl)
{
struct instruction_group *g;
uint32_t i;
/* Pipeline table instructions. */
for (i = 0; i < p->n_instructions; i++) {
struct instruction *instr = &p->instructions[i];
if (instr->type == INSTR_TABLE)
instr->type = INSTR_TABLE_AF;
if (instr->type == INSTR_LEARNER)
instr->type = INSTR_LEARNER_AF;
}
/* Pipeline custom instructions. */
i = 0;
TAILQ_FOREACH(g, igl, node) {
struct instruction *instr = &p->instructions[g->first_instr_id];
uint32_t j;
if (g->first_instr_id == g->last_instr_id)
continue;
/* Install a new custom instruction. */
p->instruction_table[INSTR_CUSTOM_0 + i] = g->func;
/* First instruction of the group: change its type to the new custom instruction. */
instr->type = INSTR_CUSTOM_0 + i;
/* All the subsequent instructions of the group: invalidate. */
for (j = g->first_instr_id + 1; j <= g->last_instr_id; j++) {
struct instruction_data *data = &p->instruction_data[j];
data->invalid = 1;
}
i++;
}
/* Remove the invalidated instructions. */
p->n_instructions = instr_compact(p->instructions, p->instruction_data, p->n_instructions);
/* Resolve the jump destination for any "standalone" jump instructions (i.e. those jump
* instructions that are the only instruction within their group, so they were left
* unmodified).
*/
instr_jmp_resolve(p->instructions, p->instruction_data, p->n_instructions);
}
static int
pipeline_compile(struct rte_swx_pipeline *p)
{
@ -12353,6 +12440,13 @@ pipeline_compile(struct rte_swx_pipeline *p)
if (status)
goto free;
/* Adjust instructions. */
status = pipeline_adjust_check(p, igl);
if (status)
goto free;
pipeline_adjust(p, igl);
free:
instruction_group_list_free(igl);