numam-dpdk/app/test-compress-perf/comp_perf_test_cyclecount.c
Artur Trybula 2695db95a1 test/compress: add cycle-count mode to perf tool
This commit adds cycle-count mode to the compression perf tool.
The new mode enhances the compression performance tool to allow
cycle-count measurement of both hardware and softwate PMDs.

Signed-off-by: Artur Trybula <arturx.trybula@intel.com>
Acked-by: Fiona Trahe <fiona.trahe@intel.com>
2020-02-05 15:20:51 +01:00

615 lines
15 KiB
C

/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(c) 2019 Intel Corporation
*/
#include <rte_malloc.h>
#include <rte_eal.h>
#include <rte_log.h>
#include <rte_cycles.h>
#include "rte_spinlock.h"
#include <rte_compressdev.h>
#include "comp_perf_test_cyclecount.h"
struct cperf_cyclecount_ctx {
struct cperf_verify_ctx ver;
uint32_t ops_enq_retries;
uint32_t ops_deq_retries;
uint64_t duration_op;
uint64_t duration_enq;
uint64_t duration_deq;
};
void
cperf_cyclecount_test_destructor(void *arg)
{
struct cperf_cyclecount_ctx *ctx = arg;
if (arg) {
comp_perf_free_memory(ctx->ver.options, &ctx->ver.mem);
rte_free(arg);
}
}
void *
cperf_cyclecount_test_constructor(uint8_t dev_id, uint16_t qp_id,
struct comp_test_data *options)
{
struct cperf_cyclecount_ctx *ctx = NULL;
ctx = rte_malloc(NULL, sizeof(struct cperf_cyclecount_ctx), 0);
if (ctx == NULL)
return NULL;
ctx->ver.mem.dev_id = dev_id;
ctx->ver.mem.qp_id = qp_id;
ctx->ver.options = options;
ctx->ver.silent = 1; /* ver. part will be silent */
if (!comp_perf_allocate_memory(ctx->ver.options, &ctx->ver.mem)
&& !prepare_bufs(ctx->ver.options, &ctx->ver.mem))
return ctx;
cperf_cyclecount_test_destructor(ctx);
return NULL;
}
static int
cperf_cyclecount_op_setup(struct rte_comp_op **ops,
struct cperf_cyclecount_ctx *ctx,
struct rte_mbuf **input_bufs,
struct rte_mbuf **output_bufs,
void *priv_xform,
uint32_t out_seg_sz)
{
struct comp_test_data *test_data = ctx->ver.options;
struct cperf_mem_resources *mem = &ctx->ver.mem;
uint32_t i, iter, num_iter;
int res = 0;
uint16_t ops_needed;
num_iter = test_data->num_iter;
for (iter = 0; iter < num_iter; iter++) {
uint32_t remaining_ops = mem->total_bufs;
uint32_t total_deq_ops = 0;
uint32_t total_enq_ops = 0;
uint16_t num_enq = 0;
uint16_t num_deq = 0;
while (remaining_ops > 0) {
uint16_t num_ops = RTE_MIN(remaining_ops,
test_data->burst_sz);
ops_needed = num_ops;
/* Allocate compression operations */
if (ops_needed && rte_mempool_get_bulk(
mem->op_pool,
(void **)ops,
ops_needed) != 0) {
RTE_LOG(ERR, USER1,
"Cyclecount: could not allocate enough operations\n");
res = -1;
goto end;
}
for (i = 0; i < ops_needed; i++) {
/* Calculate next buffer to attach */
/* to operation */
uint32_t buf_id = total_enq_ops + i;
uint16_t op_id = i;
/* Reset all data in output buffers */
struct rte_mbuf *m = output_bufs[buf_id];
m->pkt_len = out_seg_sz * m->nb_segs;
while (m) {
m->data_len = m->buf_len - m->data_off;
m = m->next;
}
ops[op_id]->m_src = input_bufs[buf_id];
ops[op_id]->m_dst = output_bufs[buf_id];
ops[op_id]->src.offset = 0;
ops[op_id]->src.length =
rte_pktmbuf_pkt_len(input_bufs[buf_id]);
ops[op_id]->dst.offset = 0;
ops[op_id]->flush_flag = RTE_COMP_FLUSH_FINAL;
ops[op_id]->input_chksum = buf_id;
ops[op_id]->private_xform = priv_xform;
}
/* E N Q U E U I N G */
/* assuming that all ops are enqueued */
/* instead of the real enqueue operation */
num_enq = num_ops;
remaining_ops -= num_enq;
total_enq_ops += num_enq;
/* D E Q U E U I N G */
/* assuming that all ops dequeued */
/* instead of the real dequeue operation */
num_deq = num_ops;
total_deq_ops += num_deq;
rte_mempool_put_bulk(mem->op_pool,
(void **)ops, num_deq);
}
}
return res;
end:
rte_mempool_put_bulk(mem->op_pool, (void **)ops, ops_needed);
rte_free(ops);
return res;
}
static int
main_loop(struct cperf_cyclecount_ctx *ctx, enum rte_comp_xform_type type)
{
struct comp_test_data *test_data = ctx->ver.options;
struct cperf_mem_resources *mem = &ctx->ver.mem;
uint8_t dev_id = mem->dev_id;
uint32_t i, iter, num_iter;
struct rte_comp_op **ops, **deq_ops;
void *priv_xform = NULL;
struct rte_comp_xform xform;
struct rte_mbuf **input_bufs, **output_bufs;
int ret, res = 0;
int allocated = 0;
uint32_t out_seg_sz;
uint64_t tsc_start, tsc_end, tsc_duration;
if (test_data == NULL || !test_data->burst_sz) {
RTE_LOG(ERR, USER1, "Unknown burst size\n");
return -1;
}
ctx->duration_enq = 0;
ctx->duration_deq = 0;
ctx->ops_enq_retries = 0;
ctx->ops_deq_retries = 0;
/* one array for both enqueue and dequeue */
ops = rte_zmalloc_socket(NULL,
2 * mem->total_bufs * sizeof(struct rte_comp_op *),
0, rte_socket_id());
if (ops == NULL) {
RTE_LOG(ERR, USER1,
"Can't allocate memory for ops strucures\n");
return -1;
}
deq_ops = &ops[mem->total_bufs];
if (type == RTE_COMP_COMPRESS) {
xform = (struct rte_comp_xform) {
.type = RTE_COMP_COMPRESS,
.compress = {
.algo = RTE_COMP_ALGO_DEFLATE,
.deflate.huffman = test_data->huffman_enc,
.level = test_data->level,
.window_size = test_data->window_sz,
.chksum = RTE_COMP_CHECKSUM_NONE,
.hash_algo = RTE_COMP_HASH_ALGO_NONE
}
};
input_bufs = mem->decomp_bufs;
output_bufs = mem->comp_bufs;
out_seg_sz = test_data->out_seg_sz;
} else {
xform = (struct rte_comp_xform) {
.type = RTE_COMP_DECOMPRESS,
.decompress = {
.algo = RTE_COMP_ALGO_DEFLATE,
.chksum = RTE_COMP_CHECKSUM_NONE,
.window_size = test_data->window_sz,
.hash_algo = RTE_COMP_HASH_ALGO_NONE
}
};
input_bufs = mem->comp_bufs;
output_bufs = mem->decomp_bufs;
out_seg_sz = test_data->seg_sz;
}
/* Create private xform */
if (rte_compressdev_private_xform_create(dev_id, &xform,
&priv_xform) < 0) {
RTE_LOG(ERR, USER1, "Private xform could not be created\n");
res = -1;
goto end;
}
tsc_start = rte_rdtsc_precise();
ret = cperf_cyclecount_op_setup(ops,
ctx,
input_bufs,
output_bufs,
priv_xform,
out_seg_sz);
tsc_end = rte_rdtsc_precise();
/* ret value check postponed a bit to cancel extra 'if' bias */
if (ret < 0) {
RTE_LOG(ERR, USER1, "Setup function failed\n");
res = -1;
goto end;
}
tsc_duration = tsc_end - tsc_start;
ctx->duration_op = tsc_duration;
num_iter = test_data->num_iter;
for (iter = 0; iter < num_iter; iter++) {
uint32_t total_ops = mem->total_bufs;
uint32_t remaining_ops = mem->total_bufs;
uint32_t total_deq_ops = 0;
uint32_t total_enq_ops = 0;
uint16_t ops_unused = 0;
uint16_t num_enq = 0;
uint16_t num_deq = 0;
while (remaining_ops > 0) {
uint16_t num_ops = RTE_MIN(remaining_ops,
test_data->burst_sz);
uint16_t ops_needed = num_ops - ops_unused;
/*
* Move the unused operations from the previous
* enqueue_burst call to the front, to maintain order
*/
if ((ops_unused > 0) && (num_enq > 0)) {
size_t nb_b_to_mov =
ops_unused * sizeof(struct rte_comp_op *);
memmove(ops, &ops[num_enq], nb_b_to_mov);
}
/* Allocate compression operations */
if (ops_needed && rte_mempool_get_bulk(
mem->op_pool,
(void **)ops,
ops_needed) != 0) {
RTE_LOG(ERR, USER1,
"Could not allocate enough operations\n");
res = -1;
goto end;
}
allocated += ops_needed;
for (i = 0; i < ops_needed; i++) {
/*
* Calculate next buffer to attach to operation
*/
uint32_t buf_id = total_enq_ops + i +
ops_unused;
uint16_t op_id = ops_unused + i;
/* Reset all data in output buffers */
struct rte_mbuf *m = output_bufs[buf_id];
m->pkt_len = out_seg_sz * m->nb_segs;
while (m) {
m->data_len = m->buf_len - m->data_off;
m = m->next;
}
ops[op_id]->m_src = input_bufs[buf_id];
ops[op_id]->m_dst = output_bufs[buf_id];
ops[op_id]->src.offset = 0;
ops[op_id]->src.length =
rte_pktmbuf_pkt_len(input_bufs[buf_id]);
ops[op_id]->dst.offset = 0;
ops[op_id]->flush_flag = RTE_COMP_FLUSH_FINAL;
ops[op_id]->input_chksum = buf_id;
ops[op_id]->private_xform = priv_xform;
}
if (unlikely(test_data->perf_comp_force_stop))
goto end;
tsc_start = rte_rdtsc_precise();
num_enq = rte_compressdev_enqueue_burst(dev_id,
mem->qp_id, ops,
num_ops);
tsc_end = rte_rdtsc_precise();
tsc_duration = tsc_end - tsc_start;
ctx->duration_enq += tsc_duration;
if (num_enq < num_ops)
ctx->ops_enq_retries++;
if (test_data->cyclecount_delay)
rte_delay_us_block(test_data->cyclecount_delay);
if (num_enq == 0) {
struct rte_compressdev_stats stats;
rte_compressdev_stats_get(dev_id, &stats);
if (stats.enqueue_err_count) {
res = -1;
goto end;
}
}
ops_unused = num_ops - num_enq;
remaining_ops -= num_enq;
total_enq_ops += num_enq;
tsc_start = rte_rdtsc_precise();
num_deq = rte_compressdev_dequeue_burst(dev_id,
mem->qp_id,
deq_ops,
allocated);
tsc_end = rte_rdtsc_precise();
tsc_duration = tsc_end - tsc_start;
ctx->duration_deq += tsc_duration;
if (num_deq < allocated)
ctx->ops_deq_retries++;
total_deq_ops += num_deq;
if (iter == num_iter - 1) {
for (i = 0; i < num_deq; i++) {
struct rte_comp_op *op = deq_ops[i];
if (op->status !=
RTE_COMP_OP_STATUS_SUCCESS) {
RTE_LOG(ERR, USER1, "Some operations were not successful\n");
goto end;
}
struct rte_mbuf *m = op->m_dst;
m->pkt_len = op->produced;
uint32_t remaining_data = op->produced;
uint16_t data_to_append;
while (remaining_data > 0) {
data_to_append =
RTE_MIN(remaining_data,
out_seg_sz);
m->data_len = data_to_append;
remaining_data -=
data_to_append;
m = m->next;
}
}
}
rte_mempool_put_bulk(mem->op_pool,
(void **)deq_ops, num_deq);
allocated -= num_deq;
}
/* Dequeue the last operations */
while (total_deq_ops < total_ops) {
if (unlikely(test_data->perf_comp_force_stop))
goto end;
tsc_start = rte_rdtsc_precise();
num_deq = rte_compressdev_dequeue_burst(dev_id,
mem->qp_id,
deq_ops,
test_data->burst_sz);
tsc_end = rte_rdtsc_precise();
tsc_duration = tsc_end - tsc_start;
ctx->duration_deq += tsc_duration;
ctx->ops_deq_retries++;
if (num_deq == 0) {
struct rte_compressdev_stats stats;
rte_compressdev_stats_get(dev_id, &stats);
if (stats.dequeue_err_count) {
res = -1;
goto end;
}
}
total_deq_ops += num_deq;
if (iter == num_iter - 1) {
for (i = 0; i < num_deq; i++) {
struct rte_comp_op *op = deq_ops[i];
if (op->status !=
RTE_COMP_OP_STATUS_SUCCESS) {
RTE_LOG(ERR, USER1, "Some operations were not successful\n");
goto end;
}
struct rte_mbuf *m = op->m_dst;
m->pkt_len = op->produced;
uint32_t remaining_data = op->produced;
uint16_t data_to_append;
while (remaining_data > 0) {
data_to_append =
RTE_MIN(remaining_data,
out_seg_sz);
m->data_len = data_to_append;
remaining_data -=
data_to_append;
m = m->next;
}
}
}
rte_mempool_put_bulk(mem->op_pool,
(void **)deq_ops, num_deq);
allocated -= num_deq;
}
}
allocated = 0;
end:
if (allocated)
rte_mempool_put_bulk(mem->op_pool, (void **)ops, allocated);
rte_compressdev_private_xform_free(dev_id, priv_xform);
rte_free(ops);
if (test_data->perf_comp_force_stop) {
RTE_LOG(ERR, USER1,
"lcore: %d Perf. test has been aborted by user\n",
mem->lcore_id);
res = -1;
}
return res;
}
int
cperf_cyclecount_test_runner(void *test_ctx)
{
struct cperf_cyclecount_ctx *ctx = test_ctx;
struct comp_test_data *test_data = ctx->ver.options;
uint32_t lcore = rte_lcore_id();
static rte_atomic16_t display_once = RTE_ATOMIC16_INIT(0);
static rte_spinlock_t print_spinlock;
int i;
uint32_t ops_enq_retries_comp;
uint32_t ops_deq_retries_comp;
uint32_t ops_enq_retries_decomp;
uint32_t ops_deq_retries_decomp;
uint32_t duration_setup_per_op;
uint32_t duration_enq_per_op_comp;
uint32_t duration_deq_per_op_comp;
uint32_t duration_enq_per_op_decomp;
uint32_t duration_deq_per_op_decomp;
ctx->ver.mem.lcore_id = lcore;
/*
* printing information about current compression thread
*/
if (rte_atomic16_test_and_set(&ctx->ver.mem.print_info_once))
printf(" lcore: %u,"
" driver name: %s,"
" device name: %s,"
" device id: %u,"
" socket id: %u,"
" queue pair id: %u\n",
lcore,
ctx->ver.options->driver_name,
rte_compressdev_name_get(ctx->ver.mem.dev_id),
ctx->ver.mem.dev_id,
rte_compressdev_socket_id(ctx->ver.mem.dev_id),
ctx->ver.mem.qp_id);
/*
* First the verification part is needed
*/
if (cperf_verify_test_runner(&ctx->ver))
return EXIT_FAILURE;
/*
* Run the tests twice, discarding the first performance
* results, before the cache is warmed up
*/
/* C O M P R E S S */
for (i = 0; i < 2; i++) {
if (main_loop(ctx, RTE_COMP_COMPRESS) < 0)
return EXIT_FAILURE;
}
ops_enq_retries_comp = ctx->ops_enq_retries;
ops_deq_retries_comp = ctx->ops_deq_retries;
duration_enq_per_op_comp = ctx->duration_enq /
(ctx->ver.mem.total_bufs * test_data->num_iter);
duration_deq_per_op_comp = ctx->duration_deq /
(ctx->ver.mem.total_bufs * test_data->num_iter);
/* D E C O M P R E S S */
for (i = 0; i < 2; i++) {
if (main_loop(ctx, RTE_COMP_DECOMPRESS) < 0)
return EXIT_FAILURE;
}
ops_enq_retries_decomp = ctx->ops_enq_retries;
ops_deq_retries_decomp = ctx->ops_deq_retries;
duration_enq_per_op_decomp = ctx->duration_enq /
(ctx->ver.mem.total_bufs * test_data->num_iter);
duration_deq_per_op_decomp = ctx->duration_deq /
(ctx->ver.mem.total_bufs * test_data->num_iter);
duration_setup_per_op = ctx->duration_op /
(ctx->ver.mem.total_bufs * test_data->num_iter);
/* R E P O R T processing */
if (rte_atomic16_test_and_set(&display_once)) {
rte_spinlock_lock(&print_spinlock);
printf("\nLegend for the table\n"
" - Retries section: number of retries for the following operations:\n"
" [C-e] - compression enqueue\n"
" [C-d] - compression dequeue\n"
" [D-e] - decompression enqueue\n"
" [D-d] - decompression dequeue\n"
" - Cycles section: number of cycles per 'op' for the following operations:\n"
" setup/op - memory allocation, op configuration and memory dealocation\n"
" [C-e] - compression enqueue\n"
" [C-d] - compression dequeue\n"
" [D-e] - decompression enqueue\n"
" [D-d] - decompression dequeue\n\n");
printf("\n%12s%6s%12s%17s",
"lcore id", "Level", "Comp size", "Comp ratio [%]");
printf(" |%10s %6s %8s %6s %8s",
" Retries:",
"[C-e]", "[C-d]",
"[D-e]", "[D-d]");
printf(" |%9s %9s %9s %9s %9s %9s\n",
" Cycles:",
"setup/op",
"[C-e]", "[C-d]",
"[D-e]", "[D-d]");
rte_spinlock_unlock(&print_spinlock);
}
rte_spinlock_lock(&print_spinlock);
printf("%12u"
"%6u"
"%12zu"
"%17.2f",
ctx->ver.mem.lcore_id,
test_data->level,
ctx->ver.comp_data_sz,
ctx->ver.ratio);
printf(" |%10s %6u %8u %6u %8u",
" ",
ops_enq_retries_comp,
ops_deq_retries_comp,
ops_enq_retries_decomp,
ops_deq_retries_decomp);
printf(" |%9s %9u %9u %9u %9u %9u\n",
" ",
duration_setup_per_op,
duration_enq_per_op_comp,
duration_deq_per_op_comp,
duration_enq_per_op_decomp,
duration_deq_per_op_decomp);
rte_spinlock_unlock(&print_spinlock);
return EXIT_SUCCESS;
}