|
|
|
|
|
|
|
|
|
|
|
|
|
import argparse |
|
import bisect |
|
import codecs |
|
import collections |
|
import os |
|
import sys |
|
import yaml |
|
import zlib |
|
|
|
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) |
|
from primes import next_prime |
|
import xngen |
|
import xnncommon |
|
|
|
parser = argparse.ArgumentParser(description="XNNPACK generator") |
|
parser.add_argument( |
|
"-s", "--spec", metavar="FILE", required=True, help="Spec (YAML) file") |
|
parser.add_argument( |
|
"-o", |
|
"--output", |
|
action="append", |
|
metavar="FILE", |
|
required=True, |
|
help="Output (C++ source) file(s)") |
|
parser.set_defaults(defines=list()) |
|
|
|
def split_ukernel_name(name): |
|
common_name, target_name = name.split("__", 1) |
|
common_parts = common_name.split("_") |
|
xw = "gemm_xw_" in common_name |
|
param_spec = common_parts[-1] |
|
if "s" in param_spec: |
|
param_spec, sr = param_spec.split("s", 1) |
|
sr = int(sr) |
|
else: |
|
sr = 1 |
|
if "c" in param_spec: |
|
param_spec, kr = param_spec.split("c", 1) |
|
kr = int(kr) |
|
else: |
|
kr = 1 |
|
mr, nr = map(int, param_spec.split("x")) |
|
arch, isa, assembly = xnncommon.parse_target_name(target_name) |
|
|
|
requantization = common_parts[-3] |
|
if requantization not in ["fp32", "rndnu"]: |
|
requantization = None |
|
|
|
return mr, nr, kr, sr, xw, requantization, arch, isa, assembly |
|
|
|
|
|
GEMM_TEST_CODE = """\ |
|
TEST(${TEST_NAME}, k_eq_${KBLOCK}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
TEST(${TEST_NAME}, strided_cn) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.cn_stride(${next_prime(NR + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
$if UKERNEL_TYPE != "IGEMM": |
|
TEST(${TEST_NAME}, k_eq_${KBLOCK}_strided_a) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.a_stride(${next_prime(KBLOCK + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
TEST(${TEST_NAME}, k_eq_${KBLOCK}_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(${KBLOCK}) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, k_eq_${KBLOCK}_subtile_m) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
|
|
TEST(${TEST_NAME}, k_eq_${KBLOCK}_subtile_n) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(${KBLOCK}) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if IS_PIPELINED: |
|
TEST(${TEST_NAME}, k_eq_${KBLOCK * 2}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK * 2}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
$if UKERNEL_TYPE != "IGEMM": |
|
TEST(${TEST_NAME}, k_eq_${KBLOCK * 2}_strided_a) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK * 2}) |
|
.a_stride(${next_prime(KBLOCK * 2 + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
TEST(${TEST_NAME}, k_eq_${KBLOCK * 2}_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(${KBLOCK * 2}) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
$if KBLOCK > 1: |
|
TEST(${TEST_NAME}, k_lt_${ADJKBLOCK}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k < ${ADJKBLOCK}; k++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if UKERNEL_TYPE != "IGEMM": |
|
TEST(${TEST_NAME}, k_lt_${ADJKBLOCK}_strided_a) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k < ${ADJKBLOCK}; k++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.a_stride(${next_prime(ADJKBLOCK + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, k_lt_${ADJKBLOCK}_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k < ${ADJKBLOCK}; k++) { |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(k) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, k_gt_${ADJKBLOCK}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = ${ADJKBLOCK + 1}; k < ${ADJKBLOCK * 10 if ADJKBLOCK == 1 else ADJKBLOCK * 2}; k++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if UKERNEL_TYPE.startswith("GEMM"): |
|
TEST(${TEST_NAME}, k_gt_${ADJKBLOCK}_strided_a) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = ${ADJKBLOCK + 1}; k < ${10 if ADJKBLOCK == 1 else ADJKBLOCK * 2}; k++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.a_stride(${next_prime(10 if ADJKBLOCK == 1 else ADJKBLOCK * 2 + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, k_gt_${ADJKBLOCK}_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = ${ADJKBLOCK + 1}; k < ${10 if ADJKBLOCK == 1 else ADJKBLOCK * 2}; k++) { |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(k) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
$if KBLOCK > 1: |
|
TEST(${TEST_NAME}, k_div_${KBLOCK}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = ${ADJKBLOCK + KBLOCK}; k <= ${KBLOCK * 10}; k += ${KBLOCK}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if UKERNEL_TYPE.startswith("GEMM"): |
|
TEST(${TEST_NAME}, k_div_${KBLOCK}_strided_a) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = ${ADJKBLOCK + KBLOCK}; k <= ${KBLOCK * 10}; k += ${KBLOCK}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.a_stride(${next_prime(KBLOCK * 10 + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, k_div_${KBLOCK}_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = ${ADJKBLOCK + KBLOCK}; k <= ${KBLOCK * 10}; k += ${KBLOCK}) { |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(k) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_gt_${NR}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${NR + 1}; n < ${NR * 2}; n++) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_gt_${NR}_strided_cn) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${NR + 1}; n < ${NR * 2}; n++) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
.cn_stride(${next_prime(NR + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
$if UKERNEL_TYPE != "IGEMM": |
|
TEST(${TEST_NAME}, n_gt_${NR}_strided_a) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${NR + 1}; n < ${NR * 2}; n++) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
.a_stride(${next_prime(KBLOCK * 5 + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_gt_${NR}_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${NR + 1}; n < ${NR * 2}; n++) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(k) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_div_${NR}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${2 * NR}; n <= ${3 * NR}; n += ${NR}) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_div_${NR}_strided_cn) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${2 * NR}; n <= ${3 * NR}; n += ${NR}) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
.cn_stride(${next_prime(NR + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
$if UKERNEL_TYPE != "IGEMM": |
|
TEST(${TEST_NAME}, n_div_${NR}_strided_a) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${2 * NR}; n <= ${3 * NR}; n += ${NR}) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
.a_stride(${next_prime(KBLOCK * 5 + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_div_${NR}_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${2 * NR}; n <= ${3 * NR}; n += ${NR}) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(k) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
$if UKERNEL_TYPE.startswith("IGEMM"): |
|
TEST(${TEST_NAME}, small_kernel) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.ks(3) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, small_kernel_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(k) |
|
.ks(3) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_gt_${NR}_small_kernel) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${NR + 1}; n < ${NR * 2}; n++) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
.ks(3) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, n_div_${NR}_small_kernel) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t n = ${2 * NR}; n <= ${3 * NR}; n += ${NR}) { |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(n) |
|
.k(k) |
|
.ks(3) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, strided_cm_subtile) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
for (uint32_t n = 1; n <= ${NR}; n++) { |
|
for (uint32_t m = 1; m <= ${MR}; m++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(n) |
|
.k(k) |
|
.cm_stride(${next_prime(NR + 1)}) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
$if UKERNEL_TYPE.startswith("IGEMM"): |
|
TEST(${TEST_NAME}, a_offset) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.ks(3) |
|
.a_offset(${next_prime(MR * KBLOCK * 5 + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
TEST(${TEST_NAME}, zero) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
for (uint32_t mz = 0; mz < ${MR}; mz++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.ks(3) |
|
.a_offset(${next_prime(MR * KBLOCK * 5 + 1)}) |
|
.zero_index(mz) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
|
|
$if ACTIVATION == "MINMAX": |
|
TEST(${TEST_NAME}, qmin) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.qmin(128) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
TEST(${TEST_NAME}, qmax) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.qmax(128) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
TEST(${TEST_NAME}, strided_cm) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.cm_stride(${next_prime(NR + 1)}) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
|
|
$if DATATYPE == "qu8": |
|
TEST(${TEST_NAME}, no_a_zero_point) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.a_zero_point(0) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if DATATYPE in ["qu8", "f32_qc4w"]: |
|
TEST(${TEST_NAME}, no_b_zero_point) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.b_zero_point(0) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if DATATYPE == "qu8": |
|
TEST(${TEST_NAME}, b_zero_point) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint16_t b_zero_point = 0; b_zero_point <= 255; ++b_zero_point) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.b_zero_point(b_zero_point) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if DATATYPE == "f32_qc4w": |
|
TEST(${TEST_NAME}, b_zero_point) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint8_t b_zero_point = 0; b_zero_point <= 15; ++b_zero_point) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.b_zero_point(b_zero_point) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if DATATYPE == "qu8": |
|
TEST(${TEST_NAME}, no_zero_point) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (size_t k = 1; k <= ${KBLOCK * 5}; k += ${KBLOCK + 1}) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(k) |
|
.a_zero_point(0) |
|
.b_zero_point(0) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
|
|
$if TEST_NAME.startswith('GENERATE') and DATATYPE in ['f32', 'f16']: |
|
TEST(${TEST_NAME}, subtile_m_upto_mr) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
for (uint32_t max_mr = 1; max_mr <= ${MR}; max_mr++) { |
|
for (uint32_t m = 1; m <= max_mr; m++) { |
|
for (size_t k = 1; k <= ${KBLOCK * 2}; k += 1) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(max_mr) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(m) |
|
.n(${NR}) |
|
.k(k) |
|
.iterations(1) |
|
$if DATATYPE == 'f32_qc4w': |
|
.b_zero_point(7) |
|
.Test(${", ".join(TEST_ARGS)}); |
|
} |
|
} |
|
} |
|
} |
|
|
|
$if TEST_NAME.startswith('GENERATE') and DATATYPE == 'f32' and POST_OP: |
|
TEST(${TEST_NAME}, hardswish) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
const std::vector<xnn_post_operation> fused_operators = { {xnn_post_operation_type_hardswish} }; |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.Test( |
|
${", ".join(TEST_ARGS)}, |
|
fused_operators); |
|
} |
|
$if MR > 1: |
|
TEST(${TEST_NAME}, hardswish_max_mr_lt_${MR}) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
const std::vector<xnn_post_operation> fused_operators = { {xnn_post_operation_type_hardswish} }; |
|
for (uint32_t max_mr = 1; max_mr < ${MR}; max_mr++) { |
|
GemmMicrokernelTester() |
|
$if EXTENDED_WEIGHTS: |
|
.extended_weights(true) |
|
.mr(max_mr) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(max_mr) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.Test( |
|
${", ".join(TEST_ARGS)}, |
|
fused_operators); |
|
} |
|
} |
|
|
|
$if TEST_NAME.startswith('GENERATE') and DATATYPE in ['f32', 'f16'] and PROTOTYPE is not None: |
|
#if XNN_ENABLE_ASSEMBLY |
|
TEST(${TEST_NAME}, matches_assembly) { |
|
$if ISA_CHECK: |
|
${ISA_CHECK}; |
|
GemmMicrokernelTester() |
|
.mr(${MR}) |
|
.nr(${NR}) |
|
.kr(${KR}) |
|
.sr(${SR}) |
|
.m(${MR}) |
|
.n(${NR}) |
|
.k(${KBLOCK}) |
|
.Test( |
|
${", ".join(TEST_ARGS)}, |
|
&${PROTOTYPE}); |
|
} |
|
#endif // XNN_ENABLE_ASSEMBLY |
|
|
|
""" |
|
|
|
|
|
def generate_test_cases(ukernel, mr, nr, kr, sr, xw, k_block, init_fn, |
|
requantization, is_pipelined, isa, jit, prototype, post_op): |
|
"""Generates all tests cases for a GEMM micro-kernel. |
|
|
|
Args: |
|
ukernel: C name of the micro-kernel function. |
|
mr: MR parameter of the GEMM micro-kernel. |
|
nr: NR parameter of the GEMM micro-kernel. |
|
kr: KR parameter of the GEMM micro-kernel. |
|
sr: SR parameter of the GEMM micro-kernel. |
|
xw: boolean indicator for microkernel with extended weights. |
|
k_block: Number of K values processed per one iteration of the main loop of |
|
the micro-kernel. |
|
init_fn: C name of the function to initialize microkernel parameters. |
|
requantization: name of the requantization scheme used by the microkernel. |
|
is_pipelined: Indicates if the micro-kernel is implemented with software |
|
pipelining. Additional test cases are generated for software pipelined |
|
micro-kernels to separately test prologue + epiloque of the pipelined loop |
|
and iteration of the pipelined loop. |
|
isa: instruction set required to run the micro-kernel. Generated unit test |
|
will skip execution if the host processor doesn't support this ISA. |
|
jit: if we are generating test code for JIT codegen. |
|
post_op: if post operation is supported (only for JIT). |
|
|
|
Returns: |
|
Code for the test case. |
|
""" |
|
_, ukernel_name = ukernel.split("_", 1) |
|
|
|
if jit: |
|
_, _, datatype, ukernel_type, _ = ukernel.split("_", 4) |
|
activation = None |
|
else: |
|
_, datatype, ukernel_type, activation, _ = ukernel.split("_", 4) |
|
if datatype == "f32" and ukernel_type in ["qc8w", "qc4w"]: |
|
_, datatype, weighttype, ukernel_type, activation, _ = ukernel.split("_", 5) |
|
datatype = datatype + "_" + weighttype |
|
if activation == "ukernel": |
|
activation = "linear" |
|
if activation in ["qs8w"]: |
|
_, _, _, _, _, activation, _ = ukernel.split("_", 6) |
|
test_args = [ukernel] |
|
if init_fn: |
|
test_args.append(init_fn) |
|
if requantization: |
|
requantization_datatype = {"qc8": "qs8"}.get(datatype, datatype) |
|
test_args.append("xnn_%s_requantize_%s" % \ |
|
(requantization_datatype, requantization)) |
|
|
|
if jit: |
|
if "minmax" in init_fn: |
|
activation = "minmax" |
|
|
|
return xngen.preprocess( |
|
GEMM_TEST_CODE, { |
|
"TEST_NAME": ukernel_name.upper().replace("UKERNEL_", ""), |
|
"TEST_ARGS": test_args, |
|
"UKERNEL_TYPE": ukernel_type.upper(), |
|
"DATATYPE": datatype, |
|
"ACTIVATION": activation.upper(), |
|
"MR": mr, |
|
"NR": nr, |
|
"KR": kr, |
|
"SR": sr, |
|
"EXTENDED_WEIGHTS": xw, |
|
"KBLOCK": k_block, |
|
"ADJKBLOCK": 2 * k_block if is_pipelined else k_block, |
|
"IS_PIPELINED": is_pipelined, |
|
"ISA_CHECK": xnncommon.generate_isa_check_macro(isa), |
|
"next_prime": next_prime, |
|
"POST_OP": post_op, |
|
"PROTOTYPE": prototype, |
|
}) |
|
|
|
|
|
def main(args): |
|
options = parser.parse_args(args) |
|
num_output_files = len(options.output) |
|
|
|
with codecs.open(options.spec, "r", encoding="utf-8") as spec_file: |
|
spec_yaml = yaml.safe_load(spec_file) |
|
if not isinstance(spec_yaml, list): |
|
raise ValueError("expected a list of micro-kernels in the spec") |
|
|
|
tests = """\ |
|
// Copyright (c) Facebook, Inc. and its affiliates. |
|
// All rights reserved. |
|
// |
|
// Copyright 2019 Google LLC |
|
// |
|
// This source code is licensed under the BSD-style license found in the |
|
// LICENSE file in the root directory of this source tree. |
|
// |
|
// Auto-generated file. Do not edit! |
|
// Specification: {specification} |
|
// Generator: {generator} |
|
|
|
|
|
#include <gtest/gtest.h> |
|
|
|
#include <xnnpack/allocator.h> |
|
#include <xnnpack/common.h> |
|
#include <xnnpack/isa-checks.h> |
|
#include <xnnpack/microparams-init.h> |
|
|
|
#include <xnnpack/gemm.h> |
|
#include <xnnpack/igemm.h> |
|
#include <xnnpack/ppmm.h> |
|
#include "gemm-microkernel-tester.h" |
|
""".format( |
|
specification=options.spec, generator=sys.argv[0]) |
|
|
|
outputs = collections.defaultdict(lambda: tests) |
|
|
|
for ukernel_spec in spec_yaml: |
|
name = ukernel_spec["name"] |
|
k_block = int(ukernel_spec["k-block"]) |
|
init_fn = ukernel_spec.get("init") |
|
pipelined = bool(ukernel_spec.get("pipelined", False)) |
|
jit = name.startswith("xnn_generate") |
|
prototype = ukernel_spec.get("prototype") |
|
post_op = ukernel_spec.get("post-op", True) |
|
mr, nr, kr, sr, xw, requantization, arch, isa, assembly = \ |
|
split_ukernel_name(name) |
|
|
|
test_case = generate_test_cases(name, mr, nr, kr, sr, xw, k_block, |
|
init_fn, requantization, pipelined, isa, |
|
jit, prototype, post_op) |
|
|
|
|
|
|
|
output_index = zlib.crc32(bytes(name, "utf-8")) % num_output_files |
|
outputs[options. |
|
output[output_index]] += "\n\n" + xnncommon.postprocess_test_case( |
|
test_case, arch, isa, assembly, jit) |
|
|
|
for output_name in options.output: |
|
txt_changed = True |
|
if os.path.exists(output_name): |
|
with codecs.open(output_name, "r", encoding="utf-8") as output_file: |
|
txt_changed = output_file.read() != outputs[output_name] |
|
|
|
if txt_changed: |
|
with codecs.open(output_name, "w", encoding="utf-8") as output_file: |
|
output_file.write(outputs[output_name]) |
|
|
|
|
|
if __name__ == "__main__": |
|
main(sys.argv[1:]) |
|
|