mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-04-24 18:00:48 +02:00
teflon/tests: Remove dependency on xtensor
Upstream has been moving headers around and breaking users. Because we don't use it for much right now, drop the dependency altogether by open coding some rand() helpers. Issue: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13681 Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37220>
This commit is contained in:
Tomeu Vizoso
parent
b148d47c3e
commit
5eab4f06d5
3 changed files with 37 additions and 24 deletions
|
|
@ -2193,7 +2193,6 @@ endif
|
|||
|
||||
with_teflon = get_option('teflon')
|
||||
if with_teflon and with_tests
|
||||
dep_xtensor = dependency('xtensor')
|
||||
dep_flatbuffers = dependency('flatbuffers')
|
||||
prog_flatc = find_program('flatc')
|
||||
endif
|
||||
|
|
|
|||
|
|
@ -10,9 +10,6 @@
|
|||
#include <stdio.h>
|
||||
#include <vector>
|
||||
#include <gtest/gtest.h>
|
||||
#include <xtensor/xrandom.hpp>
|
||||
|
||||
#include "util/macros.h"
|
||||
|
||||
#include "tensorflow/lite/c/c_api.h"
|
||||
#include "tensorflow/lite/c/common.h"
|
||||
|
|
@ -27,6 +24,28 @@ randf(float min, float max)
|
|||
return ((max - min) * ((float)rand() / (float)RAND_MAX)) + min;
|
||||
}
|
||||
|
||||
template<typename T>
|
||||
std::vector<T> rand(const std::vector<int>& shape, T min, T max) {
|
||||
size_t size = 1;
|
||||
for (int dim : shape) {
|
||||
size *= dim;
|
||||
}
|
||||
|
||||
std::vector<T> result(size);
|
||||
|
||||
if constexpr (std::is_integral<T>::value) {
|
||||
std::vector<T> result(size);
|
||||
std::generate(result.begin(), result.end(), [&]() { return rand() % (max - min + 1) + min; });
|
||||
return result;
|
||||
} else if constexpr (std::is_floating_point<T>::value) {
|
||||
std::vector<T> result(size);
|
||||
std::generate(result.begin(), result.end(), [&]() { return randf(-1.0, 1.0); });
|
||||
return result;
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
static void
|
||||
read_model(const char *file_name, tflite::ModelT &model)
|
||||
{
|
||||
|
|
@ -111,7 +130,7 @@ patch_conv2d(unsigned operation_index,
|
|||
bias_tensor->shape.data()[0] = output_channels;
|
||||
|
||||
auto bias_data = &model->buffers[bias_buffer_index]->data;
|
||||
xt::xarray<int32_t> bias_array = xt::random::randint<int32_t>({output_channels}, -20000, 20000);
|
||||
std::vector<int32_t> bias_array = rand<int32_t>({output_channels}, -20000, 20000);
|
||||
bias_data->resize(bias_array.size() * sizeof(int32_t));
|
||||
memcpy(bias_data->data(), bias_array.data(), bias_array.size() * sizeof(int32_t));
|
||||
|
||||
|
|
@ -140,7 +159,7 @@ patch_conv2d(unsigned operation_index,
|
|||
else
|
||||
weight_shape = {output_channels, weight_size, weight_size, input_channels};
|
||||
|
||||
xt::xarray<uint8_t> weights_array = xt::random::randint<uint8_t>(weight_shape, 0, 255);
|
||||
std::vector<uint8_t> weights_array = rand<uint8_t>(weight_shape, 0, 255);
|
||||
weights_data->resize(weights_array.size());
|
||||
memcpy(weights_data->data(), weights_array.data(), weights_array.size());
|
||||
|
||||
|
|
@ -250,6 +269,8 @@ add_generate_model(int input_size,
|
|||
return buf;
|
||||
}
|
||||
|
||||
|
||||
|
||||
static void
|
||||
patch_fully_connected(unsigned operation_index,
|
||||
tflite::ModelT *model,
|
||||
|
|
@ -287,7 +308,7 @@ patch_fully_connected(unsigned operation_index,
|
|||
bias_tensor->shape.data()[0] = output_channels;
|
||||
|
||||
auto bias_data = &model->buffers[bias_buffer_index]->data;
|
||||
xt::xarray<int32_t> bias_array = xt::random::randint<int32_t>({output_channels}, -20000, 20000);
|
||||
std::vector<int32_t> bias_array = rand<int32_t>({output_channels}, -20000, 20000);
|
||||
bias_data->resize(bias_array.size() * sizeof(int32_t));
|
||||
memcpy(bias_data->data(), bias_array.data(), bias_array.size() * sizeof(int32_t));
|
||||
|
||||
|
|
@ -302,7 +323,7 @@ patch_fully_connected(unsigned operation_index,
|
|||
std::vector<int> weight_shape;
|
||||
weight_shape = {output_channels, input_size};
|
||||
|
||||
xt::xarray<uint8_t> weights_array = xt::random::randint<uint8_t>(weight_shape, 0, 255);
|
||||
std::vector<uint8_t> weights_array = rand<uint8_t>(weight_shape, 0, 255);
|
||||
weights_data->resize(weights_array.size());
|
||||
memcpy(weights_data->data(), weights_array.data(), weights_array.size());
|
||||
|
||||
|
|
@ -432,7 +453,7 @@ run_model(TfLiteModel *model, enum executor executor, void ***input, size_t *num
|
|||
if ((*input)[i] == NULL) {
|
||||
(*input)[i] = malloc(input_tensor->bytes);
|
||||
|
||||
std::vector<size_t> shape;
|
||||
std::vector<int> shape;
|
||||
|
||||
shape.resize(input_tensor->dims->size);
|
||||
for (int j = 0; j < input_tensor->dims->size; j++)
|
||||
|
|
@ -440,12 +461,12 @@ run_model(TfLiteModel *model, enum executor executor, void ***input, size_t *num
|
|||
|
||||
switch (input_tensor->type) {
|
||||
case kTfLiteFloat32: {
|
||||
xt::xarray<float_t> a = xt::random::rand<float_t>(shape);
|
||||
std::vector<float> a = rand<float>(shape, -1.0, 1.0);
|
||||
memcpy((*input)[i], a.data(), input_tensor->bytes);
|
||||
break;
|
||||
}
|
||||
default: {
|
||||
xt::xarray<uint8_t> a = xt::random::randint<uint8_t>(shape, 0, 255);
|
||||
std::vector<uint8_t> a = rand<uint8_t>(shape, 0, 255);
|
||||
memcpy((*input)[i], a.data(), input_tensor->bytes);
|
||||
break;
|
||||
}
|
||||
|
|
|
|||
|
|
@ -3,13 +3,13 @@
|
|||
* SPDX-License-Identifier: MIT
|
||||
*/
|
||||
|
||||
#include <algorithm>
|
||||
#include <cstdio>
|
||||
#include <fcntl.h>
|
||||
#include <filesystem>
|
||||
#include <fstream>
|
||||
#include <gtest/gtest.h>
|
||||
#include <sys/mman.h>
|
||||
#include <xtensor/xrandom.hpp>
|
||||
|
||||
#include <iostream>
|
||||
#include <sstream>
|
||||
|
|
@ -38,13 +38,6 @@ std::vector<int> input_size{3, 5, 8, 80, 112};
|
|||
std::vector<int> fc_channels{23, 46, 128, 256, 512};
|
||||
std::vector<int> fc_size{128, 1280, 25088, 62720};
|
||||
|
||||
static void
|
||||
set_seed(unsigned seed)
|
||||
{
|
||||
srand(seed);
|
||||
xt::random::seed(seed);
|
||||
}
|
||||
|
||||
static void
|
||||
test_model(void *buf, size_t buf_size, std::string cache_dir, unsigned tolerance)
|
||||
{
|
||||
|
|
@ -175,7 +168,7 @@ test_model_file(std::string file_name, unsigned tolerance, bool use_cache)
|
|||
cache_dir << path.stem().string();
|
||||
}
|
||||
|
||||
set_seed(4);
|
||||
srand(4);
|
||||
|
||||
struct stat sb;
|
||||
int model_fd = open(file_name.c_str(), O_RDONLY);
|
||||
|
|
@ -200,7 +193,7 @@ test_conv(int input_size, int weight_size, int input_channels, int output_channe
|
|||
if (weight_size > input_size)
|
||||
GTEST_SKIP();
|
||||
|
||||
set_seed(seed);
|
||||
srand(seed);
|
||||
|
||||
if (cache_is_enabled()) {
|
||||
if (access(model_cache.str().c_str(), F_OK) == 0) {
|
||||
|
|
@ -245,7 +238,7 @@ test_add(int input_size, int weight_size, int input_channels, int output_channel
|
|||
if (weight_size > input_size)
|
||||
GTEST_SKIP();
|
||||
|
||||
set_seed(seed);
|
||||
srand(seed);
|
||||
|
||||
if (cache_is_enabled()) {
|
||||
if (access(model_cache.str().c_str(), F_OK) == 0) {
|
||||
|
|
@ -284,7 +277,7 @@ test_fully_connected(int input_size, int output_channels, bool is_signed, int se
|
|||
model_cache << cache_dir.str() << "/"
|
||||
<< "model.tflite";
|
||||
|
||||
set_seed(seed);
|
||||
srand(seed);
|
||||
|
||||
if (cache_is_enabled()) {
|
||||
if (access(model_cache.str().c_str(), F_OK) == 0) {
|
||||
|
|
@ -579,7 +572,7 @@ main(int argc, char **argv)
|
|||
int depthwise = atoi(argv[n++]);
|
||||
int seed = atoi(argv[n++]);
|
||||
|
||||
set_seed(seed);
|
||||
srand(seed);
|
||||
|
||||
buf = conv2d_generate_model(input_size, weight_size,
|
||||
input_channels, output_channels,
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue