Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

#12795: Move numpy functions.hpp #12817

Merged
merged 6 commits into from
Sep 22, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 17 additions & 17 deletions tests/tt_eager/integration_tests/test_bert.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@
#include "ttnn/operations/normalization/softmax/softmax.hpp"
#include "tt_metal/common/constants.hpp"
#include "tt_metal/host_api.hpp"
#include "tt_numpy/functions.hpp"
#include "ttnn/operations/numpy/functions.hpp"
#include "ttnn/operations/matmul/matmul.hpp"
#include "ttnn/operations/normalization/layernorm/layernorm.hpp"
#include "ttnn/operations/eltwise/binary/binary.hpp"
Expand Down Expand Up @@ -231,34 +231,34 @@ void test_bert() {
std::uint32_t hidden_size = num_heads * head_size;
std::uint32_t intermediate_size = hidden_size * 4;

auto attention_mask = tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {batch_size, 1, TILE_HEIGHT, sequence_size}, Layout::TILE).to(device, l1_memory_config);
auto attention_mask = ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {batch_size, 1, TILE_HEIGHT, sequence_size}, Layout::TILE).to(device, l1_memory_config);

auto parameters = Parameters{};
for (auto encoder_index = 0; encoder_index < num_encoders; encoder_index++) {
parameters.emplace(fmt::format("fused_qkv_weight_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, hidden_size * 3}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("fused_qkv_bias_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, hidden_size * 3}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("selfout_weight_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("selfout_bias_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("attention_layernorm_weight_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
parameters.emplace(fmt::format("attention_layernorm_bias_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff1_weight_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, intermediate_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff1_bias_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, intermediate_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff2_weight_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, intermediate_size, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff2_bias_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("feedforward_layernorm_weight_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
parameters.emplace(fmt::format("feedforward_layernorm_bias_{}", encoder_index), tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
parameters.emplace(fmt::format("fused_qkv_weight_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, hidden_size * 3}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("fused_qkv_bias_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, hidden_size * 3}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("selfout_weight_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("selfout_bias_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("attention_layernorm_weight_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
parameters.emplace(fmt::format("attention_layernorm_bias_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff1_weight_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, intermediate_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff1_bias_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, intermediate_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff2_weight_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, intermediate_size, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("ff2_bias_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, hidden_size}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(fmt::format("feedforward_layernorm_weight_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
parameters.emplace(fmt::format("feedforward_layernorm_bias_{}", encoder_index), ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::ROW_MAJOR).to(device, dram_memory_config));
};
parameters.emplace("qa_head_weight", tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, TILE_WIDTH}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace("qa_head_weight", ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, hidden_size, TILE_WIDTH}, Layout::TILE).to(device, dram_memory_config));
parameters.emplace(
"qa_head_bias",
ttnn::reshape(
tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::TILE).to(device, dram_memory_config),
ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {1, 1, TILE_HEIGHT, TILE_WIDTH}, Layout::TILE).to(device, dram_memory_config),
ttnn::Shape{tt::tt_metal::LegacyShape{{1, 1, 1, TILE_WIDTH}, {1, 1, TILE_HEIGHT, TILE_WIDTH}}}));

auto run_bert = [&]() {
tt::log_debug(tt::LogTest, "run_bert started");
auto begin = std::chrono::steady_clock::now();
auto hidden_states = tt::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {batch_size, 1, sequence_size, hidden_size}, Layout::TILE).to(device, l1_memory_config);
auto hidden_states = ttnn::numpy::random::uniform(bfloat16(-1.0f), bfloat16(1.0f), {batch_size, 1, sequence_size, hidden_size}, Layout::TILE).to(device, l1_memory_config);
for (auto encoder_index = 0; encoder_index < num_encoders; encoder_index++) {
hidden_states = encoder(std::move(hidden_states), attention_mask, parameters, encoder_index, head_size);
}
Expand Down
4 changes: 2 additions & 2 deletions tests/tt_eager/ops/test_average_pool.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@

#include "ttnn/operations/pool/avgpool/avg_pool.hpp"
#include "ttnn/operations/experimental/auto_format/auto_format.hpp"
#include "tt_numpy/functions.hpp"
#include "ttnn/operations/numpy/functions.hpp"

#include "ttnn/tensor/tensor.hpp"
#include "common/constants.hpp"
Expand All @@ -17,7 +17,7 @@ using tt::tt_metal::LegacyShape;

Tensor run_avg_pool_2d_resnet(tt::tt_metal::LegacyShape& tensor_shape, Device* device) {
using ttnn::operations::experimental::auto_format::AutoFormat;
auto input_tensor = tt::numpy::random::random(tensor_shape, DataType::BFLOAT16);
auto input_tensor = ttnn::numpy::random::random(tensor_shape, DataType::BFLOAT16);
auto padded_input_shape = AutoFormat::pad_to_tile_shape(tensor_shape, false, false);
Tensor padded_input_tensor = input_tensor;
if (!AutoFormat::check_input_tensor_format(input_tensor, padded_input_shape)) {
Expand Down
22 changes: 11 additions & 11 deletions tests/tt_eager/ops/test_bcast_op.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
#include "ttnn/operations/data_movement/bcast/bcast.hpp"
#include "common/constants.hpp"
#include "third_party/magic_enum/magic_enum.hpp"
#include <tt_numpy/functions.hpp>
#include <ttnn/operations/numpy/functions.hpp>

#include <algorithm>
#include <functional>
Expand Down Expand Up @@ -56,8 +56,8 @@ int main(int argc, char **argv) {
throw std::runtime_error("Unsupported Dim!");
}

Tensor a = tt::numpy::random::random(input_shape_a).to(Layout::TILE).to(device);
Tensor b = tt::numpy::zeros({1, 1, TILE_HEIGHT, TILE_WIDTH}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor a = ttnn::numpy::random::random(input_shape_a).to(Layout::TILE).to(device);
Tensor b = ttnn::numpy::zeros({1, 1, TILE_HEIGHT, TILE_WIDTH}, DataType::BFLOAT16).to(Layout::TILE).to(device);

for (auto bcast_math: magic_enum::enum_values<ttnn::BcastOpMath>()) {
Tensor c = ttnn::bcast(0,a, b, bcast_math, bcast_dim);
Expand All @@ -73,29 +73,29 @@ int main(int argc, char **argv) {
}

{
Tensor a = tt::numpy::random::random({1, 1, 32, 4544}).to(Layout::TILE).to(device);
Tensor b = tt::numpy::zeros({1, 1, 32, 4544}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor a = ttnn::numpy::random::random({1, 1, 32, 4544}).to(Layout::TILE).to(device);
Tensor b = ttnn::numpy::zeros({1, 1, 32, 4544}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor c = ttnn::bcast(0, a, b, ttnn::BcastOpMath::MUL, ttnn::BcastOpDim::H);
Tensor d = c.cpu();
}

{
Tensor a = tt::numpy::random::random({1, 1, 32, 4544}).to(Layout::TILE).to(device);
Tensor b = tt::numpy::zeros({1, 1, 32, 4544}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor a = ttnn::numpy::random::random({1, 1, 32, 4544}).to(Layout::TILE).to(device);
Tensor b = ttnn::numpy::zeros({1, 1, 32, 4544}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor c = ttnn::bcast(0,a, b, ttnn::BcastOpMath::ADD, ttnn::BcastOpDim::H);
Tensor d = c.cpu();
}

{
Tensor a = tt::numpy::random::random({1, 71, 32, 32}).to(Layout::TILE).to(device);
Tensor b = tt::numpy::zeros({1, 1, 32, 32}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor a = ttnn::numpy::random::random({1, 71, 32, 32}).to(Layout::TILE).to(device);
Tensor b = ttnn::numpy::zeros({1, 1, 32, 32}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor c = ttnn::bcast(0,a, b, ttnn::BcastOpMath::MUL, ttnn::BcastOpDim::HW);
Tensor d = c.cpu();
}

{
Tensor a = tt::numpy::random::random({1, 71, 32, 64}).to(Layout::TILE).to(device);
Tensor b = tt::numpy::zeros({1, 1, 32, 32}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor a = ttnn::numpy::random::random({1, 71, 32, 64}).to(Layout::TILE).to(device);
Tensor b = ttnn::numpy::zeros({1, 1, 32, 32}, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor c = ttnn::bcast(0,a, b, ttnn::BcastOpMath::MUL, ttnn::BcastOpDim::HW);
Tensor d = c.cpu();
}
Expand Down
8 changes: 4 additions & 4 deletions tests/tt_eager/ops/test_bmm_op.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
#include "ttnn/tensor/tensor.hpp"
#include "ttnn/operations/matmul/device/matmul_op.hpp"
#include "common/constants.hpp"
#include "tt_numpy/functions.hpp"
#include "ttnn/operations/numpy/functions.hpp"

#include <algorithm>
#include <functional>
Expand Down Expand Up @@ -46,9 +46,9 @@ int main(int argc, char **argv) {
tt::tt_metal::LegacyShape shapeb1 = {1, 1, Kt*TILE_HEIGHT, Nt*TILE_WIDTH};

// Allocates a DRAM buffer on device populated with values specified by initialize
Tensor a = tt::numpy::random::random(shapea).to(Layout::TILE).to(device);
Tensor b = tt::numpy::zeros(shapeb, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor b1 = tt::numpy::zeros(shapeb1, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor a = ttnn::numpy::random::random(shapea).to(Layout::TILE).to(device);
Tensor b = ttnn::numpy::zeros(shapeb, DataType::BFLOAT16).to(Layout::TILE).to(device);
Tensor b1 = ttnn::numpy::zeros(shapeb1, DataType::BFLOAT16).to(Layout::TILE).to(device);

Tensor mm = ttnn::operations::matmul::matmul(a, b, /*bias=*/std::nullopt,
ttnn::operations::matmul::Matmul{/*program_config=*/std::nullopt, /*bcast_batch=*/std::nullopt,operation::DEFAULT_OUTPUT_MEMORY_CONFIG, /*output_dtype=*/std::nullopt, /*compute_kernel_config=*/std::nullopt, /*untilize_out=*/false, /*user_core_coord=*/std::nullopt, /*user_fused_activation=*/std::nullopt, /*user_run_batched=*/true}).cpu();
Expand Down
10 changes: 5 additions & 5 deletions tests/tt_eager/ops/test_eltwise_binary_op.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
#include "ttnn/tensor/host_buffer/types.hpp"
#include "ttnn/tensor/tensor.hpp"
#include "ttnn/operations/eltwise/binary/binary.hpp"
#include "tt_numpy/functions.hpp"
#include "ttnn/operations/numpy/functions.hpp"

using tt::tt_metal::DataType;
using tt::tt_metal::Device;
Expand All @@ -32,13 +32,13 @@ Tensor host_function(const Tensor& input_tensor_a, const Tensor& input_tensor_b)

template <auto HostFunction, typename DeviceFunction, typename... Args>
bool run_test(const tt::tt_metal::LegacyShape& shape, const DeviceFunction& device_function, Device* device, Args... args) {
auto input_tensor_a = tt::numpy::random::random(shape, DataType::BFLOAT16);
auto input_tensor_b = tt::numpy::random::random(shape, DataType::BFLOAT16);
auto input_tensor_a = ttnn::numpy::random::random(shape, DataType::BFLOAT16);
auto input_tensor_b = ttnn::numpy::random::random(shape, DataType::BFLOAT16);

auto host_output = HostFunction(input_tensor_a, input_tensor_b);
auto device_output = device_function(input_tensor_a.to(Layout::TILE).to(device), input_tensor_b.to(Layout::TILE).to(device)).cpu().to(Layout::ROW_MAJOR);

return tt::numpy::allclose<bfloat16>(host_output, device_output, args...);
return ttnn::numpy::allclose<bfloat16>(host_output, device_output, args...);
}

int main() {
Expand Down Expand Up @@ -108,7 +108,7 @@ int main() {

// Allocate a tensor to show that the addresses aren't cached
auto input_tensor =
tt::numpy::random::uniform(bfloat16(0.0f), bfloat16(0.0f), {1, 1, 32, 32}).to(Layout::TILE).to(device);
ttnn::numpy::random::uniform(bfloat16(0.0f), bfloat16(0.0f), {1, 1, 32, 32}).to(Layout::TILE).to(device);

run_binary_ops();

Expand Down
Loading
Loading