Skip to content

Commit

Permalink
Merge branch 'main' into dgomezTT-16418
Browse files Browse the repository at this point in the history
  • Loading branch information
@dgomezTT
dgomezTT authored Feb 18, 2025
2 parents 6b955ed + 2d4f994 commit 2fdee2e
Show file tree
Hide file tree
Showing 36 changed files with 748 additions and 530 deletions.
4 changes: 3 additions & 1 deletion infra/data_collection/github/download_cicd_logs_and_artifacts.sh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,9 @@ download_artifacts() {
local repo=$1
local workflow_run_id=$2

if gh api --paginate /repos/$repo/actions/runs/$workflow_run_id/artifacts | jq '.artifacts[] | .name' | grep -q "test_reports_"; then
echo "[info] Downloading test reports for workflow run $workflow_run_id"
api_output=$(gh api --paginate /repos/$repo/actions/runs/$workflow_run_id/artifacts | jq -r '.artifacts[] | .name')
if echo "$api_output" | grep -q "test_reports_"; then
gh run download --repo $repo -D generated/cicd/$workflow_run_id/artifacts --pattern test_reports_* $workflow_run_id
else
echo "[Warning] Test reports not found for workflow run $workflow_run_id"
Expand Down
10 changes: 7 additions & 3 deletions tests/tt_metal/distributed/test_mesh_buffer.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,11 +25,11 @@ struct DeviceLocalShardedBufferTestConfig {
TensorMemoryLayout mem_config = TensorMemoryLayout::HEIGHT_SHARDED;
ShardOrientation shard_orientation = ShardOrientation::ROW_MAJOR;

Shape2D tensor2d_shape() {
Shape2D tensor2d_shape_in_pages() {
return {num_pages_per_core.height() * num_cores.height(), num_pages_per_core.width() * num_cores.width()};
}

uint32_t num_pages() { return tensor2d_shape().height() * tensor2d_shape().width(); }
uint32_t num_pages() { return tensor2d_shape_in_pages().height() * tensor2d_shape_in_pages().width(); }

std::array<uint32_t, 2> shard_shape() {
return {num_pages_per_core.height() * page_shape.height(), num_pages_per_core.width() * page_shape.width()};
Expand All @@ -44,7 +44,11 @@ struct DeviceLocalShardedBufferTestConfig {

ShardSpecBuffer shard_parameters() {
return ShardSpecBuffer(
this->shard_grid(), this->shard_shape(), this->shard_orientation, this->page_shape, this->tensor2d_shape());
this->shard_grid(),
this->shard_shape(),
this->shard_orientation,
this->page_shape,
this->tensor2d_shape_in_pages());
}
};

Expand Down
40 changes: 23 additions & 17 deletions ...metal/tt_metal/dispatch/dispatch_buffer/test_EnqueueWriteBuffer_and_EnqueueReadBuffer.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -56,11 +56,11 @@ class BufferStressTestConfigSharded {
this->num_cores = cores;
}

std::array<uint32_t, 2> tensor2d_shape() {
std::array<uint32_t, 2> tensor2d_shape_in_pages() {
return {num_pages_per_core[0] * num_cores[0], num_pages_per_core[1] * num_cores[1]};
}

uint32_t num_pages() { return tensor2d_shape()[0] * tensor2d_shape()[1]; }
uint32_t num_pages() { return tensor2d_shape_in_pages()[0] * tensor2d_shape_in_pages()[1]; }

std::array<uint32_t, 2> shard_shape() {
return {num_pages_per_core[0] * page_shape[0], num_pages_per_core[1] * page_shape[1]};
Expand All @@ -73,7 +73,11 @@ class BufferStressTestConfigSharded {

ShardSpecBuffer shard_parameters() {
return ShardSpecBuffer(
this->shard_grid(), this->shard_shape(), this->shard_orientation, this->page_shape, this->tensor2d_shape());
this->shard_grid(),
this->shard_shape(),
this->shard_orientation,
this->page_shape,
this->tensor2d_shape_in_pages());
}

uint32_t page_size() { return page_shape[0] * page_shape[1] * element_size; }
Expand All @@ -87,7 +91,7 @@ struct ShardedSubBufferStressTestConfig {
CoreRangeSet cores;
Shape2D shard_shape;
Shape2D page_shape;
Shape2D tensor2d_shape;
Shape2D tensor2d_shape_in_pages;
TensorMemoryLayout layout;
ShardOrientation orientation;
};
Expand Down Expand Up @@ -133,11 +137,12 @@ vector<ShardedSubBufferStressTestConfig> generate_sharded_sub_buffer_test_config
uint32_t page_shape_width_div_factor = 1;
while (page_shape_width_div_factor <= num_pages_per_shard) {
if (page_shape_width_div_factor * page_shape_height_div_factor == num_pages_per_shard) {
uint32_t tensor2d_shape_height = page_shape_height_div_factor;
while (tensor2d_shape_height <= num_pages) {
uint32_t tensor2d_shape_width = page_shape_width_div_factor;
while (tensor2d_shape_width <= num_pages) {
if (tensor2d_shape_height * tensor2d_shape_width == num_pages) {
uint32_t tensor2d_shape_in_pages_height = page_shape_height_div_factor;
while (tensor2d_shape_in_pages_height <= num_pages) {
uint32_t tensor2d_shape_in_pages_width = page_shape_width_div_factor;
while (tensor2d_shape_in_pages_width <= num_pages) {
if (tensor2d_shape_in_pages_height * tensor2d_shape_in_pages_width ==
num_pages) {
for (TensorMemoryLayout layout :
{TensorMemoryLayout::HEIGHT_SHARDED,
TensorMemoryLayout::BLOCK_SHARDED,
Expand All @@ -157,17 +162,18 @@ vector<ShardedSubBufferStressTestConfig> generate_sharded_sub_buffer_test_config
page_shape_height_div_factor,
tt::constants::TILE_WIDTH /
page_shape_width_div_factor},
.tensor2d_shape =
{tensor2d_shape_height, tensor2d_shape_width},
.tensor2d_shape_in_pages =
{tensor2d_shape_in_pages_height,
tensor2d_shape_in_pages_width},
.layout = layout,
.orientation = orientation};
configs.push_back(config);
}
}
}
tensor2d_shape_width += page_shape_width_div_factor;
tensor2d_shape_in_pages_width += page_shape_width_div_factor;
}
tensor2d_shape_height += page_shape_height_div_factor;
tensor2d_shape_in_pages_height += page_shape_height_div_factor;
}
}
page_shape_width_div_factor += 1;
Expand Down Expand Up @@ -1018,7 +1024,7 @@ TEST_F(CommandQueueSingleCardBufferFixture, TestReadWriteShardedSubBufferForL1)
tt::log_debug(
tt::LogTest,
"Device: {} buffer_size: {} page_size: {} region_offset: {} region_size: {} shard_shape: [{}, {}] "
"page_shape: [{}, {}] tensor2d_shape: [{}, {}] layout: {} orientation: {} cores: {}",
"page_shape: [{}, {}] tensor2d_shape_in_pages: [{}, {}] layout: {} orientation: {} cores: {}",
device->id(),
config.buffer_size,
config.page_size,
Expand All @@ -1028,8 +1034,8 @@ TEST_F(CommandQueueSingleCardBufferFixture, TestReadWriteShardedSubBufferForL1)
config.shard_shape.width(),
config.page_shape.height(),
config.page_shape.width(),
config.tensor2d_shape.height(),
config.tensor2d_shape.width(),
config.tensor2d_shape_in_pages.height(),
config.tensor2d_shape_in_pages.width(),
magic_enum::enum_name(config.layout).data(),
magic_enum::enum_name(config.orientation).data(),
config.cores.str());
Expand All @@ -1039,7 +1045,7 @@ TEST_F(CommandQueueSingleCardBufferFixture, TestReadWriteShardedSubBufferForL1)
{tt::constants::TILE_HEIGHT, tt::constants::TILE_WIDTH},
config.orientation,
config.page_shape,
config.tensor2d_shape);
config.tensor2d_shape_in_pages);
auto buffer =
Buffer::create(device, config.buffer_size, config.page_size, BufferType::L1, config.layout, shard_spec);

Expand Down
13 changes: 3 additions & 10 deletions tests/tt_metal/tt_metal/perf_microbenchmark/routing/kernels/tt_fabric_traffic_gen_tx.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -83,10 +83,6 @@ packet_header_t packet_header __attribute__((aligned(16)));
uint32_t target_address;
uint32_t noc_offset;
uint32_t rx_addr_hi;

uint32_t gk_interface_addr_l;
uint32_t gk_interface_addr_h;

uint32_t controller_noc_offset;

// flag to check if need to zero out notification addr
Expand Down Expand Up @@ -389,11 +385,9 @@ void kernel_main() {
src_endpoint_id = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
noc_offset = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
controller_noc_offset = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
uint32_t routing_plane = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
uint32_t outbound_eth_chan = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
dest_device = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
uint32_t rx_buf_size = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
gk_interface_addr_l = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
gk_interface_addr_h = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));

if constexpr (ASYNC_WR & test_command) {
base_target_address = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
Expand Down Expand Up @@ -462,9 +456,8 @@ void kernel_main() {
uint32_t packet_count = 0;

// initalize client
fabric_endpoint_init(client_interface_addr, gk_interface_addr_l, gk_interface_addr_h);
routing_table = reinterpret_cast<tt_l1_ptr fabric_router_l1_config_t*>(
client_interface->routing_tables_l1_offset + sizeof(fabric_router_l1_config_t) * routing_plane);
fabric_endpoint_init(client_interface_addr, outbound_eth_chan);
routing_table = reinterpret_cast<tt_l1_ptr fabric_router_l1_config_t*>(client_interface->routing_tables_l1_offset);

while (true) {
iter++;
Expand Down
12 changes: 4 additions & 8 deletions tests/tt_metal/tt_metal/perf_microbenchmark/routing/kernels/tt_fabric_tx_ubench.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -68,8 +68,6 @@ volatile fabric_client_interface_t* client_interface;
uint64_t xy_local_addr;
uint32_t target_address;
uint32_t noc_offset;
uint32_t gk_interface_addr_l;
uint32_t gk_interface_addr_h;
uint32_t controller_noc_offset;
uint32_t time_seed;

Expand All @@ -94,11 +92,9 @@ void kernel_main() {
src_endpoint_id = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
noc_offset = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
controller_noc_offset = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
uint32_t routing_plane = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
uint32_t outbound_eth_chan = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
dest_device = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
uint32_t rx_buf_size = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
gk_interface_addr_l = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
gk_interface_addr_h = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));

if constexpr (ASYNC_WR & test_command) {
base_target_address = get_arg_val<uint32_t>(increment_arg_idx(rt_args_idx));
Expand Down Expand Up @@ -140,7 +136,7 @@ void kernel_main() {
}

// initalize client
fabric_endpoint_init(client_interface_addr, gk_interface_addr_l, gk_interface_addr_h);
fabric_endpoint_init(client_interface_addr, outbound_eth_chan);

// notify the controller kernel that this worker is ready to proceed
notify_traffic_controller();
Expand All @@ -161,7 +157,7 @@ void kernel_main() {
client_interface->local_pull_request.pull_request.words_read = 0;
if constexpr (mcast_data) {
fabric_async_write_multicast<ASYNC_WR_SEND>(
routing_plane, // the network plane to use for this transaction
0, // the network plane to use for this transaction
data_buffer_start_addr, // source address in sender’s memory
dest_device >> 16,
dest_device & 0xFFFF,
Expand All @@ -173,7 +169,7 @@ void kernel_main() {
s_depth);
} else {
fabric_async_write<ASYNC_WR_SEND>(
routing_plane, // the network plane to use for this transaction
0, // the network plane to use for this transaction
data_buffer_start_addr, // source address in sender’s memory
dest_device >> 16,
dest_device & 0xFFFF,
Expand Down
Loading

0 comments on commit 2fdee2e

Please sign in to comment.