Skip to content

Commit

Permalink
Modify Slinky usage in XNNPACK to always link in Slinky when building…
Browse files Browse the repository at this point in the history 
… on Linux-x86; in this configuration, you can enable Slinky via a runtime flag rather than a compile-time flag. (On non-Linux-x86 builds the runtime flag will simply be ignored.)

PiperOrigin-RevId: 691984806
  • Loading branch information
@xnnpack-bot
xnnpack-bot committed Nov 1, 2024
1 parent 40ed254 commit 8c3d6e5
Show file tree
Hide file tree
Showing 4 changed files with 54 additions and 62 deletions.
12 changes: 10 additions & 2 deletions BUILD.bazel
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,6 @@ load(
"xnnpack_if_kleidiai_enabled",
"xnnpack_min_size_copts",
"xnnpack_slinky_deps",
"xnnpack_slinky_srcs",
"xnnpack_source_list_file",
"xnnpack_transitive_source_list",
"xnnpack_visibility",
Expand Down Expand Up @@ -1019,9 +1018,17 @@ xnnpack_cc_library(
],
)

xnnpack_cc_library(
name = "subgraph_h",
hdrs = [
"src/xnnpack/subgraph.h",
],
compatible_with = [],
)

xnnpack_cc_library(
name = "subgraph",
srcs = SUBGRAPH_SRCS + xnnpack_slinky_srcs(),
srcs = SUBGRAPH_SRCS,
hdrs = [
"src/xnnpack/memory-planner.h",
"src/xnnpack/reshape-helpers.h",
Expand Down Expand Up @@ -1050,6 +1057,7 @@ xnnpack_cc_library(
":operators",
":params",
":requantization",
":subgraph_h",
":xnnpack_h",
"@pthreadpool",
] + xnnpack_slinky_deps(),
Expand Down
3 changes: 0 additions & 3 deletions build_defs.bzl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -62,9 +62,6 @@ def xnnpack_optional_dnnl_deps():
"""Optional Intel DNNL dependencies."""
return []

def xnnpack_slinky_srcs():
return []

def xnnpack_slinky_deps():
return []

Expand Down
71 changes: 26 additions & 45 deletions src/runtime.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -553,8 +553,19 @@ enum xnn_status xnn_create_runtime_v4(
}

#ifdef XNN_SLINKY_ENABLED
runtime->slinky_pipeline = xnn_runtime_to_slinky_pipeline(runtime);
// If compiling with XNN_SLINKY_ENABLED defined, assume we always
// want Slinky enabled, regardless of the runtime flag
const bool use_slinky = true;
#else
const bool use_slinky = (flags & XNN_FLAG_SLINKY_ENABLED) != 0;
#endif
if (use_slinky) {
#ifdef XNN_SLINKY_AVAILABLE
// slinky_init_pipeline(runtime);
#else
xnn_log_warning("Slinky requested but not available");
#endif
}

for (uint32_t i = 0; i < runtime->num_values; i++) {
struct xnn_value* value = &runtime->values[i];
Expand Down Expand Up @@ -695,23 +706,9 @@ enum xnn_status xnn_setup_runtime(
}
}

#ifdef XNN_SLINKY_ENABLED
size_t input_id = 0, output_id = 0;
// Use the runtime values instead of the external values so the order is the
// same.
for (size_t i = 0; i < runtime->num_values; i++) {
struct xnn_value* value = &runtime->values[i];
if (xnn_value_is_static(value)) {
// The value is constant.
} else if (value->flags & XNN_VALUE_FLAG_EXTERNAL_INPUT) {
runtime->input_values[input_id++] = value;
} else if (value->flags & XNN_VALUE_FLAG_EXTERNAL_OUTPUT) {
runtime->output_values[output_id++] = value;
}
}
runtime->num_inputs = input_id;
runtime->num_outputs = output_id;
#endif
#ifdef XNN_SLINKY_AVAILABLE
// slinky_setup_inputs_and_outputs(runtime);
#endif

// Apply runtime state changes.
for (size_t i = 0; i < num_external_values; i++) {
Expand Down Expand Up @@ -794,23 +791,9 @@ enum xnn_status xnn_setup_runtime_v2(
value->data = external_value->data;
}

#ifdef XNN_SLINKY_ENABLED
size_t input_id = 0, output_id = 0;
// Use the runtime values instead of the external values so the order is the
// same.
for (size_t i = 0; i < runtime->num_values; i++) {
struct xnn_value* value = &runtime->values[i];
if (xnn_value_is_static(value)) {
// The value is constant.
} else if (value->flags & XNN_VALUE_FLAG_EXTERNAL_INPUT) {
runtime->input_values[input_id++] = value;
} else if (value->flags & XNN_VALUE_FLAG_EXTERNAL_OUTPUT) {
runtime->output_values[output_id++] = value;
}
}
runtime->num_inputs = input_id;
runtime->num_outputs = output_id;
#endif
#ifdef XNN_SLINKY_AVAILABLE
// slinky_setup_inputs_and_outputs(runtime);
#endif

for (uint32_t opdata_id = 0; opdata_id < runtime->num_ops; opdata_id++) {
struct xnn_operator_data* opdata = &runtime->opdata[opdata_id];
Expand Down Expand Up @@ -981,13 +964,10 @@ enum xnn_status xnn_get_runtime_profiling_info(xnn_runtime_t runtime,
enum xnn_status xnn_invoke_runtime(
xnn_runtime_t runtime)
{
#ifdef XNN_SLINKY_ENABLED
if (runtime->slinky_pipeline) {
return evaluate(runtime->slinky_pipeline, runtime->input_values,
runtime->num_inputs, runtime->output_values,
runtime->num_outputs);
}
#endif
#ifdef XNN_SLINKY_AVAILABLE
enum xnn_status status;
// if (slinky_evaluate(runtime, &status)) return status;
#endif

if (runtime->profiling) {
runtime->start_ts = xnn_read_timer();
Expand Down Expand Up @@ -1015,9 +995,10 @@ enum xnn_status xnn_delete_runtime(
xnn_runtime_t runtime)
{
if (runtime != NULL) {
#ifdef XNN_SLINKY_ENABLED
destroy_slinky_pipeline(runtime->slinky_pipeline);
#endif
#ifdef XNN_SLINKY_AVAILABLE
// slinky_destroy_pipeline(runtime);
#endif

if (runtime->opdata != NULL) {
for (size_t i = 0; i < runtime->num_ops; i++) {
for (size_t j = 0; j < XNN_MAX_OPERATOR_OBJECTS; j++) {
Expand Down
30 changes: 18 additions & 12 deletions src/xnnpack/subgraph.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -34,18 +34,23 @@
/// Disable fusion of nodes in subgraph. Fusion is enabled by default, set this flag to turn it off.
#define XNN_FLAG_NO_OPERATOR_FUSION 0x80000000

/// Enable Slinky (if available).
#define XNN_FLAG_SLINKY_ENABLED 0x40000000

#ifdef __cplusplus
extern "C" {
#endif

#ifdef XNN_SLINKY_ENABLED
struct xnn_value;
#ifdef XNN_SLINKY_AVAILABLE
/// Slinky interface -- unused unless XNN_FLAG_SLINKY_ENABLED is set
struct slinky_pipeline;
typedef struct slinky_pipeline* slinky_pipeline_t;
slinky_pipeline_t xnn_runtime_to_slinky_pipeline(xnn_runtime_t runtime);
void destroy_slinky_pipeline(slinky_pipeline_t pipeline);
enum xnn_status evaluate(slinky_pipeline_t p, struct xnn_value* const* input_values, size_t num_inputs, struct xnn_value* const* output_values, size_t num_outputs);
#endif

void slinky_init_pipeline(xnn_runtime_t runtime);
void slinky_setup_inputs_and_outputs(xnn_runtime_t runtime);
void slinky_destroy_pipeline(xnn_runtime_t runtime);
bool slinky_evaluate(xnn_runtime_t runtime, enum xnn_status* status);
#endif // XNN_SLINKY_AVAILABLE

struct xnn_shape {
size_t num_dims;
Expand Down Expand Up @@ -446,13 +451,14 @@ struct xnn_runtime {
bool has_been_setup;
bool memory_planned;

#ifdef XNN_SLINKY_ENABLED
#ifdef XNN_SLINKY_AVAILABLE
// Fields used by Slinky -- unused unless XNN_FLAG_SLINKY_ENABLED is set
slinky_pipeline_t slinky_pipeline;
size_t num_inputs;
size_t num_outputs;
struct xnn_value* input_values[XNN_MAX_SUBGRAPH_INPUT_OR_OUTPUTS];
struct xnn_value* output_values[XNN_MAX_SUBGRAPH_INPUT_OR_OUTPUTS];
#endif
size_t slinky_num_inputs;
size_t slinky_num_outputs;
struct xnn_value* slinky_input_values[XNN_MAX_SUBGRAPH_INPUT_OR_OUTPUTS];
struct xnn_value* slinky_output_values[XNN_MAX_SUBGRAPH_INPUT_OR_OUTPUTS];
#endif // XNN_SLINKY_AVAILABLE
};

enum xnn_status xnn_insert_clamp_node(xnn_subgraph_t subgraph, float output_min, float output_max, struct xnn_node *node);
Expand Down

0 comments on commit 8c3d6e5

Please sign in to comment.