experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
STT-tensorflow/tensorflow/lite/experimental/ruy/BUILD
Benoit Jacob ee25c2a3a6 Fix allocator in cases of sizes overflowing 32bit integer arithmetic 
in size_util.

Part of it was AllocateFast not checking if ptr_ is null
before using it (null deref with offset, so didn't look like a null deref).

Part of it was using round_up_pot with a large size_t value that got implicitly casted to int as round_up_pot took an int argument. This showed how it's safer to just templatize the helpers in size_util.h, make them accept either int32 or int64 (guarded in floor_log2, which is the only of these functions that cares).

I just changed allocator to use only signed types (std::size_t --> std::ptrdiff_t) because I didn't want to deal with the extra complexity of dealing with both signed and unsigned in size_util.

PiperOrigin-RevId: 264668215
2019-08-21 13:07:37 -07:00

501 lines
8.6 KiB
Python
Raw Permalink Blame History

# Ruy is not BLAS
# TODO(b/123403203) actually make TFLite use ruy.
load(":ruy_visibility.bzl", "ruy_visibility")
load(":ruy_test_ext.bzl", "ruy_test_ext_defines", "ruy_test_ext_deps")
load(":ruy_test.bzl", "ruy_benchmark", "ruy_benchmark_opt_sets", "ruy_test")
load("//tensorflow/lite:special_rules.bzl", "tflite_portable_test_suite")
package(
default_visibility = ["//visibility:private"],
licenses = ["notice"], # Apache 2.0
)
cc_library(
name = "platform",
hdrs = ["platform.h"],
)
cc_library(
name = "check_macros",
hdrs = ["check_macros.h"],
deps = ["//tensorflow/lite/kernels/internal:compatibility"],
)
cc_library(
name = "opt_set",
hdrs = ["opt_set.h"],
)
cc_library(
name = "time",
hdrs = ["time.h"],
)
cc_library(
name = "wait",
srcs = ["wait.cc"],
hdrs = ["wait.h"],
deps = [":time"],
)
cc_test(
name = "wait_test",
srcs = ["wait_test.cc"],
deps = [
":wait",
"@com_google_googletest//:gtest",
],
)
cc_library(
name = "size_util",
hdrs = ["size_util.h"],
deps = [":check_macros"],
)
cc_test(
name = "size_util_test",
srcs = ["size_util_test.cc"],
deps = [
":size_util",
"@com_google_googletest//:gtest",
],
)
cc_library(
name = "tune",
srcs = [
"tune.cc",
],
hdrs = [
"tune.h",
],
deps = [
":opt_set",
":platform",
":time",
],
)
cc_test(
name = "tune_test",
srcs = ["tune_test.cc"],
deps = [
":tune",
"@com_google_googletest//:gtest",
],
)
cc_binary(
name = "tune_tool",
srcs = ["tune_tool.cc"],
deps = [
":tune",
],
)
cc_library(
name = "allocator",
srcs = [
"allocator.cc",
],
hdrs = [
"allocator.h",
],
deps = [
":check_macros",
":size_util",
],
)
cc_test(
name = "allocator_test",
srcs = ["allocator_test.cc"],
deps = [
":allocator",
"@com_google_googletest//:gtest",
],
)
cc_library(
name = "side_pair",
hdrs = ["side_pair.h"],
deps = [":check_macros"],
)
cc_library(
name = "block_map",
srcs = [
"block_map.cc",
],
hdrs = [
"block_map.h",
],
deps = [
":check_macros",
":opt_set",
":side_pair",
":size_util",
"@gemmlowp//:profiler",
],
)
cc_library(
name = "blocking_counter",
srcs = [
"blocking_counter.cc",
],
hdrs = [
"blocking_counter.h",
],
deps = [
":check_macros",
":wait",
],
)
cc_library(
name = "thread_pool",
srcs = [
"thread_pool.cc",
],
hdrs = [
"thread_pool.h",
],
visibility = ruy_visibility(),
deps = [
":blocking_counter",
":check_macros",
":wait",
],
)
cc_library(
name = "detect_dotprod",
srcs = [
"detect_dotprod.cc",
],
hdrs = [
"detect_dotprod.h",
],
visibility = ruy_visibility(),
)
cc_library(
name = "path",
hdrs = ["path.h"],
visibility = ruy_visibility(),
deps = [
":platform",
":size_util",
],
)
cc_library(
name = "trace",
srcs = [
"trace.cc",
],
hdrs = [
"trace.h",
],
deps = [
":block_map",
":check_macros",
":side_pair",
":time",
],
)
cc_library(
name = "context",
srcs = [
"context.cc",
],
hdrs = [
"context.h",
],
visibility = ruy_visibility(),
deps = [
":allocator",
":check_macros",
":detect_dotprod",
":path",
":thread_pool",
":trace",
":tune",
],
)
cc_library(
name = "matrix",
hdrs = ["matrix.h"],
visibility = ruy_visibility(),
deps = [":check_macros"],
)
cc_library(
name = "spec",
hdrs = ["spec.h"],
visibility = ruy_visibility(),
deps = [":matrix"],
)
cc_library(
name = "internal_matrix",
hdrs = ["internal_matrix.h"],
deps = [
":check_macros",
":common",
":matrix",
":size_util",
],
)
cc_library(
name = "common",
hdrs = [
"common.h",
],
deps = [
":check_macros",
":matrix",
":opt_set",
":path",
":platform",
],
)
cc_library(
name = "kernel",
srcs = [
"kernel_arm32.cc",
"kernel_arm64.cc",
"kernel_avx512.cc",
],
hdrs = [
"kernel.h",
],
deps = [
":check_macros",
":common",
":internal_matrix",
":opt_set",
":path",
":platform",
":side_pair",
":size_util",
":spec",
":tune",
"@gemmlowp//:fixedpoint",
"@gemmlowp//:profiler",
],
)
cc_library(
name = "pack",
srcs = [
"pack_arm.cc",
"pack_avx512.cc",
],
hdrs = [
"pack.h",
],
deps = [
":common",
":internal_matrix",
":opt_set",
":path",
":platform",
":spec",
":tune",
"@gemmlowp//:profiler",
],
)
cc_library(
name = "trmul_params",
hdrs = ["trmul_params.h"],
deps = [
":context",
":internal_matrix",
":side_pair",
":tune",
],
)
cc_library(
name = "trmul",
srcs = ["trmul.cc"],
hdrs = ["trmul.h"],
deps = [
":allocator",
":block_map",
":check_macros",
":common",
":context",
":internal_matrix",
":opt_set",
":side_pair",
":spec",
":thread_pool",
":trace",
":trmul_params",
":tune",
"@gemmlowp//:profiler",
],
)
# The main library.
cc_library(
name = "ruy",
srcs = [
"dispatch.h",
"prepack.h",
],
hdrs = [
"ruy.h",
"ruy_advanced.h",
],
visibility = ruy_visibility(),
deps = [
":check_macros",
":common",
":context",
":kernel",
":matrix",
":pack",
":path",
":side_pair",
":size_util",
":spec",
":trmul",
":tune",
"@gemmlowp//:profiler",
],
)
# Usage examples.
cc_binary(
name = "example",
srcs = ["example.cc"],
deps = [
":ruy",
],
)
# Usage examples of the advanced API.
cc_binary(
name = "example_advanced",
srcs = ["example_advanced.cc"],
deps = [
":ruy",
],
)
# Small library to query PMU counters, for benchmark only
cc_library(
name = "pmu",
testonly = True,
srcs = ["pmu.cc"],
hdrs = ["pmu.h"],
deps = [":check_macros"],
)
# Testing framework.
cc_library(
name = "test_lib",
testonly = True,
hdrs = ["test.h"],
# need defines, not copts, because it's controlling a header, test.h
defines = ruy_test_ext_defines(),
linkopts = select({
"//tensorflow:windows": [],
"//conditions:default": ["-lm"],
}),
deps = [
":pmu",
":ruy",
":time",
"@com_google_googletest//:gtest",
":platform",
] + ruy_test_ext_deps(),
)
ruy_benchmark(
name = "benchmark",
srcs = ["benchmark.cc"],
lhs_rhs_accum_dst = [
("f32", "f32", "f32", "f32"),
("u8", "u8", "i32", "u8"),
("i8", "i8", "i32", "u8"),
("i8", "i8", "i32", "i8"),
("u8", "u8", "i32", "i16"),
("i8", "i8", "i32", "i32"),
],
)
ruy_test(
name = "test_fast",
srcs = ["test_fast.cc"],
lhs_rhs_accum_dst = [
("f32", "f32", "f32", "f32"),
("f64", "f32", "f64", "f32"),
("f32", "f64", "f64", "f64"),
("u8", "u8", "i32", "u8"),
("i8", "i8", "i32", "i8"),
("i8", "u8", "i32", "i8"),
("u8", "u8", "i32", "i16"),
("i8", "i8", "i32", "i32"),
],
)
ruy_test(
name = "test_slow",
srcs = ["test_slow.cc"],
lhs_rhs_accum_dst = [
("f32", "f32", "f32", "f32"),
("u8", "u8", "i32", "u8"),
("i8", "i8", "i32", "i8"),
("u8", "u8", "i32", "i16"),
("i8", "i8", "i32", "i32"),
],
tags = ["slow"],
)
ruy_test(
name = "test_special_specs",
srcs = ["test_special_specs.cc"],
lhs_rhs_accum_dst = [
("f32", "f32", "f32", "f32"),
("u8", "u8", "i32", "u8"),
("u8", "u8", "i32", "i16"),
],
)
ruy_benchmark_opt_sets(
name = "benchmark_opt_set",
srcs = ["benchmark.cc"],
lhs_rhs_accum_dst = [
("f32", "f32", "f32", "f32"),
("u8", "u8", "i32", "u8"),
],
opt_sets = [
"1",
"3",
"7",
"f",
"1f",
"3f",
"7f",
"ff",
"1ff",
"3ff",
"7ff",
],
)
tflite_portable_test_suite()
Reference in New Issue View Git Blame Copy Permalink