Change op names to linear_8bit_act_xbit_weight

Differential Revision: D63347192

Pull Request resolved: #978

Author: metascroy

torchao/experimental/CMakeLists.txt

@@ -25,14 +25,14 @@ if(NOT TORCHAO_INCLUDE_DIRS)
 endif()
 
 if (NOT TORCHAO_OP_TARGET)
-    message(FATAL_ERROR "TORCHAO_OP_TARGET is not set.  Set it to ATEN or EXECUTORCH.")
+    message(FATAL_ERROR "TORCHAO_OP_TARGET is not set.  Set it to aten or executorch.")
 endif()
 
 if (NOT TORCHAO_PARALLEL_BACKEND)
-    if (TORCHAO_OP_TARGET STREQUAL "ATEN")
-        set(TORCHAO_PARALLEL_BACKEND "ATEN_OPENMP")
-    elseif(TORCHAO_OP_TARGET STREQUAL "EXECUTORCH")
-        set(TORCHAO_PARALLEL_BACKEND "EXECUTORCH")
+    if (TORCHAO_OP_TARGET STREQUAL "aten")
+        set(TORCHAO_PARALLEL_BACKEND "aten_openmp")
+    elseif(TORCHAO_OP_TARGET STREQUAL "executorch")
+        set(TORCHAO_PARALLEL_BACKEND "executorch")
     else()
         message(TORCHAO_PARALLEL_BACKEND "TORCHAO_PARALLEL_BACKEND is not set.  Please set it directly or set TORCHAO_OP_TARGET to get a default.")
     endif()
@@ -46,9 +46,26 @@ include(CMakePrintHelpers)
 message("TORCHAO_INCLUDE_DIRS: ${TORCHAO_INCLUDE_DIRS}")
 include_directories(${TORCHAO_INCLUDE_DIRS})
 
+if(TORCHAO_OP_TARGET STREQUAL "aten")
+    add_library(torchao_ops_${TORCHAO_OP_TARGET} SHARED)
+elseif(TORCHAO_OP_TARGET STREQUAL "executorch")
+    add_library(torchao_ops_${TORCHAO_OP_TARGET} STATIC)
+else()
+    message(FATAL_ERROR "Unknown TORCHAO_OP_TARGET: ${TORCHAO_OP_TARGET}. Please choose one of: aten, executorch.")
+endif()
+
 if (CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
     # Defines target torchao_kernels_aarch64
     add_subdirectory(${TORCHAO_ROOT}/kernels/cpu/aarch64)
-    add_subdirectory(${TORCHAO_ROOT}/ops/linear)
-    add_subdirectory(${TORCHAO_ROOT}/ops/linear/linear_a8wxdq_op)
+    add_subdirectory(${TORCHAO_ROOT}/ops/linear_8bit_act_xbit_weight)
+
+    target_link_libraries(
+        torchao_ops_${TORCHAO_OP_TARGET} PRIVATE
+        torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET}
+    )
 endif()
+
+install(
+  TARGETS  torchao_ops_${TORCHAO_OP_TARGET}
+  DESTINATION lib
+)

torchao/experimental/ops/linear/benchmarks/CMakeLists.txt renamed to torchao/experimental/ops/benchmarks/CMakeLists.txt

@@ -11,8 +11,8 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_BUILD_TYPE Release)
 add_compile_options("-Wall" "-Werror")
 
-set(TORCHAO_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/../../..)
-set(TORCHAO_INCLUDE_DIRS ${CMAKE_CURRENT_SOURCE_DIR}/../../../../..)
+set(TORCHAO_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/../..)
+set(TORCHAO_INCLUDE_DIRS ${CMAKE_CURRENT_SOURCE_DIR}/../../../..)
 
 include(FetchContent)
 FetchContent_Declare(googlebenchmark
@@ -25,16 +25,20 @@ FetchContent_MakeAvailable(
 
 include_directories(${TORCHAO_INCLUDE_DIRS})
 
-set(TORCHAO_PARALLEL_BACKEND "OPENMP")
-add_subdirectory(${TORCHAO_ROOT}/ops/linear ${CMAKE_CURRENT_BINARY_DIR}/torchao_ops_linear_${TORCHAO_PARALLEL_BACKEND})
+set(TORCHAO_PARALLEL_BACKEND "openmp")
+
+include(${TORCHAO_ROOT}/Utils.cmake)
+
 add_subdirectory(${TORCHAO_ROOT}/kernels/cpu/aarch64 ${CMAKE_CURRENT_BINARY_DIR}/torchao_kernels_aarch64)
 
-add_executable(benchmark_linear_operator benchmark_linear_operator.cpp)
+add_executable(benchmark_linear_8bit_act_xbit_weight
+  benchmark_linear_8bit_act_xbit_weight.cpp
+  ${TORCHAO_ROOT}/ops/linear_8bit_act_xbit_weight/linear_8bit_act_xbit_weight.cpp
+)
+target_link_torchao_parallel_backend(benchmark_linear_8bit_act_xbit_weight "${TORCHAO_PARALLEL_BACKEND}")
 target_link_libraries(
-    benchmark_linear_operator
+    benchmark_linear_8bit_act_xbit_weight
     PRIVATE
     benchmark::benchmark
     torchao_kernels_aarch64
-    torchao_ops_linear_${TORCHAO_PARALLEL_BACKEND}
 )
-target_link_torchao_parallel_backend(benchmark_linear_operator "${TORCHAO_PARALLEL_BACKEND}")

torchao/experimental/ops/linear/benchmarks/benchmark_linear_operator.cpp renamed to torchao/experimental/ops/benchmarks/benchmark_linear_8bit_act_xbit_weight.cpp

@@ -7,13 +7,12 @@
 #include <benchmark/benchmark.h>
 #include <torchao/experimental/kernels/cpu/aarch64/linear/linear.h>
 #include <torchao/experimental/kernels/cpu/aarch64/tests/test_utils.h>
-#include <torchao/experimental/ops/linear/channelwise_8bit_activation_groupwise_lowbit_weight.h>
+#include <torchao/experimental/ops/linear_8bit_act_xbit_weight/linear_8bit_act_xbit_weight.h>
 #include <torchao/experimental/ops/memory.h>
 #include <torchao/experimental/ops/parallel.h>
 #include <vector>
 
-using namespace torchao::ops::linear::
-    channelwise_8bit_activation_groupwise_lowbit_weight;
+using namespace torchao::ops::linear_8bit_act_xbit_weight;
 
 template <int weight_nbit, bool has_weight_zeros, bool has_bias, bool has_clamp>
 UKernelConfig get_ukernel_config() {
@@ -40,8 +39,7 @@ UKernelConfig get_ukernel_config() {
 }
 
 template <int weight_nbit, bool has_weight_zeros, bool has_bias, bool has_clamp>
-static void channelwise_8bit_activation_groupwise_lowbit_weight(
-    benchmark::State& state) {
+static void linear_8bit_act_xbit_weight(benchmark::State& state) {
   int m = state.range(0);
   int n = state.range(1);
   int k = state.range(2);
@@ -150,19 +148,20 @@ static void channelwise_8bit_activation_groupwise_lowbit_weight(
     }                                                                    \
   }
 
-#define BENCHMARK_CHANNELWISE_8BIT_ACTIVATION_GROUPWISE_LOWBIT_WEIGHT( \
-    weight_nbit)                                                       \
-  BENCHMARK(channelwise_8bit_activation_groupwise_lowbit_weight<       \
-                weight_nbit,                                           \
-                false /*has_weight_zeros*/,                            \
-                false /*has_bias*/,                                    \
-                false /*has_clamp*/>)                                  \
-      ->ArgsProduct(BENCHMARK_PARAMS)                                  \
-      ->ArgNames(                                                      \
+#define BENCHMARK_LINEAR_8BIT_ACT_XBIT_WEIGHT(weight_nbit) \
+  BENCHMARK(linear_8bit_act_xbit_weight<                   \
+                weight_nbit,                 \
+                false /*has_weight_zeros*/,  \
+                false /*has_bias*/,          \
+                false /*has_clamp*/>)        \
+      ->ArgsProduct(BENCHMARK_PARAMS)        \
+      ->ArgNames(                            \
           {"m", "n", "k", "group_size", "num_threads", "num_test_cases"});
 
-BENCHMARK_CHANNELWISE_8BIT_ACTIVATION_GROUPWISE_LOWBIT_WEIGHT(3);
-BENCHMARK_CHANNELWISE_8BIT_ACTIVATION_GROUPWISE_LOWBIT_WEIGHT(4);
+BENCHMARK_LINEAR_8BIT_ACT_XBIT_WEIGHT(2);
+BENCHMARK_LINEAR_8BIT_ACT_XBIT_WEIGHT(3);
+BENCHMARK_LINEAR_8BIT_ACT_XBIT_WEIGHT(4);
+BENCHMARK_LINEAR_8BIT_ACT_XBIT_WEIGHT(5);
 
 // Run the benchmark
 BENCHMARK_MAIN();

torchao/experimental/ops/linear/benchmarks/build_and_run_benchmarks.sh renamed to torchao/experimental/ops/benchmarks/build_and_run_benchmarks.sh

@@ -17,7 +17,4 @@ cmake -DTORCHAO_LIBRARIES=${TORCHAO_LIBRARIES} \
 cmake --build ${CMAKE_OUT}
 
 # Run
-case "$1" in
-    linear_operator) ${CMAKE_OUT}/benchmark_linear_operator; ;;
-    *) echo "Unknown benchmark: $1. Please specify one of: linear_operator."; exit 1; ;;
-esac
+${CMAKE_OUT}/benchmark_linear_8bit_act_xbit_weight

torchao/experimental/ops/linear/CMakeLists.txt

@@ -0,0 +1,44 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+cmake_minimum_required(VERSION 3.19)
+
+include(${TORCHAO_ROOT}/Utils.cmake)
+
+
+if(TORCHAO_OP_TARGET STREQUAL "aten")
+    message(STATUS "Building with TORCHAO_OP_TARGET=aten")
+    find_package(Torch REQUIRED)
+    add_library(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} OBJECT
+        linear_8bit_act_xbit_weight.cpp
+        op_linear_8bit_act_xbit_weight_aten.cpp
+    )
+    target_link_torchao_parallel_backend(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} "${TORCHAO_PARALLEL_BACKEND}")
+    target_link_libraries(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE torchao_kernels_aarch64)
+    target_include_directories(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE "${TORCH_INCLUDE_DIRS}")
+    target_link_libraries(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE "${TORCH_LIBRARIES}")
+    target_compile_definitions(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE USE_ATEN=1)
+elseif(TORCHAO_OP_TARGET STREQUAL "executorch")
+    message(STATUS "Building with TORCHAO_OP_TARGET=executorch")
+    add_library(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} OBJECT
+        linear_8bit_act_xbit_weight.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w2s.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w2sz.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w3s.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w3sz.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w4s.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w4sz.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w5s.cpp
+        op_linear_8bit_act_xbit_weight_executorch/w5sz.cpp
+    )
+    target_link_torchao_parallel_backend(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} "${TORCHAO_PARALLEL_BACKEND}")
+    target_include_directories(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE "${EXECUTORCH_INCLUDE_DIRS}")
+    target_compile_definitions(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE USE_EXECUTORCH=1)
+    target_link_libraries(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE "${EXECUTORCH_LIBRARIES}")
+    target_link_libraries(torchao_ops_linear_8bit_act_xbit_weight_${TORCHAO_OP_TARGET} PRIVATE torchao_kernels_aarch64)
+else()
+    message(FATAL_ERROR "Unknown TORCHAO_OP_TARGET: ${TORCHAO_OP_TARGET}. Please choose one of: aten, executorch.")
+endif()

torchao/experimental/ops/linear/linear_a8wxdq_op/CMakeLists.txt

@@ -17,26 +17,29 @@ set(TORCHAO_INCLUDE_DIRS ${CMAKE_CURRENT_SOURCE_DIR}/../../../../..)
 
 include_directories(${TORCHAO_INCLUDE_DIRS})
 
-set(TORCHAO_PARALLEL_BACKEND "OPENMP")
-add_subdirectory(${TORCHAO_ROOT}/ops/linear ${CMAKE_CURRENT_BINARY_DIR}/torchao_ops_linear_${TORCHAO_PARALLEL_BACKEND})
+set(TORCHAO_PARALLEL_BACKEND "openmp")
 add_subdirectory(${TORCHAO_ROOT}/kernels/cpu/aarch64 ${CMAKE_CURRENT_BINARY_DIR}/torchao_kernels_aarch64)
 
 include(${TORCHAO_ROOT}/Utils.cmake)
 
-add_executable(separate_function_wrappers separate_function_wrappers.cpp)
+add_executable(separate_function_wrappers
+  separate_function_wrappers.cpp
+  ${TORCHAO_ROOT}/ops/linear_8bit_act_xbit_weight/linear_8bit_act_xbit_weight.cpp
+)
 target_link_libraries(
   separate_function_wrappers
     PRIVATE
     torchao_kernels_aarch64
-    torchao_ops_linear_${TORCHAO_PARALLEL_BACKEND}
 )
 target_link_torchao_parallel_backend(separate_function_wrappers "${TORCHAO_PARALLEL_BACKEND}")
 
-add_executable(stateful_class_wrapper stateful_class_wrapper.cpp)
+add_executable(stateful_class_wrapper
+  stateful_class_wrapper.cpp
+  ${TORCHAO_ROOT}/ops/linear_8bit_act_xbit_weight/linear_8bit_act_xbit_weight.cpp
+)
 target_link_libraries(
   stateful_class_wrapper
     PRIVATE
     torchao_kernels_aarch64
-    torchao_ops_linear_${TORCHAO_PARALLEL_BACKEND}
 )
 target_link_torchao_parallel_backend(stateful_class_wrapper "${TORCHAO_PARALLEL_BACKEND}")

torchao/experimental/ops/linear_8bit_act_xbit_weight/CMakeLists.txt

@@ -5,16 +5,15 @@
 // LICENSE file in the root directory of this source tree.
 
 #pragma once
-#include <torchao/experimental/ops/linear/channelwise_8bit_activation_groupwise_lowbit_weight.h>
+#include <torchao/experimental/ops/linear_8bit_act_xbit_weight/linear_8bit_act_xbit_weight.h>
 #include <torchao/experimental/ops/macro.h>
 #include <torchao/experimental/ops/memory.h>
 #include <cassert>
 #include <optional>
 
-namespace torchao::ops::linear::
-    channelwise_8bit_activation_groupwise_lowbit_weight {
+namespace torchao::ops::linear_8bit_act_xbit_weight {
 
-class Channelwise8BitActivationGroupwiseLowbitWeightLinearOperator {
+class Linear8BitActXBitWeightOperator {
  private:
   torchao::aligned_byte_ptr packed_weight_data_{nullptr, nullptr};
   int packed_weight_data_size_{0};
@@ -40,7 +39,7 @@ class Channelwise8BitActivationGroupwiseLowbitWeightLinearOperator {
   LinearTileSchedulingPolicy linear_scheduling_policy_;
 
  public:
-  Channelwise8BitActivationGroupwiseLowbitWeightLinearOperator(
+  Linear8BitActXBitWeightOperator(
       UKernelConfig ukernel_config,
       int n,
       int k,
@@ -195,4 +194,4 @@ class Channelwise8BitActivationGroupwiseLowbitWeightLinearOperator {
   }
 };
 } // namespace
-  // torchao::ops::linear::channelwise_8bit_activation_groupwise_lowbit_weight
+  // torchao::ops::linear_8bit_act_xbit_weight

torchao/experimental/ops/linear/examples/CMakeLists.txt renamed to torchao/experimental/ops/linear_8bit_act_xbit_weight/examples/CMakeLists.txt

@@ -6,7 +6,7 @@
 
 #include <torchao/experimental/kernels/cpu/aarch64/linear/linear.h>
 #include <torchao/experimental/kernels/cpu/aarch64/tests/test_utils.h>
-#include <torchao/experimental/ops/linear/channelwise_8bit_activation_groupwise_lowbit_weight.h>
+#include <torchao/experimental/ops/linear_8bit_act_xbit_weight/linear_8bit_act_xbit_weight.h>
 #include <torchao/experimental/ops/memory.h>
 #include <torchao/experimental/ops/parallel.h>
 #include <iostream>
@@ -22,8 +22,7 @@
 // one stateful class, but not all surfaces support this (see
 // examples/stateful_class_wrapper.cpp for an example of this).
 
-namespace torchao::ops::linear::
-    channelwise_8bit_activation_groupwise_lowbit_weight {
+namespace torchao::ops::linear_8bit_act_xbit_weight {
 
 template <int weight_nbit, bool has_weight_zeros, bool has_bias, bool has_clamp>
 UKernelConfig get_ukernel_config() {
@@ -141,11 +140,10 @@ void linear_operator(
 }
 
 } // namespace
-  // torchao::ops::linear::channelwise_8bit_activation_groupwise_lowbit_weight
+  // torchao::ops::linear_8bit_act_xbit_weight
 
 int main() {
-  using namespace torchao::ops::linear::
-      channelwise_8bit_activation_groupwise_lowbit_weight;
+  using namespace torchao::ops::linear_8bit_act_xbit_weight;
 
   torchao::set_num_threads(8);
   std::cout << "Using " << torchao::get_num_threads() << " threads."

torchao/experimental/ops/linear/examples/Channelwise8BitActivationGroupwiseLowbitWeightLinearOperator.h renamed to torchao/experimental/ops/linear_8bit_act_xbit_weight/examples/Linear8BitActXBitWeightOperator.h

@@ -6,7 +6,7 @@
 
 #include <torchao/experimental/kernels/cpu/aarch64/linear/linear.h>
 #include <torchao/experimental/kernels/cpu/aarch64/tests/test_utils.h>
-#include <torchao/experimental/ops/linear/examples/Channelwise8BitActivationGroupwiseLowbitWeightLinearOperator.h>
+#include <torchao/experimental/ops/linear_8bit_act_xbit_weight/examples/Linear8BitActXBitWeightOperator.h>
 #include <torchao/experimental/ops/parallel.h>
 #include <iostream>
 #include <vector>
@@ -22,8 +22,7 @@
 // examples/separate_function_wrappers.cpp for an example of how to split the
 // operations into two steps.
 
-using namespace torchao::ops::linear::
-    channelwise_8bit_activation_groupwise_lowbit_weight;
+using namespace torchao::ops::linear_8bit_act_xbit_weight;
 
 template <int weight_nbit, bool has_weight_zeros, bool has_bias, bool has_clamp>
 UKernelConfig get_ukernel_config() {
@@ -81,7 +80,7 @@ int main() {
       get_ukernel_config<weight_nbit, has_weight_zeros, has_bias, has_clamp>();
 
   auto linear_operator =
-      Channelwise8BitActivationGroupwiseLowbitWeightLinearOperator(
+      Linear8BitActXBitWeightOperator(
           ukernel_config,
           n,
           k,