Rename all jblas files and their usage (#7)

* pass compilation * rename common files * BesTLA ut passed * rename variables and namespaces
intel · Jan 1, 2024 · d5c26d4 · d5c26d4
1 parent b330746
commit d5c26d4
Show file tree

Hide file tree

Showing 121 changed files with 9,113 additions and 6,501 deletions.
diff --git a/.gitignore b/.gitignore
@@ -0,0 +1,6 @@
+################################################################################
+# This .gitignore file was automatically created by Microsoft(R) Visual Studio.
+################################################################################
+
+/.vs
+/out
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -70,6 +70,7 @@ if (NE_GPU)
 endif()
 
 option(NE_BUILD_TESTS            "neural_engine: build tests"    ${NE_STANDALONE})
+option(NE_BTLA_UT                "enable BesTLA's unit tests"    OFF)
 option(NE_BUILD_EXAMPLES         "neural_engine: build examples" ${NE_STANDALONE})
 if(NE_BUILD_TESTS)
   add_compile_definitions(NE_BUILD_TESTS)
@@ -139,6 +140,11 @@ if (NE_PYTHON_API)
   add_subdirectory(third_party/pybind11)
 endif()
 
-add_subdirectory(bestla jblas)
+if (NE_BTLA_UT)
+  set(BTLA_UT_ALL ON)
+endif()
+include(FindOpenMP)
+
+add_subdirectory(bestla)
 
 add_subdirectory(neural_speed)
diff --git a/CMakePresets.json b/CMakePresets.json
@@ -57,6 +57,16 @@
       "description": "Target Windows (64-bit) with the Visual Studio development environment. (RelWithDebInfo)",
       "inherits": "x64-debug",
       "cacheVariables": { "CMAKE_BUILD_TYPE": "Release" }
+    },
+    {
+      "name": "x64-bestla-UT",
+      "displayName": "x64 BesTLA unit test",
+      "description": "Target Windows (64-bit) with the Visual Studio development environment. (RelWithDebInfo)",
+      "inherits": "x64-debug",
+      "cacheVariables": {
+        "CMAKE_BUILD_TYPE": "Release",
+        "NE_BTLA_UT": "ON"
+      }
     }
   ]
 }
diff --git a/README.md b/README.md
@@ -385,7 +385,7 @@ Argument description of inference.py:
 | --keep                                            | Number of tokens to keep from the initial prompt: Int (default: 0, -1 = all)                                                                                                            |
 | --shift-roped-k                                   | Use [ring-buffer](./docs/infinite_inference.md#shift-rope-k-and-ring-buffer) and thus do not re-computing after reaching ctx_size (default: False)                                      |
 | --glm_tokenizer                                   | The path of the chatglm tokenizer: String (default: THUDM/chatglm-6b)                                                                                                                   |
-| --memory-f32 <br> --memory-f16 <br> --memory-auto | Data type of kv memory (default to auto);<br>If set to auto, the runtime will try with jblas flash attn managed format (currently requires GCC11+ & AMX) and fall back to fp16 if failed |
+| --memory-f32 <br> --memory-f16 <br> --memory-auto | Data type of kv memory (default to auto);<br>If set to auto, the runtime will try with bestla flash attn managed format (currently requires GCC11+ & AMX) and fall back to fp16 if failed |
 
 
 ### 3. Tensor Parallelism cross nodes/sockets

diff --git a/bestla/CMakeLists.txt b/bestla/CMakeLists.txt
@@ -1,40 +1,40 @@
-cmake_minimum_required(VERSION 3.5)
+cmake_minimum_required(VERSION 3.12)
 
-project(jblas LANGUAGES CXX VERSION 0.1.0)
+project(bestla LANGUAGES CXX VERSION 0.1.0)
 file(GLOB headers ${PROJECT_NAME}/*.h ${PROJECT_NAME}/*.hpp)
 file(GLOB xbyak_headers ${PROJECT_NAME}/xbyak/*.h ${PROJECT_NAME}/xbyak/*.hpp)
 
-option(JBLAS_UT_ALL "Enable all unit tests" OFF)
-option(JBLAS_UT_DEBUG "Enable debug unit tests" ON)
-option(JBLAS_UT_EPILOGUE "Enable unit test for epilogue" OFF)
-option(JBLAS_UT_PROLOGUE_A "Enable unit test for activation prologue" OFF)
-option(JBLAS_UT_PROLOGUE_B "Enable unit test for weight prologue" OFF)
-option(JBLAS_UT_GEMM "Enable unit test for micro gemm kernels" OFF)
-option(JBLAS_UT_WRAPPER "Enable unit test for parallel gemms" OFF)
-option(JBLAS_UT_PARALLEL "Enable unit test for parallel set" OFF)
-option(JBLAS_UT_KERNEL_JIT "Enable unit test for jit kernels" OFF)
-option(JBLAS_UT_KERNEL_INTRIN "Enable unit test for intrinsic kernels" OFF)
-option(JBLAS_UT_KERNEL_WRAPPER "Enable unit test for runtime ISA kernels" OFF)
-option(JBLAS_UT_NOASAN "Disable sanitize" OFF)
-option(JBLAS_UT_BENCHMARK "Benchmark ON may take a long time to finish all tests" OFF)
-option(JBLAS_UT_OPENMP "Use OpenMP" ON)
-
-if(JBLAS_UT_ALL)
-set(JBLAS_UT_EPILOGUE ON)
-set(JBLAS_UT_PROLOGUE_A ON)
-set(JBLAS_UT_PROLOGUE_B ON)
-set(JBLAS_UT_GEMM ON)
-set(JBLAS_UT_WRAPPER ON)
-set(JBLAS_UT_PARALLEL ON)
-set(JBLAS_UT_KERNEL_JIT ON)
-set(JBLAS_UT_KERNEL_INTRIN ON)
-set(JBLAS_UT_KERNEL_WRAPPER ON)
-endif(JBLAS_UT_ALL)
+option(BTLA_UT_ALL "Enable all unit tests" OFF)
+option(BTLA_UT_DEBUG "Enable debug unit tests" ON)
+option(BTLA_UT_EPILOGUE "Enable unit test for epilogue" OFF)
+option(BTLA_UT_PROLOGUE_A "Enable unit test for activation prologue" OFF)
+option(BTLA_UT_PROLOGUE_B "Enable unit test for weight prologue" OFF)
+option(BTLA_UT_GEMM "Enable unit test for micro gemm kernels" OFF)
+option(BTLA_UT_WRAPPER "Enable unit test for parallel gemms" OFF)
+option(BTLA_UT_PARALLEL "Enable unit test for parallel set" OFF)
+option(BTLA_UT_KERNEL_JIT "Enable unit test for jit kernels" OFF)
+option(BTLA_UT_KERNEL_INTRIN "Enable unit test for intrinsic kernels" OFF)
+option(BTLA_UT_KERNEL_WRAPPER "Enable unit test for runtime ISA kernels" OFF)
+option(BTLA_UT_NOASAN "Disable sanitize" OFF)
+option(BTLA_UT_BENCHMARK "Benchmark ON may take a long time to finish all tests" OFF)
+option(BTLA_UT_OPENMP "Use OpenMP" ON)
+
+if(BTLA_UT_ALL)
+set(BTLA_UT_EPILOGUE ON)
+set(BTLA_UT_PROLOGUE_A ON)
+set(BTLA_UT_PROLOGUE_B ON)
+set(BTLA_UT_GEMM ON)
+set(BTLA_UT_WRAPPER ON)
+set(BTLA_UT_PARALLEL ON)
+set(BTLA_UT_KERNEL_JIT ON)
+set(BTLA_UT_KERNEL_INTRIN ON)
+set(BTLA_UT_KERNEL_WRAPPER ON)
+endif(BTLA_UT_ALL)
 
 set(UT_BUILD FALSE)
-if(JBLAS_UT_DEBUG OR JBLAS_UT_PROLOGUE_A OR JBLAS_UT_PROLOGUE_B OR JBLAS_UT_EPILOGUE OR JBLAS_UT_GEMM
-OR JBLAS_UT_WRAPPER OR JBLAS_UT_PARALLEL OR JBLAS_UT_KERNEL_JIT OR JBLAS_UT_KERNEL_INTRIN
-OR JBLAS_UT_KERNEL_WRAPPER)
+if(BTLA_UT_DEBUG OR BTLA_UT_PROLOGUE_A OR BTLA_UT_PROLOGUE_B OR BTLA_UT_EPILOGUE OR BTLA_UT_GEMM
+OR BTLA_UT_WRAPPER OR BTLA_UT_PARALLEL OR BTLA_UT_KERNEL_JIT OR BTLA_UT_KERNEL_INTRIN
+OR BTLA_UT_KERNEL_WRAPPER)
 set(UT_BUILD TRUE)
 endif()
 
@@ -91,10 +91,7 @@ if(WIN32)
 	target_link_options(${PROJECT_NAME} INTERFACE /STACK:5242880) #Stack requires up to L2 cache size
 endif(WIN32)
 
-if(JBLAS_UT_OPENMP)
-include(FindOpenMP)
-target_link_libraries(${PROJECT_NAME} INTERFACE OpenMP::OpenMP_CXX OpenMP::OpenMP_C)
-endif()
+
 
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
@@ -109,27 +106,31 @@ target_compile_features(${PROJECT_NAME} INTERFACE cxx_std_17)
 if(UT_BUILD)
 	file(GLOB srcs ${PROJECT_NAME}/ut/*.cc ${PROJECT_NAME}/ut/*.cpp) #compile everthing even run parts of UTs
 	file(GLOB ut_headers ${PROJECT_NAME}/ut/*.h)
+  include_directories(${PROJECT_NAME})
 	add_executable(${PROJECT_NAME}_ut ${srcs} ${headers} ${ut_headers})
-
+  if(BTLA_UT_OPENMP)
+    include(FindOpenMP)
+    target_link_libraries(${PROJECT_NAME}_ut PRIVATE OpenMP::OpenMP_CXX OpenMP::OpenMP_C)
+  endif()
 	if(NOT WIN32)
-		if(NOT JBLAS_UT_NOASAN)
-		target_compile_options(${PROJECT_NAME}_ut PRIVATE -fsanitize=address)
-		target_link_options(${PROJECT_NAME}_ut PRIVATE -fsanitize=address)
+		if(NOT BTLA_UT_NOASAN)
+		  target_compile_options(${PROJECT_NAME}_ut PRIVATE -fsanitize=address)
+		  target_link_options(${PROJECT_NAME}_ut PRIVATE -fsanitize=address)
 		endif()
 		target_link_options(${PROJECT_NAME}_ut PRIVATE -lpthread)
 	endif()
 
-	add_ut_flag(JBLAS_UT_DEBUG)
-	add_ut_flag(JBLAS_UT_EPILOGUE)
-	add_ut_flag(JBLAS_UT_PROLOGUE_A)
-	add_ut_flag(JBLAS_UT_PROLOGUE_B)
-	add_ut_flag(JBLAS_UT_GEMM)
-	add_ut_flag(JBLAS_UT_PARALLEL)
-	add_ut_flag(JBLAS_UT_WRAPPER)
-	add_ut_flag(JBLAS_UT_KERNEL_INTRIN)
-	add_ut_flag(JBLAS_UT_KERNEL_JIT)
-	add_ut_flag(JBLAS_UT_KERNEL_WRAPPER)
-	add_ut_flag(JBLAS_UT_BENCHMARK)
+	add_ut_flag(BTLA_UT_DEBUG)
+	add_ut_flag(BTLA_UT_EPILOGUE)
+	add_ut_flag(BTLA_UT_PROLOGUE_A)
+	add_ut_flag(BTLA_UT_PROLOGUE_B)
+	add_ut_flag(BTLA_UT_GEMM)
+	add_ut_flag(BTLA_UT_PARALLEL)
+	add_ut_flag(BTLA_UT_WRAPPER)
+	add_ut_flag(BTLA_UT_KERNEL_INTRIN)
+	add_ut_flag(BTLA_UT_KERNEL_JIT)
+	add_ut_flag(BTLA_UT_KERNEL_WRAPPER)
+	add_ut_flag(BTLA_UT_BENCHMARK)
 
 	target_link_libraries(${PROJECT_NAME}_ut PRIVATE ${PROJECT_NAME})
 endif(UT_BUILD)

diff --git a/bestla/README.md b/bestla/README.md
@@ -52,6 +52,6 @@ Compile:
 
 Usage:
 ```cmake
-add_subdirectory(jblas)
-target_link_libraries("${YOUR_PROJECT}" jblas::jblas)
+add_subdirectory(bestla)
+target_link_libraries("${YOUR_PROJECT}" bestla::bestla)
 ```
diff --git a/bestla/jblas/jit_blas.h → bestla/bestla/bestla.h b/bestla/jblas/jit_blas.h → bestla/bestla/bestla.h
@@ -13,26 +13,26 @@
 //  limitations under the License.
 #pragma once
 #include <stdint.h>
-enum JBLAS_CODE {
-  JblasSuccess = 0,
-  JblasInvalidParam = 1,
-  JblasInvalidISA = 2,
-  JblasRuntimeError = 4,
-  JblasNotSupport = 8,
+enum class BTLA_CODE {
+  Success = 0,
+  InvalidParam = 1,
+  InvalidISA = 2,
+  RuntimeError = 4,
+  NotSupport = 8,
 };
-enum JBLAS_ISA : uint8_t {
-  JblasNoSIMD = 0,
-  JblasAVX,
-  JblasAVX2,
-  JblasAVX_VNNI,
-  JblasAVX512F,
-  JblasAVX512_VNNI,
-  JblasAMX_BF16,
-  JblasAMX_INT8,
-  JblasAVX512_FP16,
-  JblasAVX512_BF16,
+enum class BTLA_ISA : uint8_t {
+  NoSIMD = 0,
+  AVX,
+  AVX2,
+  AVX_VNNI,
+  AVX512F,
+  AVX512_VNNI,
+  AMX_BF16,
+  AMX_INT8,
+  AVX512_FP16,
+  AVX512_BF16,
 };
-enum class JBLAS_DTYPE : uint32_t {
+enum class BTLA_DTYPE : uint32_t {
   EleBitsMask = 0xff,
   EleBitsShift = 0,
   EleBitsUndef = 0,
@@ -70,15 +70,9 @@ enum class JBLAS_DTYPE : uint32_t {
   U32 = EleBits32 | TypeInt | SubType1,
 };
 
-enum JBLAS_LAYOUT { JblasRowMajor = 101, JblasColMajor = 102 };
-enum JBLAS_TRANSPOSE {
-  JblasNoTrans = 111,
-  JblasTrans = 112,
-  JblasConjTrans = 113,
-};
-enum JBLAS_ELTWISEOP { GELU, SWISH, TANH, EXP, LOW_PRECISION_EXP, RELU, LINEAR };
+enum class BTLA_ELTWISEOP { GELU, SWISH, TANH, EXP, LOW_PRECISION_EXP, RELU, LINEAR };
 
-enum class JBLAS_PROLOGUEB_IDS : uint32_t {
+enum class BTLA_PROLOGUEB_IDS : uint32_t {
   Undef = (uint32_t)-1,
   Begin = 0,
   NormalBegin = Begin,

diff --git a/bestla/jblas/jit_blas_device.h → bestla/bestla/bestla_device.h b/bestla/jblas/jit_blas_device.h → bestla/bestla/bestla_device.h
@@ -15,15 +15,15 @@
 #include <map>
 #include <thread>
 #include <vector>
-#include "jit_blas.h"
+#include "bestla.h"
 #include "xbyak/xbyak_util.h"
 #ifdef _WIN32
 #include <windows.h>
 #else
 #include <sched.h>
 #endif
 
-namespace jblas {
+namespace bestla {
 
 namespace device {
 
@@ -195,16 +195,16 @@ class SapphireRapids {
   static constexpr bool AMX_COMPLEX = 0;
 };
 
-template <JBLAS_ISA ISA_T>
+template <BTLA_ISA ISA_T>
 class isa_base {
  public:
-  static bool constexpr avx = ISA_T >= JblasAVX;
-  static bool constexpr avx2 = ISA_T >= JblasAVX2;
-  static bool constexpr avx512f = ISA_T >= JblasAVX512F;
-  static bool constexpr avx512_vnni = ISA_T >= JblasAVX512_VNNI;
-  static bool constexpr avx512_fp16 = ISA_T >= JblasAVX512_FP16;
-  static bool constexpr amx_bf16 = ISA_T >= JblasAMX_BF16;
-  static bool constexpr amx_int8 = ISA_T >= JblasAMX_INT8;
+  static bool constexpr avx = ISA_T >= BTLA_ISA::AVX;
+  static bool constexpr avx2 = ISA_T >= BTLA_ISA::AVX2;
+  static bool constexpr avx512f = ISA_T >= BTLA_ISA::AVX512F;
+  static bool constexpr avx512_vnni = ISA_T >= BTLA_ISA::AVX512_VNNI;
+  static bool constexpr avx512_fp16 = ISA_T >= BTLA_ISA::AVX512_FP16;
+  static bool constexpr amx_bf16 = ISA_T >= BTLA_ISA::AMX_BF16;
+  static bool constexpr amx_int8 = ISA_T >= BTLA_ISA::AMX_INT8;
 };
 
 class CpuDevice {
@@ -422,7 +422,7 @@ class CpuDevice {
   float P_power = 4.8, E_power = 2.3;
 };
 
-#define GetCPUDevice() auto _cd = jblas::device::CpuDevice::getInstance();
+#define GetCPUDevice() auto _cd = bestla::device::CpuDevice::getInstance();
 
 class CpuBase {
  public:
@@ -436,4 +436,4 @@ class CpuBase {
   int mNumThreads;
 };
 }  // namespace device
-}  // namespace jblas
+}  // namespace bestla