Bump asmjit to 2025-05-24

CMakeLists.txt

@@ -195,14 +195,8 @@ target_link_libraries(fbgemm_generic PUBLIC
   $<BUILD_INTERFACE:asmjit>
   $<BUILD_INTERFACE:cpuinfo>)
 
-# On Windows:
-# 1)  Adding definition of ASMJIT_STATIC to avoid generating asmjit function
-#     calls with _dllimport attribute
 if(MSVC)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4244 /wd4267 /wd4305 /wd4309")
-  if(FBGEMM_LIBRARY_TYPE STREQUAL "static")
-    target_compile_definitions(fbgemm_generic PUBLIC ASMJIT_STATIC)
-  endif()
   target_compile_options(fbgemm_avx2 PRIVATE "/arch:AVX2")
   target_compile_options(fbgemm_avx512 PRIVATE "/arch:AVX512")
   if(OpenMP_FOUND)
@@ -271,23 +265,10 @@ if(NOT TARGET asmjit)
   else()
     set(ASMJIT_STATIC ON)
   endif()
+  set(ASMJIT_NO_INSTALL ON)
 
   add_subdirectory("${ASMJIT_SRC_DIR}" "${FBGEMM_BINARY_DIR}/asmjit")
   set_property(TARGET asmjit PROPERTY POSITION_INDEPENDENT_CODE ON)
-  # Add a flag required for mac build
-  if(NOT MSVC)
-    target_compile_options(asmjit PRIVATE "-Wno-sign-conversion")
-  endif()
-
-  target_compile_options_if_supported(asmjit -Wno-deprecated-enum-enum-conversion)
-  target_compile_options_if_supported(asmjit -Wno-deprecated-anon-enum-enum-conversion)
-  target_compile_options_if_supported(asmjit -Wno-error=deprecated-enum-enum-conversion)
-
-  if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang" AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 13.0.0)
-    # See https://github.com/pytorch/pytorch/issues/74352, https://github.com/pytorch/FBGEMM/issues/1173
-    target_compile_options_if_supported(asmjit -Wno-deprecated-copy)
-    target_compile_options_if_supported(asmjit -Wno-unused-but-set-variable)
-  endif()
 endif()
 
 if(NOT TARGET cpuinfo)
@@ -302,8 +283,6 @@ if(NOT TARGET cpuinfo)
   set(CPUINFO_BUILD_MOCK_TESTS OFF CACHE BOOL "Do not build cpuinfo mock tests")
   set(CPUINFO_BUILD_BENCHMARKS OFF CACHE BOOL "Do not build cpuinfo benchmarks")
   set(CPUINFO_LIBRARY_TYPE static CACHE STRING "Set lib type to static")
-  #Select static runtime, needed for static build for MSVC
-  set(CPUINFO_RUNTIME_TYPE static CACHE STRING "Set runtime to static")
   add_subdirectory("${CPUINFO_SOURCE_DIR}" "${FBGEMM_BINARY_DIR}/cpuinfo")
   set_property(TARGET cpuinfo PROPERTY POSITION_INDEPENDENT_CODE ON)
 endif()
@@ -351,16 +330,6 @@ install(
   DESTINATION share/cmake/fbgemm
   FILE fbgemmLibraryConfig.cmake)
 
-if(MSVC)
-  if(FBGEMM_LIBRARY_TYPE STREQUAL "shared")
-    install(
-      FILES $<TARGET_PDB_FILE:fbgemm> $<TARGET_PDB_FILE:asmjit>
-      DESTINATION ${CMAKE_INSTALL_LIBDIR} OPTIONAL)
-  endif()
-  install(TARGETS fbgemm DESTINATION ${CMAKE_INSTALL_LIBDIR})
-  install(TARGETS asmjit DESTINATION ${CMAKE_INSTALL_LIBDIR})
-endif()
-
 #Make project importable from the build directory
 #export(TARGETS fbgemm asmjit FILE fbgemmLibraryConfig.cmake)
 

include/fbgemm/SimdUtils.h

@@ -10,7 +10,7 @@
 
 #include "./Utils.h" // @manual
 
-#include <asmjit/asmjit.h> // @manual
+#include <asmjit/x86.h> // @manual
 
 namespace fbgemm {
 /**

src/CodeGenHelpers.h

@@ -7,7 +7,7 @@
  */
 
 #pragma once
-#include <asmjit/asmjit.h> // @manual
+#include <asmjit/x86.h> // @manual
 #include "fbgemm/SimdUtils.h"
 #include "fbgemm/Utils.h"
 

src/DirectConv.h

@@ -7,7 +7,7 @@
  */
 
 #pragma once
-#include <asmjit/asmjit.h> // @manual
+#include <asmjit/x86.h> // @manual
 #include <cpuinfo.h>
 #include <cassert>
 #include <cstdint>

src/EmbeddingSpMDM.cc

@@ -11,7 +11,7 @@
 #include <type_traits>
 #define FBGEMM_EXPORTS
 
-#include <asmjit/asmjit.h> // @manual
+#include <asmjit/x86.h> // @manual
 #include <cpuinfo.h>
 #include <iostream>
 #include <mutex>
@@ -311,7 +311,7 @@ GenEmbeddingSpMDMLookup<
         x86::Gp scratchReg1_ = a->gpz(reg_id); // 12 or 13, also for mask
 
         ++reg_id;
-        x86::Gpd lengths_R_ = a->gpz(reg_id).r32(); // 13 or 14
+        auto lengths_R_ = a->gpz(reg_id).r32(); // 13 or 14
         ++reg_id;
         x86::Gp scratchReg2_ = a->gpz(reg_id); // 14 or 15
 
@@ -330,7 +330,7 @@ GenEmbeddingSpMDMLookup<
                   const float*, // weights
                   outType*, // out
                   const int32_t*, // compressed_indices_table and then mask
-                  const int*>(asmjit::CallConvId::kHost),
+                  const int*>(),
               a->environment());
         } else {
           func.init(
@@ -344,7 +344,7 @@ GenEmbeddingSpMDMLookup<
                   const offsetType*, // offsets or lengths
                   const float*, // weights
                   outType*, // out and then mask
-                  const int*>(asmjit::CallConvId::kHost),
+                  const int*>(),
               a->environment());
         }
 

src/EmbeddingSpMDMNBit.cc

@@ -10,7 +10,6 @@
 
 #include "fbgemm/FbgemmEmbedding.h"
 
-#include <asmjit/asmjit.h> // @manual
 #include <cpuinfo.h>
 #include <cassert>
 #include <iostream>
@@ -301,7 +300,7 @@ GenEmbeddingSpMDMNBitLookup<
                   const float*, // weights
                   float*, // out
                   const int32_t* /* compressed_indices_table */,
-                  const int* /* mask */>(asmjit::CallConvId::kHost),
+                  const int* /* mask */>(),
               a->environment());
         } else {
           func.init(
@@ -315,7 +314,7 @@ GenEmbeddingSpMDMNBitLookup<
                   const offsetType*, // offsets or lengths
                   const float*, // weights
                   float*, // out
-                  const int* /* mask */>(asmjit::CallConvId::kHost),
+                  const int* /* mask */>(),
               a->environment());
         }
 

src/FbgemmI64.cc

@@ -179,8 +179,7 @@ CodeGenBase<int64_t, int64_t, int64_t, int64_t>::getOrCreate(
     asmjit::FuncDetail func;
     func.init(
         asmjit::FuncSignature::
-            build<void, int64_t*, int64_t*, int64_t*, int64_t*, int, int>(
-                asmjit::CallConvId::kHost),
+            build<void, int64_t*, int64_t*, int64_t*, int64_t*, int, int>(),
         a->environment());
 
     asmjit::FuncFrame frame;

src/GenerateI8Depthwise.cc

@@ -8,7 +8,6 @@
 
 #include "./GenerateI8Depthwise.h" // @manual
 
-#include <asmjit/asmjit.h> // @manual
 #include <cassert>
 #include <iostream>
 #include <numeric>
@@ -273,7 +272,7 @@ GenI8Depthwise::jit_kernel_signature GenI8Depthwise::getOrCreate(
             int,
             const int*,
             int,
-            const std::int32_t*>(asmjit::CallConvId::kHost),
+            const std::int32_t*>(),
         e->environment());
 
     asmjit::FuncFrame frame;

src/GenerateKernel.h

@@ -7,9 +7,7 @@
  */
 
 #pragma once
-#include <asmjit/asmjit.h> // @manual
 #include <cpuinfo.h>
-#include <map>
 #include <mutex>
 #include <sstream>
 #include <string>

src/GenerateKernelDirectConvU8S8S32ACC32.cc

@@ -227,8 +227,7 @@ DirectConvCodeGenBase<uint8_t, int8_t, int32_t, int32_t>::getOrCreateDirectConv(
     asmjit::FuncDetail func;
     func.init(
         asmjit::FuncSignature::
-            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(
-                asmjit::CallConvId::kHost),
+            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(),
         a->environment());
 
     asmjit::FuncFrame frame;
@@ -646,8 +645,7 @@ DirectConvCodeGenBase<uint8_t, int8_t, int32_t, int32_t>::
     asmjit::FuncDetail func;
     func.init(
         asmjit::FuncSignature::
-            build<void, uint8_t*, int8_t*, int32_t*, int, int, int, int>(
-                asmjit::CallConvId::kHost),
+            build<void, uint8_t*, int8_t*, int32_t*, int, int, int, int>(),
         a->environment());
 
     asmjit::FuncFrame frame;

src/GenerateKernelU8S8S32ACC16.cc

@@ -181,8 +181,7 @@ CodeGenBase<uint8_t, int8_t, int32_t, int16_t>::getOrCreate<inst_set_t::avx2>(
     asmjit::FuncDetail func;
     func.init(
         asmjit::FuncSignature::
-            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(
-                asmjit::CallConvId::kHost),
+            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(),
         a->environment());
 
     asmjit::FuncFrame frame;

src/GenerateKernelU8S8S32ACC16Avx512.cc

@@ -143,8 +143,7 @@ CodeGenBase<uint8_t, int8_t, int32_t, int16_t>::getOrCreate(
     asmjit::FuncDetail func;
     func.init(
         asmjit::FuncSignature::
-            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(
-                asmjit::CallConvId::kHost),
+            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(),
         a->environment());
 
     asmjit::FuncFrame frame;

src/GenerateKernelU8S8S32ACC32.cc

@@ -180,8 +180,7 @@ CodeGenBase<uint8_t, int8_t, int32_t, int32_t>::getOrCreate(
     asmjit::FuncDetail func;
     func.init(
         asmjit::FuncSignature::
-            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(
-                asmjit::CallConvId::kHost),
+            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(),
         a->environment());
 
     asmjit::FuncFrame frame;

src/GenerateKernelU8S8S32ACC32Avx512VNNI.cc

@@ -134,8 +134,7 @@ CodeGenBase<uint8_t, int8_t, int32_t, int32_t>::getOrCreate(
     asmjit::FuncDetail func;
     func.init(
         asmjit::FuncSignature::
-            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(
-                asmjit::CallConvId::kHost),
+            build<void, uint8_t*, int8_t*, int8_t*, int32_t*, int, int>(),
         a->environment());
 
     asmjit::FuncFrame frame;

src/GroupwiseConv.cc

@@ -8,7 +8,6 @@
 
 #define FBGEMM_EXPORTS
 #include "./GroupwiseConv.h" // @manual
-#include <asmjit/asmjit.h> // @manual
 #include <cpuinfo.h>
 #include <array>
 #include <iostream>
@@ -226,7 +225,7 @@ jit_conv_kernel_fp GenConvKernel<SPATIAL_DIM, INST_SET>::getOrCreate() {
           int32_t,
           int32_t,
           int32_t,
-          int32_t*>(asmjit::CallConvId::kHost),
+          int32_t*>(),
       a->environment());
 
   frame_.init(func_);

src/GroupwiseConv.h

@@ -7,7 +7,7 @@
  */
 
 #pragma once
-#include <asmjit/asmjit.h> // @manual
+#include <asmjit/x86.h> // @manual
 #include <cpuinfo.h>
 #include <cassert>
 #include <cstdint>

src/GroupwiseConvAcc32Avx2.cc

@@ -7,7 +7,6 @@
  */
 
 #define FBGEMM_EXPORTS
-#include <asmjit/asmjit.h> // @manual
 #include "./CodeGenHelpers.h" // @manual
 #include "./GroupwiseConv.h" // @manual
 #include "fbgemm/Fbgemm.h"

src/GroupwiseConvAcc32Avx512.cc

@@ -7,7 +7,6 @@
  */
 
 #define FBGEMM_EXPORTS
-#include <asmjit/asmjit.h> // @manual
 #include "./CodeGenHelpers.h" // @manual
 #include "./GroupwiseConv.h" // @manual
 

src/RowWiseSparseAdagradFused.cc

@@ -9,7 +9,6 @@
 #define FBGEMM_EXPORTS
 #include "fbgemm/FbgemmEmbedding.h"
 
-#include <asmjit/asmjit.h> // @manual
 #include <cpuinfo.h>
 #include <cassert>
 #include <iostream>
@@ -180,7 +179,7 @@ typename ReturnFunctionSignature<indxType, offsetType, dataType>::
                 const int*, // lengths
                 float, // epsilon
                 float, // lr then rand_buffer
-                uint32_t*>(asmjit::CallConvId::kHost),
+                uint32_t*>(),
             a->environment());
 
         asmjit::FuncFrame frame;

src/SparseAdagrad.cc

@@ -9,7 +9,6 @@
 #define FBGEMM_EXPORTS
 #include "fbgemm/FbgemmEmbedding.h"
 
-#include <asmjit/asmjit.h> // @manual
 #include <cpuinfo.h>
 #include <cmath>
 #include <iostream>
@@ -513,7 +512,7 @@ GenSparseAdagrad<indxType, instSet>::getOrCreate(
                 const int*, // mask_avx2
                 float, // weight_decay
                 const double*, // counter then counter_halflife
-                std::int64_t>(asmjit::CallConvId::kHost),
+                std::int64_t>(),
             a->environment());
 
         asmjit::FuncFrame frame;

test/CMakeLists.txt

@@ -24,7 +24,11 @@ if(FBGEMM_BUILD_TESTS AND NOT TARGET gtest)
   endif()
 
   #build Googletest framework
+  set(TEMP_BUILD_SHARED_LIBS ${BUILD_SHARED_LIBS})
+  set(BUILD_SHARED_LIBS OFF)
   add_subdirectory("${GOOGLETEST_SOURCE_DIR}" "${FBGEMM_BINARY_DIR}/googletest")
+  # Recover the build shared libs option.
+  set(BUILD_SHARED_LIBS ${TEMP_BUILD_SHARED_LIBS})
   # add flags required for mac build
   if(NOT MSVC)
     target_compile_options(gtest PRIVATE "-Wno-deprecated-copy")