Merge branch 'main' into fix-matmul

Xilinx · May 16, 2024 · a509e4c · a509e4c
2 parents b896b7f + 71c9229
commit a509e4c
Show file tree

Hide file tree

Showing 11 changed files with 98 additions and 73 deletions.
diff --git a/.github/workflows/buildAndTestMulti.yml b/.github/workflows/buildAndTestMulti.yml
@@ -39,7 +39,7 @@ jobs:
       # Run the tests on the Cartesian product of the following
       matrix:
 
-        OS: [ ubuntu-20.04, ubuntu-22.04, macos-12 ]
+        OS: [ ubuntu-20.04, ubuntu-22.04 ]
         COMPILER: [ llvm, gcc ]
         ENABLE_ASSERTIONS: [ ON, OFF ]
         ENABLE_RTTI: [ ON, OFF ]
@@ -61,9 +61,6 @@ jobs:
             COMPILER: msvc
             ENABLE_ASSERTIONS: OFF
             ENABLE_RTTI: OFF
-        exclude:
-          - OS: macos-12
-            COMPILER: gcc
 
     steps:
       # Clone the repo and its submodules. Do shallow clone to save clone

diff --git a/.github/workflows/mlirDistro.yml b/.github/workflows/mlirDistro.yml
@@ -100,13 +100,13 @@ jobs:
             ARCH: AMD64
             ENABLE_RTTI: ON
 
-          - OS: macos-12
-            ARCH: x86_64
-            ENABLE_RTTI: ON
+          # - OS: macos-12
+          #   ARCH: x86_64
+          #   ENABLE_RTTI: ON
 
-          - OS: macos-12
-            ARCH: arm64
-            ENABLE_RTTI: ON
+          # - OS: macos-12
+          #   ARCH: arm64
+          #   ENABLE_RTTI: ON
 
           - OS: ubuntu-20.04
             ARCH: x86_64
@@ -120,13 +120,13 @@ jobs:
             ARCH: AMD64
             ENABLE_RTTI: OFF
 
-          - OS: macos-12
-            ARCH: x86_64
-            ENABLE_RTTI: OFF
+          # - OS: macos-12
+          #   ARCH: x86_64
+          #   ENABLE_RTTI: OFF
 
-          - OS: macos-12
-            ARCH: arm64
-            ENABLE_RTTI: OFF
+          # - OS: macos-12
+          #   ARCH: arm64
+          #   ENABLE_RTTI: OFF
 
     steps:
 
@@ -378,9 +378,9 @@ jobs:
             ARCH: AMD64
             ENABLE_RTTI: ON
 
-          - OS: macos-12
-            ARCH: x86_64
-            ENABLE_RTTI: ON
+          # - OS: macos-12
+          #   ARCH: x86_64
+          #   ENABLE_RTTI: ON
 
           - OS: ubuntu-20.04
             ARCH: x86_64
@@ -390,9 +390,9 @@ jobs:
             ARCH: AMD64
             ENABLE_RTTI: OFF
 
-          - OS: macos-12
-            ARCH: x86_64
-            ENABLE_RTTI: OFF
+          # - OS: macos-12
+          #   ARCH: x86_64
+          #   ENABLE_RTTI: OFF
 
     steps:
       - uses: actions/download-artifact@v3
@@ -438,13 +438,13 @@ jobs:
             ARCH: AMD64
             ENABLE_RTTI: ON
 
-          - OS: macos-12
-            ARCH: x86_64
-            ENABLE_RTTI: ON
+          # - OS: macos-12
+          #   ARCH: x86_64
+          #   ENABLE_RTTI: ON
 
-          - OS: macos-12
-            ARCH: arm64
-            ENABLE_RTTI: ON
+          # - OS: macos-12
+          #   ARCH: arm64
+          #   ENABLE_RTTI: ON
 
           - OS: ubuntu-20.04
             ARCH: x86_64
@@ -458,13 +458,13 @@ jobs:
             ARCH: AMD64
             ENABLE_RTTI: OFF
 
-          - OS: macos-12
-            ARCH: x86_64
-            ENABLE_RTTI: OFF
+          # - OS: macos-12
+          #   ARCH: x86_64
+          #   ENABLE_RTTI: OFF
 
-          - OS: macos-12
-            ARCH: arm64
-            ENABLE_RTTI: OFF
+          # - OS: macos-12
+          #   ARCH: arm64
+          #   ENABLE_RTTI: OFF
 
     permissions:
       id-token: write

diff --git a/aie_kernels/mm.cc b/aie_kernels/mm.cc
@@ -22,6 +22,8 @@
 
 #include <aie_api/aie.hpp>
 
+#include "zero.cc"
+
 template <typename T_in, typename T_out, unsigned rowA, unsigned colA,
           unsigned colB, unsigned r, unsigned s, unsigned t>
 void matmul_vectorized(const T_in *__restrict pA, unsigned offsetA,
@@ -273,6 +275,12 @@ extern "C" {
         64, 64, 64>(a_in, offsetA, b_in, offsetB, c_out, offsetC);             \
   }
 
-combos(matmul_vectorized_c_func)
+#define zero_vectorized_c_func(ctype_in, mlir_type_in, ctype_out,              \
+                               mlir_type_out, r, s, t)                         \
+  void zero_##mlir_type_out(ctype_out *c_out, unsigned offsetC) {              \
+    zero_vectorized<ctype_out, 64, 64, 32>(c_out, offsetC);                    \
+  }
+
+combos(matmul_vectorized_c_func) combos(zero_vectorized_c_func)
 
 } // extern "C"
diff --git a/aie_kernels/zero.cc b/aie_kernels/zero.cc
@@ -0,0 +1,34 @@
+//===- zero.cc --------------------------------------------000---*- C++ -*-===//
+//
+// This file is licensed under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// Copyright (C) 2024, Advanced Micro Devices, Inc.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef ZERO_CC
+#define ZERO_CC
+
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <type_traits>
+
+template <typename T, int M, int N, int r>
+void zero_vectorized(T *__restrict pC, unsigned offsetC) {
+  const aie::vector<T, r> zeros = aie::zeros<T, r>();
+  T *__restrict pC1 = pC + offsetC;
+  const T *__restrict c_end = pC1 + M * N;
+  for (; pC1 + r < c_end; pC1 += r) {
+    aie::store_v(pC1, zeros);
+  }
+  // Do a scalar write for any remainder not divisible by vector instruction
+  // size r
+  for (; pC1 < c_end; pC1++) {
+    *pC1 = 0;
+  }
+}
+
+#endif
diff --git a/lib/Targets/AIETargetCDODirect.cpp b/lib/Targets/AIETargetCDODirect.cpp
@@ -437,11 +437,6 @@ struct AIEControl {
     TRY_XAIE_API_FATAL_ERROR(XAie_UpdateNpiAddr, &devInst, NPI_ADDR);
   }
 
-  LogicalResult addErrorHandlingToCDO() {
-    TRY_XAIE_API_LOGICAL_RESULT(XAie_ErrorHandlingInit, &devInst);
-    return success();
-  }
-
   LogicalResult addAieElfToCDO(uint8_t col, uint8_t row,
                                const StringRef elfPath, bool aieSim) {
     // loadSym: Load symbols from .map file. This argument is not used when
@@ -721,12 +716,6 @@ LogicalResult generateCDOBinariesSeparately(AIEControl &ctl,
                                             const StringRef workDirPath,
                                             DeviceOp &targetOp, bool aieSim,
                                             bool enableCores) {
-  if (failed(generateCDOBinary(
-          (llvm::Twine(workDirPath) + std::string(1, ps) +
-           "aie_cdo_error_handling.bin")
-              .str(),
-          std::bind(&AIEControl::addErrorHandlingToCDO, ctl))))
-    return failure();
 
   if (!targetOp.getOps<CoreOp>().empty() &&
       failed(generateCDOBinary(
@@ -759,8 +748,6 @@ LogicalResult generateCDOUnified(AIEControl &ctl, const StringRef workDirPath,
   return generateCDOBinary(
       (llvm::Twine(workDirPath) + std::string(1, ps) + "aie_cdo.bin").str(),
       [&ctl, &targetOp, &workDirPath, &aieSim, &enableCores] {
-        if (failed(ctl.addErrorHandlingToCDO()))
-          return failure();
         if (!targetOp.getOps<CoreOp>().empty() &&
             failed(ctl.addAieElfsToCDO(targetOp, workDirPath, aieSim)))
           return failure();

diff --git a/python/compiler/aiecc/main.py b/python/compiler/aiecc/main.py
@@ -267,7 +267,6 @@ def emit_design_bif(root_path, has_cores=True, enable_cores=True):
           {{
             name=aie_image, id=0x1c000000
             {{ type=cdo
-               file={root_path}/aie_cdo_error_handling.bin
                {elf_file}
                file={root_path}/aie_cdo_init.bin
                {enable_file}

diff --git a/runtime_lib/xaiengine/CMakeLists.txt b/runtime_lib/xaiengine/CMakeLists.txt
@@ -10,26 +10,10 @@ project("xaiengine lib for ${AIE_RUNTIME_TARGET}")
 include("aiert.cmake")
 
 if (${CMAKE_CROSSCOMPILING})
-
-    #tmp path to header files since we use an older (3_0) version of xaiengine. need to fix when we upgrade to 2023.1
-    if(${Vitis_VERSION_MAJOR} EQUAL "2022")
-        set(aieRTIncludePath "${VITIS_ROOT}/data/embeddedsw/XilinxProcessorIPLib/drivers/aienginev2_v3_0/src" CACHE STRING "AIE-RT include path")
-    elseif(${Vitis_VERSION_MAJOR} EQUAL "2023")
-        set(aieRTIncludePath "${VITIS_AIETOOLS_DIR}/include/drivers/aiengine" CACHE STRING "AIE-RT include path")
-    else()
-        message(FATAL_ERROR "Unsupported Vitis version: ${Vitis_VERSION_MAJOR}")
-    endif()
-
     message("Building xaiengine for ${AIE_RUNTIME_TARGET} from Vitis at ${VITIS_ROOT}.")
-    add_aiert_headers(xaiengine
-                        ${aieRTIncludePath}
-                        ${CMAKE_CURRENT_BINARY_DIR}/include
-                        ${CMAKE_INSTALL_PREFIX}/runtime_lib/${AIE_RUNTIME_TARGET}/xaiengine/include)
-
     add_subdirectory(lib)
 
 else()
-
     message("Copying xaiengine for ${AIE_RUNTIME_TARGET} from ${AIERT_LIBS}.") 
     add_aiert_headers(xaiengine
                         ${AIERT_INCLUDE_DIR}
@@ -47,7 +31,6 @@ else()
 
     # install library and headers
     install(FILES ${libs} DESTINATION ${CMAKE_INSTALL_PREFIX}/runtime_lib/${AIE_RUNTIME_TARGET}/xaiengine/lib)
-
 endif()
 
 add_subdirectory(cdo)
diff --git a/runtime_lib/xaiengine/aiert.cmake b/runtime_lib/xaiengine/aiert.cmake
@@ -1,6 +1,6 @@
 
 function(add_aiert_headers TARGET SRCPATH BUILDPATH INSTALLPATH)
-  message("${TARGET} ${SRCPATH} ${BUILDPATH}")
+  message("Installing aie-rt includes for ${TARGET} from ${SRCPATH} in ${BUILDPATH}")
   file(GLOB libheaders ${SRCPATH}/*.h)
   file(GLOB libheadersSub ${SRCPATH}/*/*.h)
 
@@ -37,7 +37,8 @@ function(add_aiert_headers TARGET SRCPATH BUILDPATH INSTALLPATH)
 endfunction()
 
 function(add_aiert_library TARGET XAIE_SOURCE)
-  cmake_parse_arguments(ARG "STATIC" "" "" ${ARGN})
+message("Building aie-rt library for ${TARGET} from ${SRCPATH}")
+cmake_parse_arguments(ARG "STATIC" "" "" ${ARGN})
   if(ARG_STATIC)
     set(LIBTYPE STATIC)
   else()

diff --git a/runtime_lib/xaiengine/lib/CMakeLists.txt b/runtime_lib/xaiengine/lib/CMakeLists.txt
@@ -6,7 +6,25 @@
 
 include("../aiert.cmake")
 
-set(XAIE_SOURCE ${VITIS_ROOT}/data/embeddedsw/XilinxProcessorIPLib/drivers/aienginev2_v3_0/src)
+#tmp path to header files since we use an older (3_0) version of xaiengine. need to fix when we upgrade to 2023.1
+if(${Vitis_VERSION_MAJOR} EQUAL "2022")
+    set(XAIE_SOURCE ${VITIS_ROOT}/data/embeddedsw/XilinxProcessorIPLib/drivers/aienginev2_v3_0/src)
+    set(aieRTIncludePath "${XAIE_SOURCE}")
+elseif(${Vitis_VERSION_MAJOR} EQUAL "2023")
+    set(XAIE_SOURCE ${VITIS_ROOT}/data/embeddedsw/XilinxProcessorIPLib/drivers/aienginev2_v3_0/src)
+    set(aieRTIncludePath "${VITIS_AIETOOLS_DIR}/include/drivers/aiengine")
+elseif(${Vitis_VERSION_MAJOR} EQUAL "2024")
+    set(XAIE_SOURCE ${VITIS_ROOT}/data/embeddedsw/XilinxProcessorIPLib/drivers/aienginev2_v3_5/src)
+    set(aieRTIncludePath "${VITIS_AIETOOLS_DIR}/include/drivers/aiengine")
+else()
+    message(FATAL_ERROR "Unsupported Vitis version: ${Vitis_VERSION_MAJOR}")
+endif()
+
+message("Building xaiengine for ${AIE_RUNTIME_TARGET} from Vitis at ${VITIS_ROOT}.")
+add_aiert_headers(xaiengine
+                    ${aieRTIncludePath}
+                    ${CMAKE_CURRENT_BINARY_DIR}/include
+                    ${CMAKE_INSTALL_PREFIX}/runtime_lib/${AIE_RUNTIME_TARGET}/xaiengine/include)
 
 add_aiert_library(xaiengine ${XAIE_SOURCE})
 

diff --git a/tools/aie2xclbin/XCLBinGen.cpp b/tools/aie2xclbin/XCLBinGen.cpp
@@ -527,8 +527,6 @@ static LogicalResult generateXCLBin(MLIRContext *context, ModuleOp moduleOp,
                        << "\t{\n"
                        << "\t\tname=aie_image, id=0x1c000000\n"
                        << "\t\t{ type=cdo\n"
-                       << "\t\t  file=" << TK.TempDir
-                       << "/aie_cdo_error_handling.bin\n"
                        << "\t\t  file=" << TK.TempDir << "/aie_cdo_elfs.bin\n"
                        << "\t\t  file=" << TK.TempDir << "/aie_cdo_init.bin\n"
                        << "\t\t  file=" << TK.TempDir << "/aie_cdo_enable.bin\n"

diff --git a/utils/clone-llvm.sh b/utils/clone-llvm.sh
@@ -13,8 +13,8 @@
 ##===----------------------------------------------------------------------===##
 
 # The LLVM commit to use.
-LLVM_PROJECT_COMMIT=1cde1240ed6e45012d7510f4aa39badbdb4a4721
-DATETIME=2024050314
+LLVM_PROJECT_COMMIT=8a71284cb9463a90fab0d9e8edbeb5d879531e32
+DATETIME=2024051512
 WHEEL_VERSION=19.0.0.$DATETIME+${LLVM_PROJECT_COMMIT:0:8}
 
 ############################################################################################