[Windows] Enable lit tests and CXX tests (#3294)

Signed-off-by: Anatoly Myachev <[email protected]>

Author: anmyachev

.github/workflows/build-test-windows.yml

@@ -83,6 +83,13 @@ jobs:
           pip install -r scripts\requirements-test.txt
           pip uninstall pytest_forked -y
 
+      - name: Run unit tests
+        run: |
+          .venv\Scripts\activate.ps1
+          Invoke-BatchFile "C:\Program Files (x86)\Intel\oneAPI\setvars.bat"
+          cd ${{ env.NEW_WORKSPACE }}
+          ${{ env.TRITON_TEST_CMD }} --unit
+
       - name: Run core tests
         run: |
           .venv\Scripts\activate.ps1

.github/workflows/pip-test-windows.yml

@@ -100,6 +100,13 @@ jobs:
           .venv\Scripts\activate.ps1
           pip list -v
 
+      - name: Run unit tests
+        run: |
+          .venv\Scripts\activate.ps1
+          Invoke-BatchFile "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvarsall.bat" x64
+          cd ${{ env.NEW_WORKSPACE }}
+          ${{ env.TRITON_TEST_CMD }} --unit
+
       - name: Run core tests
         run: |
           .venv\Scripts\activate.ps1

CMakeLists.txt

@@ -299,6 +299,12 @@ if(TRITON_BUILD_PYTHON_MODULE)
     target_link_libraries(triton PRIVATE ${CMAKE_DL_LIBS})
     set_target_properties(triton PROPERTIES SUFFIX ".pyd")
     set_target_properties(triton PROPERTIES PREFIX "lib")
+    if(DEFINED TRITON_PYD_PATH)
+      string(TOUPPER "${CMAKE_BUILD_TYPE}" UPPER_CMAKE_BUILD_TYPE)
+      set_target_properties(triton PROPERTIES
+        RUNTIME_OUTPUT_DIRECTORY_${UPPER_CMAKE_BUILD_TYPE}
+      "${TRITON_PYD_PATH}")
+    endif(DEFINED TRITON_PYD_PATH)
   else()
     target_link_libraries(triton PRIVATE z)
   endif()

python/setup.py

@@ -454,7 +454,7 @@ def build_extension(self, ext):
 
         cmake_args += [f"-DCMAKE_BUILD_TYPE={cfg}"]
         if platform.system() == "Windows":
-            cmake_args += [f"-DCMAKE_RUNTIME_OUTPUT_DIRECTORY_{cfg.upper()}={extdir}"]
+            cmake_args += [f"-DTRITON_PYD_PATH={extdir}"]
         else:
             max_jobs = os.getenv("MAX_JOBS", str(2 * os.cpu_count()))
             build_args += ['-j' + max_jobs]

scripts/test-triton.sh

@@ -178,7 +178,7 @@ run_unit_tests() {
   echo "******       Running Triton LIT tests        ******"
   echo "***************************************************"
   cd $TRITON_PROJ/python/build/cmake*/test
-  lit -v .
+  lit -v . || $TRITON_TEST_IGNORE_ERRORS
 }
 
 run_core_tests() {

test/Conversion/intel/tritongpu_to_gen_dot.mlir

@@ -1,5 +1,5 @@
 // RUN: triton-opt %s -split-input-file --intel-allocate-shared-memory  --convert-triton-intel-gpu-to-llvm --cse -canonicalize | FileCheck %s --implicit-check-not=llvm.inline_asm --check-prefixes=CHECK,NO-AGGRESSIVE-REUSE
-// RUN: TRITON_INTEL_AGGRESSIVE_DPAS_REUSE=1 triton-opt %s -split-input-file --intel-allocate-shared-memory  --convert-triton-intel-gpu-to-llvm --cse -canonicalize | FileCheck %s --implicit-check-not=llvm.inline_asm  --check-prefixes=CHECK,AGGRESSIVE-REUSE
+// RUN: env TRITON_INTEL_AGGRESSIVE_DPAS_REUSE=1 triton-opt %s -split-input-file --intel-allocate-shared-memory  --convert-triton-intel-gpu-to-llvm --cse -canonicalize | FileCheck %s --implicit-check-not=llvm.inline_asm  --check-prefixes=CHECK,AGGRESSIVE-REUSE
 
 #dpas = #triton_intel_gpu.dpas<{repeatCount = 8, systolicDepth = 8, executionSize = 16, opsPerChan = 2, threadsPerWarp = 16, warpsPerCTA = [1, 1], repCluster = [1, 1]}>
 #dot_operand_a = #ttg.dot_op<{opIdx=0, parent=#dpas, kWidth=1}>

test/Conversion/intel/tritongpu_to_llvm_intel_advanced_path.mlir

@@ -1,4 +1,4 @@
-// RUN: TRITON_INTEL_ADVANCED_PATH=1 triton-opt %s --convert-triton-intel-gpu-to-llvm --split-input-file | FileCheck %s
+// RUN: env TRITON_INTEL_ADVANCED_PATH=1 triton-opt %s --convert-triton-intel-gpu-to-llvm --split-input-file | FileCheck %s
 
 module attributes {"triton_intel_gpu.support_sg_2d_block", "triton_intel_gpu.support_dpas", "ttg.num-warps" = 32 : i32, "ttg.threads-per-warp" = 1 : i32} {
   // CHECK-DAG: llvm.func spir_funccc @_Z38intel_sub_group_f16_f16_matrix_mad_k16Dv8_sDv8_iDv8_f(vector<8xi16>, vector<8xi32>, vector<8xf32>) -> vector<8xf32> attributes {convergent, memory_effects = #llvm.memory_effects<other = none, argMem = none, inaccessibleMem = none>, no_unwind, will_return}

test/Conversion/intel/tritongpu_to_llvm_intel_advanced_path_invalid.mlir

@@ -1,4 +1,4 @@
-// RUN: TRITON_INTEL_ADVANCED_PATH=1 triton-opt %s --convert-triton-intel-gpu-to-llvm --verify-diagnostics --split-input-file
+// RUN: env TRITON_INTEL_ADVANCED_PATH=1 triton-opt %s --convert-triton-intel-gpu-to-llvm --verify-diagnostics --split-input-file
 
 module attributes {"triton_intel_gpu.support_sg_2d_block", "triton_intel_gpu.support_dpas", "ttg.num-warps" = 32 : i32, "ttg.threads-per-warp" = 16 : i32} {
   tt.func public @matmul_kernel_with_block_pointers(%arg0: !tt.ptr<f32>, %arg1: i64, %arg2: i32) {

test/Conversion/intel/tritonintelgpu_to_llvm.mlir

@@ -8,7 +8,7 @@ module attributes { "ttg.threads-per-warp" = 16 : i32, "ttg.num-warps" = 4 : i32
   // As the assert message is shared, a single instance is emitted.
 
   // CHECK-DAG:         llvm.mlir.global internal constant @assertFunc_("unknown\00") {addr_space = 1 : i32}
-  // CHECK-DAG:         llvm.mlir.global internal constant @assertFile_("{{.*}}/test/Conversion/intel/tritonintelgpu_to_llvm.mlir\00") {addr_space = 1 : i32}
+  // CHECK-DAG:         llvm.mlir.global internal constant @assertFile_("{{.*}}tritonintelgpu_to_llvm.mlir\00") {addr_space = 1 : i32}
   // CHECK-DAG:         llvm.mlir.global internal constant @assertMessage_("assert text\00") {addr_space = 1 : i32}
   // CHECK-DAG:         llvm.mlir.global internal constant @assertMessage_3("different assert text\00") {addr_space = 1 : i32}
   // CHECK-DAG:         llvm.func spir_funccc @__assert_fail(!llvm.ptr<4>, !llvm.ptr<4>, i32, !llvm.ptr<4>)

test/TritonIntelGPU/schedule-load.mlir

@@ -1,5 +1,5 @@
-// RUN: TRITON_INTEL_ENABLE_INSTR_SCHED=1 triton-opt %s -split-input-file -tritonintelgpu-schedule-load | FileCheck %s --check-prefixes=CHECK,SINK-ACROSS-REGIONS
-// RUN: TRITON_INTEL_ENABLE_INSTR_SCHED=1 TRITON_INTEL_DO_NOT_SINK_INSTR_ACROSS_RGN=1 triton-opt %s -split-input-file -tritonintelgpu-schedule-load | FileCheck %s --check-prefixes=CHECK,DO-NOT-SINK-ACROSS-REGIONS
+// RUN: env TRITON_INTEL_ENABLE_INSTR_SCHED=1 triton-opt %s -split-input-file -tritonintelgpu-schedule-load | FileCheck %s --check-prefixes=CHECK,SINK-ACROSS-REGIONS
+// RUN: env TRITON_INTEL_ENABLE_INSTR_SCHED=1 TRITON_INTEL_DO_NOT_SINK_INSTR_ACROSS_RGN=1 triton-opt %s -split-input-file -tritonintelgpu-schedule-load | FileCheck %s --check-prefixes=CHECK,DO-NOT-SINK-ACROSS-REGIONS
 
 // -----
 // COM: Inst Schedule for Flash Attention case