Merge branch 'main' into clSetKernelArg_negative_invalid_sampler

Author: shajder

.github/workflows/presubmit.yml

@@ -10,7 +10,7 @@ jobs:
       matrix:
         build-type: [Release]
         gl: [0]
-        os: [ubuntu-22.04, macos-latest, windows-latest]
+        os: [ubuntu-22.04, macos-latest, windows-latest, windows-11-arm]
         include:
           - os: ubuntu-22.04
             gl: 1
@@ -61,6 +61,10 @@ jobs:
           cd OpenCL-Headers
           ln -s CL OpenCL # For OSX builds
           cd ..
+      - name: Fetch SPIR-V Headers
+        shell: bash
+        run: |
+          git clone https://github.com/KhronosGroup/SPIRV-Headers.git
       - name: Install Vulkan SDK
         uses: humbletim/install-vulkan-sdk@main
         with:
@@ -69,12 +73,13 @@ jobs:
       - name: Install Android NDK
         if: ${{ matrix.arch == 'android-arm' || matrix.arch == 'android-aarch64' }}
         run: |
-          wget https://dl.google.com/android/repository/android-ndk-r27c-linux.zip -O android-ndk.zip
+          wget https://dl.google.com/android/repository/android-ndk-r28c-linux.zip -O android-ndk.zip
           unzip android-ndk.zip -d $HOME
-          export ANDROID_NDK=$HOME/android-ndk-r27c
+          export ANDROID_NDK=$HOME/android-ndk-r28c
           echo "ANDROID_NDK=$ANDROID_NDK" >> $GITHUB_ENV
           export ANDROID_ARCH_ABI=${{ matrix.android_arch_abi }}
           echo "ANDROID_ARCH_ABI=$ANDROID_ARCH_ABI" >> $GITHUB_ENV
+          echo "ANDROID_PLATFORM=29" >> $GITHUB_ENV
       - name: Prepare CMake Toolchain file
         shell: bash
         run: |
@@ -107,7 +112,7 @@ jobs:
         if: ${{ matrix.arch == 'android-arm' || matrix.arch == 'android-aarch64' }}
         shell: bash
         run: |
-          echo "CMAKE_CONFIG_ARGS_ANDROID=-DCMAKE_ANDROID_ARCH_ABI=${ANDROID_ARCH_ABI}" >> $GITHUB_ENV
+          echo "CMAKE_ADDITIONAL_CONFIG_ARGS=-DCMAKE_ANDROID_ARCH_ABI=${ANDROID_ARCH_ABI} -DANDROID_PLATFORM=${ANDROID_PLATFORM}" >> $GITHUB_ENV
       - name: Fetch and build OpenCL ICD Loader
         shell: bash
         run: |
@@ -119,7 +124,7 @@ jobs:
                 -DCMAKE_BUILD_TYPE=Release \
                 -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_FILE} \
                 -DOPENCL_ICD_LOADER_HEADERS_DIR='${{ github.workspace }}'/OpenCL-Headers/ \
-                "${CMAKE_CONFIG_ARGS_ANDROID}"
+                ${CMAKE_ADDITIONAL_CONFIG_ARGS}
           cmake --build . --parallel
       - name: Fetch Vulkan Headers
         shell: bash
@@ -149,6 +154,7 @@ jobs:
           cd build
           if [[ ${RUNNER_OS} == "Windows" ]]; then
               CMAKE_OPENCL_LIBRARIES_OPTION="OpenCL"
+              CMAKE_ADDITIONAL_CONFIG_ARGS="-DD3D10_IS_SUPPORTED=ON -DD3D11_IS_SUPPORTED=ON"
           else
               CMAKE_OPENCL_LIBRARIES_OPTION="-lOpenCL"
               if [[ '${{ matrix.arch }}' != android-* ]]; then
@@ -157,8 +163,10 @@ jobs:
           fi
           cmake .. -G Ninja \
                 -DCMAKE_BUILD_TYPE=${{ matrix.build-type }} \
-                -DCMAKE_CACHE_OPTIONS="-DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache" \
+                -DCMAKE_C_COMPILER_LAUNCHER=sccache \
+                -DCMAKE_CXX_COMPILER_LAUNCHER=sccache \
                 -DCL_INCLUDE_DIR='${{ github.workspace }}'/OpenCL-Headers \
+                -DSPIRV_INCLUDE_DIR='${{ github.workspace }}'/SPIRV-Headers \
                 -DCL_LIB_DIR='${{ github.workspace }}'/OpenCL-ICD-Loader/build \
                 -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_FILE} \
                 -DCMAKE_RUNTIME_OUTPUT_DIRECTORY=./bin \
@@ -168,7 +176,7 @@ jobs:
                 -DVULKAN_IS_SUPPORTED=ON \
                 -DVULKAN_INCLUDE_DIR='${{ github.workspace }}'/Vulkan-Headers/include/ \
                 -DVULKAN_LIB_DIR='${{ github.workspace }}'/Vulkan-Loader/build/loader/ \
-                "${CMAKE_CONFIG_ARGS_ANDROID}"
+                ${CMAKE_ADDITIONAL_CONFIG_ARGS}
           cmake --build . --parallel
   formatcheck:
     name: Check code format

CMakeLists.txt

@@ -59,6 +59,12 @@ else(CL_INCLUDE_DIR AND CL_LIB_DIR)
     message(FATAL_ERROR "Either install OpenCL or pass -DCL_INCLUDE_DIR and -DCL_LIB_DIR")
 endif(CL_INCLUDE_DIR AND CL_LIB_DIR)
 
+# SPIRV_INCLUDE_DIR - path to dir with SPIR-V headers
+if(NOT SPIRV_INCLUDE_DIR)
+    message(STATUS "SPIR-V headers haven't been found!")
+    message(FATAL_ERROR "Pass -DSPIRV_INCLUDE_DIR")
+endif(NOT SPIRV_INCLUDE_DIR)
+
 # CLConform_GL_LIBRARIES_DIR - path to OpenGL libraries
 if(GL_IS_SUPPORTED AND CLConform_GL_LIBRARIES_DIR)
     link_directories(${CLConform_GL_LIBRARIES_DIR})
@@ -195,6 +201,7 @@ if(APPLE)
 endif(APPLE)
 
 include_directories(SYSTEM ${CL_INCLUDE_DIR})
+include_directories(SYSTEM ${SPIRV_INCLUDE_DIR}/include)
 include_directories(${CLConform_SOURCE_DIR}/test_common/harness
                     ${CLConform_SOURCE_DIR}/test_common/gles
                     ${CLConform_SOURCE_DIR}/test_common/gl

README.md

@@ -13,6 +13,8 @@ Compiling the CTS requires the following CMake configuration options to be set:
 
 * `CL_INCLUDE_DIR` Points to the unified
   [OpenCL-Headers](https://github.com/KhronosGroup/OpenCL-Headers).
+* `SPIRV_INCLUDE_DIR` Points to the unified
+  [SPIRV-Headers](https://github.com/KhronosGroup/SPIRV-Headers).
 * `CL_LIB_DIR` Directory containing the OpenCL library to build against.
 * `SPIRV_TOOLS_DIR` Directory containing the `spirv-as` and `spirv-val` binaries
    to be used in the CTS build process. Alternatively, the location to these binaries
@@ -31,6 +33,7 @@ a build, and compile.
 ```sh
 git clone https://github.com/KhronosGroup/OpenCL-CTS.git
 git clone https://github.com/KhronosGroup/OpenCL-Headers.git
+git clone https://github.com/KhronosGroup/SPIRV-Headers.git
 git clone https://github.com/KhronosGroup/OpenCL-ICD-Loader.git
 git clone https://github.com/KhronosGroup/SPIRV-Tools.git
 git clone https://github.com/KhronosGroup/SPIRV-Headers.git SPIRV-Tools/external/spirv-headers
@@ -50,6 +53,7 @@ cmake --build SPIRV-Tools/build --config Release
 mkdir OpenCL-CTS/build
 cmake -S OpenCL-CTS -B OpenCL-CTS/build \
       -DCL_INCLUDE_DIR=$PWD/OpenCL-Headers \
+      -DSPIRV_INCLUDE_DIR=$PWD/SPIRV-Headers \
       -DCL_LIB_DIR=$PWD/OpenCL-ICD-Loader/build \
       -DSPIRV_TOOLS_DIR=$PWD/SPIRV-Tools/build/tools/ \
       -DOPENCL_LIBRARIES=OpenCL

test_common/harness/ThreadPool.cpp

@@ -436,7 +436,14 @@ void *ThreadPool_WorkerFunc(void *p)
 
                 // drop run count to 0
                 gRunCount = 0;
+#if defined(_M_IX86) || defined(_M_X64)
                 _mm_mfence();
+#elif defined(_M_ARM64)
+                __dmb(_ARM64_BARRIER_ISHST);
+#else
+#error Architecture needs an implementation
+#endif
+
 #else
                 if (pthread_mutex_lock(&gAtomicLock))
                     log_error(
@@ -703,7 +710,13 @@ void ThreadPool_Exit(void)
     // http://gcc.gnu.org/onlinedocs/gcc/Atomic-Builtins.html#Atomic-Builtins
     __sync_synchronize();
 #elif defined(_MSC_VER)
+#if defined(_M_IX86) || defined(_M_X64)
     _mm_mfence();
+#elif defined(_M_ARM64)
+    __dmb(_ARM64_BARRIER_ISHST);
+#else
+#error Architecture needs an implementation
+#endif
 #else
 #warning If this is a weakly ordered memory system, please add a memory barrier here to force this and everything else to memory before we proceed
 #endif

test_common/harness/conversions.cpp

@@ -23,10 +23,10 @@
 
 #include <CL/cl_half.h>
 
-#if defined(__SSE__) || defined(_MSC_VER)
+#if defined(__SSE__) || _M_IX86_FP == 1
 #include <xmmintrin.h>
 #endif
-#if defined(__SSE2__) || defined(_MSC_VER)
+#if defined(__SSE2__) || _M_IX86_FP == 2 || defined(_M_X64)
 #include <emmintrin.h>
 #endif
 
@@ -110,7 +110,7 @@ static long lrintf_clamped(float f)
         volatile float x = f;
         float magicVal = magic[f < 0];
 
-#if defined(__SSE__) || defined(_WIN32)
+#if defined(__SSE__) || _M_IX86_FP == 1
         // Defeat x87 based arithmetic, which cant do FTZ, and will round this
         // incorrectly
         __m128 v = _mm_set_ss(x);
@@ -150,7 +150,7 @@ static long lrint_clamped(double f)
     {
         volatile double x = f;
         double magicVal = magic[f < 0];
-#if defined(__SSE2__) || (defined(_MSC_VER))
+#if defined(__SSE2__) || _M_IX86_FP == 2 || defined(_M_X64)
         // Defeat x87 based arithmetic, which cant do FTZ, and will round this
         // incorrectly
         __m128d v = _mm_set_sd(x);

test_common/harness/fpcontrol.h

@@ -37,36 +37,44 @@ typedef int FPU_mode_type;
 #else
 typedef int64_t FPU_mode_type;
 #endif
-#if defined(__i386__) || defined(__x86_64__) || defined(_MSC_VER)              \
-    || defined(__MINGW32__)
+#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)               \
+    || defined(_M_X64) || defined(__MINGW32__)
 #include <xmmintrin.h>
+#elif defined(_M_ARM64)
+#include <intrin.h>
 #elif defined(__PPC__)
 #include <fpu_control.h>
 extern __thread fpu_control_t fpu_control;
 #elif defined(__mips__)
 #include "mips/m32c1.h"
 #endif
+
 // Set the reference hardware floating point unit to FTZ mode
-inline void ForceFTZ(FPU_mode_type *mode)
+inline void ForceFTZ(FPU_mode_type *oldMode)
 {
-#if defined(__i386__) || defined(__x86_64__) || defined(_MSC_VER)              \
-    || defined(__MINGW32__)
-    *mode = _mm_getcsr();
-    _mm_setcsr(*mode | 0x8040);
+#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)               \
+    || defined(_M_X64) || defined(__MINGW32__)
+    *oldMode = _mm_getcsr();
+    _mm_setcsr(*oldMode | 0x8040);
 #elif defined(__PPC__)
-    *mode = fpu_control;
+    *oldMode = fpu_control;
     fpu_control |= _FPU_MASK_NI;
 #elif defined(__arm__)
     unsigned fpscr;
     __asm__ volatile("fmrx %0, fpscr" : "=r"(fpscr));
-    *mode = fpscr;
+    *oldMode = fpscr;
     __asm__ volatile("fmxr fpscr, %0" ::"r"(fpscr | (1U << 24)));
     // Add 64 bit support
-#elif defined(__aarch64__)
+#elif defined(__aarch64__) // Clang
     uint64_t fpscr;
     __asm__ volatile("mrs %0, fpcr" : "=r"(fpscr));
-    *mode = fpscr;
+    *oldMode = fpscr;
     __asm__ volatile("msr fpcr, %0" ::"r"(fpscr | (1U << 24)));
+#elif defined(_M_ARM64) // Visual Studio
+    uint64_t fpscr;
+    fpscr = _ReadStatusReg(ARM64_FPSR);
+    *oldMode = fpscr;
+    _WriteStatusReg(ARM64_FPCR, fpscr | (1U << 24));
 #elif defined(__mips__)
     fpa_bissr(FPA_CSR_FS);
 #else
@@ -75,26 +83,31 @@ inline void ForceFTZ(FPU_mode_type *mode)
 }
 
 // Disable the denorm flush to zero
-inline void DisableFTZ(FPU_mode_type *mode)
+inline void DisableFTZ(FPU_mode_type *oldMode)
 {
-#if defined(__i386__) || defined(__x86_64__) || defined(_MSC_VER)              \
-    || defined(__MINGW32__)
-    *mode = _mm_getcsr();
-    _mm_setcsr(*mode & ~0x8040);
+#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)               \
+    || defined(_M_X64) || defined(__MINGW32__)
+    *oldMode = _mm_getcsr();
+    _mm_setcsr(*oldMode & ~0x8040);
 #elif defined(__PPC__)
     *mode = fpu_control;
     fpu_control &= ~_FPU_MASK_NI;
 #elif defined(__arm__)
     unsigned fpscr;
     __asm__ volatile("fmrx %0, fpscr" : "=r"(fpscr));
-    *mode = fpscr;
+    *oldMode = fpscr;
     __asm__ volatile("fmxr fpscr, %0" ::"r"(fpscr & ~(1U << 24)));
     // Add 64 bit support
-#elif defined(__aarch64__)
+#elif defined(__aarch64__) // Clang
     uint64_t fpscr;
     __asm__ volatile("mrs %0, fpcr" : "=r"(fpscr));
-    *mode = fpscr;
+    *oldMode = fpscr;
     __asm__ volatile("msr fpcr, %0" ::"r"(fpscr & ~(1U << 24)));
+#elif defined(_M_ARM64) // Visual Studio
+    uint64_t fpscr;
+    fpscr = _ReadStatusReg(ARM64_FPSR);
+    *oldMode = fpscr;
+    _WriteStatusReg(ARM64_FPCR, fpscr & ~(1U << 24));
 #elif defined(__mips__)
     fpa_bicsr(FPA_CSR_FS);
 #else
@@ -105,16 +118,18 @@ inline void DisableFTZ(FPU_mode_type *mode)
 // Restore the reference hardware to floating point state indicated by *mode
 inline void RestoreFPState(FPU_mode_type *mode)
 {
-#if defined(__i386__) || defined(__x86_64__) || defined(_MSC_VER)              \
-    || defined(__MINGW32__)
+#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)               \
+    || defined(_M_X64) || defined(__MINGW32__)
     _mm_setcsr(*mode);
 #elif defined(__PPC__)
     fpu_control = *mode;
 #elif defined(__arm__)
     __asm__ volatile("fmxr fpscr, %0" ::"r"(*mode));
     // Add 64 bit support
-#elif defined(__aarch64__)
+#elif defined(__aarch64__) // Clang
     __asm__ volatile("msr fpcr, %0" ::"r"(*mode));
+#elif defined(_M_ARM64) // Visual Studio
+    _WriteStatusReg(ARM64_FPCR, *mode);
 #elif defined(__mips__)
     // Mips runs by default with DAZ=1 FTZ=1
 #else
@@ -125,4 +140,4 @@ inline void RestoreFPState(FPU_mode_type *mode)
 #error ForceFTZ and RestoreFPState need implentations
 #endif
 
-#endif
+#endif

test_common/harness/msvc9.c

@@ -786,7 +786,9 @@ int __builtin_clz(unsigned int pattern)
 #endif // !__has_builtin(__builtin_clz)
 
 #include <intrin.h>
+#if !defined(_M_ARM64)
 #include <emmintrin.h>
+#endif
 
 int usleep(int usec)
 {

test_common/harness/parseParameters.cpp

@@ -35,6 +35,7 @@ std::string gCompilationProgram = DEFAULT_COMPILATION_PROGRAM;
 bool gDisableSPIRVValidation = false;
 std::string gSPIRVValidator = DEFAULT_SPIRV_VALIDATOR;
 unsigned gNumWorkerThreads;
+bool gListTests = false;
 
 void helpInfo()
 {
@@ -49,6 +50,8 @@ void helpInfo()
             spir-v     Use SPIR-V offline compilation
     --num-worker-threads <num>
         Select parallel execution with the specified number of worker threads.
+    --list
+        List sub-tests
 
 For offline compilation (binary and spir-v modes) only:
     --compilation-cache-mode <cache-mode>
@@ -104,6 +107,11 @@ int parseCustomParam(int argc, const char *argv[], const char *ignore)
             // option and print its own help.
             helpInfo();
         }
+        else if (!strcmp(argv[i], "--list") || !strcmp(argv[i], "-list"))
+        {
+            delArg++;
+            gListTests = true;
+        }
         else if (!strcmp(argv[i], "--compilation-mode"))
         {
             delArg++;

test_common/harness/parseParameters.h

@@ -40,6 +40,7 @@ extern std::string gCompilationCachePath;
 extern std::string gCompilationProgram;
 extern bool gDisableSPIRVValidation;
 extern std::string gSPIRVValidator;
+extern bool gListTests;
 
 extern int parseCustomParam(int argc, const char *argv[],
                             const char *ignore = 0);