Add LSX and LASX configure detection

Adds loongarch simd extension(LSX LASX) configure test and
-feature-lsx and -feature-lasx configure options.
Add detection of LSX and LASX at run-time in qsimd.cpp.

Change-Id: I63eab2f4f45c306b672a89b376e0cbc01da0df83
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>

cmake/QtCompilerOptimization.cmake

@@ -68,6 +68,8 @@ if(GCC OR CLANG OR QCC)
     endif()
     set(QT_CFLAGS_ARM_SVE    "${__prefix}-march=armv8-a+sve")
     set(QT_CFLAGS_ARM_CRYPTO "${__prefix}-march=armv8-a+crypto")
+    set(QT_CFLAGS_LSX        "${__prefix}-mlsx")
+    set(QT_CFLAGS_LASX       "${__prefix}-mlasx")
     set(QT_CFLAGS_MIPS_DSP   "${__prefix}-mdsp")
     set(QT_CFLAGS_MIPS_DSPR2 "${__prefix}-mdspr2")
     unset(__prefix)

cmake/QtFeature.cmake

@@ -1308,6 +1308,31 @@ function(qt_config_compile_test_armintrin extension label)
     set(TEST_subarch_${extension} "${TEST_ARMINTRIN_${extension}}" CACHE INTERNAL "${label}")
 endfunction()
 
+function(qt_config_compile_test_loongarchsimd extension label)
+        if (DEFINED TEST_LOONGARCHSIMD_${extension})
+        return()
+    endif()
+
+    set(flags "-DSIMD:string=${extension}")
+
+    qt_get_platform_try_compile_vars(platform_try_compile_vars)
+    list(APPEND flags ${platform_try_compile_vars})
+
+    message(STATUS "Performing Test ${label} intrinsics")
+    try_compile("TEST_LOONGARCHSIMD_${extension}"
+        "${CMAKE_CURRENT_BINARY_DIR}/config.tests/loongarch_simd_${extension}"
+        "${CMAKE_CURRENT_SOURCE_DIR}/config.tests/loongarch_simd"
+        loongarch_simd
+        CMAKE_FLAGS ${flags})
+    if(${TEST_LOONGARCHSIMD_${extension}})
+        set(status_label "Success")
+    else()
+        set(status_label "Failed")
+    endif()
+    message(STATUS "Performing Test ${label} intrinsics - ${status_label}")
+    set(TEST_subarch_${extension} "${TEST_LOONGARCHSIMD_${extension}}" CACHE INTERNAL "${label}")
+endfunction()
+
 function(qt_config_compile_test_machine_tuple label)
     if(DEFINED TEST_MACHINE_TUPLE OR NOT (LINUX OR HURD) OR ANDROID)
         return()

cmake/QtProcessConfigureArgs.cmake

@@ -245,6 +245,7 @@ defstub(qt_config_compile_test)
 defstub(qt_config_compile_test_armintrin)
 defstub(qt_config_compile_test_machine_tuple)
 defstub(qt_config_compile_test_x86simd)
+defstub(qt_config_compile_test_loongarchsimd)
 defstub(qt_config_compiler_supports_flag_test)
 defstub(qt_config_linker_supports_flag_test)
 defstub(qt_configure_add_report_entry)

config.tests/loongarch_simd/CMakeLists.txt

@@ -0,0 +1,26 @@
+# Copyright (C) 2022 The Qt Company Ltd.
+# SPDX-License-Identifier: BSD-3-Clause
+
+cmake_minimum_required(VERSION 3.16)
+project(loongarch_simd LANGUAGES CXX)
+
+include(../../cmake/QtPlatformSupport.cmake)
+include(../../cmake/QtCompilerOptimization.cmake)
+
+# FIXME: Make the this project handle a list of SIMD entries.
+# FIXME: Make this project handle appending of the cflags (similar to the qmake project).
+# This is needed for the loongarchsimd configure test (
+# aka we test to see if setting no SIMD (-mlsx) cflags at all, will result in their implicit
+# addition by the compiler).
+string(TOUPPER "${SIMD}" upper_simd)
+
+if(NOT DEFINED "QT_CFLAGS_${upper_simd}")
+    # Don't use CMake error() because a configure error also fails the try_compile() call.
+    # Instead use a compile flag that doesn't exist to force a compiler error.
+    set(QT_CFLAGS_${upper_simd} "--qt-cflags-not-found")
+endif()
+
+add_executable("SimdTest${SIMD}")
+target_sources("SimdTest${SIMD}" PRIVATE main.cpp)
+target_compile_options("SimdTest${SIMD}" PRIVATE ${QT_CFLAGS_${upper_simd}})
+target_compile_definitions("SimdTest${SIMD}" PRIVATE QT_COMPILER_SUPPORTS_${upper_simd})

config.tests/loongarch_simd/main.cpp

@@ -0,0 +1,29 @@
+// Copyright (C) 2017 Intel Corporation.
+// SPDX-License-Identifier: BSD-3-Clause
+
+#define T(x)                      (QT_COMPILER_SUPPORTS_ ## x)
+
+#if T(LSX)
+#include <lsxintrin.h>
+void test_lsx()
+{
+    __m128i a = __lsx_vldi(0);
+    (void) __lsx_vshuf_h(__lsx_vldi(0), a, a);
+}
+#endif
+
+#if T(LASX)
+#include <lsxintrin.h>
+#include <lasxintrin.h>
+void test_lasx()
+{
+    __m256i a = __lasx_xvldi(0);
+    __m256i b = __lasx_xvadd_b(a, a);
+    (void) __lasx_xvadd_b(a, b);
+}
+#endif
+
+int main()
+{
+    return 0;
+}

configure.cmake

@@ -376,6 +376,12 @@ qt_config_compile_test_armintrin(crypto "CRYPTO")
 # arm: sve
 qt_config_compile_test_armintrin(sve "SVE")
 
+# loongarch: lsx
+qt_config_compile_test_loongarchsimd(lsx "LSX")
+
+# loongarch: lasx
+qt_config_compile_test_loongarchsimd(lasx "LASX")
+
 # localtime_r
 qt_config_compile_test(localtime_r
     LABEL "localtime_r()"
@@ -901,6 +907,18 @@ qt_feature("shani" PRIVATE
 )
 qt_feature_definition("shani" "QT_COMPILER_SUPPORTS_SHA" VALUE "1")
 qt_feature_config("shani" QMAKE_PRIVATE_CONFIG)
+qt_feature("lsx" PRIVATE
+    LABEL "LSX"
+    CONDITION ( TEST_architecture_arch STREQUAL loongarch64 ) AND TEST_subarch_lsx
+)
+qt_feature_definition("lsx" "QT_COMPILER_SUPPORTS_LSX" VALUE "1")
+qt_feature_config("lsx" QMAKE_PRIVATE_CONFIG)
+qt_feature("lasx" PRIVATE
+    LABEL "LASX"
+    CONDITION ( TEST_architecture_arch STREQUAL loongarch64 ) AND TEST_subarch_lasx
+)
+qt_feature_definition("lasx" "QT_COMPILER_SUPPORTS_LASX" VALUE "1")
+qt_feature_config("lasx" QMAKE_PRIVATE_CONFIG)
 qt_feature("mips_dsp" PRIVATE
     LABEL "DSP"
     CONDITION ( TEST_architecture_arch STREQUAL mips ) AND TEST_arch_${TEST_architecture_arch}_subarch_dsp
@@ -1301,6 +1319,12 @@ qt_configure_add_summary_entry(
     MESSAGE "ARM Extensions"
     CONDITION ( TEST_architecture_arch STREQUAL arm ) OR ( TEST_architecture_arch STREQUAL arm64 )
 )
+qt_configure_add_summary_entry(
+    TYPE "featureList"
+    ARGS "lsx lasx"
+    MESSAGE "LOONGARCH Extensions"
+    CONDITION ( TEST_architecture_arch STREQUAL loongarch64 )
+)
 qt_configure_add_summary_entry(
     ARGS "mips_dsp"
     CONDITION ( TEST_architecture_arch STREQUAL mips )

mkspecs/common/gcc-base.conf

@@ -119,6 +119,8 @@ QMAKE_CFLAGS_VAES      += -mvaes
 QMAKE_CFLAGS_NEON      += -mfpu=neon
 QMAKE_CFLAGS_MIPS_DSP  += -mdsp
 QMAKE_CFLAGS_MIPS_DSPR2 += -mdspr2
+QMAKE_CFLAGS_LSX  += -mlsx
+QMAKE_CFLAGS_LASX += -mlasx
 
 # -march=haswell is supported as of GCC 4.9 and Clang 3.6
 QMAKE_CFLAGS_ARCH_HASWELL           = -march=core-avx2

mkspecs/features/simd.prf

@@ -146,6 +146,8 @@ addSimdCompiler(rdseed)
 addSimdCompiler(neon)
 addSimdCompiler(mips_dsp)
 addSimdCompiler(mips_dspr2)
+addSimdCompiler(lsx)
+addSimdCompiler(lasx)
 
 # Haswell sub-architecture
 defineTest(addSimdArch) {

src/corelib/global/qsimd.cpp

@@ -33,7 +33,7 @@
 #  endif
 #elif defined(Q_OS_LINUX) && defined(Q_PROCESSOR_MIPS_32)
 #  include "private/qcore_unix_p.h"
-#elif QT_CONFIG(getauxval) && defined(Q_PROCESSOR_ARM)
+#elif QT_CONFIG(getauxval) && (defined(Q_PROCESSOR_ARM) || defined(Q_PROCESSOR_LOONGARCH))
 #  include <sys/auxv.h>
 
 // the kernel header definitions for HWCAP_*
@@ -95,6 +95,19 @@ static const char features_string[] =
     " dsp\0"
     " dspr2\0";
 
+static const int features_indices[] = {
+       0, 1, 6
+};
+#elif defined(Q_PROCESSOR_LOONGARCH)
+/* Data:
+ lsx
+ lasx
+*/
+static const char features_string[] =
+    "\0"
+    " lsx\0"
+    " lasx\0";
+
 static const int features_indices[] = {
        0, 1, 6
 };
@@ -192,6 +205,40 @@ static inline quint64 detectProcessorFeatures()
     return features;
 }
 
+#elif defined(Q_PROCESSOR_LOONGARCH)
+static inline quint64 detectProcessorFeatures()
+{
+    quint64 features = 0;
+#  if QT_CONFIG(getauxval)
+    quint64 hwcap = getauxval(AT_HWCAP);
+
+    if (hwcap & HWCAP_LOONGARCH_LSX)
+        features |= CpuFeatureLSX;
+    if (hwcap & HWCAP_LOONGARCH_LASX)
+        features |= CpuFeatureLASX;
+#  else
+    enum LoongArchFeatures {
+        LOONGARCH_CFG2 = 0x2,
+        LOONGARCH_CFG2_LSX = (1 << 6),
+        LOONGARCH_CFG2_LASX = (1 << 7)
+    };
+
+    quint64 reg = 0;
+
+    __asm__ volatile(
+        "cpucfg %0, %1 \n\t"
+        : "+&r"(reg)
+        : "r"(LOONGARCH_CFG2)
+    );
+
+    if (reg & LOONGARCH_CFG2_LSX)
+        features |= CpuFeatureLSX;
+    if (reg & LOONGARCH_CFG2_LASX)
+        features |= CpuFeatureLASX;
+#  endif
+    return features;
+}
+
 #elif defined(Q_PROCESSOR_X86)
 
 #ifdef Q_PROCESSOR_X86_32

src/corelib/global/qsimd.h

@@ -21,6 +21,8 @@
  *  sse4_1     | x86
  *  sse4_2     | x86
  *  avx        | x86
+ *  lsx        | loongarch
+ *  lasx       | loongarch
  *
  * Code can use the following constructs to determine compiler support & status:
  * - #if QT_COMPILER_USES(XXX) (e.g: #if QT_COMPILER_USES(neon) or QT_COMPILER_USES(sse4_1)
@@ -53,6 +55,20 @@
 #  define QT_COMPILER_USES_mips_dspr2 -1
 #endif
 
+#if defined(Q_PROCESSOR_LOONGARCH) && defined(__loongarch_sx)
+#  include <lsxintrin.h>
+#  define QT_COMPILER_USES_lsx 1
+#else
+#  define QT_COMPILER_USES_lsx -1
+#endif
+
+#if defined(Q_PROCESSOR_LOONGARCH) && defined(__loongarch_asx)
+#  include <lasxintrin.h>
+#  define QT_COMPILER_USES_lasx 1
+#else
+#  define QT_COMPILER_USES_lasx -1
+#endif
+
 #if defined(Q_PROCESSOR_X86) && defined(Q_CC_MSVC)
 // MSVC doesn't define __SSE2__, so do it ourselves
 #  if (defined(_M_X64) || _M_IX86_FP >= 2) && defined(QT_COMPILER_SUPPORTS_SSE2)

src/corelib/global/qsimd_p.h

@@ -120,6 +120,9 @@ QT_WARNING_DISABLE_INTEL(103)
 #  if !defined(__MIPS_DSPR2__) && defined(__mips_dspr2) && defined(Q_PROCESSOR_MIPS_32)
 #    define __MIPS_DSPR2__
 #  endif
+#elif defined(Q_PROCESSOR_LOONGARCH)
+#  define QT_COMPILER_SUPPORTS_HERE(x)    QT_COMPILER_SUPPORTS(x)
+#  define QT_FUNCTION_TARGET(x)
 #elif defined(Q_PROCESSOR_X86)
 #  if defined(Q_CC_CLANG) && defined(Q_CC_MSVC)
 #    define QT_COMPILER_SUPPORTS_HERE(x)    (__ ## x ## __)
@@ -383,6 +386,9 @@ enum CPUFeatures {
 #elif defined(Q_PROCESSOR_MIPS)
     CpuFeatureDSP           = 2,
     CpuFeatureDSPR2         = 4,
+#elif defined(Q_PROCESSOR_LOONGARCH)
+    CpuFeatureLSX           = 2,
+    CpuFeatureLASX          = 4,
 #endif
 };
 
@@ -411,6 +417,12 @@ static const uint64_t qCompilerCpuFeatures = 0
 #endif
 #if defined __mips_dspr2
         | CpuFeatureDSPR2
+#endif
+#if defined __loongarch_sx
+        | CpuFeatureLSX
+#endif
+#if defined __loongarch_asx
+        | CpuFeatureLASX
 #endif
         ;
 #endif