build(core) with clang-cl when possible

Build times with Visual Studio's Clang/LLVM toolset are considerably higher, but the resulting binaries are generally more efficient.
LWJGL · Jun 10, 2024 · 4ac9d90 · 4ac9d90
1 parent 233fb88
commit 4ac9d90
Show file tree

Hide file tree

Showing 11 changed files with 85 additions and 42 deletions.
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -359,9 +359,10 @@ jobs:
           Rename-Item zulu8.78.0.19-ca-jdk8.0.412-win_i686 jdk8
         shell: pwsh
         if: matrix.ARCH == 'x86'
-      - name: Build native
+      - name: Build native # TODO: remove old LLVM workaround, see https://github.com/actions/runner-images/issues/10001
         run: |
           set LWJGL_BUILD_ARCH=${{matrix.ARCH}}
+          set PATH=%PATH:C:\Program Files\LLVM\bin;=%
           ant -emacs compile-native
       - name: Run tests
         run: ant -emacs tests

diff --git a/config/windows/build.xml b/config/windows/build.xml
@@ -5,16 +5,26 @@
 <project name="native-windows" basedir="../.." xmlns:if="ant:if" xmlns:unless="ant:unless">
     <import file="../build-definitions.xml"/>
 
+    <property name="clang-target" value="i686" if:set="build.arch.x86"/>
+    <property name="clang-target" value="amd64" if:set="build.arch.x64"/>
+    <property name="clang-target" value="aarch64" if:set="build.arch.arm64"/>
+
     <macrodef name="compile">
         <attribute name="dest" default="${dest}"/>
+        <attribute name="useClang" default="true"/>
         <attribute name="flags" default=""/>
         <attribute name="simple" default="false"/>
         <attribute name="relative" default="true"/>
         <element name="source" implicit="true" optional="true"/>
         <sequential>
+            <local name="compiler"/>
+            <condition property="compiler" value="clang-cl" else="cl"><istrue value="@{useClang}"/></condition>
+
             <mkdir dir="@{dest}"/>
-            <apply executable="cl" dest="@{dest}" skipemptyfilesets="true" failonerror="true" relative="@{relative}" parallel="true" taskname="Compiler">
-                <arg line="/c /EHsc /O2 /GL /GR- /GS- /MT /MP @{flags} /nologo /DNDEBUG /DLWJGL_WINDOWS /DLWJGL_${build.arch}"/>
+            <apply executable="${compiler}" dest="@{dest}" skipemptyfilesets="true" failonerror="true" relative="@{relative}" parallel="true" taskname="Compiler">
+                <arg line="/c /EHsc /O2 /GR- /GS- /MT @{flags} /nologo /DNDEBUG /DLWJGL_WINDOWS /DLWJGL_${build.arch}"/>
+                <arg line="/GL /MP" unless:true="@{useClang}"/>
+                <arg line="-flto -fwhole-program-vtables --target=${clang-target}-pc-windows-msvc" if:true="@{useClang}"/>
 
                 <arg value="/Fo@{dest}\"/>
 
@@ -36,6 +46,7 @@
 
     <macrodef name="build">
         <attribute name="module"/>
+        <attribute name="useClang" default="true"/>
         <attribute name="flags" default="/W2"/>
         <attribute name="simple" default="false"/>
         <element name="beforeCompile" optional="true"/>
@@ -59,7 +70,7 @@
             <property name="dest" value="${bin.native}/@{module}"/>
 
             <beforeCompile/>
-            <compile flags="@{flags}" simple="@{simple}">
+            <compile useClang="@{useClang}" flags="@{flags}" simple="@{simple}">
                 <compileargs/>
                 <source/>
             </compile>
@@ -111,8 +122,12 @@ EXPORTS
 
             <mkdir dir="${lib.arch}" unless:set="lib-uptodate"/>
             <beforeLink unless:set="lib-uptodate"/>
-            <apply executable="cl" failonerror="true" parallel="true" taskname="Linker" unless:set="lib-uptodate">
+
+            <local name="linker"/>
+            <condition property="linker" value="clang-cl" else="cl"><istrue value="@{useClang}"/></condition>
+            <apply executable="${linker}" failonerror="true" parallel="true" taskname="Linker" unless:set="lib-uptodate">
                 <arg line='/LD /WX /nologo /Fe:"${bin.native}\build\${name}.dll"'/>
+                <arg line='-fuse-ld=lld' if:true="@{useClang}"/>
 
                 <fileset dir="${dest}" includes="*.obj"/>
                 <link/>
@@ -121,6 +136,7 @@ EXPORTS
                 <arg value="/link"/>
                 <arg value="/OPT:REF,ICF"/>
                 <arg value="/DLL"/>
+                <arg value="/INCREMENTAL:NO"/>
                 <arg value="/LTCG"/>
                 <arg value="/DEF:${bin.native}\build\${name}.def" if:set="undecorate"/>
 
@@ -134,8 +150,9 @@ EXPORTS
 
     <macrodef name="build_simple">
         <attribute name="module"/>
+        <attribute name="useClang" default="true"/>
         <sequential>
-            <build module="@{module}" simple="true" if:true="${binding.@{module}}"/>
+            <build module="@{module}" useClang="@{useClang}" simple="true" if:true="${binding.@{module}}"/>
         </sequential>
     </macrodef>
 
@@ -145,7 +162,7 @@ EXPORTS
         <parallel threadsPerProcessor="1">
 
         <!-- CORE -->
-        <build module="core" flags="/Wall /WX /wd4710 /wd4711">
+        <build module="core" flags="/W4 /WX /wd4710 /wd4711">
             <compileargs>
                 <arg value="/I${src.main}\libffi"/>
                 <arg value="/I${src.main}\libffi\aarch64" if:set="build.arch.arm64"/>
@@ -213,17 +230,21 @@ EXPORTS
         </build>
 
         <!-- Meow -->
-        <build_simple module="meow"/>
+        <build_simple module="meow" useClang="false"/>
 
         <!-- meshoptimizer -->
-        <build module="meshoptimizer" simple="true" if:true="${binding.meshoptimizer}">
-            <beforeCompile>
-                <compile flags="/DMESHOPTIMIZER_NO_WRAPPERS">
-                    <arg value="/I${src.main}"/>
-                    <fileset dir="." includes="${src.main}/*.cpp"/>
-                </compile>
-            </beforeCompile>
-        </build>
+        <sequential if:true="${binding.meshoptimizer}">
+            <local name="meshoptimizer.useClang"/>
+            <condition property="meshoptimizer.useClang" value="false" else="true"><isset property="build.arch.arm64"/></condition>
+            <build module="meshoptimizer" useClang="${meshoptimizer.useClang}" simple="true">
+                <beforeCompile>
+                    <compile useClang="${meshoptimizer.useClang}" flags="/DMESHOPTIMIZER_NO_WRAPPERS">
+                        <arg value="/I${src.main}"/>
+                        <fileset dir="." includes="${src.main}/*.cpp"/>
+                    </compile>
+                </beforeCompile>
+            </build>
+        </sequential>
 
         <!-- NanoVG -->
         <build module="nanovg" simple="true" if:true="${binding.nanovg}">
@@ -281,15 +302,19 @@ EXPORTS
         <build_simple module="par"/>
 
         <!-- Remotery -->
-        <build module="remotery" if:true="${binding.remotery}">
-            <compileargs>
-                <arg value="/I${src.main}"/>
-                <arg value="/I${module.lwjgl}\vulkan\src\main\c"/>
-            </compileargs>
-            <source>
-                <fileset dir="." includes="${src.generated}/*.c" excludes="**/*Metal.c"/>
-            </source>
-        </build>
+        <sequential if:true="${binding.remotery}">
+            <local name="remotery.useClang"/>
+            <condition property="remotery.useClang" value="false" else="true"><isset property="build.arch.x86"/></condition>
+            <build module="remotery" useClang="${remotery.useClang}">
+                <compileargs>
+                    <arg value="/I${src.main}"/>
+                    <arg value="/I${module.lwjgl}\vulkan\src\main\c"/>
+                </compileargs>
+                <source>
+                    <fileset dir="." includes="${src.generated}/*.c" excludes="**/*Metal.c"/>
+                </source>
+            </build>
+        </sequential>
 
         <!-- rpmalloc -->
         <build module="rpmalloc" simple="true" if:true="${binding.rpmalloc}">

diff --git a/doc/notes/3.3.4.md b/doc/notes/3.3.4.md
@@ -53,6 +53,7 @@ This build includes the following changes:
 - Linux: Added support for the RISC-V 64 architecture. (#890)
   * Maven classifier: `linux-riscv64`
 - Linux: ARM/PowerPC/RISC-V shared libraries are now built with GCC 11 (up from GCC 7).
+- Windows: Shared libraries are now built with Clang/LLVM (clang-cl toolset) when possible.
 - Vulkan: Made `VkMemoryRequirements` mutable for the `vmaAllocateMemory(Pages)` functions. (#937)
 
 #### Fixes

diff --git a/modules/lwjgl/core/src/main/c/common_tools.h b/modules/lwjgl/core/src/main/c/common_tools.h
@@ -75,7 +75,9 @@ extern JNIEnv* getEnv(jboolean* async);
             envData = (EnvData *)(uintptr_t)(*__env)->CallStaticLongMethod(__env, TLU, (*__env)->GetStaticMethodID(__env, TLU, "setupEnvData", "()J")); \
         } \
         envData->LastError = LastError;
+#endif
 
+#if defined(LWJGL_WINDOWS) && !defined(__clang__)
     #define VA_LIST_CAST &(va_list)
 #else
     #define VA_LIST_CAST (va_list *)

diff --git a/modules/lwjgl/core/src/main/c/windows/WindowsConfig.h b/modules/lwjgl/core/src/main/c/windows/WindowsConfig.h
@@ -5,14 +5,28 @@
 
 #include <stdint.h>
 
-#define DISABLE_WARNINGS() \
-    __pragma(warning(push, 0))
+#ifdef __clang__
+    #define DISABLE_WARNINGS() \
+        _Pragma("GCC diagnostic push") \
+        _Pragma("GCC diagnostic ignored \"-Wall\"") \
+        _Pragma("GCC diagnostic ignored \"-Wextra\"")
+        _Pragma("GCC diagnostic ignored \"-Wunused-value\"")
+        _Pragma("GCC diagnostic ignored \"-Wunused-function\"")
+        _Pragma("GCC diagnostic ignored \"-Wdeprecated-declarations\"")
+        _Pragma("GCC diagnostic ignored \"-Wignored-attributes\"")
 
-#define ENABLE_WARNINGS() \
-    __pragma(warning(pop))
+    #define ENABLE_WARNINGS() \
+        _Pragma("GCC diagnostic pop")
+#else
+    #define DISABLE_WARNINGS() \
+        __pragma(warning(push, 0))
 
-#ifndef __cplusplus
-    #define inline __forceinline
+    #define ENABLE_WARNINGS() \
+        __pragma(warning(pop))
+
+    #ifndef __cplusplus
+        #define inline __forceinline
+    #endif
 #endif
 
 // JNIEXPORT_CRITICAL & CRITICAL are used as a workaround for JDK-8167409 on applicable functions.

diff --git a/modules/lwjgl/nanovg/src/main/c/blendish.h b/modules/lwjgl/nanovg/src/main/c/blendish.h
@@ -1750,21 +1750,21 @@ void bndNodeBackground(NVGcontext *ctx, float x, float y, float w, float h,
         iconid, bnd_theme.regularTheme.textColor, 
         bndOffsetColor(titleColor, BND_BEVEL_SHADE), 
         BND_LEFT, BND_LABEL_FONT_SIZE, label);
-    NVGcolor arrowColor;
+    //NVGcolor arrowColor;
     NVGcolor borderColor;
     switch(state) {
     default:
     case BND_DEFAULT: {
         borderColor = nvgRGBf(0,0,0);
-        arrowColor = bndOffsetColor(titleColor, -BND_BEVEL_SHADE);
+        //arrowColor = bndOffsetColor(titleColor, -BND_BEVEL_SHADE);
     } break;
     case BND_HOVER: {
         borderColor = bnd_theme.nodeTheme.nodeSelectedColor;
-        arrowColor = bnd_theme.nodeTheme.nodeSelectedColor;
+        //arrowColor = bnd_theme.nodeTheme.nodeSelectedColor;
     } break;
     case BND_ACTIVE: {
         borderColor = bnd_theme.nodeTheme.activeNodeColor;
-        arrowColor = bnd_theme.nodeTheme.nodeSelectedColor;
+        //arrowColor = bnd_theme.nodeTheme.nodeSelectedColor;
     } break;
     }
     bndOutlineBox(ctx,x,y,w,h+1,

diff --git a/modules/lwjgl/tootle/src/main/c/RayTracer/Math/JMLSSEVec.h b/modules/lwjgl/tootle/src/main/c/RayTracer/Math/JMLSSEVec.h
@@ -10,7 +10,7 @@
 #include <iostream>
 #include <math.h>
 
-#ifdef _LINUX
+#if defined(_LINUX) || defined(__clang__)
     class SSEVec4
 #else
     class __declspec(align(16)) SSEVec4
@@ -21,7 +21,7 @@
     union
     {
         __m128 vec128;
-#ifdef _LINUX
+#if defined(_LINUX) || defined(__clang__)
         float f32[4];
 #endif
     };
@@ -42,7 +42,7 @@
     inline operator const __m128() const { return vec128; };
 
     // indexing
-#ifdef _LINUX
+#if defined(_LINUX) || defined(__clang__)
     inline const float& operator[](int i) const { return f32[i]; };
     inline float& operator[](int i) { return f32[i]; };
 #else

diff --git a/modules/lwjgl/tootle/src/main/c/tootlelib.cpp b/modules/lwjgl/tootle/src/main/c/tootlelib.cpp
@@ -22,7 +22,7 @@
 #include "Stripifier.h"
 
 #ifdef _DX11_1_
-#include "directxmesh.h"
+#include "DirectXMesh.h"
 #endif
 
 #define AMD_TOOTLE_API_FUNCTION_BEGIN try {

diff --git a/modules/lwjgl/vma/src/generated/c/org_lwjgl_util_vma_Vma.cpp b/modules/lwjgl/vma/src/generated/c/org_lwjgl_util_vma_Vma.cpp
@@ -6,7 +6,7 @@
 #include "common_tools.h"
 #include "lwjgl_malloc.h"
 DISABLE_WARNINGS()
-#if LWJGL_MACOS
+#ifdef __clang__
     _Pragma("GCC diagnostic ignored \"-Wnullability-completeness\"")
 #endif
 #define VMA_IMPLEMENTATION

diff --git a/modules/lwjgl/vma/src/templates/kotlin/vma/templates/VMA.kt b/modules/lwjgl/vma/src/templates/kotlin/vma/templates/VMA.kt
@@ -12,7 +12,7 @@ val VMA = "Vma".nativeClass(Module.VMA, "Vma", prefix = "VMA") {
     nativeDirective(
         """#include "lwjgl_malloc.h"
 DISABLE_WARNINGS()
-#if LWJGL_MACOS
+#ifdef __clang__
     _Pragma("GCC diagnostic ignored \"-Wnullability-completeness\"")
 #endif
 #define VMA_IMPLEMENTATION

diff --git a/modules/lwjgl/zstd/src/main/c/common/cpu.h b/modules/lwjgl/zstd/src/main/c/common/cpu.h
@@ -35,7 +35,7 @@ MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
     U32 f7b = 0;
     U32 f7c = 0;
 #if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
-#if !defined(__clang__)
+#if !defined(__clang__) || defined(_M_IX86)
     int reg[4];
     __cpuid((int*)reg, 0);
     {