Merge pull request #164 from taisei-project/fix-cast-align

simd/x86: fix -Wcast-align warnings (gcc/clang)

.gitignore

@@ -75,3 +75,4 @@ Default-568h@2x.png
 build/
 conftest.dir/*
 confdefs.h
+*.xcuserdatad

CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.8.2)
-project(cglm VERSION 0.7.8 LANGUAGES C)
+project(cglm VERSION 0.7.9 LANGUAGES C)
 
 set(CMAKE_C_STANDARD 11)
 set(CMAKE_C_STANDARD_REQUIRED YES)

CREDITS

@@ -70,4 +70,7 @@ Möller–Trumbore ray-triangle intersection algorithm, from "Fast, Minimum Stor
 Authors:
   Thomas Möller (tompa@clarus.se)
   Ben Trumbore (wbt@graphics.cornell.edu)
-Link to paper: http://webserver2.tecgraf.puc-rio.br/~mgattass/cg/trbRR/Fast%20MinimumStorage%20RayTriangle%20Intersection.pdf
+Link to paper: http://webserver2.tecgraf.puc-rio.br/~mgattass/cg/trbRR/Fast%20MinimumStorage%20RayTriangle%20Intersection.pdf
+
+14. ARM NEON: Matrix Vector Multiplication
+https://stackoverflow.com/a/57793352/2676533

Package.swift

@@ -0,0 +1,44 @@
+// swift-tools-version:5.2
+
+import PackageDescription
+
+let package = Package(
+    name: "cglm",
+    products: [
+        .library(name: "cglm", type: .static, targets: ["cglmHeader"]),
+        .library(name: "cglmc", targets: ["cglmCompiled"]),
+    ],
+    dependencies: [],
+    targets: [
+        .target(
+            name: "cglmCompiled",
+            path: "./",
+            exclude: [
+                "./docs",
+                "./src/swift",
+                "./include",
+                "./test",
+                "./win",
+            ],
+            sources: [
+                "./src",
+            ],
+            publicHeadersPath: "./include"
+        ),
+        .target(
+            name: "cglmHeader",
+            path: "./",
+            exclude: [
+                "./docs",
+                "./include",
+                "./test",
+                "./win",
+            ],
+            sources: [
+                "./src/swift",
+            ],
+            publicHeadersPath: "./include"
+        ),
+    ],
+    cLanguageStandard: .c11
+)

README.md

@@ -212,6 +212,38 @@ cglm_dep = dependency('cglm', fallback : 'cglm', 'cglm_dep')
 executable('exe', 'src/main.c', dependencies : cglm_dep)
 ```
 
+### Swift (Swift Package Manager)
+
+Currently only default build options are supported. Add **cglm** dependency to your project:
+
+```swift
+...
+Package( 
+  ...
+  dependencies: [
+    ...
+    .package(url: "https://github.com/recp/cglm", .branch("master")),
+  ]
+  ...
+)
+```
+
+Now add **cgml** as a dependency to your target. Product choices are:
+- **cglm** for inlined version of the library which can be linked only statically
+- **cglmc** for a compiled version of the library with no linking limitation
+
+```swift
+...
+.target(
+  ...
+  dependencies: [
+    ...
+    .product(name: "cglm", package: "cglm"),
+  ]
+  ...
+)
+...
+```
 
 ### Unix (Autotools)
 

configure.ac

@@ -7,7 +7,7 @@
 #*****************************************************************************
 
 AC_PREREQ([2.69])
-AC_INIT([cglm], [0.7.8], [info@recp.me])
+AC_INIT([cglm], [0.7.9], [info@recp.me])
 AM_INIT_AUTOMAKE([-Wall -Werror foreign subdir-objects serial-tests])
 
 # Don't use the default cflags (-O2 -g), we set ours manually in Makefile.am.

docs/source/conf.py

@@ -62,9 +62,9 @@ author = u'Recep Aslantas'
 # built documents.
 #
 # The short X.Y version.
-version = u'0.7.8'
+version = u'0.7.9'
 # The full version, including alpha/beta/rc tags.
-release = u'0.7.8'
+release = u'0.7.9'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

docs/source/vec2.rst

@@ -55,12 +55,12 @@ Functions:
 Functions documentation
 ~~~~~~~~~~~~~~~~~~~~~~~
 
-.. c:function:: void glm_vec2(vec4 v4, vec2 dest)
+.. c:function:: void glm_vec2(float \*v, vec2 dest)
 
-    init vec2 using vec3
+    init vec2 using vec3 or vec4
 
     Parameters:
-      | *[in]*  **v3**    vector3
+      | *[in]*  **v**     vector
       | *[out]* **dest**  destination
 
 .. c:function:: void glm_vec2_copy(vec2 a, vec2 dest)

include/cglm/mat4.h

@@ -358,6 +358,8 @@ void
 glm_mat4_mulv(mat4 m, vec4 v, vec4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   glm_mat4_mulv_sse2(m, v, dest);
+#elif defined(CGLM_NEON_FP)
+  glm_mat4_mulv_neon(m, v, dest);
 #else
   vec4 res;
   res[0] = m[0][0] * v[0] + m[1][0] * v[1] + m[2][0] * v[2] + m[3][0] * v[3];
@@ -476,6 +478,8 @@ void
 glm_mat4_transpose_to(mat4 m, mat4 dest) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   glm_mat4_transp_sse2(m, dest);
+#elif defined(CGLM_NEON_FP)
+  glm_mat4_transp_neon(m, dest);
 #else
   dest[0][0] = m[0][0]; dest[1][0] = m[0][1];
   dest[0][1] = m[1][0]; dest[1][1] = m[1][1];
@@ -498,6 +502,8 @@ void
 glm_mat4_transpose(mat4 m) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   glm_mat4_transp_sse2(m, m);
+#elif defined(CGLM_NEON_FP)
+  glm_mat4_transp_neon(m, m);
 #else
   mat4 d;
   glm_mat4_transpose_to(m, d);
@@ -536,12 +542,7 @@ glm_mat4_scale(mat4 m, float s) {
 #if defined( __SSE__ ) || defined( __SSE2__ )
   glm_mat4_scale_sse2(m, s);
 #elif defined(CGLM_NEON_FP)
-  float32x4_t v0;
-  v0 = vdupq_n_f32(s);
-  vst1q_f32(m[0], vmulq_f32(vld1q_f32(m[0]), v0));
-  vst1q_f32(m[1], vmulq_f32(vld1q_f32(m[1]), v0));
-  vst1q_f32(m[2], vmulq_f32(vld1q_f32(m[2]), v0));
-  vst1q_f32(m[3], vmulq_f32(vld1q_f32(m[3]), v0));
+  glm_mat4_scale_neon(m, s);
 #else
   glm_mat4_scale_p(m, s);
 #endif

include/cglm/simd/neon/mat4.h

@@ -12,6 +12,32 @@
 #include "../../common.h"
 #include "../intrin.h"
 
+CGLM_INLINE
+void
+glm_mat4_scale_neon(mat4 m, float s) {
+  float32x4_t v0;
+  
+  v0 = vdupq_n_f32(s);
+
+  vst1q_f32(m[0], vmulq_f32(vld1q_f32(m[0]), v0));
+  vst1q_f32(m[1], vmulq_f32(vld1q_f32(m[1]), v0));
+  vst1q_f32(m[2], vmulq_f32(vld1q_f32(m[2]), v0));
+  vst1q_f32(m[3], vmulq_f32(vld1q_f32(m[3]), v0));
+}
+
+CGLM_INLINE
+void
+glm_mat4_transp_neon(mat4 m, mat4 dest) {
+  float32x4x4_t vmat;
+  
+  vmat = vld4q_f32(m[0]);
+
+  vst1q_f32(dest[0], vmat.val[0]);
+  vst1q_f32(dest[1], vmat.val[1]);
+  vst1q_f32(dest[2], vmat.val[2]);
+  vst1q_f32(dest[3], vmat.val[3]);
+}
+
 CGLM_INLINE
 void
 glm_mat4_mul_neon(mat4 m1, mat4 m2, mat4 dest) {
@@ -53,5 +79,27 @@ glm_mat4_mul_neon(mat4 m1, mat4 m2, mat4 dest) {
   vst1q_f32(dest[3], d3);
 }
 
+CGLM_INLINE
+void
+glm_mat4_mulv_neon(mat4 m, vec4 v, vec4 dest) {
+  float32x4_t l0, l1, l2, l3;
+  float32x2_t vlo, vhi;
+  
+  l0  = vld1q_f32(m[0]);
+  l1  = vld1q_f32(m[1]);
+  l2  = vld1q_f32(m[2]);
+  l3  = vld1q_f32(m[3]);
+
+  vlo = vld1_f32(&v[0]);
+  vhi = vld1_f32(&v[2]);
+
+  l0  = vmulq_lane_f32(l0, vlo, 0);
+  l0  = vmlaq_lane_f32(l0, l1, vlo, 1);
+  l0  = vmlaq_lane_f32(l0, l2, vhi, 0);
+  l0  = vmlaq_lane_f32(l0, l3, vhi, 1);
+
+  vst1q_f32(dest, l0);
+}
+
 #endif
 #endif /* cglm_mat4_neon_h */

include/cglm/simd/x86.h

@@ -175,7 +175,7 @@ glmm_load3(float v[3]) {
   __m128i xy;
   __m128  z;
 
-  xy = _mm_loadl_epi64((const __m128i *)v);
+  xy = _mm_loadl_epi64(CGLM_CASTPTR_ASSUME_ALIGNED(v, const __m128i));
   z  = _mm_load_ss(&v[2]);
 
   return _mm_movelh_ps(_mm_castsi128_ps(xy), z);
@@ -184,7 +184,7 @@ glmm_load3(float v[3]) {
 static inline
 void
 glmm_store3(float v[3], __m128 vx) {
-  _mm_storel_pi((__m64 *)&v[0], vx);
+  _mm_storel_pi(CGLM_CASTPTR_ASSUME_ALIGNED(v, __m64), vx);
   _mm_store_ss(&v[2], glmm_shuff1(vx, 2, 2, 2, 2));
 }
 

include/cglm/types.h

@@ -32,6 +32,16 @@
 #  define CGLM_ALIGN_MAT CGLM_ALIGN(16)
 #endif
 
+#ifdef __GNUC__
+#  define CGLM_ASSUME_ALIGNED(expr, alignment) \
+  __builtin_assume_aligned((expr), (alignment))
+#else
+#  define CGLM_ASSUME_ALIGNED(expr, alignment) (expr)
+#endif
+
+#define CGLM_CASTPTR_ASSUME_ALIGNED(expr, type) \
+  ((type*)CGLM_ASSUME_ALIGNED((expr), __alignof__(type)))
+
 typedef float                   vec2[2];
 typedef float                   vec3[3];
 typedef int                    ivec3[3];

include/cglm/version.h

@@ -10,6 +10,6 @@
 
 #define CGLM_VERSION_MAJOR 0
 #define CGLM_VERSION_MINOR 7
-#define CGLM_VERSION_PATCH 8
+#define CGLM_VERSION_PATCH 9
 
 #endif /* cglm_version_h */

include/module.modulemap

@@ -0,0 +1,14 @@
+module cglm {
+    header "cglm/cglm.h"
+    header "cglm/struct.h"
+    
+    export *
+}
+
+module cglmc {
+    header "cglm/cglm.h"
+    header "cglm/struct.h"
+    header "cglm/call.h"
+    
+    export *
+}

meson.build

@@ -1,5 +1,5 @@
 project('cglm', 'c',
-    version : '0.7.8', 
+    version : '0.7.9', 
     license : 'mit',
     default_options : [
         'c_std=c11',
@@ -11,6 +11,7 @@ project('cglm', 'c',
 
 cc = meson.get_compiler('c')
 
+cglm_install = get_option('install')
 cglm_deps = cc.find_library('m', required : false)
 
 cglm_args = []
@@ -50,11 +51,9 @@ cglm_src = files(
     'src/vec4.c'
 )
 
-install_subdir('include/cglm', install_dir : get_option('includedir'))
-
 cglm_lib = library('cglm',
     cglm_src,
-    install : true,
+    install : cglm_install,
     dependencies : cglm_deps,
     c_args : [ build_args, cglm_args ]
 )
@@ -71,17 +70,19 @@ if meson.version().version_compare('>= 0.54.0')
     meson.override_dependency('cglm', cglm_dep)
 endif
 
+if cglm_install
+    install_subdir('include/cglm', install_dir : get_option('includedir'))
 
-pkg = import('pkgconfig')
-
-pkg.generate(
-    name : 'cglm',
-    libraries : cglm_lib,
-    extra_cflags : cglm_args,
-    version : meson.project_version(),
-    url : 'https://github.com/recp/cglm',
-    description : 'OpenGL Mathematics (glm) for C'
-)
+    pkg = import('pkgconfig')
+    pkg.generate(
+        name : 'cglm',
+        libraries : cglm_lib,
+        extra_cflags : cglm_args,
+        version : meson.project_version(),
+        url : 'https://github.com/recp/cglm',
+        description : 'OpenGL Mathematics (glm) for C'
+    )
+endif
 
 if get_option('build_tests') == true
 

meson_options.txt

@@ -1 +1,2 @@
-option('build_tests', type : 'boolean', value : false, description : 'Build tests')
+option('build_tests', type : 'boolean', value : false, description : 'Build tests')
+option('install', type : 'boolean', value : true, description : 'Include the library, headers, and pkg-config file in the install target')

src/swift/empty.c

@@ -0,0 +1 @@
+// This empty file is needed to trick swiftpm to build the header-only version of cglm as swiftpm itself does not support C targets that have no source code files