SIMD plumbing

runtime/platform/globals.h

 // Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #ifndef PLATFORM_GLOBALS_H_
 #define PLATFORM_GLOBALS_H_
 
 // __STDC_FORMAT_MACROS has to be defined before including <inttypes.h> to
 // enable platform independent printf format specifiers.
 #ifndef __STDC_FORMAT_MACROS
@@ skipping 54 matching lines @@
 #elif defined(__linux__) || defined(__FreeBSD__)
 #define TARGET_OS_LINUX 1
 #elif defined(__APPLE__)
 #define TARGET_OS_MACOS 1
 #elif defined(_WIN32)
 #define TARGET_OS_WINDOWS 1
 #else
 #error Automatic target os detection failed.
 #endif
 
+struct simd128_value_t {
+  float storage[4];
+  simd128_value_t& readFrom(const float* v) {
+    storage[0] = v[0];
+    storage[1] = v[1];
+    storage[2] = v[2];
+    storage[3] = v[3];
+    return *this;
+  }
+  simd128_value_t& readFrom(const uint32_t* v) {
+    const float* vv = reinterpret_cast<const float*>(v);
+    storage[0] = vv[0];
+    storage[1] = vv[1];
+    storage[2] = vv[2];
+    storage[3] = vv[3];
+    return *this;
+  }
+  simd128_value_t& readFrom(const simd128_value_t* v) {
+    *this = *v;
+    return *this;
+  }
+  void writeTo(float* v) {
+    v[0] = storage[0];
+    v[1] = storage[1];
+    v[2] = storage[2];
+    v[3] = storage[3];
+  }
+  void writeTo(uint32_t* v) {
+    float* vv = reinterpret_cast<float*>(v);
+    vv[0] = storage[0];
+    vv[1] = storage[1];
+    vv[2] = storage[2];
+    vv[3] = storage[3];
+  }
+  void writeTo(simd128_value_t* v) {
+    *v = *this;
+  }
+};
+
 // Processor architecture detection.  For more info on what's defined, see:
 //   http://msdn.microsoft.com/en-us/library/b0084kay.aspx
 //   http://www.agner.org/optimize/calling_conventions.pdf
 //   or with gcc, run: "echo | gcc -E -dM -"
 #if defined(_M_X64) || defined(__x86_64__)
 #define HOST_ARCH_X64 1
 #define ARCH_IS_64_BIT 1
-#include <xmmintrin.h>  // NOLINT
 #define kFpuRegisterSize 16
-typedef __m128 fpu_register_t;
-typedef __m128 simd_value_t;
-// Unaligned load.
-#define simd_value_safe_load(addr)                                             \
-  _mm_loadu_ps(reinterpret_cast<const float*>(addr))
-// Unaligned store.
-#define simd_value_safe_store(addr, value)                                     \
-  _mm_storeu_ps(reinterpret_cast<float*>(addr), value)
+typedef simd128_value_t fpu_register_t;
 #elif defined(_M_IX86) || defined(__i386__)
 #define HOST_ARCH_IA32 1
 #define ARCH_IS_32_BIT 1
-#include <xmmintrin.h>  // NOLINT
 #define kFpuRegisterSize 16
-typedef __m128 fpu_register_t;
-typedef __m128 simd_value_t;
-// Unaligned load.
-#define simd_value_safe_load(addr)                                             \
-  _mm_loadu_ps(reinterpret_cast<const float*>(addr))
-// Unaligned store.
-#define simd_value_safe_store(addr, value)                                     \
-  _mm_storeu_ps(reinterpret_cast<float*>(addr), value)
+typedef simd128_value_t fpu_register_t;
 #elif defined(__ARMEL__)
 #define HOST_ARCH_ARM 1
 #define ARCH_IS_32_BIT 1
 #define kFpuRegisterSize 8
 typedef double fpu_register_t;
-// TODO(johnmccutchan): ARM simd type.
-typedef struct {
-  uint32_t data_[4];
-} simd_value_t;
 #elif defined(__MIPSEL__)
 #define HOST_ARCH_MIPS 1
 #define ARCH_IS_32_BIT 1
 #define kFpuRegisterSize 8
 typedef double fpu_register_t;
-// TODO(johnmccutchan): MIPS simd type.
-typedef struct {
-  uint32_t data_[4];
-} simd_value_t;
 #else
 #error Architecture was not detected as supported by Dart.
 #endif
 
 // DART_FORCE_INLINE strongly hints to the compiler that a function should
 // be inlined. Your function is not guaranteed to be inlined but this is
 // stronger than just using "inline".
 // See: http://msdn.microsoft.com/en-us/library/z8y1yy88.aspx for an
 // explanation of some the cases when a function can never be inlined.
 #ifdef _MSC_VER
@@ skipping 75 matching lines @@
 
 // Types for native machine words. Guaranteed to be able to hold pointers and
 // integers.
 typedef intptr_t word;
 typedef uintptr_t uword;
 
 // Byte sizes.
 const int kWordSize = sizeof(word);
 const int kDoubleSize = sizeof(double);  // NOLINT
 const int kFloatSize = sizeof(float);  // NOLINT
-const int kSimd128Size = 16;
+const int kSimd128Size = sizeof(simd128_value_t);  // NOLINT
 #ifdef ARCH_IS_32_BIT
 const int kWordSizeLog2 = 2;
 const uword kUwordMax = kMaxUint32;
 #else
 const int kWordSizeLog2 = 3;
 const uword kUwordMax = kMaxUint64;
 #endif
 
 // Bit sizes.
 const int kBitsPerByte = 8;
@@ skipping 204 matching lines @@
 // N.B.: As the GCC manual states, "[s]ince non-static C++ methods
 // have an implicit 'this' argument, the arguments of such methods
 // should be counted from two, not one."
 #define PRINTF_ATTRIBUTE(string_index, first_to_check) \
   __attribute__((__format__(__printf__, string_index, first_to_check)))
 #else
 #define PRINTF_ATTRIBUTE(string_index, first_to_check)
 #endif
 
 #endif  // PLATFORM_GLOBALS_H_