[cfe-commits] r98587 - in /cfe/trunk: include/clang/Basic/BuiltinsX86.def lib/Headers/smmintrin.h

[Eric Christopher]

Author: echristo
Date: Mon Mar 15 18:22:58 2010
New Revision: 98587

URL: http://llvm.org/viewvc/llvm-project?rev=98587&view=rev
Log:
Add remaining sse4.1 intrinsics and builtins.

Modified:
    cfe/trunk/include/clang/Basic/BuiltinsX86.def
    cfe/trunk/lib/Headers/smmintrin.h

Modified: cfe/trunk/include/clang/Basic/BuiltinsX86.def
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/BuiltinsX86.def?rev=98587&r1=98586&r2=98587&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/BuiltinsX86.def (original)
+++ cfe/trunk/include/clang/Basic/BuiltinsX86.def Mon Mar 15 18:22:58 2010
@@ -290,6 +290,7 @@
 BUILTIN(__builtin_ia32_ptestz128, "iV2LLiV2LLi", "")
 BUILTIN(__builtin_ia32_ptestc128, "iV2LLiV2LLi", "")
 BUILTIN(__builtin_ia32_ptestnzc128, "iV2LLiV2LLi", "")
-
+BUILTIN(__builtin_ia32_pcmpeqq, "V2LLiV2LLiV2LLi", "")
+BUILTIN(__builtin_ia32_mpsadbw128, "V16cV16cV16ci", "")
 
 #undef BUILTIN

Modified: cfe/trunk/lib/Headers/smmintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/smmintrin.h?rev=98587&r1=98586&r2=98587&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/smmintrin.h (original)
+++ cfe/trunk/lib/Headers/smmintrin.h Mon Mar 15 18:22:58 2010
@@ -246,6 +246,97 @@
 #define _mm_test_mix_ones_zeros(M, V) _mm_testnzc_si128((M), (V))
 #define _mm_test_all_zeros(M, V) _mm_testz_si128 ((V), (V))
 
+/* SSE4 64-bit Packed Integer Comparisons.  */
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cmpeq_epi64(__m128i __V1, __m128i __V2)
+{
+  return (__m128i) __builtin_ia32_pcmpeqq((__v2di)__V1, (__v2di)__V2);
+}
+
+/* SSE4 Packed Integer Sign-Extension.  */
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepi8_epi16(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovsxbw128((__v16qi) __V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepi8_epi32(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovsxbd128((__v16qi) __V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepi8_epi64(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovsxbq128((__v16qi) __V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepi16_epi32(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovsxwd128((__v8hi) __V); 
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepi16_epi64(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovsxwq128((__v8hi)__V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepi32_epi64(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovsxdq128((__v4si)__V);
+}
+
+/* SSE4 Packed Integer Zero-Extension.  */
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepu8_epi16(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovzxbw128((__v16qi) __V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepu8_epi32(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovzxbd128((__v16qi)__V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepu8_epi64(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovzxbq128((__v16qi)__V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepu16_epi32(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovzxwd128((__v8hi)__V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepu16_epi64(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovzxwq128((__v8hi)__V);
+}
+
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_cvtepu32_epi64(__m128i __V)
+{
+  return (__m128i) __builtin_ia32_pmovzxdq128((__v4si)__V);
+}
+
+/* SSE4 Pack with Unsigned Saturation.  */
+static inline __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_packus_epi32(__m128i __V1, __m128i __V2)
+{
+  return (__m128i) __builtin_ia32_packusdw128((__v4si)__V1, (__v4si)__V2);
+}
+
+/* SSE4 Multiple Packed Sums of Absolute Difference.  */
+#define _mm_mpsadbw_epu8(X, Y, M) __builtin_ia32_mpsadbw128((X), (Y), (M))
+
 #endif /* __SSE4_1__ */
 
 #endif /* _SMMINTRIN_H */